age مهم جدا

Acknowledgments
The editors would like to acknowledge the support of NIH/NIA to this project
(grants AG09282, AG04581, and AG12263). Also, we would like to thank the
departments of psychology at Cleveland State University, The University of Northern
Colorado, and The University of Am~erdam for their support of this project. Finally, we
would be remiss in failing to acknowledge the contributors' fine work, and the technical
assistance of Beth Goldstein (typesetting), Lida Allen (electronic information transfer),
Scott Hayer (digital image processing), and Fabian Ferreri (Word for Windows advice).
vii
PREFACE
Component cognitive processes have played a critical role in the development of

experimental aging research and theory in psychology. A quick perusal of the articles
published in Journal of Gerontology: Psychological Sciences, Psychology and Aging, and
ExperimentalAging Research will confirm this fact. However, in the last five to ten years,
there has been a substantial increase in the number of articles attempting to isolate a single
factor (or small subset of factors) responsa'ble for age differences in information processing.
This view of aging is t~equently termed the complexity model or the generalized slowing
model. The primary assumption in this view is that age differences in cognition are due simply
to a relatively larger performance decrement on the part of older adults (compared to younger
adults) as task complexity increases. Support for generalized slowing has come almost
exclusively t~om the results of regression analyses of old and young response latencies.
Because generalized complexity theorists have questioned the utility of using component
cognitive processes as theoretical constructs, we feel it is time to re-state why component
cognitive processes are critical to any t[lorough understanding of age differences in cognition.
Thus, the present edited volume represents an attempt to demonstrate the utility of the
process-specific approach to cognitive aging. Central to this effort are illustrations of how
regression analyses may provide evidence for general slowing by maximizing explained
variance while at the same time obscuring important local sources of variance.
This volume concentrates on age differences in word and language processing, because
these factors relate to reading, and reading is a critical cognitive process used in everyday life.
Furthermore, age differences in word and language processing illustrate the importance of
taking component cognitive processes into consideration. We feel, though, that the breadth of
the coverage of the present book attests to the wide range of cognitive processes involved in
word and language processing. In the spirit of cognitive science, this book is split into three
different parts: 1) Traditional information processing approaches to age differences in word
and language processing (Part I); 2) neuropsychological approaches (Part II); and 3)
psychophysiological approaches (Part HI). Traditional information processing approaches use
reaction time and accuracy as dependent variables and employ healthy younger and older
adults as research participants. Neuropsychological approaches compare healthy aging to
abnormal aging (e.g., Alzheimefs disease). Psychophysiological approaches can examine
either healthy or abnormal subjects, but psychophysiological dependent variables such as
event-related potentials are used in addition to reaction time or accuracy. We hope that this
use of converging operations will provide a more complete picture of age differences in word
and language processing than simply examining a single approach.
Chapters within each section are organized on the basis of an information processing
continuum ranging from word sensation and perception, to word cognition, and to language
processes. Chapters consist of critical reviews, empirical reports, and position papers that are
contemporarily relevant to cognitive aging research and theory. The first chapter (actually the
first two chapters in Part I) in each Part consists of a tutorial in that particular area of the
literature. We have taken special care to make this book as thorough, yet as readable, as
possible. This book will be of interest to researchers and students of gerontology, cognitive
psychology, neuropsychology, psychophysiology, cognitive science, and health fields.
ix
ADDRESSES OF SENIOR AUTHORS:
Phil Allen, Ph.D.

Dept. of Psychology
Cleveland State University
Euclid Ave. at East 24th St.
Cleveland, Ohio 44115
Paul Amrhein, Ph.D.

Department of Psychology
University of New Mexico
Albuquerque, NM 87131
Ted Bashore, Ph.D.

Dept. of Psychology
University of Northern Colorado
Greeley, Colorado 80639
Richard Ferraro, Ph.D.

Box 7187
University of North Dakota
Grand Forks, ND 58202-7187
Donald Fisher, Ph.D.

117 Amity St.
Amherst, MA 01002
David Friedman, Ph.D.

New York State Psychiatric Institute
Cognitive Electrophysiology Laboratory
722 West 168th Street
New York, NY 10032
Grover Cfilmore, Ph.D.

Case Western Reserve University
10900 Euclid Ave.
Cleveland, OH 44106-7123
Marilyn Hartman, Ph.D.

Psychology, CB #3270
Davie Hall, University of North Carolina
Chapel Hill, NC 27599-3270
x Addresses of Senior Authors
George Kellas, Ph.D.

University of Kansas
Lawrence, KS 66045
Johnathan King, Ph.D.

Department of Cognitive Science
University of California, San Diego
D-015
La Jolla, CA 92093-0515
David Mitchell, Ph.D.

Psychology Department
SMU
Dallas, TX 75275
Beth Ober, Ph.D.

Applied Behavioral Sciences
University of California, Davis
Davis, CA 95616
Marian Patterson, Ph.D.

2520 Fairmount Blvd.
Cleveland Heights, OH 44106
Richard Ridderinkhof~ Ph.D.

University of Am~erdam
Reotersstraat 15
1018 WB Am~erdam
The Netherlands
Leann Stadtlander, Ph.D.

300 Traphagen Hall
Montana State University
Bozeman, MT 59717
Elizabeth Stine, Ph.D.

Conant Hall
University of New Hampshire
Durham, NH 03824
Age Differences in Word and Language Processing
Ph. Allen and Th.R. Bashore (Editors)
9 1995 Elsevier Science B.V. All rights reserved.
Why latent models are needed to test hypotheses about the slowing of word and
language processes in older adults*
Donald L. Fishera, Arthur D. Fiskb, and Susan A. Duffyc
a U n i v e r s i t y of Massachusetts at Amherst
bGoorgia Institute of Technology
Mount Holyoke College
1. INTRODUCTION
Younger and older adults' performance has been compared on a number of different
tasks. Many such tasks use response time as a dependent variable. Older subjects are typically
slower than younger subjects. This suggests that the cognitive processes which mediate the
behavior of older adults are slowed. A primary goal of studies of age-related slowing has been
to determine whether all processes are slowed and, if not, which particular processes are
slowed and which are spared. A knowledge of how much each process is slowed bears
critically on the evaluation of the various theoretical attempts to relate the slowing of particular
cognitive processes to a wide range of performance decrements (Salthouse, 1991) including
deficits on tests of explicit and implicit memory (Howard and Wiggs, 1993), attention
(Giambra, 1993; Madden and Plude, 1993), and intelligence (Hertzog, 1989; Schaie, 1989).
This knowledge is critical because in most cases it is not known whether the particular
processes thought to produce a given deficit are slowed or spared. A knowledge of how much
each individual process is slowed also bears critically on the more practical attempts to design
displays and interfaces for older adults which make it easier for them to perform the ordinary
tasks involved in day to day living. Specifically, this knowledge makes it possible to focus on
redesigning those aspects of the environment which reduce the slowing of the processes which
are most affected by aging (Staplin and Fisk, 1991; Walker, Philbin and Fisk, 1994).
Although knowledge of how much each individual process is slowed cannot be
obtained with the techniques currently used in aging research, it is generally agreed that these
techniques allow one to conclude that all processes are slowed at least some in older adults.
However, the agreement ends here. In some studies, the slowing has been reported as constant
across several different task domains. The conclusion is drawn that the cognitive processes are
all slowed by one and the same function (Cerella, Poon and Williams, 1980; Myerson, Hale,
Wagstait~ Pooh and Smith, 1990). In other more recent studies, the slowing is reported as
constant within a given domain (e.g.,the lexical domain), but different across domains (e.g., the
* Please send reprint requests to Donald L. Fisher, 114 Marston Hall, IEOR, University of Massachusetts,
Amherst, MA 01003. Donald L. Fisher was supported during the writing of the chapter by a National Institutes
ofHealth (NIA) Grant No. R01AG12461; Arthur D. Fisk was supported during the writing of the chapter by a
National Institutes of Health (NIA) Grant No. R01AG07654.
2 D.L. Fisher et al.
lexical and nonlexical domains). The conclusion is drawn that the cognitive processes are
slowed by different functions in the separate domains (Cerella, 1985; Lima, Hale and Myerson,
1091; Mayr and Kleigl, 1993). These conclusions rest on two primary and largely untested
assumptions. First, it is assumed that older and younger adults process information identically
in each of the tasks where their performance has been compared (the assumption of a structural
equivalence). Second, it is assumed that if one function relates the observed response times of
the older and younger adults in a given set of tasks, then one function relates the unobserved
durations of the individual processes which mediate the behavior of the older and younger
adults in the same set of tasks (the assumption of a functional equivalence). Although various
investigators have acknowledged the importance of these two assmnptions for testing theories
of cognitive slowing, they have not been directly examined.
The violation of either or both of the above assumptions can lead to potential problems.
Specifically, if the first assumption, the assumption of structural equivalence is violated, then
the differences in the time it takes younger and older adults to perform various tasks may be
due wholly to the use of different processes to perform a given task and not to prolongations
of the durations of a common set of cognitive processes. Thus, potentially, none of the
processes may be slowed. Psychologists now know that the strategies adults use to perform a
given task can vary widely from one individual to the next (e.g., Hunt, 1978). Perhaps the
existence of different strategies can explain why older adults are slower than younger adults, at
least on some tasks. Problems still remain even if the first assumption is satisfied. Specifically,
if the first assumption is satisfied and if the second assumption, the assumption of functional
equivalence, is violated then the fact that older adults' response times are well fit by a single
slowing function on a given set of tasks need not imply that the slowing of the cognitive
processes is identical across processes. In fact, some processes might not be slowed at all.
Nevertheless, investigators frequently make the decision to accept a model which assumes that
all processes are slowed identically based on the finding that the older adults' response times
are well fit by a single function of the younger adults' response times. Such a decision will be
called a false positive decision if in fact the individual processes are slowed differentially. Just
the opposite problem can arise. Specifically, the fact that older adults' response times are well
fit by different slowing functions across separate sets of tasks need not imply that the slowing
of the cognitive processes varies among processes. In fact, all processes might be slowed
identically. Nevertheless, investigators typically make the decision to reject a model which
assumes all processes are slowed identically based on the finding that older adults' response
times are best fit by different functions of younger adults' response times across task domains.
Such a decision will be called a false negative decision if the in fact the individual processes
are slowed identically.
In summary, if one or both of the above assumptions are not satisfied, then the
techniques that currently are widely used in studies of age-related slowing can easily lead the
investigator falsely to accept or reject the theory of general slowing. However, alternative
techniques do exist for testing theories of cognitive slowing when these assmnptions are not
satisfied. In particular, if the assumption of structural equivalence is violated, then techniques
have been described which can be used to identify the exact structure of older and younger
adults cognitive networks in a given task (Schweickert, 1978; Schweickert and Townsend,
1989; Steinberg, 1969; Townsend and Schweickert, 1989). Since a detailed overview of
these techniques is now available (Schweickert, Fisher and Goldstein, 1994), we will not
pursue these particular techniques further. If the assumption of functional equivalence is
Why latent models are needed to test hypotheses 3
violated, then techniques exist which can be used to derive the exact slowing of the individual
cognitive processes from the overall response times. Elsewhere we have summarized these
techniques (Fisher, 1994; Fisk, Fisher and Rogers, 1992; Fisk and Fisher, 1994). In the
remainder of this chapter, we want to describe these latter techniques in more detail. Unlike
current techniques, the techniques we will describe require that one have a detailed model of
processing for the younger and older adults in each of the tasks on which their performance is
compared.
2. GLOBAL AND LATENT MODELS
In order to develop techniques which can be used to determine whether and by how
much the component cognitive processes are slowed, we will argue that it is important, indeed
necessary, to differentiate between latent and global models of slowing (Table 1). We will
consider only two types of global models: global models of general slowing (Case I in Table 1)
in which the slowing of older adults' response times remains constant across tasks (indexed by
i) and domains (indexed by j) and global models of domain-specific slowing (Case II) in which
the slowing of older adults' response times remains constant across tasks within a domain, but
varies across domains. In both global models of slowing, it is assumed that the older adults'
average response time Oo on task i in domain j is a function of the younger adults' average
response time Yo on the same task plus some random error E 0 (E is the Greek uppercase
epsilon). And in both models, the argument of the global slowing function is the response time
of the younger adults. However, in the global models of domain-specific slowing the slowing
function is indexed by the domainj whereas no index is needed for this function in the global
models of general slowing. (Throughout, random variables will be represented by uppercase
letters; constants and variables will be represented by lowercase italicized letters, labels for
processes and other components of figures will be represented by lowercase nonitalicized
letters)
As with global models, we will consider only two types of latent models: latent models
of general slowing (Case HI) in which the time on average it takes an older adult to complete
each of the latent cognitive processes (indexed by k) is slowed by the same function and latent
models of process-specific slowing (Case IV) in which the time on average it takes an older
adult to complete each of the latent cognitive processes (also indexed by k) is slowed by a
different function. In both of the latent models of slowing, it is assumed that the time on
average Ak it takes an older adult to complete a particular latent cognitive process xk is some
function of the time on average Ak it takes a younger adult to complete the same latent
cognitive process xk. And in both models, the argument of the latent slowing function is the
average time it takes younger adults to complete a particular latent process. However, in the
latent models of process-specific slowing the slowing function is indexed by the process k.
An example can make the above distinctions between global and latent models of
slowing more transparent. In particular, consider simple lexical and nonlexical memory search
tasks. Subjects are asked to memorize either a list of n words (the lexical task) or n digits (the
nonlexical task). A probe word or digit is presented on each trial. Subjects must indicate
whether the probe is or is not present in the memory set. Response time and accuracy are the
dependent variables. Here, we want to predict the response times of the older adults.
First, consider global models of slowing. In such models, we take as a starting point
for predicting the older adults' global response times the global response times of the younger
Table 1
Models of Slowing
I. Global Models of General Slowing I
0o = flYo) + Eo
II. Global Models of Domain-Specific Slowing 2
0o = fJ(Yo) + EO
III. Latent Models of General Slowing 3
Ak =J(Ak)
IV. Latent Models of Process-Specific Slowing 4
Ak:A(Ak)
1 O0"is the time on average it takes older adults to respond on task i in domainj; Yo is the time on average it
takes younger adults to respond on the same task; E0.is the error; fis the global general slowing function.
2 The global domain-specific slowingfunction is denoted
3 Akis the time on average that it takes older adults to complete latent processXk; Ak is the time on average that
it takes younger adults to complete latent processXk; fis the latent general slowingfunction.
4 The latent process-specific slowingfunction is denotedJ~.
adults. Specifically, in a global model of general slowing (Case I), the time on average it takes
the older adults to indicate that a probe is present at each memory set size for both the lexical
and nonlexical probes can be written simply as:
Oo = f ( Y o ) + Eo . (1)
Here, i indexes the task (memory set size), j indexes the domain (lexical or nonlexical). In a
global model of domain-specific slowing (Case II), we would now need to index the slowing
function by j, the domain, so that:
Oo : f j ( Y o ) + E0. (2)
There are many different global slowing functions we could substitute into the above
equations. One of the simplest such functions (and perhaps the dominant function; see Cerella
and Hale, in press) is s i l l y a constant, say 13, times the response time of the younger adults.
Thus, we would rewrite Equation 1 as:
Oo = flYo + Eo (3)
We will refer to the above model as the global multiplicative model of general slowing.
Equation 2 is rewritten similarly with [3 now indexed by j. We will also talk about a global
#near model of general slowing. Here, we simply add an intercept to the model:
E[OI = a + #lU. (4)

Other global slowing functions have been proposed. For example, more complex power
functions will sometimes explain the relation between the older and younger adults' global
response times better than the simple multipfieative relations (Hale, Myerson and Wagstafl~
1987; Myerson, Hale, Wagstafl~ Poon and Smith, 1990). However, the multiplieative model
holds up extremely well by itself over a large range of response times (Cerella and Hale, in
press).
Next, consider latent models of slowing. In such models we take as a starting point for
predicting older adults' response times the durations of the latent processes which govern the
behavior of younger adults. Generally, in the task described above there are four latent
processes: encoding, comparison, decision and response. Let E, C, D and R represent,
respectively, the time on average it takes younger subjects to execute each of the above four
processes. Assume that these processes are executed in series: subjects first encode the probe
word, they then compare this probe with all stimuli in the memory set, they next decide
whether the probe is or is not one of these stimuli, and finally they respond (Figure la). To
keep the development as simple as possible, assume that the mean durations of these processes
do not vary across tasks (memory set size) or domains (lexieal or nonlexieal) except for the
mean duration of the comparison process which varies with the memory set size. Then, given
that processing in this task is exhaustive (Steinberg, 1966), the younger adults' response times
can now be written as a function of the durations of the various latent processes plus some
error:
Yo = E+ C, + D + R + Eo . (5)
To keep the exposition a straightforward one, assume that the latent slowing function is linear.
Then in a latent model of general slowing (Case III), the older adults' response times will
depend on the slowing of each of the latent cognitive processes by the same ftmetionf
Oo = fiE) + f ( c , ) + riD) + fiR) + E0 9 (6)
In a latent model of process-specific slowing (Case IV), the older adults' response times will
depend on the slowing of each of the component processes by a different function:
Oo = fe(E) + fc(C,) + f a ( D ) + f , ( / ~ + E0 9 (7)
It will be important to note that the latent and global models of general slowing can
sometimes make identical predictions (Fisher, 1994). So, for example, assume that the slowing
function was a multiplieative one and remained constant across the above four processes.
Then, a latent model of general slowing can be rewritten as:
Oo = f(E+ C, + D + R ) + E0 (8)
= f(Yo) + Eo.
This is identical to the expression above for a global model of general slowing.
Encoding Comparison Comparison Decision Response
0 0 v
E C~ C. D r
(a)
p ~ ~ D1
SOA P2 D2 R2
(b)
Figure 1. a) Representation of the network of processes, encoding (e), comparison (c), decision (d) and
response (r) processes, governing behavior in a memory scanning task. (Two items are in the memory set;
scanning is exhaustive), b) Representation of the double stimulationtask.
An understanding of the latent models can now be used to determine whether and by
how much each of the latent cognitive processes is slowed. For example, suppose that we
assume the four slowing functions in Equation 7 are multiplicative ones so that fe = eE,
fc(Ci) = cCi, and so on. Then, we can determine the actual values of the multiplicative
constants e, c, d and r. Additionally, we can determine whether these values differ significantly
from one another. It follows that we can differentiate between latent models of general
slowing and latent models of process-specific slowing and therefore avoid making the
assumption of functional equivalence.
We now want to describe the methods one uses to identify the separate slowing
functions. The methods depend on the type or class of latent network which is used to
represent processing in a given task. A number of different classes of latent models have been
reported in the literature. We want to describe the two major such classes. In the first section
below, we discuss PERT networks. PERT networks can be (and have been) used to represent
processing within a broad range of laboratory tasks (Fisher and Goldstein, 1983; Townsend
and Schweickert, 1989; Schweickert, 1978; Schweickert and Townsend, 1989). The simple
serial and parallel networks described throughout the information processing literature are
examples of such networks (Luce, 1986; Townsend and Ashby, 1983). We will discuss in
detail the construction and testing of PERT network models of memory search and double
stimulation tasks. PERT networks are a special case of a more general network, the Order-of-
Processing (OP) network, which can be used in situations where not all assumptions of the
PERT network are satisfied (Fisher, 1985; Goldstein and Fisher, 1991, 1992; Schweickert and
Fisher, 1987; Schweickert, Fisher and Goldstein, 1994). In the second section, we discuss
interactive inhibition models (McClelland and Rumelhart, 1981). Interactive inhibition models
have been used to model a wide range of language tasks among younger adults. Only recently
has this modeling effort been extended to older adults (Bowles, 1990, 1993; Laver and Burke,
1993; Salthouse, 1988). We will discuss in detail the construction of an interactive inhibition
model which can be used to predict the time it takes older and younger adults to name a target
word in a lexical naming task. Interactive inhibition models are a special case of connectionist
networks (McClelland and Rumelhart, 1986). Unfortunately there is not time to discuss either
these more general connectionist models or the more general OP networks referred to above.
3. MODELS OF SLOWING: PERT NETWORKS
To begin, we want to describe very briefly the class of latent models based on a
representation of processing as a PERT network. We mentioned above that simple serial and
parallel networks were examples of PERT networks. More generally, a PERT network is a
directed, acyclic (no cycles) graph with a single source (start node) and single sink (finish
node). Examples are presented in Figure 1. The arcs represent the latent processes. A path is
an unbroken sequence of arcs connecting two nodes. For example, in Figure lb the sequence
of arcs, SOA, p:, d:, and rE, represents one of the two paths between the start and finish nodes.
All nodes are A N D nodes. Thus, all processes which exit from a node cannot begin until all
processes which terminate at the node have completed. For example, in Figure lb, process dE
cannot begin until both processes d~ and p: have completed. Finally, all processes which can
begin do begin.
We want to use PERT networks to compute the time that it takes a subject to respond.
Define the duration of a path as the sum of the durations of the processes which lie along the
path. In Figure la, the response time is simply equal to the duration of the single path from the
start to the finish node. In Figure lb, it may not be clear immediately from the definition of a
PERT network how one should compute the response time. Recall that process dz cannot
begin until both processes dl and p2 have completed (which depend, respectively, on processes
pl and SOA completing). If we let P~ be a random variable representing the duration of
process pl, D~ be a random variable representing the duration of process dl, and so on, then
the time that it takes a subject to respond (measured from the moment the two processes at the
source node begin executing) will be equal to the sum of the duration of the longest path
between nodes 1 and 2 plus the duration of the path between nodes 2 and 3. Setting T equal to
the response time, we obtain:
7'2 -- m a x i m u m { p ~ + D~, SOA + P2} + 0 2 + R 2 . (9)
Of course, we want to compute the average or expected response time, not the response time
on a particular trial. We can do this analytically or computationally. Once we can predict the
response times of the older and younger adults based on the underlying network of processes,
we candetermine by just how much each of the latent processes is slowed in the older adults.
We now want to compare this technique with current techniques using tasks where the current
techniques have led either to false positive or false negative decisions.
3.1. False Positive Decisions
To begin, we compare the construction and evaluation of latent models of slowing with
standard tests of slowing using a task where the existing methods of analysis might lead an
investigator falsely to accept a model which assumes that all latent processes are slowed
identically. The existing methods are confined primarily t o an analysis of the correlation
coefficient obtained from the overall regression of the older adults' response times on the
younger adults' response times. As Fisk, Fisher and Rogers (1992) note, there exists the real
possibility that an investigator will falsely accept a latent model of general slowing when the
latent model is really one of process-specific slowing. For example, Fisk et al. construct a
latent model of process-specific slowing where the proportionate slowing of older adults in one
task is twice their proportionate slowing in a second task. Nevertheless, the global model of
general slowing still explains 97.5% of the variability. In order to avoid this particular
problem, investigators will sometimes look beyond the single measure of explained variability.
For example, Myerson, Wagstaff and Hale (1994) note that a simple visual inspection of the
points in a Brinley plot (a cross plot of younger and older adults' response times) can often
uncover separate slowing functions across tasks. Specifically, it might well be the case that the
points for one set of related tasks visually fell along one line, those for a second set of related
tasks visually fell along a second line (Myerson et al., their Figure 2a). And, even when
separate slowing functions are visually hidden in the original Brinley plot (Myerson et al., their
Figure la), it may still be possible to identify the existence of separate regressions by
examining the residual plots (Myerson et al., their Figure lb) since such plots often amplify
patterns present in the original data.
Examining both the Brinley and the residual plots is certainly important. But, it does
not always suffice. Specifically, it is still necessary to construct and then test latent models of
general and process-specific slowing. An example can make the point clear (Fisher, 1994).
The Brinley plot in Figure 2a represents the results from a study reported by Salthouse and
Somberg (1982). In that study, they asked younger and older adults to indicate whether a
probe digit was or was not present in a memory set. Salthouse and Somberg varied the
number of items in the memory set (1 or 4), the visibility of probe digit (degraded or intact),
and the difficulty of the response (simple or complex). Thus, given that both younger and
older adults participated in the study, there are a total of 32 conditions (i.e., 2 levels of memory
set size x 2 levels of probe visibility x 2 levels of response difficulty x 2 levels of probe
presence x 2 levels of age). A linear model of general slowing explains fully 98.2% of the
variance (collapsing across the probe present and absent conditions). Visual inspection of the,
Brinley plot in Figure 2a does not indicate to us any obvious problem with the model of
general slowing. The differences between the predicted and observed response times in each
condition are plotted in Figure 2b (computing these differences or residuals required estimating
younger and older adults' observed response times in each condition from Figure la in
Salthouse and Somberg since the response times were not directly reported in the article; the
predicted older adults' response times could then be computed directly fromthe regression
weights as reported). There is no clear pattern present in the residuals, that is, the residuals do
not lie along two clearly different lines. Thus, there is nothing to suggest the existence of
separate slowing fimctions in either the original Brinley plot (Figure 2a) or the plot of the
residuals (Figure 2b). Yet, the model of general slowing can be rejected in favor of a model in
which it is assumed that the rate parameters which govern the slowing of the latent cognitive
processes are not identical (Fisher, 1994).
Testing latent models of general and process-specific slowing entails a three step
process. First, for the younger adults a latent or network model must be constructed. Second,
for the older adults the latent model must be modified, the exact modification depending on the
type of slowing-- general or process-specific. Finally, the joint (combined) fit of the younger
adults' latent model and the older adults' latent model of general slowing must be compared
(a)
a
s
9 s
s
js
-J sj
~
p~
j s
s
J
J
" Old - . ~ 1 9 ~ 1 1 (Young)
s
/
s
s S
jP
sS
/
f /! , I i I 9 I . ! , ! , ! , ! . I l
91 .3 .S .lr J 1.1 1.3 11.S 1.7

( ~ )
Youl~
(b)
20( --
15( -- II
10( m II II i
5C-" II
Residuals 0-"
-5s - - i i
-10( - -
-15( - - II
! I I ! I I I I i
400 500 600 700 800 900 1000 1100 1200 1300 1400
Y o u n g R e s p o n s e T i m e s (ms)
Figure 2. a) A Brinley plot of younger and older adults response times in the memory search task run by
Salthouse and Somberg (1982). b) The residuals (older adults' observed response times minus older adults'
predicted response times) plotted as a function of the younger adults response times.
statistically with the joint fit of the younger adults' latent model and the older adults' latent
model of process-specific slowing. A detailed review of the testing techniques is beyond the
scope of the present chapter. However, given our very clear emphasis on the need to construct
and fit latent models, we will briefly present some of these details. We will use the memory
search task to illustrate just how the process works.
First, we need to construct a latent model of processing for the younger adults. Recall
here that we assumed that performance in the memory search task could be modeled as a set of
four processes arranged in series (Figure la). Thus, the latent model is a PERT network. The
younger adults' response time on a given trial can then be written as the sum of the encoding,
comparison, decision and response times [Equation 0]. To make the latent model useable, we
need to predict the average or expected response times. As noted earlier in the chapter, this
can be done either analytically or computationally. Here we take the analytic route since it is
the more straightforward one. In particular, the average or expected response time E[Yo] is
equal to the sum of the average or expected durations of the component processes, i.e.,
E[yo] = E[E] + E[C,] + E[D] + E[R] + E[Eo]. (10)
We assume that the expected error, E[E#], is equal to zero. So, we need to know only the
otheffour expectations. The expectations in the sum will presumably vary with the condition.
Table 2
Predicted Response Times of Younger Adults
Predicted Target Probe Response
Response Time Present Intact Simple
E[Y] = tae d- tac -~- tad -~- tar yes yes yes
E[Y] = ta~ + tac + tad + tar' yes yes no
E[Y] = ta~,+ ta~ + tad + tar yes no yes
E[Y] = ta~,+ ta~ + tad + tar' yes no no
E[Y] : tae d- tac -~- tad' d- tar no yes yes
E[Y] : tat d- tac d- tad' -~- tar' no yes no
E[Y] : tae' -[- tac -~- tad' -}- tar no no yes
E[Y] = ta~,+ ta~ + tad' + tar' no no no
Specifically, the average encoding time should be longer when the probe is degraded
[represented by tae,, i.e., E[E] = tae, in Equation 10] than when the probe is intact (ta~). The
average comparison time when there is one stimulus in the memory set (tat) will be exactly four
times shorter than this comparison time when four stimuli are in the memory set since the
search is exhaustive. The average decision time will be longer when the probe is absent (tad')
than when it is present (tad). And the average response time will be longer when the response
is a complex one (tar') than when it is a simple one (tar). In this case there are a total of 16
conditions and 7 parameters needed to predict the expected response time of the younger
adults. The predictions for all 8 conditions with a memory set size of one are listed in Table 2.
The predictions for all 8 conditions with a memory set size of four follow immediately by
substituting 4~tc for ~tc in each equation in Table 2. Second, we need to construct the latent
models of general and process-specific slowing for the older adults by modifying appropriately
the latent model which describes the behavior of the younger adults. This is simple enough if
we assume, not unreasonably, that the latent model of general slowing is a multiplicative one
(Cerella and Hale, in press). In this case only one extra parameter is needed to describe the
slowing of the older adults in the 16 conditions in which they participated. Thus, a total of 8
parameters are needed to fit jointly the latent model for the younger adults and the latent
multiplicative model of general slowing for the older adults to the 32 observations for both
groups (16 for the younger adults, 16 for the older adults). Similarly, if we assume that the
latent model of process-specific slowing is a multiplicative one, then we need an additional four
parameters, one slowing constant for each of the four processes. Thus, a total of 11
parameters are needed to fit jointly the latent model for younger adults and the latent
multiplicative model of process-specific slowing for the older adults, again to the 32
observations for both groups.
Third, we need to determine whether the joint fit of the latent model for the younger
adults and the latent model of process-specific slowing for the older adults explains
siL-,nificantly more variance than the joint fit of the latent model for the younger adults and the
latent model of general slowing for the older adults. This requires estimating the parameters.
In particular, values of the parameters are sought which minimize the error sum of squares,
SSE, where SSE is equal to the sum over all 32 conditions of the square of the difference
between the observations and predictions in each condition. Set SSE(PSS) equal to the error
sum of squares for the process-specific model; set SSE(GS) equal to the error sum of squares
for the general model. Set df(PSS) equal to the degrees of freedom for the process-specific
model: dJ(PSS) = 32- 11. And set df(GS) equal to the degrees of freedom for the general
slowing model: df(GS) = 32 - 8. Then, the statistic,
SSE(GS) - SSE(PSS)]
(11)
SSE(PSS)]
has an F distribution with dJ(GS) - df(PSS) degrees of freedom in the numerator and df(PSS)
degrees of freedom in the denominator if the model of general slowing is the correct one
(assuming that the errors are normally distributed). If this statistic is significant at an
appropriate level, then we can reject the model of general slowing. [Note that although this
procedure could in principle be used to test the latent models of general and process-specific
slowing that we just developed, the above procedure had to be modified so that it applied to
the specific set of results reported by Salthouse and Somberg (1982) since they did not report
the means in all 32 conditions. Fisher (1994) describes the additional modifications needed to
fit their results. In fact, as noted above, when this modified approach is applied to the
Salthouse and Somberg data, the latent model of general slowing must be rejected.]
In summary, a global model of general slowing can appear to fit the results extremely
well, both visually and statistically, as evidenced in the study of memory scanning reported by
Salthouse and Somberg (1982). Yet, as Fisher (1994) argues, this does not rule out a latent
model of process-specific slowing in which the processes are arranged in a PERT network.
One clear way to rule out a latent model of process-specific slowing is to construct the latent
representation and then compare the fit of the latent models of general and process-specific
slowing. We have described briefly how to do such above using the memory scanning task as
an example. The general characteristics of the procedure would not change across tasks:
construct a latent model for the younger adults; construct latent models of general and
process-specific slowing for the older adults; and then test the two models. Of course, the
details of the latent model will change with the task. This would be troublesome if latent
models were extremely difficult to construct and test. However, procedures exist which make
it relatively easy to do such for even complex latent models which can be represented as PERT
networks (Schweickert and Townsend, 1989, Townsend and Schweickert, 1989).
3.2. False Negatives
As noted earlier, a problem opposite to the one identified in the above section can arise
when a global model of general slowing fails to fit the results across two or more domains.
Here, the tendency is to want to reject the associated latent model of general slowing.
However, caution is required when one or more of the domains contain tasks where there exist
processes which influence the response time of a subject but are not under the direct control of
the subject. Such processes, called exogenous processes by Fisher (1994; also see Fisk and
Fisher, 1994), are frequently present in the tasks used to study the slowing of language skills
(Bowles, 1993; Balota, Black and Cheney, 1992; Balota and Duchek, 1988; Burke, White and
Diaz, 1987; Howard, Shaw and Heisey, 1986; Laver and Burke, 1993; Madden, 1989). For
example, in a lexical decision task the target is not presented simultaneously with the prime but
instead is delayed by some amount (the stimulus onset asynchrony or SOA). This delay is not
under the control of the subject, but does influence the time it takes the subject to decide
whether or not a word has been presented. Caution is required when attempting to fit a global
model of general slowing to tasks which do and do not contain exogenous processes because
one can (correctly) reject the global model of general slowing even though a latent model of
general slowing is the one actually governing behavior, a situation we have labelled a false
negative decision.
We now want to compare the construction and evaluation of latent models of slowing
with standard tests of slowing in cases where a false negative decision will occur if standard
techniques are applied. We need both a task containing exogenous and a task not containing
exogenous processes. In this section, the example we use for a task containing exogenous
processes is a double stimulation task like that described by Pashler and Johnston (1989) since
performance in such tasks is easily modeled by the PERT networks described here (in a later
section, the example we use for a task containing exogenous processes is the one described
above, the lexical naming task, since such tasks are well modeled by the connectionist
networks described in that later section). The example we use for a task which does not
contain exogenous processes is a task we have akeady described, the memory search task. For
both tasks, we develop latent models of general slowing and predict older and younger adults
response times. And for both tasks, we regress the predicted older adults' response times on
the predicted younger adults' response times. We then compare the global slowing functions
obtained from the above two regressions and show that these functions differ even though the
latent model of slowing is a general one. Thus, we will have shown that the standard
techniques can easily l"ad an investigator falsely to reject a latent model of general slowing.
To begin, we want to develop the latent model of general slowing in the double
stimulation task, the task which contains an exogenous process. To do this, we need to
describe the double stimulation task in more detail. In the double stimulation task, two stimuli
are presented, one stimulus followed some short time laterby a second stimulus (here, as in the
Icxicalnaming task, the time between the presentation of the two stimuli,the SOA, represents
the exogenous process). A separate decision and response must be made to each stimulus
(these could be Icxical decisions, though typically they are not such). The latent network
governing processing in this task can be represented most simply in Figure Ib (the details are
described in Schweickert, Fisher and Goldstein, 1994). It is assumed that the peripheral
processing of the two stimuli can go on in parallel(pl and p2 in Figure Ib). It is assumed that
the decision about the second stimulus (d2) cannot begin until the decision about the first
stimulus (d~) has been made, a constraint consistent with the notion that the central processor
has a very limited capacity (Broadbcnt, 1958).
We now want to predict the average response times to the second stimulus (the only
prediction of interest here since this response time is determined, in part, by the duration of the
SOA or exogenous process). Above, we said that such predictions could be obtained either
analytically or computationally. In the preceding example, we took the analytical approach,
deriving closed form expressions for the average response time. Here, we will take the
computational approach, estimating the average response time by using the computer to
simulate processing. To begin, we need to make an assumption about the distribution of the
various process durations and the length of the SOA. To keep things simple, we assume that
the time it takes to complete both processes pl and dl is exponentially distributed with a mean
of 100, that the time it takes to complete process p2 is exponentially distributed with a mean of
200, and that the time it takes to complete both processes d2 and r2 is exponentially distributed
with a mean of 100. Furthermore, we assume that three SOAs are used, 50, 100 and 150. At
each SOA, we then generate sample times from the distributions associated with the various
processes or sums which appear in Equation 9. We then use a slight modification of this
equation to compute the time T2 that it takes subjects to respond to the second stimulus on a
particular trial. The modification is required because the response time to the second stimulus
is measured from the moment that the second stimulus is presented (at the initiation of p2), not
from moment that the first stimulus is presented (at the initiation of p~). Thus, in order to
compute T2 We simply need to subtract SOA from the fight hand side of Equation 9. And we
average over many trials to get an estimate of the expected response time. Doing such, we
predict that the younger adults will take 352, 379 and 415 ms to respond to the second
stimulus at, respectively, SOAs of 50, 100 and 150 ms. Now, to simulate general slowing, we
increase the mean process duration by say 100% so that processes p~, d~, and r2 and d2 have a
mean of 200 and p2 has a mean of 400. Again, averaging over many trials, we predict that the
older adults will take, respectively, 683, 704 and 737 ms to respond to the second stimulus. If
we regress the older adults' predicted response times on the younger adults' predicted response
times, we find a correlation of 0.999 with a y intercept of 375 and a slope of 0.87.
We have now calculated the parameters (the slope and intercept) which researchers
have used to draw inferences about global slowing. Suppose that we now nm these same
simulated subjects in a task inwhich there is no exogenous'process.such asthe~memory search
task. Again, we assume that the mean process durations for the older subjects are slowed by
100%. We want to determine whether the slope and intercept remain unchanged in the new
simulation. If the task can be represented as a PERT network and there are no exogenous
processes, as we are now assuming, then the latent model of general slowing reduces to a
global model of general slowing (Fisher, 1994). Thus, in all conditions of the task, the
predicted response times of the older adults will be twice as long as those of the younger adults
since the mean process durations of the older adults are each slowed by 100%. Now, were we
to:regress older adults' predicted response times on younger adults' predicted response times in
a task without exogenous processes, we would find a correlation of 1.0, an intercept of 0.0,
and a slope of 2.0. This result differs considerably from the above finding of an intercept of
375 and a slope of 0.87 when the task contains an exogenous process. As we have shown,
however, this difference in the slope and intercept across the two tasks should not lead us to
conclude that the two task domains are governed by different slowing functions.
Intuitively, the reason for the different slopes in the tasks which do and do not contain
exogenous processes can be made clear if one assumes that the durations of the processes in
model of the double stimulation task are constants instead of random variables. If the ~SOA is
relatively short, the combined duration of process pl and dl will exceed the combined duration
of the SOA and process pz. Thus, the time it takes the younger adults to respond to the
second stimulus is given by the sum, pl + d l + d2 + r2- SOA, whereas the time that it takes
older adults to respond to this stimulus is given by the sum, ~(pl +dl + d2 + r2)- SOA,
assuming that the slowing is a general one. However, when the SOA is relatively long, the
combined duration of process pl and dl will be less than the combined duration of the SOA and
process p2. Thus, the time it takes the younger adults to respond to the second stimulus is
given by the sum, p2 + d2 + r2, whereas the time that it takes older adults to respond to this
stimulus is given by the sum, ~(p2 + d2 + r2). Note although the latent slowing function, ~, is a
general one, a global multiplicative model with the same slowing function does not fit the
results across long and short SOAs for, if such were the case, the ratio of the younger to the
older adults' response times when the SOA is short should equal this ratio when the SOA is
long Clearly such is not the case, i.e.,
Pl .+ d l + d2 + r 2 - S O A P2 + d2 + r2 (12)
fl(p,+ d, + d, + r, ) - S O A fl(P, + d, + r,)"
And, it is not the case because the duration of the exogenous process is not lengthened by the
slowing function whereas the durations of the endogenous processes are each lengthened by
the slowing function.
In summary, as claimed, one global slowing function relates the response times of the
older and younger adults in tasks with exogenous processes and a second global slowing
function relates the response times of the older and younger adults in tasks without exogenous
processes, even though the same latent slowing function 13relates the mean duration of each of
the processes in the older adults' latent network to the mean duration of the associated process
in the younger adults' latent network. Thus, there exists the real potential for falsely rejecting a
latent model of general slowing if one inspects only the multiplicative slowing function derived
from the fit of a global model of general slowing across different task domains when
exogenous processes govern performance in one but not all domains.
3.3. Global Models of General and Domain-Specific Slowing
Some readers may wonder at this point whether the above criticisms of global models
of general slowing apply equally to global models of domain-specific slowing. After all, the fit
of a global model of general slowing can be improved upon by considering the task domain
(Cerella, 1985; Lima, Hale and Myerson, 1991; Mayr and Kliegl, 1993). For example, Lima et
al. have recently reported a meta-analysis of both lexical and nonlexical tasks. The lexical
tasks are defined simply as tasks that use words as stimuli; nonlexical tasks as those that do not
use words. Lima et al. find that older adults are slowed significantly more in lexical tasks than
they are in nonlexical tasks. Specifically, regressing older adults' responses times on younger
adults response times, they find that within the lexical domain the slope is equal to 1,48
whereas within the nonlexical domain the slope is equal to 2.05. Cerella reports that older
adults are slowed si~ificantly more in experimental tasks than they are in control tasks
(CereUa notes that the definition of which was the experimental and which the control tasks
was largely arbitrary and varied from study to study). Finally, Mayr and Kliegl find that older
adults are slowed si~ificantly more in tasks which require subjects to coordinate several steps
in the task (tasks which they define as high in coordinative complexity) than they are in tasks
which do not require such coordination but do require the execution of multiple steps (tasks
which they define as high in sequential complexity).
Unfortunately, although it is true that global models of domain-specific slowing fit
much better than do global models of general slowing, the same criticisms apply to both sets of
models. Specifically, false positive decisions remain a real possibility because inspection of a
Brinley plot and the associated residuals within a given domain is not necessarily any more
instructive than inspection of a Brinley plot and the associated residuals across all domains.
False negative decisions remain a real possibility if the test of separate slowing functions within
a domain comes from the comparison of the slopes of global models of slowing fit to different
sets of tasks within the domain where one set of tasks contains exogenous processes and the
other does not. Again, the key to avoiding both false positive and false negative evaluations is
the construction and testing of latent models of general and process-specific slowing.
4. MODELS OF SLOWING: CONNECTIONIST NETWORKS
Not all behaviors can easily be modelled by PERT networks. In particular, behaviors
whose outcome is dependent on the outcome of activity within the lexicon are frequently better
described by the interactive inhibition (or more general connectionist) models first used
extensively by McClelland and gumelhart (1981, 1986). Recently, connectionist models have
been used to study the effects of aging, both generally (Salthouse, 1988) and in specific tasks
such as naming (Bowles, 1993; Laver and Burke, 1993). Below, we want to develop a simple
latent interactive inhibition model of slowing in a lexical naming task. We will show that
unless one specifically tests the latent interactive inhibition model it is possible falsely to
accept a latent model of general slowing when the correct model is one of process-specific
slowing; or, alternatively, it is possible falsely to reject a latent model of general slowing when
it is the correct model.
4.1 False Positive Decisions
Here we want to focus on the false positive decisions. That is, we want to show that a
global model of general slowing can explain an overwhelming percentage of the variability
even when the true model is a latent interactive inhibition model of process-specific slowing.
The specific lexical naming task that we want to consider is the one run by Balota and Duchek
(1988). Subjects were first given a prime word followed some brief time later by a target
word. The time between the presentation of the prime and target words was varied from 200
to 800 ms. In the condition we will model, the prime either was highly related to the target or
was a neutral prime (the word blank). The observed response times for younger and older
adults in the high related (HR) and neutral (HN) conditions were estimated from Balota and
Duchek (their Figure 1) at SOAs of 200, 500 and 800 ms (Table 3).
Initially, suppose that we fit a global model of general slowing. Then, if we regress the
response times of older adults on those of younger adults, we fmd:
O = -201.31 + 1.63Y (13)
where r 2 = . 9 5 . T h u s , it appears that a global model of general slowing does quite well.
However, suppose that we go on to construct a very simple, latent interactive inhibition model
of behavior in the lexical naming task for the younger adults and a latent interactive inhibition
model of process-specific slowing in this same task for older adults. The network used to
represent behavior in the task consists of several nodes and links between these nodes. To
keep the discussion a simple one, assume that there is a node in semantic memory which codes
the meaning of the prime (node 3 in Figure 3) and a node in semantic memory which codes the
meaning of the target (node 4). Similarly, assume that there is an orthographic node (node 1)
which codes the spelling of the prime and an orthographic node (node 2) which codes the
spelling of the target. The activation level of each node will vary in time. We will assume that
the activation ranges between 0 and 1 at each node. Let a~t) represent the activation of node i
at time t. Set the activation of the semantic nodes and the target orthographic node to 0 at the
start of the trial (when the prime is presented). Set the activation of the prime orthographic
node to 1.0 at the start of the trial. Set the activation of the target orthographic node to 1.0
when it is presented. Assume that the prime orthographic node (node 1) is linked with
strength ct31 to the prime semantic node (node 3), that the target orthographic node (node 2) is
linked with strength ct42 to the target semantic node (node 4), and that the prime semantic node
is linked with strength ct43 to the target semantic node. Finally, assume that the link strengths
in our simple example take on values between 0 and 1. Now, we know the activation of all
nodes at the start of a trial. We want to predict this activation at each point in time after a trial
is initiated. We want to do such because we will assume that at some point the activation of
the target semantic node will reach a level sufiiciently high that it causes the subject to initiate
the naming of the target. We will use a very simple rule to predict the increase in activation at
each node over some small interval of time fit. Specifically, the activation a~t + fit) of node i at
time t + fit will be set equal to its activation a~t) at time t plus some fraction equal to 1 - a~t)
of whatever activation has ~spread from the other nodes which are linked to it during the
interval fit. The amount of activation which spreads from nodej to node i will be set equal to a
fraction of the activation at node j determined by the link strength between node j and node i,
i.e., equal to the product u~j(t). If we assume no decay, then throughout a trial al(t) = 1,
Table 3
Lexical Naming Latencies
Observations I
SOA Older Adults Younger Adults

al~ 2 aN 2 SP 2 I-~ aN SP
200 674 678 4 537 539 2

500 643 660 17 514 526 12
800 627 645 18 510 523 13
Predictions
SOA Older Adults Younger Adults

HR HN SP HR HN SP
200 664 669 4 538 542 7
500 650 658 11 524 531 9
800 636 646 12 509 519 11
1Taken from Balota and Duchek (1988, Figure 1).

2HR: Highly Related Prime; HN: Neutral Prime; SP: Semantic Priming Effect (the difference between the
neutral and highly related latencies).
before the target is presented, a2(t) = 0: after the target is presented a2(t) = 1. Thus, we need
to compute only the activation a 3(0 and a4(t) at each of the semantic nodes. It follows from
the above, that the activation at each of the semantic nodes is computed is follows:
(14)
Cl3 a .4.- (~0 -- a3 (O -[- a31Gl (O ( 1 - a3 (O) ,
Computationally, one starts at time 0. A suitably small increment of time 8t is selected. The
activation is computed at t = St. Next the activation is computed at 28t, and so on. Of course,
we are not interested in the activation at each of the semantic nodes per se. Rather, we want
to know how long it takes a subject to name the target. Let c represent the critical level of
activation of the target node above which the subject decides to pronounce the target. A
younger subject initiates a verbal response as soon as the activation of the target exceeds the
critical activation, i.e., as soon as a4(t + fit) >_c. Set ty(SOA, related) equal to the shortest time
t + 8t such that the above condition is met when the SOA is as indicated and the prime is
related to the target. Let Vr represent the time it actually takes a subject to select and execute a
semantic semantic verbal

prime target response
(node 3) (node 4) (node 5)
(/,43 (/,54
(X31 (/,42
orthographic orthographic
prime target
(node 1) (node 2)
Figure 3. Representation of the nodes in the conneetionist network which governs behavior in a lexieal
naming tasks.
verbal response. This corresponds to the time that it takes the activation to flow from the
target semantic node (node 4) to the decision and response node (node 5). Then, the younger
adults' response time y o u n g ( S O A , r e l a t e d ) can be written as:
young(SOA, related) = t y (SOA, r e l a t e d ) + v , . (15)
As it stands, this prediction would not change across related and neutral conditions because we
have not differentiated above between the link strengths in these two conditions. Of course,
such a difference is expected. In particular, in the neutral condition, we cannot assume that the
strength ct43(neutral) of the link between the neutral prime semantic node and the target
semantic node is zero. However, we can assume that this strength is less than the strength of
the same link between the related prime semantic node and the target semantic node. In
summary, in order to model the younger adults naming latencies in the related and neutral
conditions we need to estimate the link strengths r aaz, cx43(related) and tx43(neutral), the
critical threshold c, and the response selection and execution time v,.
In order to model the older adults' naming latencies in the related and neutral
conditions we need to estimate the above parameters as well as whatever parameters reflect the
aging process. We will assume the existence of only one parameter. In particular, we will
assume that the activation between any two connected pair of nodes, except for the target
semantic and response nodes, is spread less rapidly, being reduced by a factor of re. The
exception corresponds to the assumption that the psychomotor portion of the response, whose
duration is represented by Vr in the younger adults, remains unchanged in the older adults
(Cerella, 1985). Then, we obtain:
(16)
Cl3 a "[- ~ 0 = Cl3 (O "}- I~ a 31C11(~)( 1 - G 3 (O ) ,
Now, if for the older adults we set to equal to the shortest time such that a4(t + 8 0 >_c, then:
old(SOA, related) = to(SOA, related) + vr 9 (17)
A similar equation holds in the neutral condition.

At this point, we can fit the above model [Equations 15 and l7] to the results from
Balota and Duchek (1988). We did not search the entire parameter space systematically since
we obtained such a good fit with the initial parameter settings. Specifically, we make the
predictions reported in the lower half of Table 3. These predictions correspond to the
parameter values:
a~l = .005, a~(related)= .00025, a~(neutral)= .00020,
(18)
a,t2 = .005, Jr = .78, c = .9.
Note that the latent model is a process-specific one since the duration vr of the response
selection and execution processes is identical across changes in age whereas the time it takes
the target node to reach the critical activation level c is not. This simple latent model of
process-specific slowing explains fully 99% of the variance. Thus, we see again the potential
for false positive decisions when only the fit of a global model of general slowing is examined.
That is, we have shown that a latent model of process-specific slowing can better explain the
results from a lexical naming task than can a global model of general slowing even though the
global model of general slowing explains most of the variance (95%). Of course, we have only
illustrated here the potential for a false positive decision. To show that the latent model of
process-specific slowing that we have constructed fits significantly better than a latent model of
general slowing, we would need to construct the latter model and then, as we did in the
previous section, compare statistically the fit of the two latent models. We should point out
that Balota and Duchek did not address specifically the issue of whether the latent slowing was
a general or process-specific one. We use their results simply to illustrate the potential for
false positive decisions.
4.2 False Negative Decisions
We now need to discuss the potential for false negative decisions when connectionist
models are used to describe performance both in tasks which do and in tasks which do not
contain exogenous processes. To illustrate this potential, we follow the same procedure we
used when discussing false negative decisions in cases where the underlying behavior was
represented as a PERT network. Specifically, we construct a latent model of general slowing
which can be used to predict performance in tasks with and without exogenous processes. We
will then regress the predicted response times of the older adults on the predicted response
times of the younger adults in each set of tasks and show that the slope differs across tasks
even though the latent slowing function did not change.
To begin, consider a task with an exogenous process. In this case, we can use the
above lexical naming task in the highly related condition. As noted there, the process
representing the SOA is an exogenous process. However, now we want a latent model of
general slowing, not a latent model of process-specific slowing. Here, we simply assume that
the slowing in the spread of activation affects all links, including the link between the target
semantic and response nodes. Set this slowing factor to ~ = .78, as above. To keep things
simple, we arbitrarily set all link strengths equal to .005. Furthermore, as above, assume that
activation does not spread from the target semantic node until its threshold is reached. Finally,
assume that a response and decision have completed executing as soon as the critical threshold
is reached at node 5, the same threshold as the target semantic node (c = .9). Then the times
that it takes younger and older adults at each of the three SOAs to make a response in the
highly related (HR) conditions are given in Table 4. Regressing the older adults' response
times on the younger adults' response times, we find the correlation is 0.99, the intercept is
302.9 and the slope is .97.
Next, consider a task with no exogenous processes. We imagine here a task like visual
search or memory search task which requires the repetition of a process, the number of
repetitions depending on the number of stimuli which are presented. In particular, assume that
one, two or three nodes (processes) are in series. Assume that each node is a threshold node.
Assume that the slowing reduction, link strengths and critical thresholds are set at the same
values in this task as they were in the above task. The predicted times that it takes younger
and older adults to complete one, two and three processes arranged in series are given in Table
4. If we now regress older adults' response times on the younger adults' response times, we
find the correlation is 0.99, the intercept is 637.3 and the slope is 1.30. These parameters
Table 4
Connectionist Model of Tasks With and Without Exogenous Processes
Predictions
Naming Latency Search Latency
SOA Younger Older Set Younger Older
Adults Adults Size Adults Adults
200 950 1220 1 461 591
500 936 1206 2 973 1247
800 921 1192 3 1485 1963
differ from those we calculated for the task which involved exogenous processes. Thus, we
see again that a latent model of general slowing can be the correct one even when a global
model of general slowing can be rejected across different domains.
5. DISCUSSION
We have spoken above of the methodological, theoretical and practical advantages

which derive from using latent models of slowing. We now want to summarize and extend
each of these advantages.
5.1 Methodological Importance
To begin, we noted at the outset that investigators typically assume that if one function
governs the global response times, then one function governs the latent process durations
(what we called the assumption of functional equivalence). We have seen that there are two
related violations of this assumption. Specifically, we have shown that a global model of
general slowing can fit the results from existing experiments extremely well (e.g., Balota and
Duchek, 1988; Salthouse and Somberg, 1982) even though a latent model of process-specific
slowing is the correct one. Thus, false positive decisions remain a very real possibility.
Additionally, we have shown that a global model of general slowing can fail to fit the results
across different sets of tasks (i.e., tasks with and without exogenous processes; Lima et al.,
1991) even though a latent model of general slowing is the correct one. Thus, false negative
decisions remain a real possibility. The latent models of general and process-specific slowing
that we constructed were drawn t~om the two major classes of information processing models,
PERT networks and connectionist networks. We did this to emphasize the point that the
methodological problems do not disappear as the models change.
We also noted at the outset that investigators typically assume that the structure which
governs the processing of older adults in a given task is identical to the structure which
governs the processing of younger adults in the same task (what we referred to as the
assumption of structural equivalence). We have ourselves made this assumption throughout.
However, ultimately this assumption needs to be tested for each task using the recent
techniques like those described by Schweickert (1978; Schweickert and Townsend, 1989;
Schweickert, Fisher and Goldstein, 1994). If the assumption is violated, then it would be the
case that a change in speed does not directly reflect a slowing of the processes used by the
younger adults. For example, although not in the domain of aging, consider Logan's (1988)
instance based theory of automaticity. After a great deal of practice, subjects change from a
slow algorithmic retrieval to a fast memory-based retrieval. Here, the increase in speed is not
produced by an increase in the rate with which the latent processes are executed but rather by a
substitution of one set of processes which together take a long time to complete (though each
individual process may be executed very quickly) with a second, different and perhaps smaller
set of processes which together take less time to execute. Thus, the change in speed is a by-
product of the change in the retrieval mechanisms.
In most cases, we do not have direct access to the durations of the latent processes or
to the structure of the network and thus must derive both from the pattern of overall response
times. However, recent research in cognitive neuroscience suggests that someday soon it may
be possible to measure directly the duration of certain groups of cognitive processes. We want
now briefly to comment on the relation between latent models of process-specific slowing and
the recent research in the cognitive neurosciences (for an extensive commentary on this
relation, the interested reader should consult Johnson and gybash, 1993). Our comments take
as their starting point a recent review by Bashore (1993). He argues that the evidence from
cognitive psychophysiological studies indicates that the response end of processing slows more
than the stimulus end. There are two key components to his argument. First, he considers a
meta-analysis which used both P300 latencies and response times as the dependent variables in
tasks which required only very simple motor responses (Bashore, Osman and Heffley, 1989).
The global response times of the older adults were regressed on the global response times of
the younger adults. A global multiplicative model of general slowing explained best the results
from tasks which required only simple motor responses [[3 = 1.27 in Equation 0]. However,
when the P300 latencies of the older adults were regressed on the P300 latencies of the
younger adults, a global additive model of general slowing best explained the results
[~ : 80.10 ms in Equation 0 with [3 set equal to 1]. The P300 latency is assumed to reflect the
time that it takes subjects to evaluate and categorize stimuli quite independently of the time
that it takes subjects to make a response. The time that it takes subjects to make a response is
then given by the difference between the overall latency and the P300 latency (e.g., Ford, Roth,
Mohs, Hopkins, and Kopell, 1979). When this difference for the older adults is regressed on
this difference for the younger adults, a global linear model now best explained the results
[a - 49.76 and 13= 1.32 in Equation 0]. In short, the meta-analysis ofBashore et al. suggests
that the slowing function governing stimulus encoding and categorization is of one form, that
governing response selection and organization of another form. The second set of studies
reviewed by Bashore (1993) also implicates separate slowing functions in the somewhat more
complex memory scanning tasks we described earlier in the article. In particular, a global
additive function best explains the relation between the P300 latencies of the older and younger
adults whereas a global linear function best explains the relation between the overall response
time - P300 difference in the older adults and this same difference in the younger adults.
In summary, the identification of the exact slowing of each of the latent processes can
in principle avoid whatever problems attend the violation of the assumptions of structural and
functional equivalence. In most cases, this identification can be done only indirectly
However, at least in some cases, it appears possible to measure the durations of the latent
processes themselves. In either case, the latent model of general slowing cannot be mistaken
for a latent model of process-specific slowing, or conversely, if it is known by how much each
of the latent processes is slowed. Similarly, the latent model describing the behavior of the
older adults need not be assumed identical to the latent model describing the behavior of the
younger adults if the network is tested explicitly.
5.2 Theoretical Importance
In addition to their importance to methodology, the techniques we described for

identifying the slowing of each of the latent processes have importance for the development of
a unified theory of slowing. Such a unified theory was possible when there existed just one
global model of general slowing. And, a unification appeared reachable when the global model
of general slowing was replaced by separate models of domain-specific slowing. However,
now these domains are multiplying several fold. And the global models of domain-specific
slowing may need to be replaced by still more finely tuned global models of task-specific
slowing. For example, Laver and Burke (1993) find the reduction in the semantic priming
effect (obtained by subtracting related priming latencies ~om unrelated priming latencies) for
older adults is not a proportional one. This runs counter to the meta-analysis reported by Lima
et al. (1991) where it will be recalled that within the lexical domain a proportionate slowing of
approximately 50% was observed among older adults. Similarly, Fisk, Fisher and Rogers
(1992), who reanalyze several conditions from Fisk and Rogers ( 1991), report that older adults
are slowed significantly more in the early trials of memory search tasks which use disjoint sets
of semantic categories as targets and distractors (called a consistent mapping or CM task, e.g.,
see Schneider and Shiffdn, 1977) than they are later in these tasks after considerable practice.
Fisk and Rogers also found that this was not the case for visual search even though the same
type of stimuli were to be searched by the subjects. Such findings are not consistent with the
conclusion that one multiplicative factor governs the slowing of all lexical processes in the
lexical domain. Contrariwise, Lima et al. (1991) find that evidence that one multiplicative
factor probably does not govern slowing in the nonlexical domain. Specifically, they note that
the magnitude of the proportional slowing of older adults in a n o n l e x i c a l picture naming study
(Bowles, 1990; Thomas, Fozard and Waugh, 1977) resembles more closely the magnitude of
the proportional slowing of older adults in lexical studies.
Not only does it appear that recent results may require the introduction of separate
global models of task-specific slowing, but a more detailed consideration of various tasks
suggests that separate global models of condition-specific slowing within a given task may be
needed. For example, tasks which measure the retention of skilled performance have recently
been conducted within one of our labs (e.g., See Anderson-Garlach, 1994; Fisk, Cooper,
Hertzog and Anderson-Garlach, 1994; Fisk, Hertzog, Lee, Rogers and Anderson-Gadach,
1994). We find that: (a) older and younger adults retain an impressive amount of skill even
after 16 months without exposure to the task; (b) retention performance declines within a three
month period and that decline remains stable between three and six months for both younger
and older adults; (c) older and younger adults equally retain general, task-relevant skills; (d)
older adults' performance declines more than younger adults' performance for both extensively
trained and moderately trained stimuli; (e) when an interfering processing activity is inserted
prior to the retention interval, older adults' performance declines disproportionately more than
younger adults' performance especially when compared with a task not subjected to such
interference; and (f) depending on the type of search task, for both younger and older adults
the initial retention deficit is largely attenuated by the end of the retention retraining periods
that were used. Since the slowing varies across conditions within the retention task, we can
only conclude that still more detailed global models of condition-specific slowing will be
needed.
The proliferation of slowing functions across domains and now, perhaps, tasks and
conditions within tasks is clearly not a desirable development if it is masking a more
fundamental simplicity. Cerella (1985) points to the core of the problem with the global
models and like approaches when he comments on the differential slowing of older adults in
control and experimental tasks: "However important for aging theory, the designation
experimental or control is psychologically somewhat arbitrary, because a given task may be
introduced as either one or the other depending on the tasks with which it is paired in the
context of the study (p. 78)." More generally, any definition of a set of domains, tasks within
domains, or conditions within tasks which is based on the common external characteristics of
the set (e.g., defining lexical tasks as simply tasks where words are used as stimuli) is
problematic if what is ultimately desired is a definition of a set which is based on the common
internal or latent cognitive processes operating in those tasks.
In summary, there is a growing body of empirical evidence that if global models are
pursued, the domains identified so far will need to be divided still more finally. The division
increasingly refers back to the latent cognitive processes. For example, Lima et al. (1991)
suggest that the proportional slowing of older adults in a nonlexical picture naming study more
clearly resembles the proportional slowing of older adults in lexical studies because the picture
naming study relies heavily on word retrieval, a process which presumably is a lexical one.
Thus, both the empirical evidence and the existence of real methodological problems indicate
the importance of testing latent models of general and process-specific slowing.
5.3 Applied Potential
A final and perhaps the most important, result of constructing and testing latent models
of slowing is the opportunity it provides ultimately to assist older adults performing the various
activities that make it possible to lead a relatively independent life. Many examples could be
used to illustrate this point. Wewill choose as an illustration the development of a physical
interface (e.g., a computer mouse) for an information system or an automatic teller machine
which optimizes an older adult's interaction with the device. To produce the optimal design we
need to answer questions like the following. Is it better to reduce the gain of the movement
control system (the mouse) and if so, what reduction is optimal? Or, is it better to provide
"sticky borders" such that when a pointing device gets close to an area, it is attracted and
"stuck" to it and, if so, exactly how "sticky" should the borders be? Or are there still other
solutions?
Walker, Philbin and Fisk (1994) investigated various explanations for differences
between younger and older adults in movement control tasks which might be used to answer
the above questions. These explanations for age-related differences in movement control
include the following: (a) older adults differ in the number and duration of the submovements
they make within an overall movement; (b) older adults have similar control and movement
structures but are flowed due to an inability to produce high levels of force; (c) older adults are
more error aversive than younger adults, therefore generate flower, but more accurate
movements; and (d)older adults have a higher noise-to-force ratio (i.e., the ratio of the
standard deviation of the force to the force itself) than younger adults. The explanations
Walker et al. offered took as their starting point the optimized submovement model (Meyer,
Abrams, Komblum, Wright and Smith, 1988; Meyer, Smith, Komblum, Abrams, and Wright,
1990; Walker, Meyer and Smelcer, 1993). The optimized submovement model provides a
framework for incorporating the different explanations of age-related differences in movement
control.
In order to test these various explanations, Walker et al. (1994) ran 16 older (75-70)
and 16 younger (18-23) adults in an eight session experiment. The first four sessions were
used to gather information about subjects' abilities, to familiarize the subjects with the use of
the movement control device, and to familiarize the subjects with the payoff conditions for the
speed and accuracy of movement. Also, during these sessions baseline movement times were
established for each individual subject for use in manipulating payoff conditions during the
remaining experimental sessions. The procedure required subjects to move a mouse-controlled
cursor from a starting location to a target box on a video monitor. Subjects pressed a button
on the mouse at the start of each trial and then released this button as soon as they felt
confident that the cursor had entered the target box. Response time and accuracy were
measured on each trial. Manipulation of the payoff structure successfully shifted each subject's
relative speed-accuracy trade-off function across sessions. Using the latent optimized
submovement model to analyze the results, it can be determined that older adults have a higher
noise-to-force ratio than younger adults. Thus, the older adults frequently require more
submovements to position the cursor inside the target box. Reducing the rate of travel of the
cursor with respect to the mouse could potentially decrease the total time it takes the older
adults to position the cursor since the first submovement would more frequently land the
cursor near the target box. The solution arises from an understanding of the underlying
processing required for a given task. Obviously, an analysis of global models would be of little
use here.
5.4 Summary
In summary, many investigators have used regression models in which older adults'
response times are predicted from younger adults' response times (what we have called global
models) to draw inferences about the slowing of latent processes. In particular, investigators
have sought to test the claim that all underlying processes are slowed by the same function or
alternatively, that these processes are slowed by different fimctions. We have argued that these
global models are not always adequate for testing such claims and that latent models of
slowing should be considered. These latter models have three advantages over the currently
popular regression models because the latent models, but not the regression models, allow the
investigator directly to determine the slowing function for each process involved in a particular
task. First, the investigator can avoid what we have termed false positive and false negative
errors. Second, the investigator can identify which processes are slowed and which are spared.
Theories of the relation between the slowing of particular processes and the production of
specific performance decrements can then be more thoroughly tested. Finally, the investigator
can identify which processes are slowed the most. The design of better environments for the
elderly can then take place in a principled fashion.
Arguing that it is important methodologically, theoretically and practically to construct
and then test latent models of slowing is one thing. Actually doing such is quite another. In
fact, we suspect that a great many readers may find somewhat cumbersome, if not daunting,
the actual constructing and testing of models for each of the tasks that they run. Of course, for
many tasks the models already exist. Such models have been reviewed in recent texts (Luce,
1986; Townsend and Ashby, 1983). To test the various theories of cognitive slowing, one
needs only to modify the existing models by introducing into these models general or process-
specific slowing parameters. However, there do exist many tasks, especially applied tasks, for
which latent models of slowing still need to be constructed. Numerous techniques exist for
identifying the structure of the latent models. The recent techniques described by Schweickert
(1978; Schweickert and Townsend, 1989; Schweickert et al., 1994) are very powerful and, we
believe, deserve more attention. A number of techniques exist for testing a latent model of
slowing once its structure has been identified. We believe that several of the recently described
techniques could prove very useful here (Fisher, 1986; Fisher and Goldstein, 1983; Goldstein
and Fisher, 1991, 1992; Kliegl, Mayr and Krampe, 1994; Schweickert et al.). We hope that
there will arise out of the detailed testing of latent models a much better understanding of just
which cognitive processes are and are not slowed. And we hope that this understanding will
bring some clarity given the increasing proliferation of global models of domain-, task-, and
now condition-specific slowing. Whereas there are potentially an infinite number of domains,
tasks and conditions, it has been argued that there are only a finite number of elementary
information processes (Newell and Simon, 1972, page 5).
REFERENCES
Anderson-Garlach, M. M. and Fisk, A. D. (1994, April). Age-related retention of skilled

performance: Within-subject examination of visual search, memory search, and lexical
decision. Presented at the Fifth Cognitive Aging Conference, Atlanta.
Balota, D. A., Black, S. and Cheney, M. (1992). Automatic and attentional priming in young
and old adults: Reevaluation of the two-process model. Journal of Experimental
Psychology: Human Perception and Performance, 18, 489-502.
Balota, D. A. and Duchek, J. M. (1988). Age-related differences in lexical access, spreading
activation, and simple pronunciation. Psychology and Aging, 3, 84-93.
Bashore, T. 1L (1993). Differential effects of aging on neurocognitive functions subserving
speeded mental processing (pp. 37-76). In J. Cerella, J. Rybash, W. Hoyer, and M. L.
Commons (Eds.), Adults information processing: Limits on loss. San Diego: Academic
Press.
Bashore, T. 1L, Osman, A. and Heffley, E. F. (1989). Mental slowing in elderly persons: A
cognitive psychophysiological analysis. Psychology and Aging, 4, 235-244.
Bowles, N. L. (1993). Semantic processes serving picture naming (pp. 303-326). In J.
CereUa, J. Rybash, W. Hoyer, and M. L. Commons (Eds.), Adults information processing:
Limits on loss. San Diego: Academic Press.
Bowles, N. L. (1990, April). Semantic activation in young and older adults during early
processing in a primed picture naming task. Paper presented at the Cognitive Aging
Conference, Atlanta, GA.
Brinley, J. F. (1965). Cognitive sets, speed and accuracy of performance in the elderly. In A.
T. Welford and J. E. Birren (Eds.), Behavior, aging and the nervous system. Charles C.
Thomas, Springfield, IL.
Broadbent, D. E. Perception and communication. Elmsford, NY: Pergamon Press.
Burke, D., White, H. and Diaz, D. (1987). Semantic priming in young and older adults:
Evidence for age constancy in automatic and attentional processes. Journal of
Experimental Psychology: Human Perception and Performance, 1 13, 79-88.
Cerella, J. (1985). Information processing rates in the elderly. Psychological Bulletin, 98, 67-
83.
Cerella, J. and Hale, S. (in press). The rise and fall of information-processing rates over the
fife-span. Acta Psychologica.
Cerella, J., Poon, L. W. and Williams, D. M. (1980). Age and the complexity hypothesis. In
L. W. Pooh (Ed.), Aging in the 1980s: Psychological issues (pp. 332-340). Washington,
DC: American Psychological Association.
Fisher, D. L. (1985). Network models of reaction time: The generalized OP diagram In G.
d~dewalle (Ed.), Cognition, Information Processing and Motivation (Volume 3).
Ammerdam: North-Holland Press, 229-254.
Fisher, D. L. (1994). Cognitive aging: Models of general, task-specific and process-specific
slowing. Submitted for publication, Psychological Bulletin.
Fisher, D. L. and Goldstein, W. M. (1983). Stochastic PERT networks as models of

cognition: Derivation of the mean, variance and distribution of reaction time using order-
of-Processing (OP) diagrams. Journal of Mathematical Psychology, 27, 121-151.
Fisk, A. D. and Fisher, D. L. (1994). Brinley plots and theories of aging: The explicit,
muddled and implicit debates. Journal of Gerontology: Psychological Sciences, 49, P81-
P89~
Fisk, A. D. and Rogers, W. A. (1991). Toward an understanding of age-related visual search
effects, dournal of Experimental Psychology: General, 120, 131-149.
Fisk, A. D., Fisher, D. L. and Rogers, W. A. (1992). General slowing alone cannot explain
age-related search effects: A reply to Cerella (1991). Journal of Experimental
Psychology: General, 121, 73-78.
Fisk, A. D., Cooper, B. P., Hertzog, C. and Anderson-Garlach, M. M. (1994a, submitted).
Age-related retention of skilled memory search: Examination of associative learning,
interference, and task-specific skills. Journal of Gerontology: Psychological Sciences.
Fisk, A. D., Hertzog, C., Lee, M. D., Rogers, W. A. and Anderson-Gadach, M. M. (1994b).
Long-term retention of skilled visual search: Do young adults retain more than old adults?
Psychology and Aging, 9, 206-215.
Ford, J. M., Roth, W. T., Mobs, l~ C., Hopkins, W. F. and Kopell, B. S. (1979). Event-
related potentials recorded from young and old adults during a memory retrieval task.
Electroencephalography and Clinical Neurophysiology, 47, 450-459.
Giambra, L. M. (1993). Sustained attention in older adults: performance and processes (pp.
259- 272). In J. Cerella, J. Rybash, W. Hoyer and M. L. Commons (Eds.), Adult
information processing: Limits on loss. San Diego: Academic Press.
Goldstein, W. M. and Fisher, D. L. (1992). Stochastic networks as models of cognition:
Deriving predictions for resource constrained mental processing. Journal of Mathematical
Psychology, 36, 129-145.
Goldstein, W. M. and Fisher, D. L. (1991). Stochastic networks as models of cognition:
Derivation of response time distributions using the Order-of-Processing method. Journal
of Mathematical Psychology, 35(2), 214-241.
Hale, S., Lima, S. D. and Myerson, J. (1991). Global cognitive slowing in the nonlexical
domain: An experimental validation. Psychology andAging, 6, 512-521.
Hale, S., Myerson, J. and Wagstafl~ D. (1987). General slowing of nonverbal information
processing: Evidence for the power law. Journal of Gerontology, 42, 131-136.
Hertzog, C. (1989). Influences of cognitive slowing on age differences in intelligence.
Developmental Psychology, 5, 636-651.
Howard, D. V. and Wiggs, C. L. (1993). Aging and learning: Insights l~om implicit and
explicit tests (pp. 512- 528). In J. Cerella, J. Rybash, W. Hoyer and M. L. Commons
(Eds.), Adult information processing: Limits on loss. San Diego: Academic Press.
Howard, D. V., Shaw, 1L J. and Heisey, J. G. (1986). Aging and the time course of semantic
activation. Journal of Gerontology, 41, 195-203.
Hunt, E. (1978). Mechanisms of verbal ability. Psychological Review, 85, 109-130.
Johnson, S. H. and Rybash, J. M. (1993). A cognitive neuroscience perspective on age-related
slowing: Developmental changes in the functional architecture (pp. 143-173). In J. Cerella,
J. Rybash, W. Hoyer, and M. L. Commons (Eds.), Adults information processing: Limits
on loss. San Diego: Academic Press.
Kliegl, P~, Mayr, U. and Krampe, IZ T. (1994). Time-accuracy functions for determining
process and person differences: An application to cognitive aging. Cognitive Psychology,
26, 134-164.
Laver, G. D. and Burke, D. M. (1993). Why do semantic priming effects increase in old age?
A recta-analysis. Psychology and Aging, 8, 34-43.
Lima, S. D., Hale, S. and Myerson, J. (1991). How general is general slowing? Evidence
from the lexical domain. Psychology and Aging, 6, 416-425.
Logan, G. D. (1988). Toward an instance theory of automatization. Psychological Review,
95, 492-527.
Luce, It D. (1986). Response times: Their role in inferring elementary mental organization.
New York: Oxford University Press.
Madden, D. J. (1989). Visual word identification and age-related slowing. Cognitive
Development, 4, 1-29.
Madden, D. J. and Plude, D. J. (1993). Selective preservation of selective attention (pp. 273 -
302). In J. Cerella, J. Rybash, W. Hoyer and M. L. Commons (Eds.), Adult information
processing." Limits on loss. San Diego: Academic Press.
Mayr, U. and Kleigl, 1L (1993). Sequential and coordinative complexity: Age-based
processing limitations in figural transformations. Journal of Experimental Psychology:
Learning, Memory and Cognition, 19, 1297-1320.
McClelland, J. L. an Rumelhart, D. E. (1981). An interactive activation model of context
effect in letter perception: Part 1. Psychological Review, 88, 375-407.
McClelland, J. L. and Rumelhart, D. E. (1986). Explorations in parallel distributed
processing: A handbook of models, programs and exercises. Cambridge: MIT Press.
Meyer, D. E., Abrams, 1L A., Kornblum, S., Wright, C. E., and Smith, J. E. K. (1988).
Optimality in human motor performance: Ideal control of rapid aimed movements.
Psychological Review, 95, 340-375.
Meyer, D. E., Smith, J. E. K., Kornblum, S., Abrams, 1L A. and Wright, C. E. (1990). Speed-
accuracy tradeoffs in rapid aimed movements: Toward a theory of rapid voluntary action.
In M. Jeannerod (Ed.), Attention and performance XIV. Hillsdale, NJ: Erlbaum.
Myerson, J., Wagstaff~ D. and Hale, S. (1994). Brinley plots, explained variance, and the
analysis of age differences in response latencies. Journal of Gerontology: Psychological
Sciences, 49, P72-P80.
Myerson, J., Hale, S., Wagstaff, D., Pooh, L. and Smith, G. (1990). The information loss
model: A mathematical theory of age-related cognitive slowing. Psychological Review, 97,
475-487.
Newell, A. and Simon, H. A. (1972). Human problem solving. Englewood Cliffs, NJ:
Prentice-Hall.
Pashler, H. and Johnston, J. C. (1989). Chronometric evidence for central postponement in
temporally overlapping tasks. Quarterly Journal of Experimental Psychology, 41A, 19-45.
Salthouse, T. A. (1988). Resource-reduction interpretations of cognitive aging.
Developmental Review, 8, 238-272.
Salthouse, T. A. (1991). Theoretical perspectives on cognitive aging. Hillsdale, NJ:
Lawrence Erlbaunl
Salthouse, T. A. and Somberg, B. L. (1982). Isolating the age deficit in speeded performance.
Journal of Gerontology, 37, 349-357.
Schaie, K~ W. (1989). Perceptual speed in adulthood: Cross-sectional and longitudinal studies.

Schneider, W. and Shill'in, 1L (1977). Controlled and automatic human information
processing: I. Detection, search and attention. Psychological Review, 84, 1-66.
Schweickert, 1L (1978). A critical path generalization of the additive factor method: Analysis
of a Stroop task. Journal of Mathematical Psychology, 18, 105-139.
Schweickert, 1L and Fisher, D. L. (1987). Stochastic network models. In G. Salvendy (Ed.),
Handbook of human factors. New York: Wiley.
Schweickert, 1L and Townsend, J. T. (1989). A trichotomy: Interactions of factors prolonging
sequential and concurrent mental processes in stochastic discrete mental (PERT) networks.
Journal of Mathematical Psychology, 33, 328-347.
Schweickert, 1L, Fisher, D. L. and Goldstein, W. M. (1994). General latent network theory:
Structural and quantitative analysis of networks of cognitive processes. Revision
requested, Psychological Review.
Staplin, L. and Fisk, A. D. (1991). A cognitive engineering approach to improving signalized
left turn intersections. Human Factors, 33, 559-571.
Sternberg, S. (1966). High-speed scanning in human memory. Science, 153, 652-654.
Sternberg, S. (1969). The discovery of processing stages: Extension of Donders' method. In
W. G. Koster (Ed.), Attention and Performance//(pp. 276-315). Amsterdam: North-
Holland.
Thomas, J. C., Fozard, J. L. and Waugh, N. C. (1977). Age-related differences in naming
latency. American Journal of Psychology, 90, 499-509.
Townsend, J. T. and Ashby, F. G. (1983). Stochastic modeling of elementary psychological
processes. Cambridge: cambridge University Press.
Townsend, J. T. and Schweickert, 1L (1989). Toward a trichotomy method of reaction times:
Laying the foundation of stochastic mental networks. Journal of Mathematical
Walker, N., Meyer, D. E., and Smelter, J. B. (1993). Spatial and temporal characteristics of
rapid cursor-positioning movements with electromechanical mice in human-computer
interaction. Human Factors, 35, 431-458.
Walker, N., Philbin, D. and Fisk, A. D. (1994, April). Optimization and movement control
among older adults. Presented at the Fifth Cognitive Aging Conference, Atlanta.
30 Ph. Allen and Th.R. Bashore (Editors)
Visual w o r d encoding and the effect of adult age and w o r d frequency*
Philip A. Allen a, David J. Madden b, and Steve Slane a
aCleveland State University
bCenter for the Study of Aging and Human Development, Duke University Medical Center
The visual processing of print is a fundamental human skill. Indeed, reading represents
one of the primary forms of human communication. The first step in the process of reading--
word encoding--is the topic of the present paper. We are particularly interested in the effect of
increased adult age on visual word encoding. By word encoding, we mean the transduction of
fight information from a word stimulus into a code that is then used for lexical access. This
specialized form of pattern perception involves both the formation of an input code and lexical
access. In order to address this issue of age differences in visual word encoding, we will first
review the literature on basic word encoding (both the transduction and the lexical access
components). After developing a general framework with which to conceptualize the
processes of word encoding, we will then review theories of aging with regard to visual word
recognition. Finally, the aging literature on visual word recognition and word frequency will
be reviewed and integrated with general theories of word recognition and cognitive aging.
There are two primary goals for this paper. First, we wanted to provide a relatively in-
depth review of the basic literature in visual word encoding and lexical access so as to
familiarize aging researchers with this vast and daunting body of work. Secondly, after
describing the component processes invOlved in visual word recognition, we wanted to
determine if age affected these processes equivalently across task complexity, or whether
certain processes were affected more by increased adult age than were others. This is
equivalent to raising the issue of whether a single (presumably biological) process can account
for age differences in information processing (e.g., Cerella, 1990, 1991; Myerson, Hale,
Wagestaff~ Poon,& Smith, 1990; Salthouse, 1991), or whether local cognitive processes (i.e.,
different stages of processing) modulate age differences due to general biological processes
(e.g., Allen, Madden, Weber, & Groth, 1993; Allen, Madden, Cerella, Jerge, & Betts, 1994;
Amrhein & Theios, 1993; Balota & Ferraro, 1993; Bashore, Osman, & Hefley, 1989).
1. PART I: BASIC THEORIES OF WORD ENCODING
Word recognition research in psychology has a long history (e.g., Cattell, 1886;
Pillsbury, 1897). Nevertheless, over the past century, researchers have failed to resolve a
fimdamental issue in word recognition. Namely, what is the basic unit of analysis in the pattern
perception process in word recognition? Cattell (1886) found that subjects could identify
component letters presented in a word context more readily than letters presented in a
*This research was supported in part by NIH grant AG09282to the first author. Email: p.allen @ csuohio.edu
Visual word encoding and the effect of adult age and word frequency 31
nonword context when using tachistoscopic presentation (i.e., a word superiority effect). This
finding led researchers to propose that the word encoding process was analytic in nature. That
is, these researchers believed that words were formed from their component letters (e.g.,
Adams, 1979, McCldland & Rumelhart, 1981). We will term this view of word encoding
"analytic." However, Pillsbury (1897) found for a reaction time (RT) task that words relative
to nonwords tended to conceal their component letters (i.e., a word inferiority effect). These
results led Pillsbury to propose that individual words were encoded as a single unit
independent of letters. We will term this view of word encoding "holistic." (Ironically, Cattdl
personally believed that the basic unit of analysis was the "total word picture," see
Woodworth, 1938, even though more recent theorists interpreted his data as being supportive
of an analytic basic unit of analysis, e.g., Krueger, 1989.)
For the last century, experimental psychologists have been debating whether initial
visual word encoding is analytic (letter-level encoding, e.g., Adams, 1979; Humphreys, Evett,
& Quinlan, 1990; McClelland & gumdhart, 1981; Paap, Newsome, & Noel, 1984) or holistic
(word-levd encoding, e.g., Johnson, 1975; Masson, 1986; Wheeler, 1970). Furthermore, there
is evidence that under certain conditions individuals encode words as component syllables
(e.g., Hansen & godgers, 1968; Lima & Pollatsek, 1983) or morphemes (e.g., Forster, 1976;
Taft, 1979). Clearly, however, it is not possible for each of the three or four different
proposed encoding schemes to be the single "basic unit of analysis" in visual word recognition.
In order to resolve this dilemma, some word recognition theorists have proposed hybrid
models (e.g., Allen & Madden, 1990; Allen & Emerson, 1991; Allen, Wallace & Weber, in
press; Besner & Johnston, 1989; Carr & Pollatsek, 1985; Coltheart, Curtis, Atkins, & Hailer,
1993; Healy, Oliver, & McNamara, 1987; Healy, Conboy, & Drewnowski, 1987). These
hybrid models can predict both a word superiority effect and a word inferiority effect under the
appropriate task conditions (Allen & Emerson, 1991; Allen et al., in press; Besner & Johnston,
1989). However, the controversy concerning holistic versus analytic encoding is still not
completely resolved, because hybrid models still tend to be either holistically (e.g., Allen et al.,
in press) or analytically (e.g., Besaer & Johnston, 1989) biased.
Hybrid models of visual word recognition predict that word-levd (i.e., the word as a
single unit), syllable-level (i.e., the word as component syllables), and letter-level (i.e., the
word as component letters) representations of words are formed separately, but in parallel.
These different levels of representation are assumed to be involved in a processing horse race
to the central executive. Depending on the task (e.g., letter identification, lexical decision, or
naming), a given level of representation may be optimal. For example, when subjects are
required to identify letters within words or nonwords (a letter identification task, Allen &
Madden, 1990), the letter-level code is optimal, because word-levd and syllable-levd codes
tend to conceal component letters (Allen & Emerson, 1991). Alternatively, the word-levd
channel tends to be optimal for a lexical decision task (a task in which subjects are required to
determine whether a letter string forms a real word, or not--see Allen et al., in press), and the
syllable-level channel tends to be optimal for a naming task (a task in which subjects are
required to pronounce letter strings, see Allen, Madden, Cerella, Jerge, & Betts, in press).
It should be noted, though, that word-levd and letter-levd input codes are traditionally
thought to be orthographic in nature (e.g., Allen & Madden, 1990; Healy et al., 1987), whereas
the syllable-level input code is frequently assumed to be phonological in nature (e.g., Coltheart
et al., 1993). Thus, in many hybrid models of visual word recognition, there is a phonological
channel in which codes are formed using grapheme-to-phoneme correspondence rules (GPC
32 Ph.A. Allen et al.
rules, e.g., Coltheart, Davelaar, Jonasson, & Besaer, 1977) and there is at least one
orthographic channel (letter-level and/or word-level). An extended version of this "dua! route"
model is the hybrid model of Allen et al. (in press; Allen et al., 1994). This model contains
both orthographic and phonological pathways. Furthermore, both pathways receive input from
up to three different input channels (or modules, see Fodor, 1983): word-level, syllable-level,
and letter-level (refer to Figure 1). A more detailed discussion of these "dual-route" models
will be presented in the upcoming section on lexical access.
Word-Level ~ Word
Identification
Orthographic Letter-Level ~ Superposition 1
Pathway
(Addressed)
Syllable-Level ~
VisualInput
Syllable-Level ~ Pronunciation
Phonological
Pathway
(Assembled)
Letter-Level ~ Pronunciation t
Figure1
All of the models of visual word recognition discussed so far are nile-based, "non-
distributed" models. That is, they are really information processing variants of a "Turing
machine" (Turing, 1936). A Turing machine is a symbol-manipulating system (Forster, 1989).
The goal of a Turing machine is to emulate human information processing with a system (e.g.,
algorithms) consisting of a relatively small number of properties and capabilities. One
important aspect of a Turing machine is that the state of the machine is known at any given
point in time.
However, another fundamentally different form of information processing architecture
that is becoming increasing more prevalent today is the connectionist, or parallel distributed
processing (PDP) framework (e.g., Rumelhart & McClelland, 1986; McClelland & Rumelhart,
1986). While Turing machine-based architectures are based upon symbolic, rule-based
representations, PDP architectures are based upon sub-symbolic models in which information
is represented as a pattern of activation across a set of processing "nodes." These nodes can
be conceptualized as neurons, thus, PDP models really contain theoretical entities (i.e., nodes)
that resemble the "cell assemblies" referred to by Hebb (1949). Hence, although it is still a
controversial issue as to whether PDP architectures will perform visual word recognition as
well as rule-based models (e.g., Besner, Twilley, McCann, Seergobin, 1990; Joordens &
Besner, 1994), one positive aspect of PDP models is that they do appear to mimic biological
brain processes more closely than do rule-based models. Because there are hidden layers in
PDP models, though, one cannot be sure of the state of the machine at any given point in time.
WORDS
SYLLABLES
LETTERS
FEATURES
IMAGE
Figure 2
The interactive activation model (IAM) of McClelland and Rumelhart (1981) and the
developmental model of Seidenberg and McClelland (1989) are two examples of PDP models
of visual word recognition (although the IAM uses local lexical representation, so this model is
not a truly "distributed" PDP model). Ironically, although these models include the term
"parallel" in their names, neither of the models use parallel encoding schemes. That is, the
IAM always encodes words as component letters (McClelland & Rumelhart, 1981), and the
developmental model always encodes words as "Wickelgraphs" (a Wickelgraph is a three-letter
sequence, or triplet, Seidenberg & McClelland, 1989). Thus, these models do not use multiple
levels of encoding initially, and this design characteristic can be problematic (see Allen et al., in
press). However, there is no intrinsic reason that a PDP model cannot encode words using
multiple input channels.
The PDP models are parallel in the sense that letter-level activation feeds forward to
the word level before the letter level has completed processing (i.e., a cascaded system, see
Figure 2). The feedback between the word and letter levels is the interactive, or connectionist,
aspect of the models. This allows the word level to affect processing at the lower tier of the
letter level. For example, this sort of model predicts a word superiority effect, because the
word level would facilitate processing at the letter level, however, nonwords (especially
random letter strings) would not provide this top-down facilitation of the letter level. It is
unclear, though, how the aforementioned PDP models could predict a word inferiority effect.
One seemingly simple solution would be to develop a PDP model in which there were separate
letter-level and word-level (and, perhaps, syllable-level) input channels that interacted
horizontally, but such a model has yet to be formalized.
1.1. The Nature of the Functional Stimulus
We have noted so far that even though more sophisticated hybrid nile-based modds
and interactive PDP models have been formulated to explain visual word recognition, it is still
unclear how individuals actually visually encode words during reading. Allen et al. (in press)
have argued that from a systems design standpoint, holistic encoding is inherently more
efficient than analytic encoding because holistic encoding requires less processing capacity.
That is, if one encodes a six-letter word holistically, then one represents the word as a single
unit (or one total word picture). However, if one encodes a six-letter word analytically, then
one represents a word as six separate units (or letter pictures), and this would be equivalent to
encoding six objects instead of one. Furthermore, the letter units are smaller, and therefore,
they require a more detailed "grain" of processing resolution than required by a single word-
level unit. Thus, it would seem to be more efficient to encode words holistically (see Allen et
al., in press).
Holistic models have frequently been assumed to use template matching as a pattern
recognition mechanism (a replica of the actual visual stimulus is stored in memory). That is,
words act like "pictures" that directly access the mental lexicon in order to be reco~ized or
identified. The inherent problem with this approach, though, is that there is considerable
stimulus variability in patterns (in this case, words) that are reco~aized as being identical
(Neisser, 1967). For example, we recognize "word," "WORD," and "wOrD" as being the
same word. To overcome this problem, some researchers have proposed that individuals use a
small subset of visual features to recognize visual objects (e.g., Gibson, 1965; SelfiJdge, 1959).
However, such models (e.g., the "Pandemonium" model of Selfridge, 1959) have limited
generalizability in a manner not unlike template matching models. That is, the Pandemonium
model can recognize component letters, but the model cannot recognize objects in a painting.
Furthermore, in is unclear how models of this type would predict empirically obtained results
for a mixed-case disadvantage for visual word recognition (when it takes longer to reco~ize a
word presented in mixed case than in consistent case on a lexical decision or naming task).
The Pandemonium model cannot parsimoniously predict a mixed-case disadvantage that varies
as a function of exposure duration and lexicality (i.e., word vs. nonword) because it has a
single analytic input channel (Allen et al., in press).
A more general method of recognizing a wide variety of objects is termed
"identification by components" (Biederman, 1987). This approach uses a primitive set of
visual, geometric components to account for all possible objects (Biederman, 1987; Marr,
1982). These simple visual geometric properties tend to remain constant in both static and
dynamic perceptual processing. The geometric components are all variations of a cylinder, and
Biederman (1987) estimated that selection from a set of no more than 36 different geons (using
three-dimensional processing) could be used to form all known objects. The identification by
components approach, though, would still have difficulty in predicting a mixed-case
disadvantage unless geons were used as building blocks for multiple input channels (e.g., Allen
& Emerson, 1991).
It is quite likely that PDP theorists used the theoretical construct of visual features to
form letters, because they felt that this avoided the potential problems inherent in some
template matching models--which are that they require an unlimited supply of templates, and
they have extremely limited generalizability. Note that models of reading that use letters to
form words need only to be able to recognize the 26 different letters of the alphabet in order to
recognize every possible English word. Unfortunately, as noted earlier, constraining models to
recognize words using only component letters slows down the potential encoding speed of
words, and also makes such models (e.g., the IAM ofMcClelland & Rumelhart, 1981) unable
to account for data in a straightforward manner in which a word context inhibits the
recognition of component letters relative to nonwords (i.e., a word inferiority effect, e.g., Allen
& Emerson, 1991; Healy et al., 1987).
An alternative approach to visual pattern recognition other than identification-by-
components is termed "model-based identification" (e.g., Brooks, 1981). Applying this general
information processing architecture to visual word recognition, word stimuli are encoded
simultaneously as both words (holistically) and letters (analytically) using an encoding process
termed "normalization" (e.g., Allen, Wallace, & Weber, in press). By doing so, the model-
based identification can overcome much of the bottleneck problems of models that form words
solely from component letters. By normalization we mean a "transformation which would give
the same output for every member of certain well-defined categories" (Neisser, 1967, p. 63).
By applying normalization to visual word recognition, we need at least two input channels--
word level and letter level--and we need to assume that input channels encode letters and
words using the spatial frequency pattern of the stimuli (although the latter assumption is not
required, the model could theoretically encode stimuli as visual features rather than spatial
frequency patterns). The inpm channels then normalize slight variations in the spatial
frequency pattern of word-consistent font type or case type so that the same code is formed for
"word" and "WORD." This can be accomplished by using a normalization algorithm such as
Fourier synthesis (Allen & Emerson, 1991; Allen, Wallace, & Weber, in press). Although
Fourier synthesis uses a linear algorithm and the human visual system is clearly non-linear at
times, it appears that Fourier synthesis is still the best norma~ation option available to visual
recognition researchers at the present time (Graham, 1981).
It should be noted that using a model-based approach (spatial frequency encoding)
rather than an identification-by-components approach (encoding by visual features) to visual
word encoding is a marked departure from the models of pattern perception that are typically
used in psychology. This is probably because of the influence of Mart (e.g., 1982) who
persuasively argued for the visual feature detector approach. However, more recent evidence
suggests that the brain initially uses spatial frequency filtering to encode visual information (see
Van Essen, Anderson, & Felleman, 1992). It now appears that what researchers think of as
feature detection occurs much later in neural processing than does spatial filtering (Van Essen
et al., 1992).
In both models of visual encoding (Mart, 1982, and Van Essen et al., 1992), it is
important to keep in mind that this process involves multiple stages. Initially, visual stimuli are
encoded as a set of primitive features (Mart, 1982) or broad spatial frequency tuning
(Derrington & Lennie, 1984; DeValois & DeValois, 1987; Van Essen et al., 1992). From a
psychological perspective, we might term this first stage of encoding the detection stage. After
detection, the second stage of visual encoding involves interpretation of the "blob-level"
representation (either by edge detection, Mart, 1982, or by spatial frequency filtering,
DeValois & DeValois, 1987). For a discussion of the neural pathways involved in visual word
recognition, see Cart (1992), Cart and Posner (1992), or Allen, Wallace, and Weber (in press).
1.2. A Hybrid Horse Race Model
A model based upon the spatial filtering design principles mentioned in the preceding
paragraphs has been proposed by Allen and Emerson (1991) and Allen et al. (in press). In the
most basic form, the model assumes the existence of two orthographic input channels used for
visual word recognition/identification. Specifically, this hybrid model uses a fast word-level
input channel and a typically slower letter-level channel to recognize and identify words.
These input channels (or modules, Fodor, 1983) are involved in a stochastic (i.e., probabilistic)
horse race to the central processor. The word-level channel encodes words using the spatial
frequency pattern of entire words (holistic encoding) and the letter-level channel encodes
words using the spatial l~equency pattern of component letters (analytic channel). In this sort
of encoding architecture, the words "word" and "WORD" are actually normalized to be
recognized as the concept "word." However, Fourier synthesis would not be successful in
forming a holistic code of "word" from "wOrD" because mixing case results in a pattern that
cannot be normalized by Fourier synthesis (Allen et al., in press). Thus, words presented in
mixed case must be processed using the typically slower letter-level input channel. The
empirical finding of a mixed-case disadvantage for words on a lexical decision task (e.g., Allen,
Madden, Weber, & Groth, 1993; Allen et al., in press) or on a naming task (Allen, Madden,
Cerella, Jerge, & Betts, 1994) is consistent with the aforementioned hypothesis that consistent-
case words tend to be encoded first by the word-level channel (assuming that they are
familiar), but that mixed-case words t a d to be encoded by the letter-level channel.
The hybrid model (Allen et al., in press) further assumes that the word-level input
channel accesses the holistic-semantic lexicon. (The holistic-semantic lexicon is actually a sub-
module of the word-level input channel.) In this manner, the model predicts word t~equency
advantages for words processed using the word-level channel. However, the hybrid model
recognizes unfamiliar words by using the typically slower letter-level channel, because the
spatial frequency pattern of a new word would be definitionally unfamiliar (thereby preventing
the word-level channel from producing a code). Since all English words can be formed from
the component spatial frequency patterns of the 26 letters of the English alphabet (albeit at a
slower rate than encoding words using holistic methods), the model can create new entries in
the mental lexicon through the use of "superposition." That is, the central processor
"superposes" the letter-level code into a pseudo-word-level code. The model assumes that
once an individual has encountered a new word a sufficient number of times, the letter-level
representation of that word is "superposed" into a word-level template so that this word can
now be directly recognized holistically (Allen & Emerson, 1991). It should be noted, though,
that the superposition process is rather time consuming and requires a considerable allocation
of processing resources. Therefore, lexical access using a superposition process is only used
when such access cannot be accomplished via the word-level input channel (Allen & Emerson,
1991). In this manner, the hybrid model can account for how children and adults learn to read
new words holistically.
After the word-level and letter-level (and probably syllable-level) input channels output
their respective codes, this information is sent to the central processor (Allen & Madden,
1990). The central processor then selects (i.e., switches attention to) the first available code in
an attempt to solve the task at hand (and the other codes that were output more slowly are
stored temporarily in an output buffer). For a letter identification task, this allows the hybrid
model to predict a word frequency disadvantage (when higher-frequency words conceal their
component letters relative to lower-frequency words, Allen & Madden, 1990; Healy et al.,
1987) and a word inferiority effect (e.g., Allen & Emerson, 1991). That is, for a letter
identification task, subjects decide whether a target letter matches the initial letter of a
subsequently presented word or nonword. The hybrid model hypothesizes that the word-level
channel wins the race to the central processor for very-high and medium-high frequency
words, but that the letter level channel wins the race to the central processor for lower-
frequency words (Allen & Madden, 1990). Furthermore, very-high-frequency words are
output so rapidly that the central processor has time to determine that this code is the wrong
one for the task at hand and to switch attention to the letter-level input channel before that
code has been output (i.e., letter identification requires letter-level information that is not
readily available from a word-level code). However, for medium-high-frequency words, the
word-level channel just barely outputs its code before the letter-level channel. Thus, by the
time the central processor determines that it cannot use the word-level code to solve the task
at hand, the letter-level code has been output into its buffer, and this results in longer access
time. Because the letter-level channel wins the race to the central processor for lower-
frequency words, there is no delay in central-processor access to the letter-level code for these
data. Therefore, the model predicts longer letter identification RT for medium-high frequency
words than for high-frequency or lower-frequency words (Allen & Emerson, 1991; Allen &
Madden, 1989, 1990; Johnson et al., 1989).
For lexical decision and naming tasks, the hybrid model predicts a word frequency
advantage and a mixed-case disadvantage (Allen, Wallace, & Weber, in press). That is, the
model predicts that the word-level code is needed to solve the task at hand (or a pseudo-word-
level code for words presented in mixed case), and that the lexical access component of word-
level processing results in a word frequency advantage. However, it should be noted that there
are different types of word-level codes depending upon the processing pathway. For example,
the hybrid model would typically use the rule-based phonological pathway to conduct a naming
task, whereas the model would use the orthographic pathway to conduct a lexical decision task
(Allen et al., in press). There is considerable evidence for this dual-route architecture (e.g.,
Carr & Pollatsek, 1985; Coltheart et al., 1993). Finally, the dual encoding architecture within
a given pathway (e.g., word-level and letter-level) of the hybrid model allows it to account for
the differential pattern of mixed-case disadvantage effects found across different stimulus
exposure durations for words and nonwords on a lexical decision task (Allen et al., in press).
It is not clear how models that always encode words from component letters (e.g., McClelland
& Rumelhart, 1981; Paap, Newsome, McDonald, & Schvaneveldt, 1982; Paap, Newsome, &
Noel, 1984) can account for this finding.
1.3. R u l e - B a s e d versus P D P models
Now we are ready to re-examine the issue of whether rule-based or sub-symbolic, PDP
models are preferable processing architectures for the optimal model of visual word
recognition. It was noted earlier that the IAM (McClelland & Rumelhart, 1981) was a
cascaded, PDP model that always encoded words initially as component letters, but that
allowed the word-level and the letter-level tiers to interact via excitatory and/or inhibitory
feedback (vertical interaction). Alternatively, we also described a rule-based, hybrid horse race
model (Allen et al., in press) that encoded words initially as both words and component letters.
Research by Allen et al. (in press) indicated that the dual encoding mechanism of the rule-based
model was better able to account for certain aspects of the mixed-case disadvantage and the
word inferiority effect. However, there is no good design reason for why a PDP model cannot
encode words holistieally as well as analytically (this would result in horizontal interaction
between the word and letter levels). Thus, in terms of implementing a model of visual word
encoding, it appears that both rule-based models and sub-symbolic PDP models can account
for a good deal of available data--as long as the sub-symbolic model uses at least two different
types of initial encoding (e.g., word level and letter level, see Allen et al., in press), and as long
as there is some sort of local representation of words (i.e., a lexicon, see Coltheart et al.,
1993).
In summary, the strength of rule-based models is that we know the state of the machine
at all times (this increases model precision) (Turing, 1936). This class of model can emulate all
known logical and mathematical operations (Minsky, 1967), and human cognition appears to
be rule based (e.g., Coltheart et al., 1993; Pinker & Prince, 1988). Alternatively, the strengths
of sub-symbolic PDP models are that they more closely resemble neural processing than do
rule-based models (Rumelhart & McClelland, 1986), and that one can observe how new
patterns are learned (Seidenberg & McClelland, 1989).
2. PART H: BASIC THEORIES OF LEXICAL ACCESS
Up to this point, we have been primarily concerned with examining how word and
letter stimuli are transduced into a form that the human visual information processing system
can use. However, this initial transduction process is only the first step of the pattern
perception of words. That is, to be of functional significance, the transduced codes need to be
recognized and/or identified. By recognition, it is meant that the transduced code representing
a word, syllable, or letter is familiar (Besner & Johnston, 1989; Besaer & McCann, 1987).
Alternatively, identification refers to the process of determining the name of the transduced
word, syllable, or letter (Besner & Johnston, 1989). With regard to words, a further
assumption of many models of visual word recognition is that identification not only allows
individuals to name words, but also to know their meaning (see Cart & Pollatsek, 1985). This
is especially the case for what are termed "lexical instance models" of visual word recognition,
because these models assume that lexical access occurs. Lexical access using the orthographic
route(s) (word-leveL syllable-level, or letter-level) is the process of addressing the mental
lexicon, or mental dictionary. That is, individuals code the transduced word stimulus
(regardless of whether it was initially encoded using word-leveL syllable-level, or letter-level
units of analysis) into a prototype representation, and then use this representation to access the
name and meaning of that word in the mental lexicon. Alternatively, lexical access using the
phonological route is the process of using grapheme-to-phoneme correspondence (GPC) rules
(or more holistic encoding schemes, see Allen et al., 1994) to generate, or assemble, a
phonological representation of a word that is then used to access the lexicon.
We will discuss three classes of lexical access models: rule-based, lexical-instance

models; rule-based, dual-route models; and sub-symbolic, distributed models. The first two
classes of models use rules to reco~mfize and identify words and have "local" representations of
words in a mental lexicon, whereas the latter class of models does not use rules to recognize
and identify words and does not possess a real lexicon (i.e., words are represented in a
distributed manner).
2.1. Rule-Based Lexical-lnstance Models
The lexical-instance models all assume the existence of a memory store termed a
lexicon, and that individual words, or prototypes, are represented within this store. This class
of models assumes that a visually encoded word addresses an entry (or entries) in the mental
lexicon with the outcome being either recognition or identification of a given word. Contrary
to the general rules used in an assembled, phonological system, lexical instance models assume
that individuals know many instances of words from experience, and that these experiences can
be described by rules (Cart & Pollatsek, 1985). There are three sub-types of lexical-instance
models: logogen models, lexical search models, and verification models (Cart & Pollatsek,
1985).
Logogen models (e.g., Morton, 1969) assume that an encoded stimulus addresses all
items stored in the lexicon in parallel. The words stored in the mental lexicon have different
threshold levels. Specifically, as words become increasingly more familiar, their activation
threshold decreases. This allows logogen models to account for word frequency advantages
fotmd for lexical decision (e.g., Allen, McNeal, & Kvak, 1992; Dobbs, Friedman, & Lloyd,
1985) and naming tasks (Monsell, Doyle, & Haggart, 1989), because the activation threshold
for higher-frequency words is lower than the activation threshold for lower-frequency words.
This sort of lexical architecture can produce the partial activation of multiple lexical entries.
However, to be tenable, this sort of model must typically produce activation of the appropriate
entry in the lexicon.
Probably the most widely cited example of a logogen model of visual word recognition
is the IAM of McClelland and Rumelhart (1981; Rumelhart & McClelland, 1982).
Interestingly, the IAM is typically classified as a PDP model. However, words are individually
represented in the lexicon of the IAM, so representation in this model is local rather than
distributed (i.e., in a truly distributed model, all words are distributed simultaneously as a
pattern of activation across a set of processing nodes, Seidenberg & McClelland, 1989).
Another subtype of lexical-instance models is the lexical search model (e.g., Forster,
1976; TaR, 1979). This model assumes that the lexicon is organized as a function of word
frequency so that higher-frequency words tend to have higher priority access than do lower-
frequency words. The model uses an access code made up of either morphemes (Tat~, 1979)
or syllables (Lima & Pollatsek, 1983) to activate a candidate set of word representations stored
in the lexicon. The recoded input stimulus is then compared serially to the candidate set of
words that contains similar or equivalent syllables or morphemes, and the model then selects
the best-fitting candidate to recognize or identify. The design goal of forming a candidate set
is to limit the size of the search set, because the search process is serial in nature (Carr &
Pollatsek, 1985).
A slight variation of the lexical search model is the verification class of models.
According to verification models (e.g., Becker, 1976; Paap et al., 1982), there is a top-down
selection process that verifies that the word initially picked during lexical access is, in fact, the
correct word. Contrary to the lexical search model in which word frequency effects result
directly from lexical access, word frequency effects in verification models result solely from
order of verification effects (e.g., Paap et al., 1982). Indeed, Paap et al. (1982) proposed that
no word frequency effects should occur when verification is prevented. However, it is clear
that word frequency effects continue to occur even in situations in which the activation-
verification model ofPaap et al. (1982) predicts that verification is impossible (see Allen et al.,
1992; Dobbs et al., 1985). Although verification models can account for a wide variety of
phenomena (e.g., Becker, 1976; Paap et al., 1982), it appears that the activation-verification
version of the model (Paap et al., 1982) cannot account for replicated word frequency
experiments that manipulate exposure duration.
2.2. Rule-Based Dual-Route Models
Another class of lexical access models are the dual-route models (e.g., Carr &
Pollatsek, 1985; Coltheart et al., 1977, 1993). Dual route models provide a mechanism that
allows the prommciation (or naming) of both familiar and unfamiliar words (or nonwords),
while still providing a visually accessible route to the lexicon (Carr & Pollatsek, 1985). These
models propose that visual word recognition is accomplished through the use of two
processing pathways. One pathway is orthographic in nature, and the lexicon is addressed
using an orthographic code. This pathway of the dual-route model is equivalent to the lexical-
instance models. The other pathway uses GPC rules (or, perhaps, multiple grapheme-to-
syllable or multiple grapheme-to-whole word correspondence rules, Allen et al., 1994), is
phonological in nature, and accesses the lexicon with a phonological code (Coltheart et al.,
1993). The original versions of the dual-route model did not provide the phonological channel
with the ability to access a lexicon (Carr & Pollatsek, 1985), however, more recent versions of
phonological processing models do assume that this channel involves lexical access (e.g.,
Coltheart et al., 1993; Levelt, Schiefers, Vorberg, Meyer, Pechmann, & Havinga, 1991).
Because of their inclusion of multiple pathways, dual-route models can account for a multitude
of phenomena (Carr & Pollatsek, 1985). However, because of their complexity, these models
are painfiflly difficult to understand, at times. Unfortunately, this complexity aspect
underscores a major reason why after over 100 years of scrutiny we still do not completely
understand visual word recognition. Clearly, there is no simple explanation for the
phenomena.
2.3. Revisiting the Hybrid Horse Race Model
Earlier in the paper, we outlined a hybrid, horse race model of visual word recognition
(and identification) that was holistically biased (e.g., Allen & Emerson, 1991; Allen & Madden,
1990; Allen et al., in press). Much of the initial encoding architecture of the model was
discussed earlier, however, we are now ready to address how the model performs lexical
access. Although the model is architecturally dual-route in nature (see Allen et al., in press,
1994), the orthographic pathway of the model has been tested much more extensively than the
phonological route (Allen et al., 1994). Thus, we need to address the lexical-instance
characteristics of this model. This is no simple task, because the hybrid model includes both
logogen and lexical search characteristics.
The lexical access characteristics of the hybrid model resemble a logogen model
because lexical entries are assumed to have progressively lower activation thresholds as the
frequency of the word entry increases (Allen, McNeal, & Kvak, 1992). However, the hybrid
model assumes that the coded spatial frequency pattern of the word stimulus (the holistic code)
is used as an access code. This considerably lessens the number lexical entries that need to be
partially activated per encoded word. This aspect of the hybrid model resembles lexical search
models because an access code is used to limit the number of comparisons that need to be
made between the encoded representation of the spatial frequency pattern of a word and
similar lexical entries. Note that words in this model are encoded in the lexicon as "pictures."
Once a picture, or icon, is accessed, then the name and/or meaning of the word can be
determined (actually, meaning is probably represented in a sub-module of the orthographic
lexicon). However, the lexical access process is parallel in the hybrid model, and word
frequency is encoded by assuming that higher frequency words are actually closer in
psychological space to the lexical accessing retrieval "pointer" (and this assumption is
consistent with Forster's, 1976, 1989, lexical search model). When multiple entries, or icons,
in the lexicon have been partially activated because they are visually similar to the word
stimulus, then this is resolved using normalization procedures until a lexical entry is activated
sufficiently to surpass its threshold. Note that the hybrid model avoids many of the problems
associated with having a lexicon represented as component letters. Instead of representing
words as component letters, words are represented holistically. Such an architecture can
account for why orthographic neighborhood effects (see Stadtlander's chapter in the present
volume) tend to be occur primarily for nonwords (e.g., Coltheart et al., 1977). That is,
nonwords must be superposed in order to be input into the lexicon, and the superposition
process is analytic. Thus, letter-level characteristics such as orthographic neighborhood effects
should affect performance for nonwords (or, perhaps, very low frequency words that are not
sufficiently activated to pass firing threshold). Finally, the hybrid model rejects nonwords
when the input stimulus fails to adequately activate any items in the lexicon (Allen et al., in
press).
2.4. Sub-Symbolic Distributed Models
Our final class of lexical access models is quite different from the first two classes,
because the final class of models uses distributed representation rather than the local
representation used by the two previously mentioned classes of models. As noted earlier, this
sort of connectionist (or distributed) model resembles a neural net. We will limit our
discussion to the developmental model of Seidenberg and McClelland (1989), because this
model has been tested more extensively than any other distributed PDP model of visual word
recognition (e.g., Besner et al., 1990; Fera & Besner, 1992; Joordens & Besner, 1994;
Coltheart et al., 1993). Every time the neural net is activated, the orthographic layer (i.e.,
letter detectors) is activated across its complete set of units. The model encodes letter strings
as Wickelgraph triplets. This pattern of activation from the orthographic layer is then sent (or
"spread") to the next layer--the hidden units. The hidden units are also connected to all the
orthographic layer units in a feedback loop. The activation strength of the connections
between the orthographic and the hidden units is then modified across processing cycles so that
the hidden layer units can learn the input pattern. An orthographic error score is then
computed by comparing the pattern of activation across the orthographic units after feedback
from the hidden layer units to the pattern of activation across the orthographic units when the
letter string was originally presented to the orthographic layer. The smaller the error score, the
better the hidden units have learned the pattern.
To perform a lexical decision task, the developmental model (Seidenberg &
McClelland, 1989) uses orthographic error scores to index the level of familiarity. If the
orthographic error score is lower than some preset criterion, then this letter string is assumed
to be familiar, and a word decision is made. However, if the orthographic error score is
greater than some preset criterion, then this letter string is classified as being unfamiliar, and a
nonword decision is made. This model is essentially a connectionist revision of the decision
model proposed by Balota and Chumbley (1984). Note that this model does not actually
conduct lexical access in order to perform a lexical decision task. Indeed, the model does not
even contain the local representations that define a lexicon.
In order to name words, the developmental model projects the activation across units
from the hidden layer to a phonetic layer (Seidenberg & McClelland, 1989). This allows
individuals to name input letter strings. Phonological error scores are formed by comparing
the pattern of activation across phonetic units as the result of letter string stimulation to the
pattern of activation across phonetic units as the result of direct phonemic specification (when
the correct pronunciation is input into the net) (Seidenberg & McClelland, 1989). The model
then assumes that this pattern of activation across the phonological units is used as input to
form an articulatory-motor program which can then be implemented by the motor system--
resulting in naming.
It is important to note that the developmental model uses the same network to process
words both orthographically and phonologically (Seidenberg & McClelland, 1989). Thus, this
distributed model is at odds with dual-route models, because the dual-route models use
separate orthographic and phonological routes (Coltheart et al., 1993). Although the
parsimony and apparent precision of the distributed, developmental model are impressive (the
developmental model settled on the correct pronunciation for 2,820 out of 2897 words which
were input into the model, Seidenberg & McClelland, 1989), the lack of a local lexicon
presents some serious problems for the model. In general, the distributed model makes more
errors under certain conditions than do humans. For example, the developmental model has
difficulty in recognizing exception words not included in the original set of words used by
Seidenberg and McClelland (see Besner et al., 1990; Coltheart et al., 1993; Fera & Besner,
1992; Joordens & Besner, 1994). However, distributed models with local lexical
representation are not susceptible to this same criticism (e.g., Coltheart et al., 1993). If such
models were to employ more than one encoding route (e.g., word level and letter level, or
syllable level and letter level), then these PDP models might be able to account for human
visual word recognition performance as well as rule-based models (see the chapter in the
present volume by Kellas, Paul, & Vu for an example of the IAM applied to aging). Forster
(1989), though, has noted that distributed models with local lexical representation still have
problems in resolving orthographic neighborhood effects when the lexical entries in the same
neighborhood have different frequencies.
2.5. Post-Lexical Effects of Word Frequency?
One final issue concerning lexical access is that of the locus of word frequency effects.
The finding that subjects respond more rapidly to higher-frequency words than to lower-
frequency words for both lexical decision and naming tasks has provided fundamental support
for idea that individuals possess at least one mental lexicon. The aforementioned rule-based
models all assume that word frequency effects are the result of faster lexical access for higher-
frequency words. However, Balota and Chumbley (1984, 1985) proposed that a substantial
portion of the word frequency effect was due to decision processes that occurred aider lexical
access. Although the work of Balota and Chumbley does suggest that s o m e of the word
frequency effect is not the result of lexical access processes, considerable research suggests
that a large portion of the word frequency effect for lexical decision and naming tasks is the
result of lexical access processes (Allen, McNeal, & Kvak, 1992; Connine, Mullennix,
Shernofl~ & Yelen, 1990; Monsell et al., 1989).
2.6. Empirical Measures of Lexical Access
In the basic visual word recognition literature, lexical access speed has been examined
by manipulating two different factors: word frequency and semantic priming. Word frequency
refers to how common a word is in written American English (e.g., Kucera & Francis, 1967).
Semantic priming refers to the facilitation of a target word by a semantically related prime
word relative to a semantically unrelated prime word (e.g., Neely, 1990). With regard to word
frequency, the assumption is that the mental lexicon is organized so as to allow more speedy
access to higher-frequency words than to lower-frequency words (e.g., Allen, McNeal, &
Kvak, 1992; Carr & Pollatsek, 1985; Dobbs, Friedman, & Lloyd, 1985) (although as noted
earlier, connectionist, or distributed, models implement a lexicon somewhat differently). With
regard to semantic priming, the relation to lexical access speed is somewhat more circuitous.
That is, the activation of the prime word is assumed to spread to other semantically related
words stored in the mental lexicon (i.e., "spreading activation") so that target words that are
semantically related to prime words are already partially activated (relative to target words that
are not semantically related to the prime word) (Collins & Quillian, 1969; Nelson, Schreiber, &
McEvoy, 1992). It is important to remember, though, that semantic priming was designed
primarily to measure how meaning affects lexical access. Lexical access speed is a secondary
concern for semantic priming studies.
2.7. Summary of the Word Encoding Tutorial
In the preceding sections, we have reviewed theories of how words and letters are
coded by the information processing system for later use (i.e., the transduction process). We
came to the conclusion that it is probably necessary to assume that visual word encoding uses
multiple input channels (e.g., Allen & Emerson, 1991; Allen et al., in press; Healy et al., 1987;
Johnson et al., 1989). Also, we reviewed the literature on lexical access and concluded there is
good evidence for the existence of a mental lexicon (Allen, McNeal, & Kvak, 1992; Cart &
Pollatsek, 1985; Forster, 1989; Monsell et al., 1989). It appears that this lexicon has both
logogen (Morton, 1969) and lexical search (Forster, 1976, 1989) characteristics.
We also discussed in some detail the difference between symbolic (role-based) and sub-
symbolic (distributed) architectures of visual word recognition/identification. We noted that
rule-based models are based on a Turing machine and seem to capture the human predilection
to mimic (or actually use) role-based information processing (e.g., Pinker & Prince, 1988).
Alternatively, we mentioned that distributed models are based upon Hebb's (1949) notion of a
"cell assembly" and are particularly effective at illustrating how a system learns a new task
(McClelland & Rumelhart, 1986). However, even though we readily admit that distributed
models have some interesting attributes, there is still reason for concern that they may not
work without adding a local lexicon (e.g., Besner et al., 1990; Fera & Besner, 1992; Joordens
& Besner, 1994; Coltheart et al., 1993) as well as an attentional selection mechanism (Forster,
1989). When these attributes are added to a distributed model, then this model is for all intents
and purposes an embellished rule-based model (Forster, 1989).
Consequently, In the next section, we will use the hybrid model (i.e., rule based) as a
guide to interpreting the aging literature because (1) it has multiple input - channels not
contained by extant distributed models (and these multiple input channels appear to be
necessary to account for visual word recognition data, Allen et al., in press), (2) it has been
applied to all three of tasks discussed below, and (3) the model can account for age differences
in t~equency effects for a letter identification task that distributed models without a selective
attention mechanism (i.e., a central processor) cannot account for.
3. PART IH: THE EFFECT OF AGING ON VISUAL WORD RECOGNITION
Now that we have reviewed the literature on how word stimuli are coded into the
visual system and then how these words are reco~ized and/or identified through lexical access
(or some other analogous process for distributed models), we are ready to address how
increased adult age affects visual word recognition for studies that have manipulated word
t~equency. To begin, we will discuss some models of cognitive aging. Next, the literature on
age differences in visual acuity will be briefly examined. Finally, we will review the aging
literature on letter identification, lexical decision, and naming tasks, and examine whether these
results can be accounted for by the hybrid model applied to aging, and whether these data
require the inclusion of cognitive constructs (localized models) or simply require a behavioral
description bereft of any such cognitive constructs (generalized models).
3.1. General Models of Cognitive Aging
As noted earlier, a major goal of the present paper was to determine whether age
differences in visual word recognition are the result of a single general factor (e.g., Allen,
1990, 1991; Salthouse, 1991), a small number of general factors (e.g., Cerella, 1985, 1990;
Lima, Hale, & Myerson, 1991; Myerson, Ferraro, Hale, & Lima, 1992), or whether some
stages of processing are more affected by aging than others (e.g., Allen, Madden, & Crozier,
1991; Allen, Madden, Weber, & Groth, 1993; Allen, Madden, Cerella, Jerge, & Betts, 1994;
Amrhein, this volume; Bashore et al., 1989; Fisher, Fisk, & Duffy, this volume; Fisk & Rogers,
1990; Fisk & Fisher, 1991; Balota & Ferraro, 1993; Madden et al., 1993). This is a crucial
issue for the psychology of aging, because the class of models used by researchers and
theoreticians will have a major impact upon whether cognitive processes remain relevant
constructs in experimental aging research. Process-based models emphasize how aging has
differential effects across processing stages (e.g., Fisk & Fisher, 1994). This sort of model
assumes that we cannot understand how aging affects information processing without alluding
to component cognitive processes (Allen et al., 1993). However, aging models that emphasize
a single factor (e.g., Allen, 1991, Salthouse, 1991) or just a few general factors (e.g., Cerella,
1985; Myerson et al., 1990) do not emphasize the importance of cognitive constructs. Indeed,
entropy-based models of aging (e.g., Allen, 1990, 1991; Myerson et al., 1990) are really
neurologically based.
3.2. Generalized Models
In order to address this issue of generalized models of aging (single-factor models and
models that include a small number of factors) versus localized modds of aging (models that
emphasize local cognitive processes), we need to develop these two views of aging in more
detail. Generalized models of aging assume that both young and older adults use the same
processing stages, but that older adults take longer to process information at each stage than
do young adults (Cerella, 1985, 1991). In order to test this view of aging, investigators
frequently plot older adults' data across task conditions along the y axis and plot younger
adults' data across the same task conditions along the x axis. Then, the best-fitting slowing
function is determined by using least-squares regression (or some other curve-fitting routine).
This approach to examining age differences is typically termed a Brinley plot (Brinley, 1965).
In Brinley-plot analyses, the single best-fitting "slowing function" frequently accounts for over
90% of the experimental variance. Investigators have found linear, multiplicative, additive, and
non-linear functions that resulted in the best-fitting function for a Brinley plot (e.g., Cerella,
1985, 1991; Myerson et al., 1990). Cerella (1985) proposed that a linear function would best
predict older adults' RT from young adults' analogous RT. This linear function was based
upon a multiplicative component for central processes and an additive component for
peripheral processes.
Advocates of Bfinley plots have emphasized how such regression techniques are
particularly effective at finding commonalties in a data set (e.g., Cerella, 1994; Myerson,
Wagsta~ & Hale, 1994). It should be noted, though, that such regression methods are not
overly sensitive in detecting interactions compared to an analysis of variance (e.g., McClelland
& Judd, 1993; Kliegl, 1994; Perfect, 1994). Thus, one should take particular care in
interpreting Brinley plots as conclusive evidence for a single factor (or small subset of factors)
accounting for age differences in a given task or group of tasks (see Fisher et al, in the present
volume, or Fisk & Fisher, 1994). Also, it is important to remember that the proportion of
variance accounted for is not the only relevant factor. For example, a basic assumption of the
experimental method is that interactions should be interpreted before main effects (e.g.,
Keppel, 1991). Thus, a situation could occur in which an interaction accounted for less
variance than a main effect (and this is typically the case for each separate interaction), but that
in which the interaction qualified the main effect. On the other hand, one should also take
particular care in concluding that localized age differences are present simply on the basis of
interactions present in an analysis of variance, because the presence of an interaction with age,
by itself; does not preclude localized age differences (e.g., Cerella, 1991).
Another method of testing generalized models of cognitive aging is to use partial
correlation (Madden, 1992; Salthouse, 1985) or hierarchical regression (Salthouse, 1991)
methods. Typically, these models assume that age differences in information processing speed
are to a large extent the result of a single processing speed factor (Salthouse, 1985, 1991). In
order to test for this possibility, advocates of this method examine age differences in task
performance on at least two tasks or multiple levels of a single task. For example, Madden
(1992) had subjects participate in both a lexical decision/priming task and the Wechsler Adult
Intelligence Scale-Revised (WAIS-R) digit symbol task. Digit symbol task performance was
used as a general measure of processing speed, lexical decision performance was assumed to
measure lexical access time, and stimulus degradation (placing asterisks between letters of
words on some trials) was assmned to measure word encoding time.
When digit symbol task performance was partialed out in the Madden (1992) study, the
correlation between age and mean RT decreased from .49 to .29. Also, the correlation
between age and stimulus degradation decreased from .57 to .49 when digit symbol task
performance was partialed. This suggested that the effect of age was partially due to
processing speed. However, even after digit symbol performance was partialed out, the
aforementioned correlations still remained reliable. Furthermore, the correlation between age
and stimulus degradation after digit symbol performance had been extracted (r = .49) appeared
to be greater than the correlation between age and overall mean RT after digit symbol
performance had been extracted (r = .29). This finding implied that age differences in encoding
(as measured by stimulus degradation effects) were greater than overall age differences
(encoding, lexical access, response selection, response execution--as measured by mean RT
over all task conditions). This suggests that there are localized (at encoding) age differences in
addition to generalized age differences.
Although the preceding paragraphs have described two methods of testing a
generalized model of aging (Brinley plots and partial correlation), these are methods of analysis
rather than theoretical frameworks. Indeed, without additional theoretical development, the
concept of generalized slowing reduces to a tautology--that older adults slow down relative to
young adults because older adults are slower. Thus, in order for generalized slowing to be a
tenable scientific model of cognitive aging, it is necessary for the model to explain what general
factor (or small subset of factors) causes age-related slowing. The first attempt to do this was
the complexity hypothesis (e.g., Birren, 1965; Cerella, Pooh, & Williams, 1980; Salthouse,
1985). The complexity hypothesis predicts that older adults are slower than young adults at
information processing tasks because of age-related changes in the central nervous system
(Birren, 1965). The complexity model predicts that as task complexity increases, age
differences should become progressively larger.
However, the complexity hypothesis, alone, would appear to be a post-hoc explanation.
That is, this hypothesis predicts that a given task condition is more complex because it takes
longer to complete, and a given task condition takes longer to complete because it is more
complex. What is necessary is an additional factor that can be used to define complexity
independent of post-hoc task performance. Thus, such a framework must provide a construct
that predicts an antecedent cause. The extensive research on cognitive processes provides
aging researchers with a rich source of potential candidates for such a factor that predicts task
complexity. For example, Salthouse (1991) has proposed that age differences in processing
speed are the result of an age difference in working memory (also see Stine's contribution to
the present volume). Because much is known about the processing resource limitations of
working memory (e.g., Baddeley & Hitch, 1974), researchers can define task condition
complexity before an experiment is conducted.
Another method of defining processing complexity independent of examining task
performance RT in a post hoc manner is to use the concept of entropy as an antecedent cause
(e.g., Allen, Patterson, & Propper, 1994). The concept of entropy is based upon a scientific
law--the Second Law of Thermodynamics. The basic assumption is that entropy must increase
across the lifespan, and that this should increase levels of neural (or internal) noise (Allen,
1990, 1991; Allen & Coyne, 1988; Allen, Madden, Cerella, Jerge, & Betts, 1994, Experiment
4; Allen, Patterson, & Propper, 1994; Krueger & Allen, 1987; Welford, 1958). For example,
from the predictions of entropy, we know that information should be represented in memory
more variably with increases in adult age, and this predicted effect has been empirically verified
(e.g., Allen, 1991; Allen, Katffman, & Propper, 1994, Part I & Part II). Furthermore, we can
predict levels of entropy across age using methods borrowed from statistical dynamics in
physics. These levels of entropy can then be compared to behavioral data such as RT (e.g.,
Allen, Kaufman, & Propper, 1994; Parts I & II). Interestingly, entropy data suggest that
processing variability may be an even more general factor than processing speed in accounting
for age differences (Allen, Kaufinan, & Propper, 1994, Part II).
3.3. Localized Slowing
In addition to generalized models of aging, there is another class of models that is

localized, or process-specific, in nature. As a simplifying assumption, these models also
assume that both young and older adults process information using the same number and order
of processing stages (although see Fisher et al., in this volume, for a cautionary note on this
assumption). However, an important assumption of localized models is that some cognitive
processes are affected more than others by increases in adult age (e.g., Allen, Madden, Cerella,
Jerge, & Betts, 1994; Allen et al., 1993; Balota & Ferraro, 1993; Bashore et al., 1989; Fisk &
Fisher, 1994; Madden et al., 1993). One manner of conceptualizing localized models of
cognitive aging is to assume that local cognitive processes modulate (i.e., qualify) the effect of
relatively general biological processes. For example, a general factor such as entropy may not
impact uniformly upon different levels of information processing, because these different levels
of information processing may require different levels of processing resources (and processes
that require more processing resources would be particularly affected in a deleterious manner
by entropy). Thus, age-related increases in entropy may affect visual encoding more than
lexical access in word recognition tasks (Allen et al., 1993; Madden, 1992). A fundamental
assumption of localized models is that one must examine component cognitive processes in
order to understand how aging affects information processing. Hence, this approach holds that
general curve-fitting descriptions of age differences in information processing across a wide
variety of tasks (e.g., Cerella, 1985, 1991) will not tell us the whole story of cognitive aging,
because they do not adequately address the fundamental cognitive processes (or internal
computations) that are involved (e.g., Allen et al., 1993; Balota & Ferraro, 1993). For
example, Allen, Madden, Cerella, Jerge, and Betts (1994) have demonstrated using localized
curve-fitting procedures that one needs two different slowing factors to describe age
differences in visual word encoding but only one slowing factor to describe age differences in
lexical access.
As noted earlier, one prediction of localized models of cognitive aging is that a single
psychological or biological cause cannot directly explain age differences in information
processing across different processing stages and different tasks. This is because cognitive
constructs define task difficulty in an a priori manner that cannot be accomplished
parsimoniously without alluding to such cognitive constructs. Thus, the emphasis in localized
models is placed upon illustrating how component cognitive processes (or the different
processes involved in different cognitive tasks) qualify the effect of general factors on aging.
However, the localized view does not necessarily assume that no single law exists that
can explain age differences in information processing. Instead, the localized view of aging
assumes that there is a qualitative difference in representational levels between biological

processes and cognitive processes. This difference in representational levels results in
emergent characteristics of cognition such that there is no simple one-to-one mapping between
biological levels and cognitive levels. (This is similar to the Gestalt dictum that "the whole is
different from the sum of its parts.") Indeed, one could argue that physics, chemistry, biology,
and experimental psychology all study the same basic phenomena--but at different levels on the
molecular-to-molar continuum Thus, according to the localized framework, cognitive
constructs are meaning~l independent of biological constructs even though neural (biological)
processes clearly have a major impact upon cognitive processes. Consequently, localized
models of aging predict that cognitive aging cannot be reduced to a single biological (or
psychological) factor--unless that biological factor can explain behavior (or emergent
characteristics) at different levels of the cognitive factor or factors.
In order to test for localized age differences, Madden, Pierce, and Allen (1992)
proposed a method that combined Brinley plot and ANOVA procedures. First, an ANOVA is
used to determine if there are Age x Task interactions. Next, the task condition RT means for
young adults were used as an independent variable and older adults' task condition RT means
formed the dependent variable using linear regression methods (i.e., a Brinley plot). The best-
fitting slowing function derived from the Brinley plot was then used to transform young adults'
raw latencies. Finally, the transformed latencies of the young adults and the untransformed
latencies of the older adults were used as the dependent variable in an ANOVA. Transforming
young adults' data using the best-fitting slowing function results in young adults being
transformed into predicted "older adults." If any interactions between age and task remain in
the transformed ANOVA, this indicates that a single general slowing model cannot adequately
predict the observed age differences (i.e., an interaction exists for the Bfinley plot data).
Localized models of cognitive aging predict that Age x Task interactions should remain in the
transformed analysis. Alternatively, general slowing models of cognitive aging predict that all
Age x Task interactions should be eliminated in the transformed analysis. Although this
transformed analysis does eliminate some Age x Task interactions, the procedure typically does
not eliminate all Age x Task interactions (e.g., Allen et al., 1993; Allen, Madden, Cerella,
Jerge, & Betts, 1994; Madden et al., 1992, 1993). Furthermore, the transform procedure can
also form new Age x Task interactions (e.g., Allen et al., 1993). Such results imply that
although some generalized effects are present, localized age differences do exist. Of course, it
should be noted that multiple-factor (i.e., process-specific) slowing functions such as those
used in Allen, Madden, Cerella, Jerge, and Betts, (1994) will eliminate all interactions with age
if these models can account for all interactions present in the Brinley plot.
3.4. Visual Acuity
It is important to consider whether age differences in other factors such as visual acuity
might affect our understanding of age differences in visual word encoding. This is because age
differences in visual acuity could affect stimulus encoding processes across age. Previous
research has demonstrated that older adults have poorer visual acuity than do young adults
(e.g., Owsley, Sekular, & Siemson, 1983; Pitts, 1982; Weale, 1986). A major problem is that
older adults typically have only one-third as much retinal illumination as young adults (Weale,
1986). Thus, one potential explanation for why older adults take longer to recognize words
than do young adults (e.g., Allen, Madden, & Crozier, 1991, 1993; Cerella & Fozard, 1984;
Madden, 1992) is that older adults have poorer visual acuity (Owsley et al., 1983, Pitts, 1982).
However, Allen, Madden, Cerella, Jerge, & Betts (1994, Experiment 3) recently tested
this hypothesis using a naming task that varied the case type and the exposure duration of
word and pronounceable nonword stimuli. Previous research has documented that subjects
take longer to recognize/identify words presented in mixed case than in consistent case (i.e.,
the mixed-case disadvantage), because mixed-case stimuli force subjects to encode stimuli
using a slower input channel (e.g., Allen et al., 1993, in press). Allen, Madden, Cerella, Jerge,
and Betts (1994) proposed that brief exposure durations should be especially difficult for older
adult subjects who had poorer visual acuity than did young adults (e.g., it would take more
time for enough visual information to accumulate for older adults if'less light reached the retina
within each sample, Madden & Allen, 1994). Thus, if older adults' typically larger mixed-case
disadvantage (Allen et al., 1993; Allen, Madden, Cerella, Jerge, & Betts, 1994) was the result
of a lack of rentinal illumination, then older adults should have revealed a relatively larger age
difference for the mixed-case disadvantage for briefer exposure durations than for
presentation-until-response. The Allen et al. experiment, though, did not find an Age x Case
Type x Exposure Duration interaction even though the Age x Case Type interaction was
reliable. This result suggests that older adults' larger mixed-case disadvantage was the result of
an encoding decrement that occurred after initial sensory transduction. Indeed, several recent
studies have found that at best, age differences in visual acuity can account for main effects,
but not for Age x Task interactions (e.g., Allen, Weber, & Madden, 1994; Long & Crambert,
1990). Consequently, although older adults' poorer visual acuity may slow down their
accumulation rate thereby affecting main affects for age, it does not appear that age differences
in visual word encoding can be accounted for by simply alluding to age differences in visual
acuity.
However, even though age differences in visual acuity do not adequately account for
age differences in encoding, there is evidence that a later stage(s) of encoding may be relevant
to this issue. Earlier in the present paper, it was proposed that visual stimulus encoding
involves more than one stage of processing. A recent study by Allen, Weber, and Madden
(1994) illustrates the distinction between two such encoding stages. These researchers
hypothesized that visual stimulus encoding involves both an initial detection stage and a
subsequent identification stage. Initially, a "blob-level" representation is formed that is used to
detect the presence of a stimulus (e.g., DeValois & DeValois, 1987; Mart, 1982; Pashler,
1987). In order to identify objects within the blob-level representation (e.g., words), though, it
is necessary to focus attention upon different sections of the blob-level representation (i.e.,
attentional selection) so that this information can be interpreted (Allen & Madden, 1990;
Pashler, 1987). Allen, Weber, and Madden (1994) using a visual search task involving letters
as stimuli found evidence that older adults had difficulty in selecting stimuli during the
identification stage of processing. However, when illumination was varied (40 cd/m: vs. 18
cd/m:) between two groups of young adults in an attempt to simulate older adults'
performances by decreasing retinal illumination in younger adults, this reduction in illumination
did not result in the young adults in the lower illumination condition (18 cd/m2) resembling
older adults in the higher illumination condition (40 cd/m:). Consequently, the Allen et al.
(1994) study suggested that age differences at the identification stage of encoding rather than
the detection stage of encoding accounted for age differences in visual search.
4. PART IV: A REVIEW OF THE AGING AND VISUAL WORD RECOGNITION

LITERATURE W H E N WORD FREQUENCY IS VARIED
In order to determine if age differences in visual word encoding are generalized or

localized, experiments necessarily must manipulate at least two different stages of the word
recognition/identification process. These stages include: detection (forming the initial blob-
level representation), selection (focusing on a particular portion of the blob-level
representation), lexical access (looking a word up in the "mental dictionary"), decision (e.g.,
"word" or "nonword" for a lexical decision task), response selection (mapping the decision on
to the appropriate response), and response execution (directing muscles to carry out the
appropriate response). Word encoding involves the first three of the six listed stages of word
recognition/identification. The implicit assumption of this stages framework is that additive
factors logic holds (e.g., Sternberg, 1966). This simplifying assumption allows one to assume,
for example, that case mixing effects and word frequency effects measure different processing
stages (e.g., Allen et al., 1993).
4.1. Age differences in Lexical Access?
As noted earlier, lexical access is typically indexed by manipulating word frequency or

semantic priming. Researchers have examined age differences in lexical access using both
indices. With regard to semantic priming, considerable aging research has been conducted
(e.g., Burke, White, & Diaz, 1987; Howard, Shaw, & Heisey, 1986; Madden, 1989, 1992;
Madden et al., 1993). In the present volume, Kellas, Paul, and Vu (normal aging) and Ober
and Shenaut (Alzheimer's disease) will address semantic priming. The gist from these studies
suggest that there are few appreciable age differences in semantic priming (at least for healthy
older adults). Because of the already existing literature reviewing semantic priming, the
present examination of visual word processing will focus upon aging studies that have
measured lexical access by manipulating word frequency.
In reviewing aging studies that have manipulated word frequency, it is important to
address the factors of education and vocabulary ability (frequently measured using WAIS-R
Vocabulary subscale scores). For example, Tainturier, Tremblay, and LeCours (1989, 1992)
found that individuals showed a progressively larger word frequency effect as their number of
years of education decreased. With regard to age, the controversy concerns whether
researchers should match their young adult and older adult groups on educational and verbal
ability levels. This is an issue, because older adults tend to score higher than young adults on
vocabulary tests. Balota and Ferraro (1993, 1994) found that when verbal ability was matched
across age, older adults showed slightly larger word frequency effects than did young adults.
This suggested that older adults actually took longer than young adults to perform lexical
access for low-frequency words. However, Allen, Madden, Cerella, Jerge, and Betts (1994)
and Tainturier et al. (1989, 1992) found no age differences in word frequency when verbal
ability was controlled. An assumption of aging research is that older adults will tend to have
more exposure to words than will young adults because older adults have lived longer.
In addition to focusing our review on research that has manipulated word frequency,
the present review will also focus on three common word processing tasks: letter
identification, lexical decision, and pronunciation (naming). Letter identification consists of
determining whether a visually present letter matches the initial letter of a subsequently visually
presented word (i.e., a pre-cue task) or whether the first letter of a word matches a
subsequently presented letter (i.e., a post-cue task) (see Allen & Madden, 1990). Letter
identification involves word processing because word frequency effects are present for this
task (e.g., Allen & Madden, 1990; Johnson, Allen, & Strand, 1989). A lexical decision task
requires subjects to decide whether visually presented letter strings form actual English words
(subjects either respond "word" or "nonword," see Carr & Pollatsek, 1985). Finally, a naming
task requires subjects to pronounce visually presented words or pronounceable nonwords (see
Carr & Pollatsek, 1985). Note that the naming task requires individuals to form a specific
speech production output code for each named stimulus, but, contrary to a lexical decision
task, individuals do not have to determine if a visual stimulus forms an actual English word on
a naming task.
4.2. Letter Identification Studies.
There have been two letter identification studies conducted that examined aging and
word frequency. Allen and Madden (1989) used a pre-cue task in which a letter target was
presented, followed by the presentation of a word. Subjects were instructed to decide whether
the pre-cued letter matched the initial letter of the subsequently presented word. As would be
expected, Allen and Madden (1989) found that older adults took longer, in general, than young
adults to perform the task (644 ms vs. 412 ms, respectively). However, this study revealed
that the pattern of word frequency effects varied across age. The word frequency factor was
based on the Kucera and Francis (1967) norms, very-high-frequency (VHF) = 240-660
instances, medium-high-frequency (MHF) = 155-235, low-frequency (LF) = 40-54, and very-
low-frequency (VLF) = 1-5. For the young adults, there was a non-monotonic function across
word frequency (VHF = 406 ms, MHF = 429 ms, LF = 408 ms, VLF = 405 ms), but older
adults showed a monotonic function across word frequency (VHF = 636 ms, MHF = 633 ms,
LF = 638 ms, VLF = 671 ms). That is, young adults showed a partial word frequency
disadvantage for a letter identification task that is consistent with horse race models of visual
word recognition (e.g., Allen & Madden, 1990; Allen & Emerson, 1991; Johnson et al., 1989).
Older adults, though, showed a partial word frequency advantage for a letter identification
task. Clearly, these data indicate that young and older adults were processing the letter
identification stimuli in a different mariner (see Fisher et al., in the present volume).
In the next section (Part V), we will argue that both young and older adults used the
same basic processing architecture, but that encoding difficulties for the letter-level input
channel forced older adults to use the rather circuitous word-level input channel for conducting
a letter identification task. These letter identification data provided support for a localized
model of cognitive aging. To illustrate this point, one needs only to test a single slowing
function for these data. When the best-fitting slowing function is computed for these data
using a Brinley plot [Y = 1.39(X) + 72], the amount of explained variance is only 36 percent.
Allen, Madden, and Crozier (1991) replicated the letter identification results of Allen
and Madden (1989). However, Allen and colleagues tested subjects on both a pre-cue letter
identification task and a lexical decision task. In this letter identification study, Allen et al.
(1991) used lower-frequency words for the VLF category (1-5 instances using the Kucera &
Francis, 1967, norms) than did Allen and Madden (1989, 40-55 instances). Thus, the results of
Allen et al. (1991) could not have been due to using words that were too high in word
frequency. These letter identification data from Allen et al. (1991) replicated the results of
Allen and Madden (1989). That is, the pattern of word frequency effects were different for
young (VHF = 414 ms, MI-IF = 450 ms, LF = 407 ms, VLF = 412 ms) than for older adults
(VHF = 724 ms, MHF = 735 ms, LF = 766 ms, VLF = 775 ms). Indeed, young adults
revealed a significant quadratic trend (but no linear trend), however, older adults revealed a
siL,nificant linear trend (but no quadratic trend) (Allen et al., 1991). The best-fitting slowing
function for these letter identification data [Y = 1.36(X) + 178] (r' = .38) was also similar to
that found by Allen and Madden (1989). Consequently, both of the available aging studies
conducted on a letter identification task found a quadratic trend across word frequency for
young adults (also see Allen & Emerson, 1991; Allen & Madden, 1990; Johnson et al., 1989)
but found a linear trend across word frequency for older adults. Given the poor fit of the best-
fitting slowing functions for these experiments (r 2 = .36, .38, respectively), and the different
trend functions for young and older adults across word frequency, the letter identification data
are more consistent with a localized model of age differences than with a generalized model.
4.3. Lexical Decision Studies.
Instead of requiring an individual to decide whether a letter matches the initial letter of
a subsequently presented word (i.e., a letter identification task), a lexical decision task requires
an individual to decide whether a letter string forms an English word. The Allen et al. (1991)
study involved a within subjects manipulation of both a letter identification and a lexical
decision task. However, for the lexical decision task, both young and older adults showed a
linear trend across word frequency (young: VHF = 485, MHF = 517, LF = 572, VLF = 613;
Older = VHF-- 852, MHF = 878, LF = 923, VLF = 1031). When these lexical decision data
were plotted, the best-fiRing slowing function was Y = 1.70(X) -6 (r 2 = .83). When both the
lexical decision and the letter identification data from Allen et al. (1991) were plotted, the best-
fitting slowing function was Y = 1.48(X) + 118 (r 2 =. 84).
The overall slowing function for Allen et al. (1991) could account for 84 percent of the
experimental variance for both letter identification and lexical decision tasks. However, since
Brinley plots are not particularly sensitive in detecting interactions (Perfect, 1994), it was
necessary to test for interactions. In order to do so, we used the procedure suggested by
Madden et al. (1992). In this procedure, the task condition RT means for young adults are
used as an independent variable and older adults' task condition RT means form the dependent
variable using linear regression methods (i.e., a Brinley plot). Next, the best-fiRing slowing
function derived from the Brinley plot is used to transform young adults' raw latencies. Finally,
the transformed latencies of the young adults and the untransformed latencies of the older
adults were used as the dependent variable in an ANOVA. When this transformed analysis was
conducted on the lexical decision and letter identification data of Allen et al. (1991) using the
best-fitting slowing function to transform young adults' data, the results demonstrated that the
Age x Task x Word Frequency interaction still remained significant [F(3, 138) = 3.59, p < .02].
Given the presence of this interaction in the transformed data, it was necessary to assume the
existence of a different slowing function for a lexical decision task than for a letter
identification task.
One interpretation o f the aforementioned finding of different slowing functions for
letter identification and lexical decision tasks is that these tasks simply map onto different
slowing domains (e.g., Lima et al., 1991, Myerson et al., 1992). Indeed, Lima et al. (1991)
classified letter processing as a non-lexical domain task but classified a lexical decision task as
a lexical domain task. However, the present letter identification task clearly involves lexical
access as is indicated by the presence of the word frequency effect for this task (see Allen &
Emerson, 1991; Allen & Madden, 1990, for an in-depth discussion ofthis topic). Although the
letter matching portion of the letter identification task does not involve lexical access, the
word-level input channel that interferes with the letter-level input channel's access to the
central processor does involve lexical access. Thus, both the letter identification and the
lexical decision task data from Allen et al. (1991) are clearly affected by lexical access. Hence,
it would seem necessary to classify both of these tasks within the lexical domain--even though
both tasks reveal different slowing functions. Consequently, the present letter identification
data provide a disconfirmation of the notion that a single slowing function can adequately
describe age differences in the lexical domain.
In another quasi-lexical decision experiment, Bowles and Poon (1981) also manipulated
word frequency. In their study, however, subjects keep their fight and left index fingers on
separate response keys. If both of the two letter strings that were presented simultaneously
were real words, then subjects were instructed to lift a finger from one key. However, if one,
or both, of the letter strings did not form real English words, then the subjects were instructed
to lift the other finger from the response key. Thus, for "word" trials, there could be two
"high-frequency" words (HH), two "low-frequency" words (LL), or one high-frequency word
and one low-frequency word (ILL). For the "nonword" trials, there could be one nonword and
one high-frequency word (HN), one nonword and one low-frequency word (LN), or two
nonwords (NN). The results indicated that there was a word frequency advantage for both age
groups, but age did not interact with word frequency. When a Brinley plot was computed for
these data, the best-fitting least-squares regression equation was Y = 1.65(X)-396 (r 2 = .97).
Based upon the slowing function data, the Bowles and Poon (1981) data are consistent with a
generalized slowing model. However, the age differences for stimulus type (word vs.
nonword) appear to be larger than the age differences for word frequency in the Bowles and
Poon experiment (at least, based upon cell means). Although the ANOVA data were not
available for additional analysis, it would have been useful to determine whether the transform
analysis used above would have resulted in an interaction (i.e., evidence for localized slowing).
Allen, Madden, Weber, and Groth (1993) reported three aging experiments that
involved a lexical decision task and that manipulated word frequency. All three of the Allen et
al. (1993) experiments presented a single letter string on each trial, and subjects were
instructed to decide whether this letter string formed a real English word. In the first
experiment, case type, stimulus type, and word frequency were manipulated. It was assumed
that mixed-case presentation would handicap the word-level channel (because the mixed-case
stimulus would result in an unfamiliar spatial frequency pattern), resulting in indirect lexical
access occun~g indirectly through the letter-level input channel (Allen & Emerson, 1991). If
age differences in visual word recognition occurred during the transduction stage of
processing, then age should have interacted with case type. Alternatively, if age differences in
visual word recognition occurred at lexical access, then age should have interacted with word
frequency. Also, for lowercase presentation, it was hypothesized that the word-level channel
would typically be used to achieve lexical access. The results indicated that older adults
showed a larger mixed-case disadvantage than did young adults (an Age x Case Type
interaction), but that both groups showed comparable word frequency advantages. Thus, these
data suggested that older adults took longer to visually encode words, but that there were no
appreciable age differences in lexical access. The best-fitting slowing function for these data
was Y - 2.08(X) - 5 3 0 ( r 2 = .94). When the transform analysis was conducted on the data
from Experiment 1, the Age x Case Type x Word Length interaction was still present.
In Experiment 2 of Allen et al. (1993), word frequency, case type and spacing (0 vs. 1
space between adjacent letters)were manipulated to further examine whether age differences
in visual word recognition performance were the result of encoding or of lexical access effects.
There was a significant Age x Case Type x Spacing Type interaction but age did not interact
with word frequency. Again, these results implied that age differences were occurring during
the initial transduction process rather than during the later lexical access process. The best-
fitting slowing function for Experiment 2 was Y = 1.93(X) - 495 (r 2 = .91). When this slowing
function was used to carry out the Madden transform of these data, this analysis revealed that
the Age x Case Type x Spacing Type interaction was eliminated, but that an Age x Word
Length interaction was created.
In Experiment 3 of Allen et al. (1993), case type was not manipulated. Instead,
response selection load (go/no-go vs. two-choice tasks) and word frequency were
manipulated. For the "go/no-go" portion of the task, subjects responded "word" if the stimulus
formed a real word (a "go" trial). However, if the stimulus did not form a word, then subjects
did not respond (a "no-go" trial). Alternatively for the two-choice trials, subjects always
responded whether a letter string was a "word" or a "nonword." The results showed that older
adults exhibited a relatively larger response selection load effect (two-choice - go/no-go RT)
than did young adults, although both age groups showed the same magnitude of word
frequency advantage. The best-fitting slowing function for Experiment 3 was Y = 1.60(X) -
269 (r 2 = .83). When the best-fitting slowing function was used to conduct the transform
analysis, the Age x Task Type interaction was eliminated, but an Age x Word Frequency
interaction was created. This Age x Word Frequency interaction was especially surprising,
because older adults actually showed a smaller word frequency advantage for the transformed
analysis than did young adults. Such a result indicates that there was an interaction across task
conditions and age in the Brinley plot. That is, more than one slowing function was needed to
adequately describe age differences in Experiment 3 of Allen et al. (1993).
Overall, the results of the three lexical decision experiments of Allen et al. (1993)
indicated that a single-factor slowing function could not account for the greater age-related
slowing found for stimulus transduction and response selection stages than for the lexical
access stage of visual word recognition. It should be noted, though, that this finding is not
necessarily inconsistent with a slightly less strict interpretation of generalized slowing.
Namely, a domain-specific slowing model could predict a multi-factor slowing function that
included different components for lexical and non-lexical domains (e.g., Lima et al., 1991;
Myerson et al., 1992; Myerson et al., 1994). Although domain-specific slowing is not
particularly generalized in the sense that it uses more than one slowing factor, this sort of
approach is generalized in the sense that the same slowing functions should work for lexical
stimuli across different tasks. The Allen et al. (1993) data, however, appear to be inconsistent
with even domain-specific slowing. This is because subjects were processing words in all three
of the Allen et al. (1993) experiments (i.e., within the lexical domain), yet multiple slowing
functions were still needed to adequately describe the data (as demonstrated by the existence
of task interactions with age for all three experiments when using the transform procedure of
Madden et al., 1992).
In a recent lexical decision experiment, Balota and Ferraro (1994) manipulated word
frequency and repetition for young and older adults. (Very old adults as well as Alzheimer's
patients were also tested, but we will limit our present examination to the young adult and the
older adult age groups). This study matched young and older adults on Vocabulary subscores
of the Wechsler Adult Intelligence Scale-Revised (WAIS-R) (although the Vocabulary scores
of both young and older adults reported in this experiment were considerably lower than the
WAIS-R Vocabulary scores reported by Allen et al., 1991, 1993). Balota and Ferraro found a
slightly larger word frequency effect for healthy older adults (compared to young adults), and
this effect was even larger for non-repeated than for repeated words. It is important to note,
though, that this effect was relatively small--older adults' word frequency effect was only 19
ms larger than young adults' word frequency effect. When the Balota and Ferraro (1994) data
are analyzed using a Brinley plot, the best-fitting, least-squares regression function is Y -
1.3(X) - 36 (r 2 - .99).
Tainturier et al. (1992) conducted a lexical decision task that manipulated word
frequency, but these researchers also manipulated years of education. Tainturier et al. (1992)
found a larger word frequency effect for individuals with fewer years of education, but age did
not interact with word frequency. That is, young and older adults showed the same word
frequency advantage, but it took individuals with fewer years of education longer to access
their mental lexicon (compared to individuals with more years of education). When the
Tainturier et al. (1992) data are analyzed using a Brinley plot, the best-fitting slowing function
is Y = 1.72(X) - 269 (r 2 = .91).
Although it appears that the results of Allen, Madden, and Crozier (1991, Allen et al.,
1993) and Tainturier et al. (1992) are inconsistent with those of Balota and Ferraro (1994), the
discrepancy may actually be the result of differences in reading level and/or verbal ability. For
example, Frederiksen (1978) found that poor readers showed larger word frequency effects
than did average or superior readers. We suspect that, in general, healthy older adults tend to
have larger vocabularies than do healthy young adults. Thus, if the Balota and Ferraro (1994)
sample of young and older adults were relatively poor readers (especially the older adults who
had WAIS-R Vocabulary scores that were one-third lower than those reported by Allen et al.,
1991, 1993, in four different samples of participants), then this might explain the discrepancy
across studies.
4.4. Pronunciation Studies.
Balota and Ferraro (1993) reported an aging study that examined pronunciation onset
(or naming) performance. Word frequency and regularity of orthographic-phonological (O-P)
correspondence were manipulated (e.g., "pint" is irregular whereas "mint" is regular). The
Balota and Ferraro naming study found that older adults showed a slightly larger word
frequency effect than did young adults, but that there were no age differences in O-P
correspondence (regularity). Balota and Ferraro (1993) did not report RT means so it was not
possible to conduct a Bfinley analysis or a Madden transform analysis. However, these authors
used a similar approach to that used by Allen et al. (1993). Namely, Balota and Ferraro
hypothesized that the regularity effect measured a different stage of processing than did the
word frequency effect. Thus, these data appear to provide yet another example in which age-
related slowing apparently occurred during one stage of processing, but not at another.
The Balota and Ferraro (1993) study matched young and older adults on Boston
Naming Test performance, so it is difficult to compare subjects' verbal ability in this study to
subjects in other studies that used the WAIS-R Vocabulary subscale test. It is probably a fair
generalization, though, that, older adults tend to have larger vocabularies than young adults
because of older adults' greater experience with words (as determined by the WAIS-R
Vocabulary subscale scores in numerous aging studies). Thus, if we compared a sample of
older adults at the 50th percentile rank for WAIS-R vocabulary scores for the population of
older adults to an equally educated sample of young adults at the 50th percentile rank for
WAIS-R vocabulary scores for the population of young adults, it is quite likely that there
would be no age differences in word frequency effects, or that older adults would even show
slightly smaller word frequency effects. However, if we match for vocabulary performance, it
is quite likely that we are dealing with a less-intelligent sample of older adults than young
adults in terms of their percentile ranks within their respective age groups (i.e., a regression
artifact). Thus, although aging studies that match on the basis of vocabulary scores are useful,
one should keep in mind that such matching will probably unfairly penalize older adults.
Allen, Madden, Cerella, Jerge, and Betts (1994) recently reported the results of four
naming experiments that varied case type, word frequency, and number of perturbed pixels. In
all four experiments, subjects pronotmced 5-letter and 6-letter, two-syllable words and
pronounceable nonwords (although subjects also performed a lexical decision task in
Experiment 2). In Experiment 1, letter strings were presented in consistent lowercase (LC,
e.g., "hello"), mixed-case by syllables (SYL, e.g., "HELlo"), and mixed-case by adjacent letters
(MC, e.g., "hElLo"), and the same four word frequency categories were used as were used in
Allen et al. (in press). The results revealed a larger mixed-case disadvantage for older adults
than for young adults, and this was especially so for MC versus LC trials (LC and SYL RTs
were similar). However, there was no age difference for the word frequency advantage. When
these data were analyzed using Brinley plots, it was determined that two slowing functions
were necessary--one for the MC condition [Y = 1.23(X) - 78], and one for the combined LC
and SYL conditions [Y = 1.23(X) - 135]. When this three-factor model (i.e., one slope, but
two intercepts) was used in the transform analysis, there were no Age x Task interactions.
Allen et al. (1994) hypothesized that the slope parameter measured central (or more cognitive)
processes, but that the intercepts measured perceptual-motor processes (in this case, the
amount of perceptual normalization that needed to be conducted).
Experiment 2 of Allen et al. (1994) was the same as Experiment 1, except that
Experiment 2 contained both naming and lexical decision tasks (this was a within subjects
factor). There was an Age x Case Type interaction, but age did not interact with either word
frequency or task type. Because a naming task and a lexical decision task have different
decision stage demands (e.g., Balota & Chumbley, 1984), the finding that the Age x Case Type
interaction did not further interact with task type indicated that the locus of the age difference
in the mixed-case disadvantage was not at the decision stage but, rather, at the hypothesized
encoding stage (e.g., Allen et al., 1993). Furthermore, it was necessary to use a separate
slowing function for the MC condition than for the combined LC and SYL conditions for both
the naming [MC: Y = 1.30(X)- 8); LC & SYL: Y - 1.30(X)- 119] and the lexical decision
tasks [MC: Y = 1.28(X) + 106); LC & SYL = Y = 1.28(X) - 56]. When three-factor models
were used to transform both the naming and lexical decision data of young adults, the
transform analyses revealed no Age x Task interactions. Consequently, the results of
Experiment 2 of Allen et al. (1994) required an interpretation based upon the use of process-
specific slowing because two different slowing functions were required to adequately describe
the data. Experiment 3 of Allen et al. (1994) was the same as Experiment 1, except Experiment
3 used both brief presentation and presentation-until-response exposure durations
(Experiments 1, 2, and 4 used all presentation-until-response exposure durations). The results

from Experiment 3 revealed an Age x Case Type interaction, but exposure duration did not
further interact with these variables. Also, age did not interact with word frequency. The
finding that exposure duration did not affect the magnitude of the age difference for the mixed-
case disadvantage suggested that this mixed-case disadvantage effect was not the result of age
differences in retinal illuminance, but rather due to some latter-occurring perceptual process
(Allen et al., 1994). As was the case for earlier experiments, a separate slowing function was
required for the LC condition and for the MC and SYL conditions for both exposure durations
[100 ms exposure duration: LC: Y = 0.92(X) + 222; MC & SYL: Y = 0.92(X) + 157;
presentation-until-response: LC: Y = 0.92(X) + 229; MC & SYL: Y = 0.92(X) + 162]. When
these slowing functions were used to transform the young adults' RT from Experiment 3, the
transformed ANOVA eliminated all Age x Task interactions for the brief exposure duration
and all but one Age x Task interaction for the longer exposure duration condition (this lone
interaction was the result of a non-monotonic cell for older adults in the presentation-until-
response condition). Thus, the data from Experiment 3 of Allen et al. (1994) also suggested
that older adults snow greater slowing at the encoding stage (i.e., the perceptual-motor
parameters) than at the lexical access stage (i.e., the central stage).
Finally, Experiment 4 of Allen et al. (1994) manipulated the number of perturbed pixels
(0, 2, or 4) rather than case type. The notion was that this procedure would result in the word-
level channel being able to output a code even as the number of perturbations increased. That
is, case mixing prevents the hybrid model from outputting a word-level code because the
holistic pattern is non-linear, and, therefore, Fourier synthesis would not be successful (Allen et
al., in press). However, pixel perturbation would result in a linear transformation, thus, the
word-level channel would be able to form a code even when four pixels per letter were
perturbed. The results from Experiment 4 revealed an Age x Pixel Perturbation Type
interaction, but no Age x Word Frequency interaction. These findings suggested that even
when just the word-level channel was used for processing, there were still age differences in
encoding, but not age differences in lexical access. It was necessary to use separate slowing
functions for the four-pixel perturbation condition [Y = 0.98(X) + 294] and for the combined
zero-pixel and 2-pixel perturbation conditions [Y = 0.98(X) + 147]. As was the case for all
three earlier experiments, when a three-factor slowing function based upon these two two-
factor slowing functions was used to transform the young adults' data, the resulting
transformed ANOVA eliminated all interactions with age.
In summary, all four of the Allen et al. (1994) naming experiments required more than
one slowing fimction to account for the encoding data. Allen et al. used the F-test suggested
by Fisk, Fisher, and Rogers (1991) to test for whether a given model accounted for more
variance than another. Specifically, in all four experiments, the three-factor model (i.e., two
two-factor models that shared the same slope but required different intercepts) accounted for
significantly more variance than did the best-fitting single two-factor model. However, the
best-fitting four-factor model (i.e., two two-factor models--each with different slopes and
intercepts) did not account for significantly more variance than did the three-factor model.
These data indicated that a single factor can account for lexical access speed differences across
age, but that two factors are necessary in order to account for encoding performance across
age.
5. PART V: W H E R E DO WE STAND?
After having reviewed the literature on age differences in visual word recognition when
word frequency was manipulated, we now can ask two central questions. First, can we
account for the letter identification, lexical decision, and naming data using a single
framework? Secondly, can these data be described adequately with a single factor, or is it
necessary to allude to process-specific factors? The answer to the first question appears to be
"yes." That is, the age differences data from letter identification (Allen & Madden, 1989; Allen
et al., 1991), lexical decision (e.g., Allen et al., 1993; Balota & Ferraro, 1993; Tainturier et al.,
1992), and naming (e.g., Allen et al., 1994; Balota & Ferraro, 1994) tasks can be accounted
for using the hybrid model of Allen et al. (in press) (refer to Figure 1).
5.1. Letter Identification
For a letter identification task using visual presentation, the model can account for
older adults showing a word frequency advantage and the young adults showing a non-
monotonic effect across word frequency by assuming that the older adults have difficulty in
forming letter-level codes (Allen, Madden, & Crozier, 1991). This is because letter-level codes
require more processing resolution that do word-level codes (Allen & Emerson, 1991). Thus,
older adults must use the segmentation process (Allen & Emerson, 1991) via the central
processor to break down a word-level code into a pseudo-letter-level code so that the letter
matching process of the letter identification task can be conducted. Because older adults
initially form a word-level code (and this involves lexical access), the model predicts a word
frequency advantage for older adults. Indeed, Allen, Madden, and Crozier found that older
adults showed approximately the same linear trend across word frequency for a letter
identification task as for a lexical decision task.
However, young adults revealed a quadratic trend for a letter identification task (as
would be predicted if the word-level and the letter-level input channels were involved in a
processing horse race), but a word frequency advantage for a lexical decision task (Allen,
Madden, & Crozier, 1991). Hence, older adults tend to rely upon the word-level input channel
information for both letter identification and lexical decision tasks. For young adults, though,
the word-level input channel interfered with the letter-level input channel for a letter
identification task, although the letter-level channel information was typically used to conduct
this task (Allen & Emerson, 1991; Allen & Madden, 1990). Alternatively for a lexical decision
task, young adults typically used the word-level input channel to conduct the task (Allen et al.,
1992, 1993) unless case mixing or some other stimulus manipulation made the word-level
(holistic) stimulus unfamiliar (Allen et al., in press).
5.2. Lexical Decision and Naming Tasks
~he hybrid model predicts a larger mixed-case disadvantage for older adults than for
young adults on a lexical decision task because the a mixed-case word must be initially
encoded using the letter-level input channel (Allen et al., 1993, in press). This letter-level code
of the word must then be superposed into a pseudo-word-level code by the central processor
so that lexical access can occur. However, the superposition task is processing resource
intensive, and this is particularly difficult for older adults who already have fewer processing
resources than young adults (e.g., Allen et al., 1993; Madden, 1986, 1990). Note that the
superposition task is really a normalization process conducted on the letter-level code. When
pixels of letters are perturbed (e.g., Allen, Madden, Cerella, Jerge, & Betts, 1994, Experiment
4), the hybrid model predicts that this sort of stimulus can be normalized by the word-level
channel. However, in both cases (Age x Case Type or Age x Perturbation Type), the hybrid
model accounts for task interactions with age by assuming that older adults are especially
affected by the processing resource-intensive normalization procedures.
For a naming task, both young and older adults typically used the phonological
pathway rather than the orthographic pathway used for a lexical decision task (e.g., Allen et
al., 1994; Balota & Ferraro, 1993). The syllable-level input channel typically wins the race to
the central processor (rather than the letter-level, GPC channel, see Figure 1), and this code is
used to form a motor code for naming. However, this holistic input channel can be
handicapped by presenting an unfamiliar syllable-level stimulus (e.g., by mixing case within a
syllable). Under such circumstances, the letter-level input channel is used (which uses
grapheme-to-phoneme correspondence, or GPC rules) to perform a naming task (Allen,
Madden, Cerella, Jerge, & Betts, 1994). As was the case for the lexical decision data, the
hybrid model accounts for age differences in naming as a function of stimulus case type by
alluding to an age difference in normalization performance. Namely, the model predicts that
older adults will be adversely affected by increased normalization requirements of a task.
Thus, when case is mixed within a word, the syllable-level, phonological channel is
handicapped, and the input code from the slower GPC channel must be superposed so that
lexical access can occur (i.e., GPC rules allow naming but not direct lexical access). The GPC
channel deals with smaller pieces of information than does the syllable-level channel, thus, the
GPC channel requires greater processing resources in order to normalize a stimulus. When the
stimulus is particularly unfamiliar as is the situation for nonwords, the age difference in
encoding should even be larger.
Also, when pixels are perturbed within letters of a word on a naming task, then a
normalization process must be carried out on the input stimulus pattern in order to reco~ize
the word. That is, the syllable-level channel is not prevented from outputting a code, but
larger levels of normalization are required in order to cohere the percept. Once again, an age
difference in processing resources would result in a larger normalization cost for older adults.
5.3. General or Process-Specific Slowing?
The aging literature on visual word recognition for which word frequency is
manipulated provides a solid argument for the idea that process-specific slowing occurs (Allen
& Madden, 1989, 1991; Allen et al., 1993; Allen, Madden, Cerella, Jerge, & Betts, 1994;
Balota & Ferraro, 1993). However, there is evidence suggesting that there exists a generalized
slowing component, as well (Allen, Madden, Cerella, Jerge, & Betts, 1994). In the present
section, we will outline a theory that can account for this seeming paradox. However, before
we present this theoretical account, the basic findings will be summarized from the three tasks
(letter identification, lexical decision, and naming tasks) using the three methods of analysis for
determining whether localized or generalized slowing is present (i.e., Brinley plots, the
transform proposed by Madden et al., 1992, and hierarchical regression).
5.4. Letter Identification
The letter identification data reviewed (Allen & Madden, 1989; Allen, Madden, &
Crozier, 1991) are qualitatively different from the lexical decision and naming data because
there is strong evidence indicating that older adults typically used a different processing
channel (word-level) to conduct a letter identification task than did young adults (letter-level).
This is the case even though both age groups possessed the same basic processing architecture
(i.e., both word-level and letter-level input channels and a central processor, see Allen et al.,
1991). The word frequency effects found for this task demonstrated that lexical access did
occur (Allen & Emerson, 1991; Allen & Madden, 1990), therefore, the task falls within the
lexical domain (Lima et al., 1991). The Brinley plots for both of the reported experiments
(Allen & Madden, 1989; Allen et al., 1991) accounted for less than 40% of the total variance,
and older adults showed a significant linear trend while young adults revealed a significant
quadratic trend for the letter identification task (Allen et al., 1991). Consequently, the most
parsimonious manner to interpret these data is to assume that older adults tend to use top-
down processing via the word-level input channel to conduct letter identification, whereas
YOung adults tend to use bottom-up processing via the letter-level input channel for a letter
identification task. Clearly these data support the concern of Fisher et al. (the present volume)
that young and older adults may not always use the same type of processing for a given task.
5.5. Lexical Decision Task
The seven experiments reviewed that examined age differences for a lexical decision
task that manipulated word frequency (Allen, Madden, & Crozier, 1991; Allen et al., 1993,
Experiments 1-3; Balota & Ferraro, 1994; Bowles & Pooh, 1981; Tainturier et al., 1992)
suggested that older adults and younger adults did use the same processing stages. For
example, Allen et al., 1991, found significant linear trends for both age groups for a lexical
decision task. Also, there were no appreciable age differences in word frequency effects--
especially if the older adult sample was not limited to individuals with rather low vocabulary
scores (based upon percentile rank within an age group).
These lexical decision data did reveal Brinley plots for individual experiments with
relatively high r-squared values (typically < .90). Also, when the transformed analysis
suggested by Madden et al. (1992) was conducted on the data from Allen et al. (1991, 1993),
some (but not all) task interactions with age were eliminated. (It should be noted that the
necessary data was not available in order to conduct the transformed analyses on the other
experiments.) These results suggest that there is a component of generalized slowing that does
occur for older adults on a lexical decision task. However, the transformed analyses still
revealed Age x Task interactions for all four experiments. This finding indicated that there
were significant interactions for these Brinley plots within the same lexical processing domain,
thus, these data indicated that process-specific slowing occurred for these lexical decision data
(in addition to the previously mentioned generalized slowing).
5.6. Naming
The five reviewed naming studies (Allen, Madden, Cerella, Jerge, & Betts, 1994,
Experiments 1-4; Balota & Ferraro, 1993) showed quite similar results as did the previously
discussed lexical decision experiments. Both age groups showed word frequency advantages,
and these word frequency effects were quite similar. The Brinley plots for the individual
naming experiments revealed r-squared values that tended to account for at least 90% of the
total variance (excluding the Balota & Ferraro, 1993, study for which no word frequency
means were reported). However, the Allen et al. (1994) Experiments (1-4)required a three-
factor model (two different slowing functions with the same slopes but different intercepts).
Based on Cerella (1985), slopes are assumed to measure age differences in central (lexical
access, retrieval, and decision) processes and y-intercepts are assumed to measure peripheral
(encoding, response selection, and response execution) processes. A slope of approximately
1.00 suggests little age appreciable differences at the central processing stage, and a positive
intercept suggests age-related slowing at the peripheral stage (Cerella, 1985). Therefore, the
present naming data from Allen et al. (1994, Experiments 1-4) suggested slight general slowing
for the central processes (i.e., a slope of slightly greater than 1.00), but differential slowing for
the peripheral processing stages depending upon processing load. These data provided
concrete support for the idea that there is process-specific age-related slowing for naming
tasks.
Finally, hierachical regression conducted on Experiments 1-2 of Allen et al. (1994)
continued to find that age was a significant predictor of mixed-case performance even when
the variance associated with lowercase performance was extracted first (i.e., before the
variance associated with age was entered into the model). However, age no longer
si~ificantly predicted mixed-case performance (or the 4-pixel perturbation condition of
Experiment 4) for Experiments 4 of Allen et al. (1994). (When SYL RT instead ofLC RT was
used in Experiment 3, age continued to predict MC RT even when SYL RT variance was
extracted first. However, the aberrant cell for older adults' presentation-until-response, MHF,
LC condition prevented age from predicting MC RT when LC RT variance was extracted
first.) The hybrid model (Allen et al., in press) predicted the obtained results for Experiment 4
(because the model predicts that all levels of pixel perturbation will be encoded by the same
channel, but that lowercase and mixed-case words will be encoded by different channels).
Thus, the finding that the age variable's ability to predict mixed-case performance was
attenuated (Experiments 1-3) suggests that there was some general factor accounting for a
si~ificant portion of the total variance (Salthouse & Coon, 1994). However, the finding from
Experiments 1-3 that age still predicted mixed-case performance even when the variance
associated with lowercase or mixed-case by syllable performance was extracted first provided
evidence that a generalized factor, alone, was not sufficient to account for these results (also
see Madden, 1992, for a similar finding). Consequently, the extant naming data that include a
manipulation of word frequency suggest that there are both generalized and process-specific
components to age differences in visual word processing.
5.7. The "Big Picture"
So far, we have discussed only Brinley plots for individual experiments. Lima et al.
(1991) and Myerson et al. (1992) have proposed that all lexical domain tasks can be described
using a single slowing function of the form: Y = 1.50(X) - 68 (in which "Y" refers to older
adults' predicted RT and "X" refers to young adults' actual RT). To test this prediction, we
entered the data from all 13 presently reviewed experiments that contained cell means into a
single Brinley plot (see Figure 3) (two letter identification experiments: Allen & Madden,
1989; Allen et al., 1991; seven lexical decision experiments: Allen et al., 1991, 1993,
Experiments 1-3; Balota & Ferraro, 1994; Bowles & Pooh, 1981; Tainturier et al., 1992; and
four naming studies: Allen et al., 1994, Experiments 1-4). The best-fitting linear slowing
function for all 13 data sets (190 data points) was: Y = 1.12(X) + 158 (r2 = .79). However,
the six cell means from the Bowles and Pooh (1981) experiment contained RTs that were 700
ms longer than any of the latencies from other experiments. Both Fisk, Fisher, and Rodgers
(1991) and Perfect (1994) have argued that such extreme outliers can result in artificially high
r-squared values on Brinley plots. Therefore, we re-ran the overall Brinley analysis with 12
data sets (i.e., excluding the Bowles & Pooh, 1981, data). The resulting best-fitting linear
slowing fimction (184 data points) was: Y = .96(X) + 250 (r2 = .67).
The finding that the single best-fitting slowing function could account for only 67% of
the total variance of 12 aging experiments that manipulated word frequency is clearly
inconsistent with the results ofLima et al. (1991) and Myerson et al., (1992) who found that a
single slowing function could account for over 90% of the total variance of lexical domain
aging
Word Frequency Meto-Anolysis
1 600 -
1 5O0
1400
OmoO
1300
_ ~
1200
CL~
L.
(D llO0 ~176
-(3
0
1000 ~oo
900 o'~l 9
8OO
7O0
6OO
5OO
3OO ,100 500 600 700 800 900 1000 1100 1200 1300 1400
Young RT
Figure 3.
data. With 33% of the variance unexplained, one would be hard-pressed to claim that a single
two-factor model (i.e., a model with one slope and one intercept) adequately accounted for
these data. Also, both of the overall Brinley plots (using 13 and 12 data sets, respectively)
revealed slowing fimctions with slopes approaching 1.00 and large, positive intercepts. Thus,
these data examining age differences in visual word recognition for which word frequency was
manipulated suggest that there is little or no age-related slowing for central processes, but that
there is considerable slowing occurring at the peripheral stages of processing (e.g., encoding,
response selection, and response execution). The results of Allen et al. (1993; Allen, Madden,
Cerella, Jerge, & Betts, 1994), in particular, suggest that age differences are especially
pronounced for visual word encoding. Again, note that the presently obtained slopes and
intercepts are inconsistent with slowing function for the lexical domain proposed by Lima et al.
(1991). That is, the present slopes were approximately 1.00 whereas the slope found by Lima
et al. (1991) and Myerson et al. (1992) was 1.50. Also, the present intercepts for both the 12-
and 13-experiment data sets were positive and greater than 150 ms, whereas the intercept for
the Lima et al. (1991) slowing function was slightly negative (-68 ms). The differences
between the present overall Brinley plot and that proposed by Lima et al. (1991) and Myerson
et al. (1992) suggest that a single slowing fimction cannot account for all lexical domain data
for age differences. Indeed, the overall Brinley plot data combined with the previously
discussed Madden et al. (1992) transform data from Allen et al. (1991, 1993, 1994), and the
hierarchical regression data from Allen et al. (1994) all indicate the process-specific age-related
slowing occurs for visual word recognition studies in the lexical domain.
5.7. A General Model
We now have evidence from the data of nearly 500 subjects from visual word
recognition studies that manipulated word frequency (i.e., the present set of experiments) that
age differences are essentially additive in nature (slopes of close to 1.00 and large, positive
intercepts). However, this finding seems to be at odds with the Cerella (1985) meta-analysis,
because Cerella found larger age differences for central processing stages than for peripheral
processing stages. This seeming paradox can be accounted for, though, if we assume that the
35 experiments analyzed in the Cerella meta-analysis (i.e., none of these were lexical domain
tasks) examined episodic and procedural memory tasks, but that the present experiments
examined semantic memory processing. We propose that tasks that involve lexical access
(e.g., letter identification, lexical decision, and naming) or the retrieval of highly memorized
arithmetic facts (e.g., Allen, Ashcrafi, & Weber, 1992; Geary & Wiley, 1991; Geary, Frensch,
& Wiley, 1993) will tend to show little age differences in lexical access or arithmetic fact
retrieval, but that there will still be substantial age differences at the encoding and/or response
selection stages of processing. Alternatively, in tasks for which decision-stage processing does
not involve semantic memory (e.g., visual search, memory search, or letter matching), we
propose that there will be larger decision stage age decrements than encoding and/or response
selection age decrements (e.g., Cerella, 1985). Of course, this is a preliminary formulation (but
also see Allen et al., 1992). Furthermore, it does appear that semantic priming may complicate
the issue (the Myerson et al., 1992, results that found a slope of 1.50 were based upon a meta-
analysis of semantic priming studies--although also see Laver & Burke, 1993, for a process-
specific interpretation of the semantic priming data on aging). However, this basic model of
semantic memory expertise nullifying most age differences in decision stages does have
considerable heuristic appeal, and can also account for results such as those reported by Stine
(in the present volume) in which older adults apparently encode words more slowly yet still
evidence comparable reading times to young adults.
REFERENCES
Adams, M.J. (1979). Modds of word recognition. Cognitive Psychology, 11, 133-176.
Allen, P.A. (1990). Influence of processing variability on adult age differences in memory
distribution of order information. Cognitive Development, 5, 177-192.
Allen, P.A. (1991). On age differences in processing variability and scanning speed. Journal of
Gerontology: Psychological Sciences, 46, P19 l-P201.
Allen, P.A., & Coyne, A.C. (1988). Age differences in primary organization or processing
variability? Part II: Evidence for processing variability. Experimental Aging Research, 14,
150-156.
Allen, P.A., & Madden, D.J. (1989). Adult age differences in the effects of word t~equency
during visual letter identification. Cognitive Development, 4, 283-294.
Allen, P.A., & Emerson, P.L. (1991). Holism revisited: Evidence for parallel independent
word-level and letter-level processors during word recognition. Journal of Experimental
Allen, P.A., & Madden, D.J. (1990). Evidence for a parallel input serial analysis model of word
processing. Journal of Experimental Psychology: Human Perception and Performance,
16, 48-64.
Allen, P.A., Madden, D.J., & Crozier, L.C. (1991). Adult age differences in letter-level and
word-level processing. Psychology and Aging, 6, 261-271.
Allen, P.A., Wallace, B, & Waag, E. (1991). Effect of imagery ability on letter-level and word-
level processing. Perception & Psychophysics, 49, 295-300.
Allen, P.A., Ashcrafi, M.H., Weber, T.A. (1992). On mental multiplication and age.
Allen, P.A., McNeal, M., & Kvak, D. (1992). Perhaps the lexicon is coded as a function of
word frequency. Journal of Memory and Language, 31, 826-844.
Allen, P.A., Madden, D.J., Weber, T.A., & Groth, I~E. (1993). Influence of age and
processing stage on visual word recognition. Psychology and Aging, 8, 274-282.
Allen, P.A., Kaufman, M., & Propper, 1~ (1994). The psychophysics of aging and entropy
Part I: Very-short-term memory. Submitted manuscript.
Allen, P.A., Kaufman, M., & Popper, 1~ (1994). The psychophysics of aging and entropy Part
II: Memory storage and retrieval. Submitted manuscript.
Allen, P.A., Patterson, M.B., Propper, R.E. (1994). Influence of letter size on age differences
in letter matching. Journal of Gerontology: Psychological Sciences, 49, P24-P28.
Allen, P.A., Wallace, B., & LoSchiavo, F. (1994). Influence of imaging ability on word
transformation. Memory & Cognition, 5, 565-574.
Allen, P.A., Weber, T.A., & Madden, D.J. (1994). Adult age differences in attention: Filtering
or selection? Journal of Gerontology: Psychological Sciences, 49, P213-P222.
Allen, P.A., Cerella, J., Madden, D.J.,Jerge, K., & Betts, L. (1994). Age differences in naming
as a function of wordfrequency and case type. Unpublished manuscript.
Allen, P.A., Wallace, B., & Weber, T.A. (in press). Influence of case type, word t~equency,
and exposure duration on visual word recognition. Journal of Experimental Psychology:
Human Perception and Performance.
Amrhein, P.C., & Theios, J. (1993). The time it takes elderly and young individuals to draw
pictures and to write words. Psychology and Aging, 8, 197-206.
Visual word encoding and the effect of adult age and wordfrequency 65
Amrhein, P.C. (in press). Evidence for task specificity in age-related slowing: A review of
speeded picture-word processing studies. In P. Allen and T. Bashore (Eds.), Age
differences in word and language processing. New York: North-Holland.
Baddeley, A.D., & Hitch, G. (1974). Working memory. In G. Bower (Ed.), Recent advances
in learning and motivation (Vol. 8). New York: Academic Press.
Balota, D.A., & Chumbley, J.I. (1984). Are lexical decisions a good index of lexical access?
The role of word frequency in the neglected decision stage. Journal of Experimental
Balota, D.A., & Chumbley, J.I. (1985). The locus of word frequency effects in the
pronunciation task: Lexical access and/or production? Journal of Memory and Language,
24, 340-357.
Balota, D.A., & Duchek, J.M. (1988). Age-related differences in lexical access, spreading
Balota, D.A., & Ferraro, F.1L (1993). A dissociation of frequency and regularity effects in
pronunciation performance across young adults, older adults, and individuals with senile
dementia of the Alzheimer's type. Journal of Memory and Language, 32, 573-592.
Balota, D.A., & Ferraro, F.1L (1994). Lexical, sublexical, and implicit memory processes in
healthy young, healthy older adults and in individuals with senile dementia of the
Alzheimer's type. Manuscript submitted for publication.
Bashore, T.1L, Osman, A., & Heffley, E.F. (1989). Mental slowing in elderly persons. A
cognitive psychophysiological analysis. Psychology and Aging, 4, 235-244.
Becker, C.A. (1976). Semantic context effects in visual word recognition: An analysis of
semantic strategies. Journal of Experimental Psychology: Human Perception and
Performance, 2, 556- 566.
Besaer, D., & McCann, 1LS. (1987). Word frequency and pattern distortion in visual word
identification and production: An examination of four classes of models. In M. Coltheart
(Ed.), Attention and Performance XII: The psychology of reading. Hillsdale, N.J.:
Erlbaum.
Besner, D., & Johnston, J.C. (1989). Reading and the mental lexicon: On the uptake of visual
information. In W. Marslen-Wilson (Ed.), Lexical representation and process. (pp. 291-
316). Cambridge, MA: MIT Press.
Besner, D., Twilley, L., McCann, R.S., & Seergobin, K. (1990). On the association between
connectionism and the data: Are a few words necessary? Psychological Review, 97, 432-
446.
Biederman, I. (1987). Recognition-by-components: A theory of human image understanding.
Birren, J.E. (1965). Age changes in the speed of behavior: Its central nature and physiological
correlates. In A.T. Welford & J.E. Birren (Eds.), Behavior, aging, and the nervous system.
Springfield, IL: Charles C. Thomas.
Bowles, N.L., & Poon, L.W. (1981). The effect of age on speed of lexical access.
Experimental Aging Research, 7, 417-426.
Brinley, J.F. (1965). Cognitive sets, speed and accuracy of performance in the elderly. In A.T.
Welford & J.E. Birren (Eds.), Behavior, aging, and the nervous system. Springfield, IL:
Charles C. Thomas.
Brooks, 1LA. (1981). Symbolic reasoning among 3-D models and 2-D images. Artificial
Intelligence, 17, 205-244.
Burke, D.M., White, H., & Diaz, D.L. (1987). Semantic priming in young and old adults:
Experimental Psychology: Human Perception and Performance, 13, 79-88.
Cart, T.H. (1992). Automaticity and cognitive anatomy: Is word recognition "automatic?"
American Journal of Psychology, 105, 201-237.
Cart, T.H., & Pollatsek, A. (1985). Recognizing printed words: A look at current models. In
D. Besner, T.G. Waller, & G.E. MacKinnon (Eds.), Reading research: Advances m theory
and practice (vol. 5). New York: Academic Press.
Cart, T.H., & Posner, M. (1992). The impact of learning to read on the functional anatomy of
language processing. In B. de Gelder and J. Morais (Eds.), Language and literacy:
Comparative approaches. Cambridge, MA: MIT Press.
Cattell, J.M. (1886). The time taken up by cerebral operations. Mind, 11, 220-242.
83.
Cerella, J. (1990). Aging and information processing rate. In J.E. Bitten and ICW. Schaie
(Eds.) Handbook of the psychology of aging (3rd ed., pp. 201-221). San Diego, CA:
Academic Press.
Cerella, J. (1991). Age effects may be local, not global: Comment on Fisk and Rogers (1991).
Journal of Experimental Psychology: General, 120, 235-243.
Cerella, J. (1994). Generalized slowing in Brinley plots. Journal of Gerontology:
Psychological Sciences, 49, P65-P71.
Cerella, J., & Hale, S. (1994). The rise and fall of information processing rates over the life
span. Acta Psychologica, 86, 109-197.
Cerella, J., & Fozard, J.L. (1984). Lexical access and age. Developmental Psychology, 20,
235-243.
Cerella, J., Poon, L., & Williams, D.M. (1980). A quantitative theory of mental processing
time and age. In L.W. Poon (Ed.), Aging in the 1980's: Psychological issues. Washington,
D. C.: American Psychological Association.
Collins, A.M. & Quillian, M.R. (1969). Retrieval time from semantic memory. Journal of
Verbal Learning and Verbal Behavior, 8, 240-247.
Coltheart, M., Davelaar, E., Jonasson, J., & Besner, D. (1977). Access to the internal lexicon.
In S. Dornic (Ed.), Attention and performance V1. London: Academic Press.
Coltheart, M., Curtis, B., Atkins, P., & Hailer, M. (1993). Models of reading aloud: Dual-
route and Parallel-distributed-processing approaches. Psychological Review, 100, 589-608.
Connine, C.M., Mullennix, J., Shemoff~ E., & Yelen, J. (1990). Word familiarity and frequency
in visual and auditory word recognition. Journal of Experimental Psychology: Learning
Memory and Cognition, 16, 1084-1096.
Derrington, A., & Lennie, P. (1984). Spatial and temporal contrast sensitivities of neurons in
the lateral geniculate nucleus of the macaque. Journal of Physiology, 357, 219-240.
DeValois, ILL., & DeValois, I~I~ (1987). Spatial vision. New York: Oxford University Press.
Dobbs, A.R., Friedman, A., & Lloyd, J. (1985). Frequency effects in lexical decisions: A test
of the verification model. Journal of Experimental Psychology: Human Perception and
Performance, 11, 81-92.
Fera, P., & Besner, D. (1992). The process of lexical decision: More words about a parallel
distributed processing model. Journal of Experimental Psychology: Learning, Memory,
and Cognition, 18, 749-764.
Fisher, D.L., Fisk, A.D., & Duffy, S.A. (in press). Why latent models are needed to test
hypotheses about the flowing of word and language processes in older adults. In P. Allen
and T. Bashore (Eds.), Age differences in word and language processing. New York:
North-Holland.
Fisk, A.D., & Fisher, D.L. (1994). Brinley plots and theories of aging: The explicit, muddled,
and implicit debates. Journal of Gerontology: Psychological Sciences, 49, P81-P89.
Fisk, A.D., Fisher, D.L., & Rodgers, W.A. (1992). General slowing alone cannot explain age-
related search effects: A reply to Cerella (1991). Journal of Experimental Psychology:
General, 121, 73-78.
Fisk, A.D., & Kodgers, W.A. (1991). Toward an understanding of age-related memory and
visual search effects. Journal of Experimental Psychology: General, 120, 131-149.
Fodor, J. A. (1983). The modularity of the mind. Cambridge, MA: MIT Press.
Forster, I(I. (1976). Accessing the mental lexicon. In 1LJ. Wales & E. Walker (Eds.), New
approaches to language mechanisms (pp. 89-109). Amsterdam: North-Holland.
Forster, I~I. (1979). levels of processing and the structure of the language processor. In W.E.
Cooper & E. Walker (Eds.), Sentence processing: Psycholinguistic studies presented to
Merrill Garrett (pp. 27-85). Hillsdale, NJ: Erlbaum.
Forster, I~I. (1989). Basic issues in lexical processing. In W. Marslen-Wilson (Ed.), Lexical
representation andprocess (pp. 75-107). Cambridge, MA: MIT Press.
Frederiksen, J.1L (1978). Assessment oflexical decoding and lexical skills and their relation to
reading proficiency. In A.M. Lesgold, J.W. Pellegrino, S.D. Fokkema, and 1L Glaser
(Eds.), Cognitive psychology and instruction (pp. 153-169). New York: Plenum
Geary, D.C., & Wiley, J.G. (1991). Cognitive addition: Strategy choice and speed-of-
processing differences in young and elderly adults. Psychology and Aging, 6, 474-483.
Geary, D.C., Frensch, P., & Wiley, J.G. (1993). Simple and complex mental subtraction:
Strategy choice and speed-of-processing differences in younger and older adults.
Gibson, E. (1965). Learning to read. Science, 148, 1066-1072.
Graham, N. (1981). The visual system does a crude Fourier analysis of patterns. SIAM-AMS
Proceedings, 13, 1-16.
Hansen, D., & Rodgers, T.S. (1968). An exploration ofpsycholinguistic units in initial reading.
In I~S. Goodman (Ed.), The psycholinguistic nature of the reading process. Detroit:
Wayne State University Press.
Healy, A.F. (1976). Detection errors on the word "the": Evidence for reading units larger than
letters. Journal of Experimental Psychology: Human Perception and Performance, 2,
235-242.
Healy, A. F., Oliver, W. L., & McNamara, T. P. (1987). Detecting letters in continuous text:
Effects of display size. Journal of Experimental Psychology: Human Perception and
Healy, A.F., Conboy, G.L., Drewnowski, A. (1987). Characterizing the processing units of
reading: Effects of intra-and interword spaces in a letter detection task. In B. Britton and
S. Glynn (Eds.), Executive control processes in reading (pp. 279-296). Hillsdale, NJ:
Erlbaum.
Hebb, D.O. (1949). The organization of behavior. New York: Wiley.
Howard, D.V., Shaw, R.J., & Heisey, J.G. (1986). Aging and the time course of semantic
Humphreys, G.W., EveR, L.J., & Quinlan, P.T. (1990). Orthographic processing in visual
word identification. Cognitive Psychology, 22, 517-560.
Johnson, N.F. (1975). On the function of letters in word id~atification: Some data and a
preliminary model. Journal of Verbal Learning and Verbal Behavior, 14, 17-29.
Johnson, N. F., Allen, P. A., & Strand, T. L. (1989). On the role of word frequency in the
detection of component letters. Memory & Cognition, 17, 474-482.
Joordens, S., & Besner, D. (1994). When banking on meaning is not (yet) money in the bank:
Explorations in connectionist modeling. Journal of Experimental Psychology: Learning,
Memory, and Cognition, 20, 1051-1062.
Kellas, G., Paul, S.T., & Vu, H. (in press). Aging and language performance: From isolated
words to multiple sentence contexts. In P. Allen & T. Bashore (Eds.), Age differences in
word and language processing. New York: North-Holland.
Keppel, G. (1991). Design and analysis: A researcher's handbook. Englewood Cliffs, NJ:
Prentice-Hall.
Kliegl, 1L (1994). Individual differences in cognitive aging: An experimental perspective.
Paper presented at the 1994 Cognitive Aging Conference.
Krueger, L.E. (1989). Detection of intraword and interword letter repetition: A test of the
word unitization hypothesis. Memory & Cognition, 17, 48-57.
Krueger, L.E., & Allen, P.A. (1987). Same-different judgments of foveal and parafoveal letter
pairs by older adults. Perception & Psychophysics, 41, 329-334.
Kucera, H., & Francis, W. N. (1967). Computational analysis of present-day American
English. Providence, RI: Brown University Press.
Laver, G.D., & Burke, D.M. (1993). Why do semantic priming effects increase in old age?
Levelt, W., Schiefers, H., Vorberg, D., Meyer, A., Pechman, T., & Havinga, J. (1991). The
time course of lexical access in speech production: A study of picture naming.
Lima, S.D., & Pollatsek, A. (1983). Lexical access via an orthographic code? The Basic
Orthographic Syllabic Structure (BOSS) reconsidered. Journal of Verbal Learning and
Verbal Behavior, 22, 310-332.
Lima, S.D., Hale, S., & Myerson, J. (1991). How general is general slowing? Evidence from
the lexical domain. Psychology and Aging, 6, 416-425.
Long, G.M., & Crambert, R.F. (1990). The nature and basis of age related changes in dynamic
visual acuity. Psychology and Aging, 5, 138-143.
Madden, D.J. (1986). Adult age differences in the attentional capacity demands of visual
search. Cognitive Development, 1, 335-363.
Madden, D.J. (1989). Visual word identification and age-related slowing. Cognitive
Develpoment, 4, 1-29.
Madden, D.J. (1990). Adult age differences in attentional selectivity and capacity. European
Journal of Cognitive Psychology, 2, 229-252.
Madden, D.J. (1992). Four to ten milliseconds per year: Age-related flowing of visual word
recognition. Journal of Gerontology: Psychological Sciences, 47, P59-P68.
Madden, D.J., Pierce, T.W., & Allen, P.A. (1992). Adult age differences in attentional
allocation. Psychology and Aging, 7, 594-601.
Madden, D.J., Pierce, T.W., & Allen, P.A. (1993). Age-related flowing and the time course of
semantic priming in visual word recognition. Psychology and Aging, 8, 490-507.
Madden, D.J., & Allen, P.A. (in press). Aging and the speed/accuracy relation in visual search:
Evidence for an accumulator model. Optometry and Vision Science.
Marr, D. (1982). Vision. San Francisco: Freeman.
Masson, M.E.J. (1986). Identification of typographically transformed words: Instance-based
skill acquisition. Journal of Experimental Psychology: Learning, Memory, and Cognition,
12, 479-488.
McClelland, G.H., & Judd, C.M. (1993). Statistical difficulties of detecting interactions and
moderator effects. Psychological Bulletin, 114, 376-390.
McClelland, J.L., & Rumelhart, D.E. (1981). An interactive activation model of context effects
in letter perception: Part I. An account of basic findings. Psychological Review, 88, 375-
407.
McClelland, J.L., & Rumelhart, D.E. (1986). Parallel distributed processing: Explorations in
the microstructure of cognition, VoL 2: Psychological and biological models. Cambridge,
MA: MIT Press.
Minsky, M. (1967). Computation: finite and infinite machines. Englewood Cliffs, NJ:
Prentice-Hall.
MonselL S., Doyle, M.C., & Haggard, P.N. (1989). Effects of frequency on visual word
recognition tasks: Where are they? Journal of Experimental Psychology: General, 118,
43-71.
Morton. J. (1969). Interaction of information in word recognition. Psychological Review, 76,
165-178.
Myerson, J., Hale, S., Wagstaff~ D., Poon, L.W., Smith, G.A. (1990). The information loss
475-487.
Myerson, J., Ferraro, K, Hale, S., Lima, S. (1992). General slowing in semantic priming and
word recognition. Psychology and Aging, 7, 257-270.
Myerson, J., Wagsta~ D., & Hale, S. (1994). Brinley plots, explained variance, and the
Neely, J. (1991). Semantic priming effects in visual word recognition: A selective review of
current findings and theories. In D. Besner and G.W. Humphreys (Eds.), Basic processes
in reading: Visual word recognition. (pp. 264-336). Hillsdale, NJ: Erlbaunl
Neisser, U. (1967). Cognitive psychology. New York: Appleton-Century-Crofts.
Nelson, D., Schreiber, T., & McEvoy, C. (1992). Processing implicit and explicit
representations. Psychological Review, 99, 322-348.
Ober, B.A., & Shenaut, G.I~ (in press). Semantic priming in Alzheimer's disease: Meta-
analysis and theoretical evaluation. In P. Allen & T. Bashore (Eds.), Age differences in
word and language processing. New York: North-Holland.
Owsley, C., Sekular, K, & Siemsen, D. (1983). Contrast sensitivity throughout adulthood.
Vision Research, 23, 689-699.
Paap, I~K, Newsome, S.L., & Noel, KW. (i984). Word shape's in poor shape for the race to
the lexicon. Journal of Experimental Psychology: Human Perception and Performance,
10, 413-428.
Paap, K.K, Newsome, S.L., McDonald, J.E., Schvaneveldt, KW. (1982). An activation-
verification model for letter and word recognition. Psychological Review, 89, 573-594.
Pashler, H. (1987). Target-distractor discriminability in visual search. Perception &

Psychophysics, 41, 285-291.
Perfect, J.T. (1994). What can Brinley plots tell us about cognitive aging? Journal of
Gerontology: Psychological Sciences, 49, P60-P64.
Pillsbury, W.B. (1897). A study in apperception. American Journal of Psychology, 8, 315-393.
Pinker, S., & Prince, A. (1988). On language and connectionism: Analysis of a parallel
distributed processing model of language acquisition. Cognition, 28, 73-193.
Pitts, D.G. (1982). Visual acuity as a function of age. Journal of the American Optometric
Association, 53, 117-124.
Rumelhart, D.E., & McClelland, J.L. (1982). An interactive activation model of context effects
in letter perception: Part I. An account of basic findings. Psychological Review, 89, 60-94.
Rumelhart, D.E., & McClelland, J.L. (1986). Parallel distributed processing: Explorations in
the microstructure of cognition, Vol. 1: Foundations. Cambridge, MA: MIT Press.
Salthouse, T.A. (1985). Speed of behavior and its implications for cognition. In J.E. Birren and
& K.W. Schaie (Eds.), Handbook of the psychology of aging (2nd Ed., pp. 400-426). New
York: Van Nostrand Reinhold.
Salthouse, T.A. (1991). Mediation of adult age differences in cognition by reductions in
working memory and speed of processing. Psychological Science, 2, 179-183.
Salthouse, T.A., & Coon, V. (1994). Interpretation of differential deficits: The case of aging
and mental arithmetic. Journal of Experimental Psychology: Learning, Memory, and
Cognition, 20, 1172-1182.
Seidenberg, M.S., & McClelland, J.L. (1989). A distributed, developmental model of word
recognition and naming. Psychological Review, 96, 523-568.
Selfridge, O.G. Pandemonium: A paradigm for learning. (1959). In Symposium on the
mechanization of though processes. London: HM Stationary Office.
Stadtlander, L. (in press). Age differences in orthographic and frequency neighborhoods. In P.
Allen and T. Bashore (Eds.), Age differences in word and language processing. New
York: North-Holland.
Steinberg, S. (1966). High-speed scanning in human memory. Science, 153, 652-654.
Stine, E. (in press). Aging and the distribution of resources in working memory. In P. Allen &
T. Bashore (Eds.), Age differences in word and language processing. New York: North-
Holland.
Taft, M. (1979). Lexical access via an orthographic code: The Basic Orthographic Syllable
Structure (BOSS). Journal of Verbal Laerning and Verbal Behavior, 18, 21-39.
Taft, M., & Forster, K.I. (1976). Lexical storage and retrieval of polymorphemic and
polysyllabic words. Journal of Verbal Learning and Verbal Behavior, 15, 607-620.
Tainturier, M-J., Trembley, M., & Lecours, A.R. (1989). Aging and the word frequency effect.
Neuropsychologica, 27, 1197-1203.
Tainturier, M-J., Trembley, M., & Lecours, A.R. (1992). Educational level and the word
frequency effect: A lexical decision investigation. Brain and Language, 43, 460-474.
Turing, A.M. (1936). On computable numbers, with an application to the
Entscheidungsproblen~ Proceedings of the London Mathematics Society (Series 2), 230-
265.
Van Essen, D., Anderson, C., & Felleman, D. (1992). Information processing in the primate
visual system: An integrated systems perspective. Science, 419-423.
Weale, R.A. (1986). Aging and normal vision. Vision Research, 26, 1507-1512.
Wheeler, (1970). Processes in word recognition. Cognitive Psychology, 1, 59-85.

Welford, A. (1958). Ageing and human skill. Oxford: Oxford University Press.
Woodworth, 1LS. (1938). Experimental Psychology. New York: Holt.
Age differences in orthographic and frequency neighborhoods

Leann M. Stadtlander*
Montana State University, Bozeman, MT
Previous researchers have examined young adults' responses in a lexical decision task to
words and nonwords which are similar to other words (i.e., the "neighborhood" of the word).
They have also explored the effect of high- and low-frequency stimulus words and high- and
low-frequency in the neighborhood of the target word. The present study extended this work
by examining differences in the responses ofyotmger (age 18-32) and older adults (age 60-75)
to items which varied in the frequency of the target word as well as the size and frequency of
the words in the target's neighborhood. Older adults were found to respond in a qualitatively
different manner than the younger adults. This finding was interpreted through an internal noise
model of aging.
1. INTRODUCTION
The average adult reader of English, has at least 50,000 words in their vocabulary
(Monsell, Doyle & Haggard, 1989). These 50,000 words are constructed from the set of 26
letters in the alphabet. Further, all of the 26 letters are constructed from a small set of features
(Andrews, 1992; Gibson, 1969). Logically, many words must share features and letters with
other words. How is it that the reader selects the correct word (i.e., lexical entry) from the
many possible candidates in memory?
Selection of the correct lexical entry, from all other possible candidates, is one of the
most fimdamental issues in models of word recognition. The selection process provides an
insight into how lexical memory (i.e., the lexicon) is structured and organized. All models or
theories of the recognition process speculate on the way that the lexicon is organized
(typically, by visual or phonological features), and in turn, suggest a method of finding a
specific item that has been "filed" in the lexicon. Such theoretical frameworks provide
hypothetical points in the selection process that may be tested experimentally. For example, if
the mental lexicon is organized by the visual or phonological features of the lexical entries,
then an experimenter should be able to influence a subject's response to a stimulus word which
has similar features to many lexical entries in memory.
This logic was applied in an experiment by Coltheart, Davelaar, Jonasson, and Besner
(1977, Experiment 2). Coltheart et al. controlled the similarity of the features of a series of
word and nonword stimuli which varied by what they called "N" or "orthographic
neighborhood" size. "N" was defined as the number of English words that could be produced
* Author Notes: The author would like to offer her appreciation to Lester E. Krueger for his contributionto the
design of this experiment. My thanks also to Susan Boardman and Emily Hoffman for assistance in the testing
of subjects.
Age differences in orthographic and frequency neighbourhoods 73
by replacing one letter, while preserving letter position for all other letters. The final stimulus
list consisted of words and nonwords with either a high or low value of N.
Coltheart et al. (1977) used a lexical decision task (LDT). In this task the subject is
required to decide if a stimulus item, shown on a computer screen, is a word or a nonword.
The authors found an effect of N only for nonword trials; that is, having many similar (word)
neighbors considerably slowed the "nonword" response. ~There was no effect for words, which
led Coltheart et al. to conclude that the number of items physically similar to a given word is
not related to the response time (RT) to reco~ize it as a word.
However, new evidence suggests that other properties of a neighborhood do affect
visual word recognition.* Andrews (1989; 1992) reported a study in which not only
neighborhood size was controlled, but also, the frequency of the stimulus word, i.e., the
"target" word. Target words were either common (i.e., high-frequency) words or were less
common (i.e., low-frequency) words. Andrews, using a LDT, found that low-frequency
target words with large neighborhoods resulted in a faster RT than those from small
neighborhoods. No neighborhood size effect was evident when the target items were high-
frequency words.
Additional data of Grainger (1990; Grainger, O'Regan, Jacobs, & Segui, 1989), are
suggestive of an even more complicated picture. Grainger (1990), using large
neighborhoods with low- and medium-frequency target words in a LDT, developed four
classes of words: (1) words with zero neighbors; (2) words with neighbors of only lower
frequencies; (3) words with one neighbor of higher frequency; (4) words with more than
one neighbor of higher frequency. Grainger found that when the target had one or more
neighbors of higher frequency, RT was slowed.
Although these studies used different types of stimuli, some consistencies between
the experiments are apparent. Consistent with both experiments are the class of stimuli
"large neighborhoods with low-frequency target words." In Andrews' (1989; 1992)
experiments these items were found to result in a faster RT when compared to similar target
words with small neighborhoods. Grainger (1990), using this class of items, found that
items within the target word's neighborhood could be an influencing factor. If a single
neighbor was of higher frequency than the target word, RT was slowed compared to a
neighborhood of all low-frequency words. These findings are important in how we
conceptualize visual word recognition. The data indicate that at some point in the search for
a low-frequency word in the mental lexicon, many similar low-frequency words do not
hinder the recognition process while words of higher frequency slow the process. It would
seem appropriate at this point to examine the locus of the "typical" frequency effect in
models of visual word recognition.
1.1. Frequency Effects in the Mental Lexicon
Many theories of word recognition assume that the mental lexicon is organized as a
function of word frequency (e.g., Cart & Pollatsek, 1985; Forger, 1976, 1979; Monsell et
al., 1989; Morton, 1969). Such models assume that visual features of stimuli are matched to
representations stored in the mental lexicon. Further, the models suggest that word
Neighborhood effects have also been reported in studies of: accuracyof identification of masked words (Luce,
1986), children's naming (Laxon, Coltheart, & Keating, 1988), and naming latencyfor German words
(Gunther & Greese, 1985; Scheerer, 1987).
74 L.M. Stadtlander
frequency influences access to the items in the lexicon, by permitting easier access to more
common words (i.e., ones of high-frequency in a language) than to less common words.
Thus, it is generally assumed that the locus of the word frequency effect is at the lexical
access stage in processing (though see Balota & Chumbley, 1984, for a contrary view).
Since the first use of the LDT, RT has been shown to be a sensitive measure of
frequency (Forster & Chambers, 1973; Rubenstein, Garfield, & Millikan, 1970; Taft, 1979;
Whaley, 1978). Thus, common words (i.e., ones of high-frequency in a language), such as
door, are responded to more quickly in the LDT than are uncommon words (i.e., ones of
low-frequency in a language), such as cask.
Taking into account the information gained from Andrews' (1989; 1992) and
Grainger's (1990) studies reported above, in which it became apparent that at some point in
the search for a low-frequency word in the mental lexicon, similar words of higher
frequency slow the process, a clearer picture of the recognition process evolves. We can
conceptualize the recognition process, through a "genetic" model, as one in which a
continuum of activation is present. First, the stimulus word is encoded into a set of visual
features. Second, words composed of similar features are activated, with more common
items more strongly activated than less common ones. Based upon the Grainger (1990)
work, it appears that the high-frequency neighbors, in some way interfere with the
recognition of a lower frequency word. Let us now take this knowledge and apply it to the
issue of age-related changes in lexical processing.
1.2. Age-Related Changes in Processing
Several LDT studies, in which word frequency was manipulated as a function of adult
age, have been previously reported (e.g., Allen, Madden, & Crozier, 1991; Allen, Madden,
Webber, & Groth, 1993; Bowles & Pooh, 1981). These studies found no age differences in
word frequency; i.e., both age groups showed a frequency effect, whereby, high-frequency
words were recognized faster than low-frequency. Allen et al. (1991) also reported a longer
response time but less errors for older adults, as compared to younger adults, in a standard
LDT (see also, Bowles & Pooh, 1985; Cerella & Fozard, 1984; Mueller, Kausler, & Faherty,
1980; Waugh & Barr, 1982). However, Allen et al. found no qualitative difference in response
to different frequency items between the two age groups.
Although older adults appear to show similar frequency effects when compared with
younger adults, they nonetheless, consistently take longer to respond than do younger adults.
Two theoretical views have been suggested to explain age-related differences in response
speed and accuracy. The first view, is that with age there is an increase in neural (internal)
noise (Allen, Namazi, Patterson, Crozier, & Groth, 1992; Cremer & Zeet~ 1987; Welford,
1958).
Many previous researchers have proposed that aging increases the level of spurious
neural activity or internal noise, thereby reducing the signal-to-noise ratio and producing age-
related deficits (Crossman & Szafran, 1956; Gregory, 1959; Krueger & Allen, 1987; Layton,
1975; Vickers, Nettlebeck, & Willson, 1972; Welford, 1965, 1977). Internal noise is assumed
to randomly change visual features during perception (Krueger, 1978; Krueger & Allen, 1987).
Krueger (1978) postulated three dements in a theory of internal noise. We can apply
these dements to the issue of processing words which differ in the size of their orthographic
neighborhoods.
1. The perceiver bases the decision on how many mismatching features a difference
counter records. Internal noise should affect input from the stimulus item by introducing
random features in the perceptual image of the stimulus item A theory of internal noise
suggests that the perceiver must compare differences (i.e., a difference counter) between the
lexical entry of the item in memory and the developing perceptual image.
2. Internal noise affects the comparison process. This element suggests that internal
noise affects the process of selecting the correct lexical entry from similar entries, by increasing
the potential candidates or neighbors. This leads to the hypothesis that as the size of the
neighborhood increases (i.e., for a large neighborhood as compared to a small neighborhood) it
should increase the difficulty of selecting the correct entry.
3. I f the difference count is not sufficiently high or low, the decision is postponed so
that a second or third glance may be made. Krueger (1978) theorized that the perceiver
continually resamples or rechecks the stimulus until sufficient evidence has been acquired. This
element suggests that if internal noise does increase with age, then older individuals should on
average, take longer to respond (which as discussed earlier, is indeed the case) in order to
recheck the stimulus. Further, although older adults may be slower in their responses, they may
also be more accurate due to the rechecking step.
The second theoretical view, is that with age there is a general slowing in processing
(Birren, 1974; Cerella & Fozard, 1984; Salthouse, 1985; also see Hasher & Zacks, 1988).
There are two subtly different versions of cognitive slowing, which can be labeled the
strong and weak versions (Hartley, 1992). A strong theory specifies a physiological
mechanism that is responsible for the slowing, such as uniform slowing of synaptic
transmission or information loss at each transmission (Myerson, Hale, Wagstafl~ Poon, &
Smith, 1990). This theoretical viewpoint would suggest that there should be no qualitative
change with age. In effect, the responses have been mathematically transformed resulting in
an overall slower response (Allen et al., 1993). According to this view, errors should be
equivalent in the two age groups. By contrast, a weak theory of slowing does not specify a
mechanism, but rather simply attempts to identify a function (i.e., a mathematical
transformation) that provides a good fit to the relationship between the younger and older
adults' performance. Thus, it provides a description, not an explanation.
1.3. The Present Experiment
In the present experiment, a series of stimulus items were developed which varied in
the l) frequency of the target items (high- or low-frequency), 2) frequency of the
neighborhood (high-frequency or low-frequency), and 3) size of the neighborhood ( large or
small; and is equivalent to Coltheart's "N"). These categories of items should permit a finer
examination of the interactions of age, frequency, and size of the neighborhood in a LDT.
Predictions. The internal noise model suggests that internal noise affects the process
of selecting the correct lexical entry from similar entries by increasing the potential
candidates or neighbors. This suggestion leads to the hypothesis that as the size of the
neighborhood increases (i.e., for a large neighborhood as compared to a small
neighborhood) the difficulty of selecting the correct entry should also increase. This would
predict that older adults should have a great deal of problems with large neighborhoods,
particularly for the conditions with a low-frequency target word and high-frequency
neighbors. It should be difficult to resolve these items, suggesting a longer RT should be
76 L.M. Stadtlander
evident. Nonword trials with large neighborhoods should also be difficult to resolve,
particularly for older adults due to the presence of many similar words requiring the
opposite response, again, suggesting a longer RT and perhaps more errors.
The slowing model suggests that older adults should be considerably slower than the
younger adults as predicted by a single best-fitting slowing fimction. However both age
groups should be approximately equivalent in errors. In order to examine the effect of
slowing, a method proposed by Madden, Pierce and Allen (1992) will be used. In this
method a best-fitting regression function is determined between the older and younger
adults. Assuming a si~ificant r 2 is found, the regression ftmction is then used to transform
the younger adults' raw data. An analysis of variance (ANOVA) is then performed on the
data (transformed for younger adults but untransformed for older adults). Because the
younger adults' raw latencies have been transformed by the function representing
generalized slowing, siL_,nificant Age x Condition interactions for this ANOVA would
represent those effects beyond a slowing model (Allen et al., 1993).
2. M E T H O D
Subjects. The younger subjects were volunteers from Introductory Psychology classes
at Montana State University. Older subjects were volunteers from the Bozeman, Montana
community, contacted through ads in newsletters and the local Senior Citizen Center. All
subjects were required to have visual acuity of at least 20/40 (corrected). Fiiteen older (M
age = 67.0 years; 7 males and 8 females) and 15 younger (M age = 19.2 years; 8 males and 7
females) subjects participated. Older subjects received an average of 65 on the Vocabulary
Subscale and 17 on the Information Subscale of the WAIS-R and had completed an average of
16 years of schooling. The younger subjects averaged 43 on the Vocabulary Subscale and 12
on the Information Subscale and had completed an average of 13 years of schooling.
Stimulus materials. Lowercase letters were presented on a 486 DEC color computer
screen. The letters, were presented as thin, illuminated lines on a dark screen. All of the 240
words contained five letters and all were presented once to each subject. The word lists
contained 43 instances of items with repeated letters (e.g., guess, sales); the nonword list was
equated for this factor.
The word lists were devised relative to three criteria: 1) frequency (based upon Kucera
& Francis, 1967, and Thorndike & Lorge, 1944, norms) of the target word (high-frequency
target [HFT] vs. low-frequency target [LFT]); 2) frequency of the neighborhood members of
the target word (two or more neighbors of higher-frequency [HFN], all low-frequency
neighbors [LFI~); 3) size of the target word's neighborhood (large or small). High-frequency
words occurred more than 50 times per million words in printed English ("A" in Thorndike &
Lorge norms), whereas, low-frequency words occurred less than 15 per million. The mean
frequency for high-frequency words was 213.48 per million, and for low-frequency words was
6.05 per million. The criterion for large-neighborhood items was that at least five different
words could be formed by changing single letters; the mean size for large neighborhoods was
8.09. For small-neighborhood items the criterion was that no more than three such alternatives
could be formed; the mean size for small neighborhoods was 1.83.
The neighborhoods were constructed primarily through the use of the WordPerfect
spell checker. Specifically, in this word-processing package, a "wild card" character can be
used to replace a letter in a word. The spelling-check program then lists all words from its
20,000-word lexicon whose letters are positionally consistent within the word. The total
number of consistent words across all positions constitutes the neighborhood for a particular
word, as defined in this experiment. For example, the target word blank would generate a
neighborhood consisting of." clank, flank, plank, blink, black, and bland.
Nonword lists were similarly devised, based upon the frequency of the (word)
neighborhood of the nonword (high-frequency neighborhood [HFN] or low-frequency
neighborhood [LFI~), and the size of the (word) neighborhood (large or small). As an
example, the target nonword bason would generate a neighborhood consisting of jason,
mason, bison, bacon, baron, baton, and basin.
The 480 regular trials (240 words, 240 nonwords) were randomly intermixed, and
formed into 16 blocks of 30 trials each. Two practice trials preceded each block, and there was
an initial practice block, resulting in a total of 544 trials. Four different random orderings of
trials were used in the experiment.
Procedure. On each trial a fixation mark appeared alone for .7 sec, and then the word
or nonword appeared just above the fixation mark until a response was made. Subjects were
instructed to respond as rapidly as possible without sacrificing accuracy. Half of the subjects
pressed a right-hand button if the target was a word, and the leit-hand button if it was a
nonword. The other half of the subjects had the reverse hand assignment.
3. RESULTS
As is typical in lexical decision tasks, subjects responded faster to words (M = 720

ms) than to nonwords (M = 912 ms; F[1,28] = 20.95, p < .0001). Words also resulted in
fewer errors (M = 7.0%) than nonwords ( M = 12.72%; F[1,28] = 33.49, p < .0001).
3.1. Analyses of Words
As is consistent with Allen et al. (1991), older adults had a longer response time but
had less errors. In a 2 x 2 x 2 x 2 Analysis of Variance (ANOVA), which examined age
(young vs. old) by size of neighborhood (large vs. small) by target word frequency (low vs.
high) by frequency of the word neighborhood (low vs. high); a main effect of age was evident
for RT (F[1,28] = 6.379,/2 < .01]. Older subjects responded slower (M = 766 ms) than
younger subjects (M = 673.5). Older subjects were also more accurate than younger subjects
(M's : 5.4% and 18.8% respectively; F[1,28] = 7.31,p < .01).
There was also an effect of target frequency, whereby, low-frequency target items
were responded to slower (M = 747 ms) than high-frequency target items (M = 693 ms;
F[1,28] = 91.59, p < .0001). A (non-significant) interaction with age was evident for RT
(older adults / high frequency target = 735 ms; older adults / low frequency target = 784
ms; younger adults / high frequency target = 627 ms; younger adults / low frequency target
= 701 ms; F[1,28] = 3.53,p = .07).
78 L.M. Stadtlander
Additionally, low-frequency (M = 10.1%) items generated more errors than did

high-frequency (M = 4.05%; F[1,28] = 33.49, p < .0001) items. However, contrary to Allen
et al. (1991) who found no qualitative difference between the two age groups for target
frequency, in the present experiment an interaction was evident for errors (F[ 1,28] = 6.26,
Figure 1. Percent Errors (PE) as a function of age (young vs. old) and frequency of target word (low vs.
high).
p < .01. As shown in Figure 1, a greater difference was present between the two age groups
for low-frequency target items as opposed to high-frequency targets.
There was no main effect of neighborhood size (large vs. small; p > . 10) for RT or
errors, nor an interaction with age for RT. However, an interaction was present for errors
between neighborhood size and age (F[ 1,28] = 7.61, p < .01). Whereby, older adults were
more accurate on small neighborhoods (4.9% vs. 6.2% for large neighborhoods) as
compared to young adults (7.5% for small neighborhoods vs. 7.8% for large
neighborhoods).
A priori analyses, based upon Andrews' (1992) work, were conducted for the two
age groups which examined neighborhood size and frequency of the target word and
neighborhood (see Tables 1 and 2). Differences were found for RT and errors between large
and small neighborhoods on items with low-frequency target words and low-frequency
neighborhoods for the older adults (RT: large neighborhoods = 792.35 ms, small
neighborhoods = 756.63 ms; t[14] = 2.39, p < .05; PE: large neighborhoods = 1.67%, small
neighborhoods = 1.07%; t[14] = 2.08, p < .05). Note, however, that these findings are in
the opposite direction from Andrews' data with young adults. Andrews reported large
neighborhoods resulted in faster RT.
For young adults, differences were found for RT between large and small
neighborhoods on items with high-frequency target words and low-frequency
Table 1.
Response Time (RT) and Percent Errors (PE) for Older Adults by Neighborhood Size and
Frequency of Target Word and Neighborhood; Standard Error in Parentheses.
Older Adults Large neighborhood Small NeigJaborhood

Low-Frequency Target / RT:792.34ms (34.00)** RT:756.63ms (33.82)**
Low-Frequency PE: 8.3 (2.8)* PE: 5.3 (1.3)*
Neighborhood
Low-Frequency Target / RT: 781.29 ms (35.42) RT: 818.79 ms (45.56)
High-Frequency PE: 7.65 (1.8) PE: 7.70 (1.7)
Neighborhood
High-Frequency Target / RT: 730.06 ms (37.19) RT: 734.63 ms (33.09)
Low-Frequency PE: 3.3 (1.5) PE: 2.0 (1.1)
Neighborhood
Neighborhood
* p < .05 **p<.01
neighborhoods (RT: large neighborhoods = 605.23 ms, small neighborhoods = 631.79 ms;
t[14] = 2.38, p < .05). For errors, young adults showed differences between large and small
neighborhoods on items with low-frequency target words and high-frequency
neighborhoods (PE: large neighborhoods = 3.85%, small neighborhoods = 2.14%; t[14] =
2.96, p < .01).
gemming to the overall analysis for words, no effect of neighborhood frequency
was present for RT or errors. Nor were interactions evident with age.
There was no interaction between age and the target word and neighborhood
frequency (low-frequency target with high-frequency neighborhood, high-frequency target
with high-frequency neighborhood, high-frequency target with low-frequency
_
neighborhood, and low-frequency target with low-frequency neighborhood). There was

however, an effect of the target/neighborhood frequency (F[3,112] = 3.72, p < .01). Post-
hoc t-tests revealed a difference only between low-frequency target with high- frequency
neighborhood (M = 706.52 m s ) v s , high-frequency target with low-frequency
neighborhood. (M = 633.21 ms; t[28] -- 2.56, p < .05). There were no further interactions.
Additional analyses, which examined each age separately, revealed an interesting bit
of information. For older adults, a one-way ANOVA across the four target word and
neighborhood frequency types (low-frequency target with high-frequency neighborhood,
high-frequency target with high-frequency neighborhood, high-frequency target with low-
frequency neighborhood, and low-frequency target with low-frequency neighborhood)
revealed no difference between them (F[3,28] = 0.85, ns). For young adults, a difference
was evident (F[3,28] = 4.41, p < .01). Post-hoc t-tests showed that the differences were
between the two high-frequency target sets and the low-frequency targets sets.
80 L.M. Stadtlander
Table 2.
Response Time (RT) and Percent Errors (PE) for Younger Adults by Neighborhood Size
and Frequency of Target Word and Nei~borhood; Standard Error in Parentheses.
Younger Adults Large neighborhood Small Neighborhood
Neighborhood
High-Frequency PE: 15.00 (2.6)* PE: 10.00 (1.9)*
Neighborhood
High-Frequency Target / RT: 605.23 ms (19.52)* RT: 631.79 ms (18.93)
Neighborhood
Neighborhood
* p < .05 **p <.01
An examination of errors revealed an effect of age (see Figure 2), whereby older
subjects were more accurate than younger subjects (F[1,112] = 4.86, p < .02). There was
also an effect of target/neighborhood frequency (F[3,112] =9.02, p < .001. Post-hoc t-tests
revealed differences for the following groups:
9 Low-frequency target with high-frequency neighborhood (M = 10%) vs. high-frequency
target with high-frequency neighborhood. (M = 4.3%; t[28] = 3.99, p < .01)
9 Low-frequency target with high-frequency neighborhood (M = 9.45%) vs. high-
frequency target with low-frequency neighborhood. (M = 3.5%; t[28] = 2.56, p < .05)
9 High-frequency target with high-frequency neighborhood ( M - 4.05%) vs. low-
frequency target with low-frequency neighborhood (34 = 8.25%; t[28] = 3.02, p < .01)
9 High-frequency target with low-frequency neighborhood (M = 3.5%) vs. low-frequency
target with low-frequency neighborhood (M = 8.25%; t[28] - 3.37, p < .01)
The interaction of these two variables did not reach significance (/713,112] = 1.87, p > .10).
However, analyses of the two age groups separately again showed that the effect
was primarily a result of the variance in the young adults' errors (older: F[ 1,28] - 2.58, p -
.06; younger:/711,28] = 7.63,p < .001).
Figure 2. Percent Errors (PE) as a function of age (young vs. old) and frequency of target word and
neighborhood.
3.2. Analyses of Nonwords
A 2 x 2 x 2 ANOVA examined age (older vs. younger adults) by size of the

nonword's neighborhood (large vs. small) by frequency of the neighborhood (high-
frequency vs. low-frequency). For RT, none of the effects reached si~ificance, p's > .10.
Thus, the age groups did not differ in their RT's nor was there an effect of the size or
frequency of the items in the neighborhood.
However, as shown in Figure 3, errors revealed a different pattern. An effect of age
was evident, in which older adults (M = 8.8%) were more accurate than younger adults (M
= 16.62%; F[1,28] = 9.18, p < .01). An effect was also present for neighborhood size.
Large neighborhoods generated more errors ( M = 15.6%) than did small neighborhoods (M
: 9.79%; F[1,28] = 5.17, p < .02). There was no effect of frequency of the neighborhood,
nor was there an interaction between the variables.
A single best-fiRing regression function was sought between the older and younger
adults' data (Allen et al., 1993). However, contrary to an aging model of generalized
slowing, the r 2 proved to be non-significant (p= .14, y = -73.43(x) + 854.72). This appears
to be due to the small amount of variance in the older adults' responses, as compared to the
wide range seen in the younger adults. Given that the best-fitting regression function was
not si~ificant, the transformed analysis (Madden et al., 1992) was not conducted. This is
because the lack of a reliable slowing function already indicates that no single linear slowing
function can account for the relationship between younger and older adults' latencies across
task conditions.
82 L.M. Stadtlander
Figure 3. Percent Errors (PE) as a function of age (young vs. old) and size of neighborhood for
nonwords.
4. DISCUSSION
Considerable new evidence emerged with the present experiment. Consistent with
Andrews (1989, 1992), neighborhood size appeared to be an important factor for low-
frequency target words, particularly in the case of older adults. However, in opposition to
Andrews' findings, older adults responded faster to small neighborhoods.
One result that was quite apparent, was the strong effect of target word ~equency.
Low-frequency words generated slower RT and more errors for both age groups as
compared to high-~equency words. These data provide evidence that the slower RT, of
older subjects, is not necessarily due to a speed-accuracy trade-off(Fitts, 1966).
One result that was quite surprising was the small amount of variance, for both RT
and errors, between the different word types for the older adults. There was little difference
between low-and high-frequency target words for the older subjects while considerable
difference for the younger ones. This suggests that the theoretical view of internal noise,
may indeed, have some validity. If older adults must wait to resolve internal noise through a
rechecking step, it may well reduce any differences between the different word types. An
advantage of this rechecking step may be that it provides the necessary delay to avoid the
many errors seen with the young adults on these types of stimuli. This was particularly
evident with the nonword stimuli, where again, older adults showed little difference
between large and small neighborhoods. However, the number of errors for young adults
for nonwords with large neighborhoods reached a startling 20%! This number of errors
suggests that the young adults may be relying on insufficient information to respond,
perhaps a heuristic may be involved, such as, many similar words = word response.
Let us reconsider the genetic conceptual model of the recognition process proposed
earlier, in which it was suggested that a continuum of activation is present. It was suggested
that first, the stimulus word is encoded into a set of visual features. Second, words
composed of similar features are activated, with more common items more strongly
Age differences in ort,'zographicandfrequency neighbourhoods 83
activated than less common ones. Further, based upon the Grainger (1990) work, it
appeared that the high-frequency neighbors, in some way, interfere with the recognition of a
lower frequency word.
One scenario feasible from the present data and consistent with the model above,
would be that for older adults, there may be less of a distinction between the activation
levels of low- and high-frequency words. If internal noise is indeed affecting the recognition
process, it would seem that it would be most prevalent during the activation stage, which
suggests that additional time would be required to recheck the stimulus input against items
in memory. If it were the case, that rechecking of the stimulus is required in order to reduce
the internal noise, then it may well be that the recognition process is fully completed before
making the lexical decision. Such a mechanism could easily account for the few errors seen
with the older adults. It may be that the differences present between older and younger
adults, are due to the younger subjects relying on less information (i.e., a heuristic) to make
their decision, as opposed to the older adult who must fully complete the recognition
process. Obviously, considerable research in this area is required.
In this empirical study, there was not a great deal of support for the overall slowing
model. This model predicted that older adults would be slower than younger adults (which
they were), but that there should be a similar pattern to the older and younger adults' data
for both RT and errors. However, a best-fitting slowing function for RT indicated that the
age groups differed qualitatively by the evidence of a non-si~ificant r 2. The age groups
differed on a number of elements, particularly for the error data. There seems to be no easy
way of resolving the issue based upon this model.
5. CONCLUSIONS
The present data support the idea that there is a qualitative difference in how older
and younger adults respond to words and nonwords which differ in the composition of their
neighborhoods. The older adults appear to wait to respond until processing is complete,
which is compatible with the internal noise model of aging. While younger adults appear to
respond before complete processing has taken place, resulting in the high error rates evident
in the present experiment.
REFERENCES
Allen, P.A., Madden, D.J., & Crozier L.C. (1991). Adult age differences in letter-level and
word-level processing. Psychology and Aging, 6, 261-271.
Allen, P.A., Madden, D.J., Weber, T.A. & Groth, K.E. (1993). Influence of age and processing
stage on visual word recognition. Psychology and Aging, 8, 274-282.
Allen, P.A., Namazi, ICH., Patterson, M.B., Crozier, L.C. & Groth, K.E. (1992). Impact of
adult age and Alzheimer's disease on levels of neural noise for letter matching, dournal of
Gerontology, 47, 344-349.
Andrews, S. (1989). Frequency and neighborhood size effects on lexical access: Activation or
search? Journal of Experimental Psychology: Learning, Memory, and Cognition, 15, 802-
814.
84 L.M. Stadtlander
Andrews, S. (1992). Frequency and neighborhood effects on lexical access: Lexical similarity
or orthographic redundancy? Journal of Experimental Psychology: Learning, Memory,
Balota, D.A. & Chumbley, J.I. (1984). Are lexical decisions a good measure of lexical access?
The role of word frequency effects in the neglected decision stage. Journal of
Birren, J.E. (1974). Psychophysiology and speed of response. American Psychologist, 29, 808-
815.
Bowles, N.L. & Poon, L. W. (1985). Aging and the retrieval of words in semantic memory.
Cart, T.H. & Pollatsek, A. (1985). Recognizing printed words: A look at current models. In D.
Besaer, T.G. Waller, & G.E. MacKinnon (Eels.), Reading research: Advances in theory
and practice (Vol. 5 pp. 1-82). San Diego, CA: Academic Press.
Cerella, J. & Fozard, J.L. (1984). Lexical access and age. Developmental Psychology, 20, 235-
243.
Coltheart, M., Davelaar, E., Jonasson, J.T., & Besner, D. (1977). Access to the internal
lexicon. In S. Domic (Ed.), Attention and performance VI (pp. 535-555). Hillsdale, NJ:
Erlbaum
Cremer, J. & Zeef~ E.J. (1987). What kind of noise increases with age? Journal of
Crossman, E.R.F.W. & Szafran, J. (1956). Changes with age in the speed of information
intake and discrimination. Experientia Supplementum, 4, 128-135.
FiRs, P.M. (1966). Cognitive aspects of information processing: m. Set for speed versus
accuracy. Journal of Experimental Psychology, 71, 849-857.
Forger, K.I. (1976). Accessing the mental lexicon. In ILL Wales & E. Walker (Eds.), New
approaches to language mechanisms (pp. 89-109). Am~erdam: North-Holland.
Forger, K.I. (1979). Levels of processing and the structure of the language processor. In
W.E. Cooper & E. Walker (Eds.), Sentence processing: Psycholinguistic studies
presented to Merrill Garrett (pp. 27-85). Hillsdale, NJ: Erlbaunl
Forger, K.I. & Chambers, S.M. (1973). Lexical access and naming time. Journal of Verbal
Learning and Verbal Behavior, 12, 627-635.
Gibson, E.J. (1969). Principles of pereeptual learning and development. NY: Appleton.
Grainger, J. (1990). Word frequency and neighborhood frequency effects in lexical decision
and naming. Journal of Memory and Language, 29, 228-244.
Grainger, J., O'Regan, J.IC, Jacobs, A.M., & Segui, J. (1989). On the role of competing word
units in visual word recognition. Perception and Psyehophysics, 47, 191-198.
Gregory, ILL. (1959). "Neurological noise" as a factor in aging. Proceedings of the Fourth
Congress of the International Association of Gerontology, 1, 314-324.
Gunther, H. & Greese, B. (1985). Lexical hermits and the pronunciation of visually presented
words. Forschungsberichte des Instimts fur Phonetik and Spaehliche Kommunikation des
universitats Munehen, 21, 25-52
Hartley, A.A. (1992). Attention. In F.I.M. Craik & T.A. Salthouse (Eds.). The handbook of
aging and cognition (pp. 3-50). Hillsdale, NJ: Erlbaum
Hasher, L. & Zacks, R.T. (1988). Working memory, comprehension, and aging: A review and
a new view. In G.H. Bower (Ed.), The psychology of learning and motivation: Advances
in research and theory. Vol. 22. NY: Academic Press.
Age differences in orthographicandfrequency neighbourhoods 85
Krueger, L.E. (1978). A theory of perceptual matching. Psychological Review, 85, 278-304.
Krueger, L.E. & Allen, P.A. (1987). Same-different judgements of foveal and parafoveal letter
pairs by older adults. Perception & Psychophysics, 41, 329-334.
Kucera, H. & Francis, W.N. (1967). Computational analysis of present-day American
English. Providence, RI: Brown Univ. Press.
Laxon, V.J., Coltheart, V., & Keating, C. (1988). Children find friendly words friendly too:
Words with many orthographic neighbors are easier to read and spell. British Journal of
Educational Psychology, 58, 103-119.
Layton, B. (1975). Perceptual noise and aging. Psychological Bulletin, 82, 875-883.
Luce, P. A. (1986). Neighborhoods of words in the mental lexicon. Research on Speech
Perception (Tech. Rep. No. 6). Bloomington: Indiana University.
Madden, D.J., Pierce, T.W. & Allen, P.A. (1992). Adult age differences in attentional
allocation during memory search. Psychology and Aging, 7, 594-601.
Monsell, S. Doyle, M.C. & Haggard, P.N. (1989). Effects of t~equency on visual word
recognition tasks: Where are they? Journal of Experimental Psychology: General, 118,
43-71.
Morton, J. (1969). Interaction of information in word recognition. Psychological Review, 76,
165-178.
Mueller, J.H., Kausler, D.H, & Faherty, A. (1980). Age and access time for different memory
codes. Experimental Aging Research, 6, 445-449.
Myerson, J., Hale, S., Wagstaff, D., Pooh, L.W., & Smith, G.A. (1990). The information loss
475-487.
Rubenstein, H., Garfield, L. & Millikan, J.A. (1970). Homographic entries in the internal
lexicon. Journal of Verbal Learning and Verbal Behavior, 9, 487-492.
Salthouse, T.A. (1985). Speed of behavior and its implications for cognition. In J.E. Birren &
K.E. Schaie (Eds.). Handbook of the psychology of aging (2nd ed., pp. 400-426). NY:
Van Nostrand Reinhold.
Scheerer, E. (1987). Visual word recognition in German. In D.A. Allport, D. Mackay, W.
Prinz, & E. Scheerer (Eds.), Language perception and production: Shared mechanism in
listening, speaking, reading, and writing (pp. 227-244).
Taft, M. (1979).Lexical Access via an orthographic code: The Basic Orthographic Syllabic
Structure (BOSS). Journal of Verbal Learning and Verbal Behavior, 18, 21-39.
Thorndike, E.L. & Lorge, I. (1944). The teacher's word book of 3 0, OOOwords. NY: Bureau of
Publications.
Vickers, D., Nettlebeck, T. & Willson, R.J. (1972). Perceptual indices of performance: The
measurement of"inspection time" and "noise" in the visual system. Perception, 1, 263-295.
Waugh, N.C. & Barr, 1LA. (1982). Encoding deficits in aging. In F.I.M. Craik & S. Treheb
(Eds.). Aging and cognitive processes. (pp. 183-190). NY: Plenum-
Welford, A.T. (1958). Aging and human skill. Oxford: Oxford University Press.
Welford, A.T. (1965). Performance, biological mechanisms and age: A theoretical sketch. In
A.T. Welford & J.E. Birren (Eds.), Behavior, aging and the nervous system (pp. 3-20).
Springfield, IL: Thomas
Welford, A.T. (1977). Motor performance. In J.E. Birren & K,W. Schaie (Eds.), Handbook of
the psychology of aging (pp. 450-496). NY: Van Nostrand Reinhold.
86 L.M. Stadtlander
Whaley, C.P. (1978). Word-nonword classification time. Journal of Verbal Learning and
Verbal Behavior, 17, 143-154.
9 1995 Elsevier Science B.V. All rights reserved. 87
Aging and language performance: From isolated words to multiple sentence

contexts
George Kellas~, Stephen T. Paulb, and Hoang Vu c
aUniversity of Kansas
bWashington University
CUniversity of Kansas
A review of age-related language research indicates that older individuals evidence a

decrement in language abilities when memory tasks are employed (Light, 1992). However,
because of the off-line nature of these measures, it is not possible to attribute this decline to
language comprehension. That is, the results from research using memory tasks may be due to
processes occurring at the time of initial encoding, during retrieval, or both. If our goal is to
examine changes in language comprehension with age, it is necessary to evaluate
comprehension processes at the moment of their occurrence or immediately thereafter.
Furthermore, in order to clarify how language is processed, it is necessary to accept that this
clarification is dependent upon a number of levels of analysis. During reading, for example,
words must be identified and associated with their meanings. The grammatical structure of
sentences must be analyzed to determine the semantic roles played by each word. The
sentences must be integrated into a coherent discourse representation. And, a mental model
representing the discourse must be constructed relative to our knowledge of the world. Only
by a comprehensive, on-line examination of these potentially age-related processes can we
begin to make strong claims regarding changes across the life-span.
Fundamental to the above issues is the question of the architecture of the language
processor. Is the language processor comprised of a series of independent processing stages
that require completion prior to fuU comprehension? Or, are the stages involved highly
interdependent such that each stage can influence and be influenced by other processing
stages? We begin our discussion with a brief tutorial on current views of the language
comprehension system
1. MODULAR VERSUS INTERACTIVE PROCESSOR
At present, there are two prominent positions regarding the organization of the
language processor. The more popular view, modularity, holds that the language processor is
autonomous, being comprised of multiple subsystems, or modules (e.g., lexicon, syntax,
Acknowledgments: Preparation of this manuscript was supported by Biomedical Research Grant #R15045
R05, awarded to George Kellas and Post-doctoral fellowship support from NIA Grant #AG00030 to Stephen T.
Paul. Correspondenceshould be addressed to George Kellas, Universityof Kansas, Department of Psychology,
426 Fraser Hall, Lawrence, Kansas, 66045-2160.
88 G. Kellas et al.
semantics), which function independently of one another to ultimately converge on

comprehension (Fodor, 1983). Processing within any given subsystem proceeds to completion
without influencing, or being influenced by, other modules. Once a module completes a task,
the output is then passed to the next processor in the sequence. Although this model appears
to require a serial processing of information, there is no requirement that processing within
modules must wait for output from a prior system Rather, parallel processing of information
is allowed so long as operations within any module currently under use are unaffected by and
unable to affect other modules that have yet to complete their tasks.
The modular approach contrasts with the view that processing at one level of the
system contributes to processes occurring simultaneously at other levels of the system (e.g.,
McClelland, 1987). This interactive-activation position allows processes to proceed in parallel
based on information from both bottom-up (i.e., the inputs to the system) and top-down (i.e.,
expectancies, world knowledge) sources. For example, when a sentence is processed, the
individual words that make up the sentence are subjected to bottom-up processing by way of
stimulus analysis. As each successive word is processed, its meaning will be integrated with
previous words. The end product is a specific pattern of activation representing the sentence.
This message level representation will serve as a top-down influence on the processing of
subsequent material. Thus, the integration of prior information (semantic and syntactic) can
place constraints on upcoming information (e.g., Taraban & McClelland, 1990).
Neither the interactive nor the modular position explicitly addresses changes in
performance with age. However, an examination of age differences might allow for a
preference of one model over the other to emerge. As a means of illustrating the underlying
differences between views, it will be useful to first elaborate specific models within both the
autonomous (modular) and interactive-activation frameworks.
From an early processing or on-line perspective, autonomous models are frequently
referred to as context-independent. In other words, access to a word meaning proceeds in the
same manner regardless (independent) of the context in which it occurs. This may seem
reasonable at first since words are typically thought of as having distinct dictionary-like
meanings. An intuitive expectation is that a word should have the same meaning no matter the
circumstances in which it is encountered. However, with English, many words frequently
encountered have distinctly unrelated interpretations. Consider the following sentences from
Johnson-Laird (1983, p. 128):
The pilot put the plane into a stall just before landing on the strip.
He got it out of it in time.
The sentences read easily enough and appear to make sense with little cognitive effort. A
quick glance through the unique words (including fimction words) which compose the
sentences reveals that virtually all the words are ambiguous. That is, the words have multiple
interpretations, some of which are unrelated to the message of the overall text. The point is,
ambiguous words (homographs) are frequently encountered in English, yet are rarely
troublesome to comprehension processes.
Obviously context influences our interpretations of the meanings of words. However,
according to modularity, context is useful only in selecting a single meaning after one or more
have become activated. In an exhaustive access view, multiple meanings of an ambiguous
word always become activated initially. It is only later that a single interpretation is selected
Aging and language performance 89
based on its congruence with the context (e.g., Lucas, 1987; Onifer & Swinney, 1981;
Seidenberg, Tanenhaus, Leiman, & Bienkowski, 1982). On the other hand, according to an
ordered access version (e.g., Hogaboam & Perfetti, 1975), multiple meanings of ambiguous
words are activated serially, according to frequency of use, and checked for consistency with
the context. Activation and checking of individual meanings proceeds one at a time from high
(dominant meaning) to low (subordinate meaning) frequency until a match is obtained, in
which case, subsequent meanings are never activated.
Interactive-activation, or context-dependent, models make nearly opposite predictions
concerning the initial activation of ambiguous word meanings. According to a selective access
version, for example, a biasing context will result in access of only a single meaning of an
ambiguous word (e.g., Schvaneveldt, Meyer, & Becker, 1976; Simpson, 1981; Tabossi, 1988).
No activation should accrue for word meanings that are not consistent with the context.
Because contexts seem variable with regard to strength of bias, another context-dependent
position has been proposed that allows activation to accumulate for multiple interpretations of
an ambiguous word despite biasing context. This context-sensitive view, however, holds that
the contextually appropriate interpretation will always receive the greatest activation (e.g.,
Paul, Kellas, Martin, & Clark, 1992; Van Petten & Kutas, 1987).
Lexical ambiguity has become a common tool for examining language comprehension
processes as well as to differentiate autonomous from interactive processing models. The
extent to which a context can be utilized successfidly to affect initial access of a meaning of an
ambiguity can serve as a measure of on-line comprehension ability. Typically, comprehension
is probed indirectly using word recognition paradigms in which subjects respond as quickly as
possible to stimuli which vary in semantic relatedness to prior contexts (e.g., naming, lexical
decision). The underlying assumption regarding these measures is that reading comprehension
results in activation of word meanings relevant to the context. To the extent that responses to
contextually appropriate words are facilitated relative to unrelated words, it can be concluded
that at least some degree of comprehension has occurred. If context does not affect
recognition of (or responses to) contextually appropriate relative to unrelated conditions,
comprehension processes can be inferred to be impaired.
Although predictions concerning context-dependent and context-independent language
processors appear easy to test, evidence supporting either view can be found. Conclusive
evidence favoring one model over the other has not been demonstrated. Research in support
of autonomous models can usually be criticized for the stimuli used. There is an implicit
assumption in this research that word meanings do not change as a function of context (see
similar arguments by Kellas, Paul, Martin, & Simpson, 1991). This assumption requires that
what is or is not related to an ambiguous word in isolation should also be related or unrelated
to the same ambiguous word when it occurs in a biasing context. There are sufficient reasons
to question this assumption (e.g., Barclay, Brantford, Franks, McCarrell, & Nitsch, 1974;
Barsalou, 1982; Kellas et al., 1991; Olson, 1970; Paul et al., 1992).
On the other hand, there is evidence that context constrains initial meaning activation
(e.g., Schvaneveldt et al., 1976; Simpson, 1981; Simpson & Krueger, 1991). But opponents
of this position argue either that the task is not sensitive to access processes (e.g., Balota &
Chumbley, 1984; Chumbley & Balota, 1984), or that meaning activation was not assessed early
enough during processing (McClelland, 1987; Onifer & Swinney, 1981). However, taking into
account several of the potential problems discussed above, recent research appears to favor an
early influence, context-sensitive system, at least for college students (e.g., Kellas et al., 1991;
90 G. Kellas et al.
Paul et al., 1992; Simpson & Krueger, 1991; Van Petten & Kutas, 1987). In order to
generalize these findings to older adults, as well as to determine the extent to which
comprehension declines with age, it is necessary to provide a thorough analysis of language
performance emphasizing on-line processing.
2. AGING AND LANGUAGE PROCESSING
The present chapter will chart aging and language processing of younger and older
adults from isolated word identification to the cumulative effects of connected discourse.
Within the domain of context effects, we will evaluate word priming, sentence priming, the
scope of meaning activation, the time course of meaning activation, and discourse priming.
Another goal of the chapter will be to integrate these findings within an interactive-activation
system It will be argued that interactive-activation models of language processing are
preferable to autonomous models, especially in fight of the results to be presented. In addition,
we hold that the same processing architecture is useful for describing comprehension
performance of younger as well as older adults.
2.1 Isolated Word Recognition
The syntactic and semantic analyses required for language comprehension cannot begin
without at least an initial identification of individual words. Therefore, we begin by examining
possible age differences in the recognition of isolated words. Kellas, Simpson, and Ferraro
(1988) utilized dual-task methodology (cf. Becker, 1976) to estimate the attentional demands
associated with ambiguous relative to unambiguous word recognition. The authors found that
responses to ambiguous words (primary task) were faster than responses to unambiguous
words which were faster than pseudowords. In particular, older adults identified ambiguous
words nearly 74-ms faster than unambiguous words. Whereas younger adults similarly
benefited from multiple meanings by only 29-ms. It was determined as well that older adults as
a group were deficient in allocating attention effectively for word recognition as compared
with college students.
The Kellas et al. (1988) restilts can be accounted for by both autonomous as well as
interactive-activation views of performance. The finding that identification of ambiguous
words is facilitated relative to unambiguous or pseudowords can be explained within an
autonomous perspective by assuming that presentation of an ambiguous word automatically
results in activation of multiple lexical entries in the mental lexicon. This will, in turn, provide
greater evidence that the target is a word as compared with unambiguous words which will
only provide a single source of lexical activation. The more information available for subjects
to make a decision, the greater the reduction in the amount of time required to process the
information to a subsequent recognition threshold.
The same outcome, however, can be explained by positing an interactive network of
connections feeding simultaneously between lexical and semantic levels. As can be seen in
Figure 1, this can be represented simply as the result of multiple semantic level nodes feeding
more activation to the lexical level when two meanings of a word become activated (as with
homographs). This contrasts with cases in which unambiguous lexical entries are only
strengthened by single semantic level nodes.
r
Outpul
Systems
Word
Figure 1. Simpleinteractive-activation model of word recognition.
It is unclear as to why older adults benefited more from multiple meanings than
younger adults. It has been proposed that word recognition processes are slower for older
adults than younger adults (e.g., Balota & Duchek, 1988; Bowles & Pooh, 1985). In the
present case, the differences in the ambiguity effect could be attn'buted to an encoding
deficiency in older adults. Because younger adults are efficient in stimulus analysis, their
reliance on top-down processes is not as great. Efficiency in stimulus analysis could lead to
word recognition thresholds being reached sooner, resulting in smaller ambiguity effects as
well as faster, overall response latencies. On the other hand, if older adults have encoding
difficulties, they may compensate by relying more heavily on top-down influences. A
consequence of multiple sources of activation will be a larger difference in response latencies
between ambiguous and unambiguous words for older adults. Overall, however, encoding
difficulties will result in slower response latencies because it takes older adults more time to
access semantic representations. Another possible explanation is that the number-of-meanings
manipulation differentially affected the age groups. Older individuals may have more
knowledge regarding word meanings than younger adults and are, therefore, more influenced
by multiple semantic representations. Along these lines, it is noteworthy that older adults are
consistently reported to have higher vocabulary scores.
Whatever the explanation for finding quantitative differences in the ambiguity effect
between younger and older adults, it is clear that the underlying mechanisms contributing to
the word identification process are similar across age groups. Both younger and older subjects
identified ambiguous words more quickly than unambiguous words when in isolation.
92 G. Kellas et al.
2.2 Word Priming
Although the preceding evidence provided no support for qualitative differences in

isolated word recognition with age, it can be argued that age related declines in processing
ability need not be restricted to meaning access of isolated words. Efficient reading ability
requires that the reader make use of prior context in order to form a coherent representation of
a text. Language processing declines with age could be manifested at least in three ways.
First, a deficiency (i.e., slowing) in activating a word's meaning could be compensated for by
increasing one's reliance on previous context (hence, increasing priming effects relative to
young subjects). Second, if older adults are deficient in sustaining activation of prior context
in order of facilitate identification of subsequent words, priming may be reduced or nonexistent
relative to young adults. Finally, prior context could potentially interfere with later word
identification if older adults have a reduced processing capacity or if older adults activate too
much information as a result of processing the context (ef. Hasher & Zaeks, 1988). Therefore,
it is important to determine whether, or to what extent, older adults show simple priming
effects.
There has been some accumulation of evidence which suggests that older adults show
greater priming effects than younger adults (e.g., Balota & Duehek, 1988; Bowles & Poon,
1985; Nebes, Boller& Holland, 1986). This evidence appears to support the position that
older adults rely more heavily on context to access word meaning than younger adults.
However, there has been a greater trend toward finding statistically equivalent priming
between young and older adults (e.g., Burke, White & Diaz, 1987; Burke & Yee, 1984;
CereUa & Fozard, 1984; Howard, MeAndrews, & Lasaga, 1981; Madden, 1988; Nebes, Brady
& Hufl~ 1989). Although these latter studies did not consistently find statistically significant
evidence for greater priming effects, the absolute priming effects observed support the
conclusion that older adults do show greater priming than young adults (ef. Laver & Burke,
1993).
An investigation of individual differences in priming effects with age was performed by
Lyons, Kellas, and Martin (in press). Lyons et al. used a lexieal decision task in which each
word target was preceded either by a related prime (e.g., DOG- BONE), an unrelated prime
(e.g., CAR- BONE), or a neutral prime (e.g., BLANK- BONE). They found that older adults
as a group showed larger priming effects than younger adults (by approximately 33- ms),
although substantial individual differences were reported. This outcome is in agreement with
the results ofKellas et al. (1988), and supports the inference that access processes are slowed
for old adults relative to young adults (ef. Balota & Duehek, 1988). As will be discussed later,
an attentional mechanism is implicated in these results. An alternative interpretation offered by
S. Paul is that older adults attempt to integrate the semantic level representations of the prime
and target prior to responding, whereas, younger adults do not.
In defense of this speculation, consider that if word recognition and access processes
become routinized with experience, there is no reason to expect that language automatieity
does not develop for higher order processes as well. ff the simple act of putting letters
together results in automatic access to the resulting word's meaning, why would the equally
simple (and common) act of placing words together not result in automatically integrating their
respective semantic representations? This level of language expertise would seem a likely
consequence of the increased experience with language that accompanies advanced age. In
addition, there is no need to posit an additional level of processing that is unavailable to
younger adults, but rather, that younger adults do not make use of such processes in the given
task. The addition of an integration process could account for the increased magnitude of
priming, and to some extent, the overall speed differences between young and old subjects.
Responses based on an integration process (i.e., older adults) would be expected to be slower
than those which typically omit such levels of processing (i.e., younger adults), but that rely
only on information made available from spreading activation.
Similar to the results of Kellas et al. (1988), the Lyons et al. (in press) results do not
differentiate autonomous or interactive models of language processing. Nevertheless, it is
useful to briefly describe how some version of each model could account for the results.
Autonomous models typically do not allow for context (or higher level modules) to influence
the speed at which a word's meaning becomes activated. Under the present set of
circumstances, activation at the meaning level for the prime should not feed down to the lexical
level to boost activation of the lexical form of the related target. However, a modular account
does not need to go beyond the lexical level to handle these priming effects. According to
such a model, priming effects can result not from semantic associations, but from lexical level
associations (i.e., direct connections among associatively related entries in the lexicon).
Alternatively, the speed of initial meaning activation should be affected by a prior
related context according to an interactive-activation model (e.g., Figure 1). Exposure to a
word (prime) should result in activation spreading beyond the lexical to the semantic level.
Activation at the semantic level spreads to related concepts. Once these related concepts are
activated, activation spreads directly to nodes representing their lexical level counterparts
(potential related targets). Hence, identification of a target will be facilitated to the extent that
it was pre-activated by a semantically or associatively related prime.
With regard to the ambiguity literature, it is important to note that Balota and Duchek
(1991) have provided evidence that meaning activation for ambiguous words can be influenced
by prior word contexts for older adults. Specifically, they demonstrated that older adults show
typical semantic priming effects when word triplets were utilized. Responses to concordant
word triplets, where the third word was consistent with the biased meaning of a homograph
(e.g., MUSIC - ORGAN - PIANO), were facilitated relative to unrelated trials (e.g., KIDNEY
- C E I L I N G - PIANO). Conversely, discordant conditions, where the third word was
inconsistent with the biased interpretation of the homograph, produced no differences
compared with unrelated trials (e.g., KIDNEY - ORGAN - PIANO). Again, connections
among lexical entries or connections between lexical and semantic entries can account for this
outcome.
These results, together with Lyons et al. (in press), suggest that older adults do not
have difficulty accessing meaning, or utilizing context so as to constrain meaning activation of
subsequently presented words. The findings also suggest that older adults should not have
difficulty using context to comprehend ambiguous words when encountered in sentence
contexts. However, in order to fully evaluate any claim that language comprehension declines
with age, as well as to differentiate autonomous from interactive-activation models of
processing, it is necessary to extend our examinations beyond isolated words. It is probable
that examinations of age differences in processing isolated words is not analytic to similar
questions regarding more typical forms of language processing such as sentences. Especially
when one considers that sentences can provide syntactic and pragmatic cues.
94 G. Kellas et al.
2.3 Sentence priming
Hopkins, Kellas, and Paul (in press) examined the ability of young and old subjects to
inhibit inappropriate senses of ambiguous words that were used to complete biasing contexts
(e.g., It began to roll or She handed him a roll). In addition, Hopkins et al. (in press)
manipulated the strength of the semantic relationship between the prime context and a
subsequent target.
Two hypotheses were tested by utilizing this combination of stimuli. The primary
hypothesis was based on inhibition deficit predictions (Hasher & Zacks, 1988) which hold that,
with age, the ability to inhibit irrelevant information deteriorates. One consequence of this
change in processing ability with age is that goal-relevant information will be difficult to
discriminate from information that is unnecessary for comprehension. Older adults, then,
should have greater comprehension difficulties than younger adults. In the case of ambiguous
word processing, an inefficient inhibitory mechanism might allow multiple interpretations of an
ambiguous word to become activated in working memory. Obviously, comprehension will
suffer if the precise meanings of ambiguous words in a text cannot be quickly activated and
inappropriate senses inhibited. Therefore, it was predicted that if inhibition becomes less
efficient with age, then older adults should provide evidence for activation of multiple
meanings of ambiguous words regardless of sentence bias.
The second hypothesis tested was that decrements in comprehension with age may be
due to changes in the amount of information available for processing. And, it can be argued
from this point of view that prior tests of on-line meaning activation via semantic priming have
implicitly assumed meaning activation to be an all-or-none process (e.g., Balota & Duchek,
1988, 1989; Bowles & Poon, 1985; Cerella & Fozard, 1984; Howard, 1983). By using only
words that are strongly related to a prime it can be determined whether one or more senses of
an ambiguous word have been activated. It can not be determined, however, whether or to
what degree, activation spreads to other related concepts. The view that the scope of meaning
activation changes with age is not all-or-none. If older adults do not process a context as fidly
as younger adults, it is reasonable to expect that the range of semantic activation would be less
for older adults than younger adults.
Empirically, an impoverished semantic representation, in this respect, might allow for
no age differences in priming if meaning activation were probed with words highly related to a
previous context. However, age differences in priming might very well be observed if meaning
activation were probed using targets not as strongly related to a prior context. One implication
of such a processing deficit with age would be a decrease in the ability to generate accurate
inferences from a text (e.g., Cohen, 1979, 1981; Cohen & Faulkner, 1983; Light, Zelinski, &
Moore, 1982; Zacks & Hasher, 1988; but see Belmore, 1981, and Burke and Yee, 1984, for
failures to find age differences in inference generation). This position assumes that a direct
consequence of being able to focus attention on a text is the spread of activation to information
beyond the explicit semantic boundaries of a text (e.g., general world knowledge).
Hopkins et al. (in press) employed a naming task to evaluate the extent to which older
adults use context to prevent contextually inappropriate senses of ambiguous words from
receiving activation. In addition, the scope of meaning activation for these senses of
ambiguous words was evaluated to determine whether older adults are able to access as
complete a representation as younger adults. The scope of activation was examined by using
strongly and weakly related (salient) targets. This measure of semantic salience was derived
Aging and languageperformance 95
l~om Kellas, Martin, Paul, and Lyons (1989) who had college students generate words related
to the prime sentences. Those generated by a high percentage of the subjects were assumed to
be high salient to the context. Those words generated by few subjects (i.e., less than five
percent) were considered to be related but low salient to the context.
As can be seen in Figure 2, the results fail to reflect age differences of any kind. For
both age groups, contextually appropriate prime-target conditions (i.e., sentences biased
toward the dominant meaning of an ambiguous word followed by dominant related targets, and
sentences biased toward a subordinate meaning followed by subordinate related targets)
resulted in faster responses than unrelated controls (i.e., same stimuli but randomly re-paired
so as to be unrelated). Responses to contextually inappropriate target words (i.e., target
dominance opposite of the sentence bias) were not siL_,nificanfly different from unrelated. The
only age-related effect obtained was that older adults responded about 150-ms more slowly
than younger adults.
Younger Adults Older Adults

35
[] Dominant Target [] Dominant Target
~: 30 ~ [] Subordinate Target 30 [] Subordinate Target
~ 25
~
/'////1
~'l///I or 25
~///A
/////
/1"///
.~ 20
~ /////
o i##../ o
/////
//I// u_ 15
/////
/////
> /////
~ 10 ///// "~ 10
/////
/////
n- 5 ///// rr 5
0
-5 -5
Dominant Subordinate Dominant Subordinate
Prime Dominance Prime Dominance
Figure 2. Mean relative facilitation scores (unrelatedminus related response times) for young and old adults as
a function of prime and target dominance from Hopkins, Kellas, and Paul (in press).
Unlike the results from Kellas et al. (1988) and Lyons et al. (in press), these results did
not indicate a greater magnitude of facilitation for older adults relative to younger adults.
Given the outcome of this experiment along with observations that priming differences cited in
prior research did not, in general, reach statistical significance, one could argue that older
adults do not differ from younger adults. Tentatively, however, we speculate that priming is
less age-sensitive when both syntactic and semantic constraints are made available to the
subjects. Perhaps when multiple linguistic devices are offered, both age groups are on equal
terms.
In any case, the above results lend credence to the position that both young and old
adults are equally able to use context to restrict processing to only contextually appropriate
96 G. Kellas et al.
information. In addition, no evidence for differences in the amount, or scope, of activation

with age were indicated. Apparently, then, declines in language comprehension ability with
age are not due either to a failure to inhibit contextually irrelevant information, or to an
incomplete specification of semantic information.
An interactive-activation framework can explain these results if we include a higher,
sentence level, influence. Intra-lexical priming accounts (Le., autonomous or modular view)
are not feasa'ble because the priming results are based on the overall semantic context. A
modular model does not allow for activation of only one contextually determined sense of an
ambiguous word if no associatively related words are present in the context (as was the case
with Hopkins et al_). Because priming effects were dependent on context, top-down influences
from a word level alone are inmxtiicient. As the ambiguous word was present in all cases, it is
necessary to postulate an influence of the context as a whole to facilitate processing of
contextually related, rather than lexically related, information.
f wo~d
(~ Meaning 9
Lexicai ~ OutpuL
RepresenLation .~ Systems
w!,
Figure 3. Elaborated interactive-activation model to account for Hopkins et al. and Paul (1994) results.
Although the present data do not speak to the specifics underlying construction of
sentence level representations, we would be remiss not to acknowledge the work of others
who have addressed such issues (e.g., Gemsbacher, 1990; Taraban & McClelland, 1985,
1990). Construction of a sentence level representation of a text must reflect processes devoted
both to overall parsing as well as thematic role assi~ments (assi~ing the constituents of the
text to appropriate semantic roles). In addition, because in some cases these two processes
depend on information from the other (especially when dealing with ambiguity), it is important
to reflect this parallel architecture as lower-level representations to the sentence-level where
these processes converge (see Figure 3). Although bottom-up information from the
homograph may result in activation of multiple meanings, the constraints imposed by the
context as a whole provide greater activation for the contextually appropriate sense. This
general processing framework has been used by others to account for similar results obtained
for young adults (e.g., Kellas et al., 1991; Paul et al., 1992; Simpson & Krueger, 1991; Van
Petten & Kutas, 1987).
2.4 Time Course of Activation
Although the evidence from Hopkins et al. (in press) appears to contradict major
predictions derived from an inhibition deficit position, there are additional considerations which
must be elaborated. According to Hasher and Zacks' (1988) description, there are two roles of
the inhibitory mechanism that may be deduced. The first is that a failure in an inhibitory
mechanism will result in the entrance of irrelevant information into working memory. The
second role is that inhibitory failure will result in sustaining irrelevant information within
working memory for an extended period of time. Hopkins et al. (in press) addressed the first
role of the inhibitory mechanism and found that older adults do not have difficulties inhibiting
inappropriate information (i.e., inappropriate word meaning) from entering working memory.
The results, however, should not be taken as evidence to suggest that other inhibitory deficits
could not contribute to language processing difficulties with age. A failure in the second role
of inhibition mentioned above would likely impair comprehension processes if activation of
contextually related but less relevant information is sustained along with contextually important
information. Failure to de-activate peripheral information may possibly tax working memory
which would render later integration processes less efficient.
The ability to attend to the relevant aspects of a text is likely dependent on the ability to
ignore or inhibit information that is either redundant or not directly pertinent to the intended
discourse message. The issue here is the extent to which relevant information can be retained
and whether tangential information can be inhibited once activated. It is this failure of the
inhibitory mechanism that was examined by Paul (1994).
Using a naming task, Paul examined the time course of meaning activation for both
high and low salient information with interstimulus intervals (ISis) of 0, 500, and 1000 ms
between sentence offset and target onset. Paul (1994) found that for older adults, both high
and low salient information were sustained in memory over the time course examined. For
younger adults, however, only high salient information was sustained over the entire time
course. Responses to low salient target words were no different from unrelated target words
500 ms after the sentence prime.
There are two possible explanations for the pattern of outcomes. First, the results can
be considered as compatible with the second role of the inhibitory mechanism: to suppress
98 G. Kellas et al.
irrelevant information that has gained entrance to working memory. This position, however, is
only viable if it is assumed that low salient information does not make an important
contribution to discourse comprehension and is therefore, irrelevant. However, that low
salient concepts are relevant was established by normative data (Kellas r al., 1989). Younger
adults provided these words as responses to the sentence primes. Consequently, it is unclear
as to why a n y r e l a t e d information would be considered as irrelevant and subsequently
discarded from working memory.
A more provocative interpretation of the results can be offered in terms of older adults'
ability to fully utilize prior context. This argument supports the notion that older adults show
greater priming effects due to greater reliance on context. Older adults, through experience,
have a greater wealth of general world knowledge than younger adults. The consequences of
being more knowledgeable is that younger and older adults may differ in terms of the products
of language comprehension, as well as how such products are handled. Specifically, a target
word is simply one probe, among many, that can be used to ascertain the domain of
information that has been activated, or analogously, the products of comprehension. Perhaps,
then, presentation of a sentence prime may activate a larger domain of information for older
than younger adults. Consequently, even if there was not a quantitative difference between the
two age groups, it is quite possible that older adults will maintain activation of all related
concepts, regardless of associative strength, because all information is potentially important in
extended discourse. In contrast, while younger adults activate less salient meanings initially, it
may be the case that they only attend to information that has an obvious relationship to the
central meaning of the sentence. This line of reasoning would suggest that older and younger
adults differ in terms of dispersal of attentional resources, not the inability to inhibit irrelevant
information. While both age groups initially activate a range of related information, younger
adults will more narrowly focus attention on the most salient information while older adults
will more broadly focus their attention to include related concepts that are not immediately
crucial for comprehension.
To the extent that older adults maintain more information in working memory than
younger adults, one would expect a corresponding reduction in the capacity of working
memory free for processing subsequent information. This is a reasonable assumption that
implies that processing difficulties for older adults may become more apparent when they have
to process more information after the capacity of working memory has been exceeded.
It is important to emphasize that the present results are not necessarily conflicting with
the results from Hopkins et al. (in press). Rather, the present findings extend the research of
Hopkins et al. to situations which examine the fate of information already activated in working
memory by prior context. Apparently, older adults are able to use context to effectively inhibit
activation of contextually inappropriate information as demonstrated by Hopkins et al.
However, once appropriate information is activated, older adults fail to eliminate less salient
information from memory. Again, however, rather than interpret the results as indicating a
failure to inhibit less salient information, it appears equally feasible to assume that young and
older adults use different strategies for allocating attentional resources to the products of
comprehension.
Due to the similarity in processing requirements in the Paul (1994) and Hopkins et al.
(in press) research, the same conclusions can be drawn favoring the interactive-activation
theory of language processing over the theory of modularity in accounting for the Paul data. A
modularity position would not restrict activated concepts to those related to a single meaning
of a homograph. Clearly, sentence-level constraints immediately influence access to word

meanings independent of age as expected ~om a fully functioning interactive network.
2.5 Discourse Priming
Although the model depicted in Figure 3 can sufficiently explain the outcomes of the
previous experiments, it is clearly inadequate for describing the ongoing process of reading. It
is rarely the case that everyday reading entails the processing and comprehension of a series of
unrelated sentences. Typically, information given in a previous sentence must be integrated
with a subsequent sentence in order for comprehension to occur. Linguistic devices such as
anaphodc reference or other inference processes will bridge otherwise disconnected sentences.
Furthermore, even if sentences are not obviously related to each other, there is at least a
common theme or topic which connects the sentences. An integration failure would produce
incoherent discourse.
The results t~om Hopkins et al. (in press) and Patti (1994) provided evidence that
language comprehension processes in older adults are intact when reading isolated sentences.
However, the extent to which older adults can successfully process extended discourse has not
been fully explored with on-line measures. As speculated earlier, older adults' tendency to
sustain activation of more information than younger adults may interfere with the processing of
upcoming information because the processing capacity of working memory may be exceeded.
Similar to the paradigms of isolated word identification and single word priming, it is possible
that the processing of single sentences does not provide an adequate assessment of language
comprehension changes with increasing age. Perhaps under more usual circumstances where
the processing of extended discourse is required, older adults will show decrements in
comprehension ability compared to younger adults. Based on research indicating declines in
memory ability with age (Light, 1992), it seems reasonable to predict that to the extent that on-
line processing requires retrieval of earlier information for comprehension, older individuals
may be disadvantaged.
The effects of prior context in language processing is most apparent when
interpretation and understanding of a subsequent sentence depends on preceding information.
According to constructionist theories of language comprehension (e.g., Gemsbacher, 1990),
initial information is important because it provides the building blocks to which subsequent
information can be attached. For example, consider the sentence, "He cleaned his arms."
There are at least two interpretation of this sentence which rely on the meaning of the
homograph "arms". One meaning refers to washing the limbs that are attached to a person's
torso, and the other meaning may refer to the treatment of weapons. Without additional
context, either interpretation is appropriate. However, the sentence can be disambiguated to
bias only one meaning by integrating it with other relevant information. For example, the
ambiguous sentence could be preceded either by the sentence "The physician finished his
duties", or "The marksman got a rag." Whether the HANDS or WEAPONS meaning of the
ambiguous sentence "He cleaned his arms" is activated will depend not only on the initial
sentence, but more importantly, that the sentence couplet is integrated to produce a text or
higher-level representation. Specifically, in order to understand the couplet "The physician
finished his duties. He cleaned his arms.", it is necessary to understand that "he" is indeed the
"physician". Through the use of pronominal reference, the subject noun "physician" may be
reinstated in the ambiguous sentence. In this manner, the first sentence is integrated with the
1 O0 G. Kellas et al.
ambiguous sentence and the discourse representation constrains the meaning of the terminal
homograph.
Discourse priming of the above nature was examined by Vu, Kellas, Herman, and
Martin (1994) for young adults. Using sentence couplets in which the second sentence was
ambiguous (taken from Vu, Kellas, and Paul, 1994), the researchers hypothesized that if the
sentences are integrated through pronominal reference, then the sentence couplet as a whole
would immediately constrain the meaning of a paragraph-final homograph. Therefore, naming
latencies for contextually appropriate target words should be facilitated relative to
inappropriate and unrelated target words. It~ however, subjects do not integrate the sentences,
then target words related to both meanings of the ambiguous sentence should be facilitated.
This latter prediction was derived from the research of Vu, Kellas, and Paul (1994), in which
presentation of isolated ambiguous sentences resulted in facilitation of naming latencies to
target words related to both meanings of the sentences compared to unrelated target words.
The results from Vu, Kellas, Herman, and Martin (1994) indicated that younger adults
did integrate the sentence couplet to influence naming latencies for target words. As the left
panel of Figure 4 shows, following a dominant biasing couplet, only dominant targets were
facilitated compared to unrelated targets while subordinate targets were not. Conversely,
following a subordinate biasing couplet, the subordinate targets were facilitated while
dominant targets were not. An argument could be made that the pattern of results could be
due to information arising from the first sentence alone (i.e., intra-lexical priming), and not the
bias of the sentence couplet. However, a second naming experiment was conducted in which
the couplets were separated into individual sentences and subsequently presented to subjects as
isolated sentence primes. The experiment produced two interesting outcomes. First, the initial
sentences did not facilitate previously related targets compared to unrelated targets. For
example, the sentences "The physician finished his duties" and "The marksman got a rag" failed
to facilitate naming latenr for the target words HANDS and WEAPONS. Secondly,
presentation of the ambiguous sentences facilitated naming latencies for target words related to
both meanings of the sentence. That is, following "He cleaned his arms", both HANDS and
WEAPONS were facilitated. This latter pattern of data converged with the findings of Vu,
Kellas, and Paul (1994) where it was demonstrated that both meanings of a sentence-final
homograph were facilitated following presentation of an ambiguous sentence.
Extending the research ofVu, Kellas, Herman, and Martin (1994), Herman, Kellas, and
Vu (1994), have gathered preliminary data on the effects of discourse priming in older adults
using the same set of stimuli. The purpose of the experiment was to determined if older
subjects could integrate the sentence couplet in order to construct a semantic representation of
the couplet as a whole or whether they treat the couplet as two unrelated sentences. It was
predicted that if older adults do integrate the sentences through pronominal reference, then the
pattern of results should converge on the results obtained for younger subjects. Alternatively,
as mentioned earlier, the greater processing load required to comprehend two sentences may
exceed the capacity of working memory for older adults in which case, they may have difficulty
establishing the antecedent of the pronoun. If true, then target words related to both meanings
of the ambiguous word couplets would be facilitated relative to unrelated target words.
The preliminary results obtained by Herman et al. (1994) deny the possibility that older
adults cannot use pronominal reference to construct a specific mental representation of the
couplet as a whole. As the right panel of Figure 4 indicates, naming latencies for dominant
targets were facilitated following dominant biasing couplets compared to unrelated targets
Younger Adults Older Adults

35 35
171 Dominant Target 171 Dominant Target
~. 30
E
,- 25
~////,I
~/////I
~/////
[] Subordinate Target
,•30- 25-
[] Subordinate Tal
/ / / / / i O
0
~'////A ".,~
".~
_~
~
2o Y///~ 20!
-/////
/////,
~
0
/ / / / / /
/ / / / / )
,,.////j
u. is!
/ / / / / /
~.///11
.>_
.,.,,
10
~ s
O"
Dominant Subordinate Dominant Subordinate
Prime Dominance Prime Dominance
Figure 4. Relative facilitation scores (unrelated minus related naming response times) from Vu, Kellas,
Herman, and Martin (1994, leit panel) and Herman, Kellas, and Vu (1994, fight panel) using discourse
priming.
while subordinate targets were not. Conversely, for the subordinate biasing couplets, naming
latencies for subordinate targets were facilitated while dominant targets were not. The results
converged with the pattern of data obtained for younger adults, indicating that, at least in
reference to processing of sentence couplets, older adults' language processing abilities are
equivalent to younger adults. However, a second experiment is necessary in order to evaluate
the possibility that the above preliminary data can be attributed to the first sentence of the
couplets. It is speculated that the results of this control experiment should replicate the second
experiment ofVu, Kellas, Herman, and Martin (1994).
Again, the pattern of data obtained for younger and older adults using on-line measures
of discourse priming is compatible with expectations derived from interactive-activation
models. According to modularity, the processing of an ambiguous word initially is unaffected
by prior context. Thus, presentation of the first sentence in the couplets should not constrain
the activation of the meanings of the sentence-final homograph in the second sentence. Only
after all meanings of the homograph are activated will prior context select an appropriate
meaning. In contradiction, the results obtained here follow dkectly from the emergent
discourse representation in a multiple-constraint based model of interactive-activation. As
each word is encountered, it will be subjected simultaneously to higher and lower-levels of
analyses. The meaning of each word will be constrained by prior discourse, its grammatical
function and thematic role, as well as general world knowledge. As each clause or sentence is
processed, a higher-level representation will be constructed that will, in turn, constrain the
meaning of upcoming words. Following a sentence, the message-level representation in
102 G. Kellas et al.
combination with content-based expectancies for constituent structure and thematic role
assignment (Taraban & McClelland, 1990) will guide the interpretation of the first word in the
next clause. When a pronoun is encountered, its antecedent is reinstated and serves as the
agent of the current action. The process continues word-by-word to develop a running
discourse-level representation. When the terminal homograph is encountered, the meaning
activated is dominated by the prior discourse events. When biased, these events can
overwhelm any activation of inappropriate meanings arising from stimulus analysis. The time-
course of interactive processing is illustrated in Figure 5. In our view, text processing is a
matter of a concatenation of individual, interactive processing episodes.
6enerai World Knowledge/Long-Term Memory

(Silualional Representation)
,.
1
\
Sentence 1 SenLence 2
TIME COURSE
Figure 5. A constraint-based interactive model of discourse processing, Note: The acronyms TRA and CBE
denote Thematic Role Assignment and Context-BasedExpectancies respectively.
2.6 Individual Differences
One final issue concerns the contributions of individual differences to age group
comparisons in language processing. This issue has evolved from the growing concern that
older adults as a group do not represent a homogeneous population (cf. Bakes, 1987). Recall
the first study described in this chapter. One of the goals of the KeUas et at (1988) research
Table 1. Mean attentional scores (difference between probe response times in dual task
conditions relative to probe responses in isolation) for each stimulus condition across probe
onset delays (estimated from Figure 1 ofKellas et al., 1988).
Stimulus Onset Asynchrony

Subject Group Stimulus Type 90 180 270
Young Adults: Ambiguous 445 350 280

Unambiguous 500 400 325
Pseudoword 575 500 425
Older Adults:
High Consistent Ambiguous 210 150 90
Low Consistent Ambiguous 550 525 475

was to specifically question the apparent widespread assumption that older adults represent a
homogeneous group. They provided evidence that older adults vary significantly from one
another in terms of the consistency with which attentional resources are allocated to a
psycholinguistic task (in this case simple word/nonword discrimination). This generally
ignored difference in the allocation of attention within the older adult population could account
for observations of language processing deficits with age (e.g., Kausler, 1982; Kynette &
Kemper, 1987).
Because Kellas et al. (1988) were interested in potential differences in performance
within the subject groups, it was hypothesized that changes in attention allocation could result
in more variable patterns of responding for some, but not necessarily all, older individuals. The
older adults' data were subsequently grouped according to a median split of individual standard
deviations to detect an auditory probe (secondary task). The half of the subjects who had the
smallest standard deviations were considered to be high consistent responders. The remaining
half of the older subjects were considered to be inconsistent responders.
As can be seen in Table 1, among older adults, the most consistent responders actually
outperformed younger subjects in terms of the estimated amount of attention required to
identify words. That is, high consistent older adults do not require as much attention as
younger adults in order to access word meanings. Low consistent older adults, on the other
hand, appear to require more attentional resources in order to perform the same task than
either the high consistent older individuals or young adults.
KeUas et al. (1988) concluded that older adults as a group do not necessarily suffer in
comparison to younger adults in allocating attention to word recognition in general. Rather,
some older adults appear to be less able to sustain attention to a task over extended periods
(el. Stollery, Rabbitt, & Moore, 1990). Indeed, Kellas et al. (1988) were able to isolate a
subset of older individuals who required fewer resources to identify a word than even college
students. When the performance of this subset is not taken into account, older individuals, as a
group, appear to perform less efficiently than younger adults. Yet, a close examination of the
low consistent older adults' trial-to-trial performance yielded evidence that, on occasion, even
they performed approximately equal to, if not better than, younger adults.
With regard to a general slowing hypothesis, recall that Kellas et al. did find that older
adults as a group responded more slowly than younger adults. However, examination of high
consistent individuals contradict this general conclusion. In addition, examinations of optimal
performance of individuals within the low consistent group did not support a more liberal view
that even s o m e older adults are comprehensively slower than younger adults.
As an extension to the Kellas et al. (1988) findings, Lyons et al. (in press) also
examined individual differences in semantic priming across age groups. They hypothesized,
however, that the often reported absolute difference in priming effects could be related either
to a subset of older adults who are unable to sustain attention (Kellas et al., 1988), or to a
general slowing of responding with age. This second hypothesis was derived from Stanovieh
(1980), who argued that greater reliance on top down processing is descriptive of
compensatory processes in the ease of poor readers. The greater need for contextual
information results in slower responding (as the reader waits for sufficient information to
accumulate) and larger context effects. The inference to be drawn from this hypothesis was
that larger priming effects for older adults relative to younger adults would reflect a general
decline in performance with age.
When subjects were grouped according to variability of responding (as determined by
each subject's response time standard deviation for nonwords), a more complicated pattern of
priming effects emerged. Although even consistent older adults responded more slowly than
the younger subjects, a comparison of absolute priming effects (as displayed in Figure 6)
revealed a dramatic increase in magnitude of priming o n l y for inconsistent older adults. No
significant differences in the magnitude of priming were observed for consistent older adults
relative to consistent (or inconsistent) young adults. These data provide additional support for
the position that deficits in sustaining attention within a subset of the older individuals may
occasionally contribute to observed age differences in on-line language performance.
It was earlier reported that the results from Paul (1994) indicated that overall, older
subjects sustained activation of high and low salient information across the time course
examined, while younger subjects sustained only high salient information. Nonetheless,
examination of individual differences in performance between younger and older adults
according to response consistency revealed that only low consistent older adults maintained the
low salient information in working memory (Figure 7).
Similar to low and high consistent younger adults, high consistent older adults
evidenced a pattern of deactivation of low salient information by 1000 ms ISI. These findings
fit speculations that older adults rely more heavily on context quite well. A greater reliance on
context would entail maintaining activated concepts in working memory for longer periods of
time. Consequently, at shorter ISis, the activated information in older adults' working memory
has not yet dissipated. Apparently, however, some older subjects are able to suppress low
salient information; they just require more time to do so. We speculate that examinations of
longer interstimulus intervals may show that even low consistent older adults can deactivate
low salient information in working memory. It is noteworthy that the overall magnitude of
priming was greater as well for older subjects in the Paul study. However, the greater priming
160
150 [] Younger Adults

[] Older Adults
140
130
E 120
r
o 110
.m
100
r
LI., 90
.> 80
rr 70
60
50
40
30
High Low
Response Consistency
Figure 6. Relative facilitation scores (response times to a target following a neutral prime minus response
times to the same targets following related primes) for young and old subjects across response consistency
groups from Lyons, Kellas, and Martin (in press).
effect was restricted to low consistent older adults, which is in agreement with the results of
Lyons et al. (in press).
3. C O N C L U D I N G REMARKS
One goal of the present chapter was to briefly outline and elaborate a progression of
experiments dealing with adult age differences in on-line language processing. Second, a
general interactive-activation processing model was proposed and expanded to account for the
outcomes observed for the word recognition studies descn'bed. In particular, these studies
represent a systematic evaluation of adult age differences in performance under a variety of
priming contexts. One point we can make with absolute certainty is that we have raised more
questions than we have answered.
The main purpose of the interactive-activation model outlined was to descn'be a
processing framework that can handle language performance for young as well as older adults.
It is our opinion that such a model will be useful for understanding and describing discourse
comprehension across the life-span. At this point, we must conclude that the mechanisms
underlying language processing appear to remain age-invariant within the constraints of the
present research.
High Consistent Low Consistent

60 60
~ ,50-
v
Au) 50
E
Older
c 40" c 40 o ................. o ................. o

Low salient
O
.,,. o
,e..,
... 30" High~allent _m 3o
.=.
... ~
o Older~ ......... "'~ 6...
~176 Low salient Younger "**-....
..... 1%
,, 20! ~. 2o High salient
~
= 1 Younger
e.~
~176176176176176
*'~
High Salient
LOW salient
~1o
o
~176176176
~
%
~
~149
~ ................. 9 Low sallen|
-10 -10
o ,ooo o ,ooo
Interstimulus Interval (ms) Interstimulus Interval (ms)
Figure 7. Younger and older adults' mean fac~litatl'onscores for higl~lnd low salient target conditions across
interstimulus interval and response consistencyfrom Paul (1994).
Although the present series of experiments represent a relatively systematic

examination of language processes with age across a variety of priming contexts, it is by no
means a complete investigation of age differences in language processing or of language
comprehension in general. These studies are not comprehensive in that on-line aspects have
been emphasized to the exclusion of off-line contributions to language comprehension. For
example, the couplet study described above involved processing and integration of only two
sentences. Normal discourse situations, however, require substantially increased amounts of
information to be processed within a given topic. Depending on the nature and organization of
textual materials, breakdowns in memory performance could easily result in a loss of
information presented earlier that is critical to comprehending subsequent material. Such
deterioration in memory performance typifies current views of the aging process.
Since language comprehension is the consequence of processes that occur
simultaneously at several levels of analyses, considerable research remains to be conducted.
Thus far, research efforts have overwhelmingly concentrated on the word-level. On-line
investigations into processes that combine word meanings to produce message, text and
situational-level representations are needed. It is important, as well, to gain more knowledge
regarding the time-course of the processes involved in order to provide an explanatory
framework for addressing the ongoing controversy concerning age-related slowing. Although
it is understood that time per se can not serve as an explanatory construct, the source of such
slowing has not been determined. That older individuals, on average, require more time to
complete the processes underlying language comprehension is indisputable. It is becoming
evident, however, that such slowing may not characterize all older individuals. And, indeed, it
is quite possible that the ultimate explanation will surface only as a result of greater
understanding of the individual differences that increase with age.
REFERENCES
Balota, D. A. & Chumbley, J. I. (1984). Are lexical decisions a good measure of lexical
access? The role of the neglected decision stage. Journal of Experimental Psychology:
Human Perception and Performance, 3, 340-357.
Balota, D. A. & Duchek, J. M. (1988). Age-related differences in lexical access, spreading
Balota, D. A. & Duchek, J. M. (1989). Spreading activation in episodic memory: Further
evidence for age independence. The Quarterly Journal of Experimental Psychology, 41A,
849-876.
Balota, D. A. & Duchek, J. M. (1991). Semantic priming effects, lexical repetition effects, and
contextual disambiguation effects in healthy aged individuals and individuals with senile
dementia of the Alzheimer type. Brain and Language, 40, 181-201.
Baltes, P. B. (1987). Theoretical positions of lifespan developmental psychology: On the
dynamics between growth and decline. Developmental Psychology, 23, 611-626.
Barclay, J. R., Bransford, J. D., Franks, J. J., McCarrell, N. S., & Nitsch, K. (1974).
Comprehension and semantic flexibility. Journal of Verbal Learning and Verbal Behavior,
13, 471-481.
Barsalou, L. W. (1982). Context-independent and context- dependent information in
concepts. Memory & Cognition, 10, 82-93.
Becker, C. A. (1976). Allocation of attention during visual word recognition. Journal of
Experimental Psychology: Human Perception & Performance, 2, 556-566.
Belmore, S. M. (1981). Age-related differences in lexical access, spreading activation, and
simple pronunciation. Psychology and Aging, 3, 84-93.
Bowles, N. L. & Pooh, L. W. (1985). Aging and retrieval of words in semantic memory.
Burke, D. M., White, H., & Diaz, D. L. (1987). Semantic priming in young and older adults:
Experimental Psychology: Human Perception and Performance, 13, 79-88
Burke, D. M. & Yee, P. L. (1984). Semantic priming during sentence processing by young
and older adults. Developmental Psychology, 20, 903-910.
Cerella, J. & Fozard, J. L. (1984). Lexical access and age. Developmental Psychology, 20,
235-243.
Chumbley, J. I. & Balota, D. A. (1984). A word's meaning affects the decision in lexical
decision. Memory & Cognition, 12, 590-606.
Cohen, G. (1979). Language comprehension in old age. Cognitive Psychology, 11, 412-429.
Cohen, G. (1981). Inferential reasoning in old age. Cognition, 9, 59-72.
Cohen, G. & Faulkner, D. (1983). Word recognition: Age differences in contextual facilitation
effects. British Journal of Psychology, 74, 239-251.
Fodor, J. A. (1983). The modularity ofmind. Cambridge, MA: MIT Press.
Gernsbacher, M. (1990). Language comprehension as structure building. Hillsdale, NJ:
Lawrence ErlbaurrL
Hasher, L. & Zacks, 1L T. (1988). Working memory, comprehension, and aging: A review
and a new view. In G. H. Bower (Ed.), The Psychology of Learning and Motivation, Vol.
22. San Diego, CA: Academic Press.
Herman, 1L, Kellas, G., Vu, H. (1994). Discourse-level contributions to lexical ambiguity
resolution for younger and older adults. In preparation.
Hogaboam, T. W. & Perfetti, C. A. (1975). Lexical ambiguity and sentence comprehension.
,Journal of Verbal Learning and Verbal Behavior, 14, 265-274.
Hopkins, IC A., Kellas, G., & Paul, S. T. (in press). Scope of word meaning activation during
sentence processing by young and older adults. Experimental Aging Research.
Howard, D. V. (1983). The effects of aging and degree of association on the semantic priming
oflexical decisions. Experimental Aging Research, 9, 145-151.
Howard, D. V., McAndrews, M. P., & Lasaga, M. I. (1981). Semantic priming of lexical
decisions in young and old adults. Journal of Gerontology, 36, 707-714.
Johnson-Laird, P. N. (1983). Mental models: Towards a cognitive science of language,
inference, and consciousness. Cambridge, MA: Harvard University Press.
Kausler, D. H. (1982). Experimentalpsychology and human aging. New York: Wiley.
Kellas, G., Martin, M., Paul, S. T., & Lyons, K. E. (1989). Semantic features of 150
ambiguous words in context. (unpublished raw data).
Kellas, G., Paul, S. T., Martin, M., & Simpson, G. B. (1991). Contextual feature activation
and meaning access (pp. 47-71). In G. B. Simpson (Ed.), Understanding word and
sentence. North-Holland, Elsevier.
Kellas, G., Simpson, G. B., & Ferraro, F. 1L (1988). Aging and performance: A mental
workload analysis. In P. Whitney, & t t B. Ochsman (Eds.), Psychology and productivity.
New York, NY: Plenunl
Kynette, D. & Kemper, S. (1987). Aging and loss of grammatical forms: A cross-sectional
study of language performance. Language & Communication, 6, 43-59.
Laver, G. D. & Burke, D. M. (1993). Why do semantic priming effects increase in old age? A
meta-analysis. Psychology and Aging, 8, 34-43.
Light, L. L. (1991). Memory and aging: Four hypotheses in search of data. Annual Review of
Light, L. L. (1992). The organization of memory in old age. In F. I. M. Craik & T. A.
Salthouse (Eels.), The handbook of aging and cognition (pp. 111-165). Hillsdale, NJ:
Lawrence Erlbaum
Light, L. L., Zelinski, E. M., & Moore, M. (1982). Adult age differences in reasoning from
new information. Journal of Experimental Psychology: Learning, Memory, and
Cognition, 8, 435-447.
Lucas, M. M. (1987). Frequency effects on the processing of ambiguous words in sentence
context. Language and Speech, 30, 25-46.
Lyons, K. A., Kellas, G., & Martin, M. (in press). Inter- and intra-individual differences in
semantic priming among young and older adults. Experimental Aging Research.
Madden, D. J. (1988). Adult age differences in the effects of sentence context and stimulus
degradation during visual word recognition. Psychology and Aging, 3, 167-172.
McClelland, J. L. (1987). The case for interactionism in language processing. In M. Coltheart
(Ed.), Attention and performance Xll. Hillsdale, NJ: Erlbaun~
Nebes, 1L D., Boiler, F., & Holland, A. (1986). Use of semantic context by patients with
Alzheimer's disease. Psychology and Aging, 1, 261-269.
Nebes, 1L D., Brady, C. B., & Hufl~ F. J. (1989). Automatic and attentional mechanisms of
semantic priming in Alzheimer's disease, dournal of Clinical and Experimental
Neuropsychology, 11, 219-230.
Olson, D. 1L (1970). Language and thought: Aspects of a cognitive theory of semantics.
Psychological Review, 77, 257- 273.
Onifer, W. & Swinney, D. A. (1981). Accessing lexical ambiguities during sentence
comprehension: Effects of frequency of meaning and contextual bias. Memory &
Cognition, 15, 225-236.
Paul, S. T. (1994, April). Search for age related semantic inhibitory failure during on-line
sentence comprehension. Poster presented at the fifth Cognitive Aging Conference.
Atlanta, GA.
Paul, S. T., Kellas, G., Martin, M., & Clark, M. B. (1992). The influence of contextual
features on the activation of ambiguous word meanings, dournal of Experimental
Psychology: Learning, Memory, and Cognition, 18, 703-717.
Schvaneveldt, 1L W., Meyer, D. E., & Becket, C. A. (1976). Lexical ambiguity, semantic
context, and visual word recognition. Journal of Experimental Psychology: Human
Perception and Performance, 2, 243-256.
Seidenberg, M. S., Tanenhaus, M. K., Leiman, J. M., & Bienkowski, M. (1982). Automatic
access of the meanings of ambiguous words in context: Some limitations of knowledge-
based processing. Cognitive Psychology, 14, 489-537.
Simpson, G. B. (1981). Meaning dominance and semantic context in the processing oflexical
ambiguity. Journal of Verbal Learning and Verbal Behavior, 20, 120-136.
Simpson, G. B. & Krueger, M. A. (1991). Selective access of homograph meanings in
sentence contexts. Journal of Memory and Language, 30, 627-643.
Stanovich, K. E. (1980). Toward an interactive-compensatory model of individual differences
in the development of reading fluency. Reading Research Quarterly, 16, 32-71.
Stollery, B. T., Rabbitt, P. M. A., & Moore, B. J. (April, 1990). Speed and concentration in
healthy oM age. Paper presented at the third Cognitive Aging Conference. Atlanta, GA.
Tabossi, P. (1988). Accessing lexical ambiguity in different types of sentential context.
dournal of Memory and Language, 27, 324-340.
Taraban, 1L & McClelland, J. L. (1985). Constituent attachment and thematic role assi~ment
in sentence processing: Influences of content-based expectations. Journal of Memory and
Language, 27, 597-632.
Taraban, IL & McClelland, J. L. (1990). Parsing and comprehension: A multiple-constraint
view. In D. A. Balota, G. B. Flores d'Arcais, & IC Rayner (Eds.), Comprehension
processes in reading. Hillsdale, NJ: Lawrence Erlbaum.
Van Petten, C. & Kutas, M. (1987). Ambiguous words in context: An event-rdated potential
analysis of the time course of meaning activation, dournal of Memory and Language, 26,
188- 208.
Vu, H., Kellas, G, Herman, 1L, & Martin, C. (1994). Discourse-level effects on lexical
ambiguity resolution. In preparation.
Vu, H., Kellas, G., & Paul, S. T. (1994). Components of context specificity and the resolution
oflexical ambiguity. Manuscript submitted for publication.
Zacks, P,. T. & Hasher, L. (1988). Capacity theory and the processing of inferences. In L. L.
Light & D. M. Burke (Eds.), Language, memory, and aging (pp. 154-170). New York:
Cambridge University Press.
110
Semantic Processes in Implicit Memory: Aging with Meaning*
David B. Mitchell
Hebrew University and Southern Methodist University
Moses was 120 years when he died,

but his eyes had not dimmed,
and his natural powers had not left him_ (Deuteronomy 34:7)
Unlike the Mosaic vitality, the experiential reality of shortcomings in our "natural
powers" of memory is usually felt long before we near the maximum lifespan. At age 74,
Donald Olding Hebb (1978) experienced "a dimini~qhing effective vocabulary" (p. 21), and
at 78, Burrhus Frederick Skinner's (1983) response was that "in old age ... verbal behavior
becomes less and less accessible" (p. 242). In spite of these testimonials by famous
septuagenarian psychology professors, we shall see that widespread memorial decline is
not inevitable in old age. This chapter will review one type of memory in particular--
implicit memory--that may be invulnerable to the effects of aging. Indeed, a recent study
in France (Kitchie, Ledsert, & Touchon, 1993) found evidence for robust memory
performance through age 100! As shown in Figure 1, their findings contrast sharply with
lifespan data fi'om Colorado (Davis, Cohen, Gandy, Colombo, Van Dusseldorp, Simolke,
& Romano, 1990). I will argue in this chapter that the pattern of unswerving stability is
the accurate portrayal of the relation between normal aging and implicit memory. I will
also present evidence that the alternative standard pattern of decline is the product of a
different memory system (not a different country). First, it will be necessary to define both
"memory" and "aging."
1. AGING
The operational definition for this term is necessarily arbitrary, as there is no

consensus as to when aging begins (Hayflick, 1984). Similarly, "there is a surprising lack
of agreement about ... the relation between age and cognitive functioning" (Salthouse,
1991, p. 32). Salthouse (1991) compiled a list of quotations regarding this relationship,
and found opinions for cognitive decline starting in the early 20s, middle 20s, after 30, at
40, 50, 60, 65, and 70, in the 60s, and in "the post-retirement phase." However, the great
majority of psychology studies tend to compare only two cross-sectional cohorts. In the
gerontological research literature, and in this chapter as well, the standard "older" group is
AUTHOR NOTES: Preparation of this chapter was facilitated by a leave granted by Dedman College,
Southern Methodist University, a grant (AG07854) from the National Institute on Aging, and by Hebrew
University, Jerusalem. Correspondence should be addressed to David Mitchell at the Psychology
Department, S.M.U., Dallas, TX 75275, or via e-mail: dmitchel@sun.cis.smu.edu.
Semantic processes in implicit memory 111
Implicit Memory: Two Aging Patterns

Davis et al. Priming (%) Ritchie et al. Priming (%)
60
50- 25
40 \ 2o
\
\
30 \
~'~7
20-
10- I
I"~ Davis et al., 1990 "~" Rilchie et al., 1993 [
0 .... I I I I I I I I 0
22 34 44 55 64 74 84 94 .
MIDPOINT OF AGE GROUP
Figure 1. Two cross-sectional aging patterns of implicit memory performance from two different studies
(Davis, Cohen, Gandy, Colombo, Van Dusseldorp, Simolke, & Romano, 1990, Exp. 2, Ritchie, Led~sert,
& Touchon, 1993). Two data points were omitted for smoother interpolation (Davis et al.: age 55 =
39%, Ritchie et al.: age 84 = 38%).
made up of individuals over the age of 60 (i.e., the3A,e passed the midpoint of the
maximum lifespan). The "young" group is most often made up of university students
ranging from 18 to 25. (The astute reader has noticed that the data plotted in Figure 1 are
already exceptions to this rule.)
2. M E M O R Y
Memory has been vigorously and rigorously researched since 1885, when Hermann
Ebbinghaus published the first known scientific memory experiments, so there is more than
one definition of the term PsycLit, a computerized database of journal articles, lists 1,016
entries for the keywords "aging AND memory" from 1984 though mid-1994. With "man's
best friend" exhibiting age-related memory deficits paralleling his own, not even sleepy old
dogs have been allowed to lie (Milgram, Head, Weiner, & Thomas, 1994).
The memory phenomena discussed in this chapter can be circumscribed by
excluding those types of memory that will not be dealt with. Thus, we are concerned with
neither sensory memory nor working memory (e.g., Baddeley, 1992), but with long term
memory. William James' (1890) definition is still apt, viz., "the knowledge of a former
state of mind after it has already once dropped from consciousness" (p. 648). But there's
more to it: Within long term memory, we can now distinguish three to five subsystems
112 D.B. Mitchell
(Tulving, 1985, 1991; Tulving & Schacter, 1990). Although these theoretical distinctions
are somewhat controversial and the terminology and classification schemes are still
evolving (of. Roediger, 1990a; Schacter, 1992; Squire, 1992), Tulving's description of
multiple memory systems and their monohierarchical arrangement provide--at the very
least--a neat conceptual framework for organizing and understanding age-related memory
patterns (cf. Mitchell, 1989, 1993). The major systems of interest here are episodic
memory, semantic memory, and implicit memory.
3. EPISODIC M E M O R Y
Episodic memory is synonymous with autobiographical memories in space and

time, memories that include conscious recollection (Tulving, 1985, 1991). Standard
memory tasks such as free recall, cued recall, and recognition tap episodic memory. These
tasks have received the most research to date, and with very few exceptions, there is a
broad consensus that episodic memory declines systematically from young adulthood
through old age (see Craik & Jennings, 1992; Howard, in press; Kausler, 1994; Light,
1991; Mitchell, 1993).
4. SEMANTIC M E M O R Y
Semantic memory is also available to conscious introspection, but lacks the spatial,
temporal, and autobiographical components of episodic memory (Tulving, 1985).
Semantic memory represents general knowledge about the world. All of the following
tasks involve semantic memory: naming pictures, defining words, general knowledge,
judging the fame of a person's name or face, generating category exemplars, free
associating to words, solving anagrams, solving word puzzles, completing a sentence,
making lexical decisions, recalling names of U.S. presidents, geography, etc. Thus,
whereas an episodic memory task requires one to recall a specific prior episode (e.g., what
word from the list presented previously started with the letters ele?), a semantic memory
task asks for retrieval without a specific temporal/spatial context (e.g., generate a word
that starts with ele). The different age-related patterns in these two kinds of tasks has
been noted for some time (Perlmutter, 1980, dubbed it "an apparent paradox"): While
episodic memory tasks reveal age-related declines, semantic memory tasks generally reveal
stability across adulthood (Kausler, 1994). Results from a number of studies suggest that
aging does not adversely affect the organization of the lexicon (e.g., word association,
Burke & Peters, 1986; category exemplar generation, Brown & Mitchell, 1991). Likewise,
automatic activation seems to be unaltered by aging (e.g., flanker task, Shaw, 1991).
When vocabulary or general knowledge tests are used as an index, older adults most often
score higher than young adults (see Chapter 12 in Kausler, 1991).
4.1. Access
An exception to the stability of semantic memory functioning appears when

specific words must be retrieved. Even though it appears that aging does not affect the
activation of a lexical item or concept, activation per se does not guarantee successful
retrieval. In tasks that require highly specific retrieval, a cue is presented and the subject
must deliver a particular target (e.g., fetch a word given a definition). People's names also
fitthis category, as the cue is usually pictorial (i.e., a face or picture) or descriptive (e.g.,
the history teacher in 10th grade, the actor in King of Hearts, the Israeli professor at a
Psychonomic Society conference). Thus, Bowles and Pooh (1985) gave their subjects
only the definitions of low frequency words (e.g., unicorn) and found that older adults had
a lower target retrieval success rate than younger adults. Heller and Dobbs (1993) found
age-related decrements in word-finding, both in discourse (labeling characters and objects
in a video) and in a category fluency task. Albert, Heller, and Millberg (1988) found a
slight (7%) but statistically reliable drop in naming relatively low frequency pictures (e.g.,
trellis, Boston Naming Test). Thus, consistent with the introspective observations of
Hebb (1978) and Skinner (1983), aging seems to be associated with increased access
ditficulties.
The most dramatic access failure experienced occasionally by everyone is the "Tip-
of-the-Tongue" state, when the inaccessibility of an known word can take on a dimension
of frustration beyond mere cognitive disappointment. Studies have not found age
differences in ToTs for objects (Maylor, 1990a; Mitchell, 1989), but have turned up age-
related increases for names of people, places, and movies (Burke, MacKay, Worthley, &
Wade, 1991; Maylor, 1990b). The eventual resolution of a ToT (i.e., finding that word!)
can and does happen in older folks (Finley & Sharp, 1989). As Skinner (1983) noted,
"When I have time--and I mean something on the order of half an hour--I can almost
always recall a name" (p. 240).
4.2. Beyond Access
A number of investigators have been concerned with the issue of age-related

"cognitive slowing," and whether the change in response speed is general or task-specific
(see Schulz, 1994, and related articles). If the slowing is in fact cognitive (i.e., not just
perceptual-motor speed) and global, then semantic retrieval efficiency would also be
affected. In an ongoing mega-meta-analysis debate, one group of investigators concluded
that semantic priming effects are qualitatively equivalent across age, differing only (and
predictably) in slope and intercept (Myerson, Ferraro, Hale, & Lima, 1992), whereas
another team concluded that there are process-specific age differences (Laver & Burke,
1993). If the slowing is task-specific, then we can ask whether the speed of access and/or
the rate of spreading activation are slower in older adults.
Balota and Duchek (1988) used a delayed pronunciation task: When a word
appeared, the subject had to hold off pronouncing it until given a cue. With a short delay
(150 msec) between the word and pronunciation cue (i.e., insufficient time for retrieval),
the age difference was largest (about 200 msec); a longer delay (1200 msec) cut the age
difference in half(about 105 msec). The latter difference was interpreted as "probably due
to differences in output processes" (p. 91), whereas the initial (remaining) difference (i.e.,
95 msec) was interpreted as an age difference in lexical access time. However, Cerella
and Fozard (1984) found a larger age difference in vocalifing alone (94 msec) than in
pronunciation (34 msec), leading them to conclude that lexical access speed was not
impaired by aging. Using a similar procedure in picture naming, we (Mitchell & Schmitt,
114 D.B. Mitchell
1994) have also found a larger age difference for vocalization (96 msec) than for lexical
retrieval (19 msec).
As for the rate of the spread of activation (i.e, by manipulating the stimulus onset
asynchrony), some studies have found age differences (Balota, Black, & Cheney, 1992)
and others have not (Madden, Pierce, & Allen, 1993). While the issues of age differences
in semantic access and semantic speed will have to be resolved ultimately, we won't
examine them any further here. The larger debate on general age-related slowing is
already 30 years old and still raging (of. Schulz, 1994; For a brief and excellent discussion
of slowing, see the relevant section in the chapter by Ober & Shenaut, this volume). My
view is that the notion of general slowing cannot account for the differential effects of
aging on episodic, semantic, and implicit memory (for similar conclusions based on
different data, see the chapters by Allen, Madden, & Slane and by Amrhein in this
volume).
5. IMPLICIT M E M O R Y
Our major concern is with a third type of memory phenomenon. Once a concept
has been accessed in semantic memory, can it be more readily or more quickly accessed
again in the future? This facilitation is neither episodic--for conscious reconection is not
required--nor is it solely semantic, for the "activation" ignited by a single retrieval has a
very long lasting procedural impact. This indelible cognitive footprint is called priming.
In the early 1980s, many reports of long-lasting priming emerged. For instance,
Jacoby and Dallas (1981) gave subjects an identification task in which words were
presented for 35 msec followed by a mask. Although familiar English words were very
diflficult for subjects to perceive, a single exposure of a word was enough to make the
word appear to "jump out" on its second presentation. They found that 1) once primed,
low-frequency words could be identified just as well as unprimed high-frequency words,
and 2) a primed word could be identified with the same facility 24 hours later as it was in
an immediate test. Jacoby and Dallas called this phenomenon "perceptual learning." In a
similar vein, Tulving, Schacter and Stark (1982) presented their subjects with a ditticult
word fragment completion task (e.g., make a word by filling in the missing letters:
o o ut, d _ f n str t on). Again, if the words had been presented in a prior list, subjects
were much more likely to be able to complete them relative to unexposed words.
Furthermore, the magnitude of the priming effect (calculated by the difference between the
proportions of studied vs. unstudied word fragments completed) did not dissipate over a
1-week interval. This priming pattern was in stark contrast to episodic memory
performance: When yes/no recognition decisions had to be made--with the words
(defenestration and coconut) in full view--performance declined precipitously between the
immediate and the 1-week tests. Years earlier, Warrington and Weiskrantz (1968) had
reported equally dramatic dissociations in amnesic patients, and Kolers (1976) had
reported 1-year priming in an inverted text reading paradigm But either the facts or their
implications for memory theory were lost to mainstream cognitive psychology until the
1980s.
Prior to this time, priming was regarded only as a temporary activation of a node
(Collins & Loflus, 1975) or logogen (Morton, 1979) in semantic memory, lasting no more
than a few hundred milliseconds or perhaps several seconds at the outside. Thus, Tulving
et al. (1982) viewed the dissociation data as a "theoretically pregnant finding" and
suggested that priming effects "may be mediated by a cognitive system other than episodic
and semantic memory" (p. 336). Furthermore, priming was not only dissociated from
episodic performance, but from semantic memory as well--even though the priming tasks
themselves do involve semantic memory! This theoretical speculation was motivated by
and supported by modality effects (e.g., auditory stimuli produce little or no priming to
visual targets) and cross-format deficits (e.g., the usually robust picture superiority effect
vanishes on a word t~agment completion task; cs Weldon & Roediger, 1987). Thus, we
have at least two types of functional dissociations between episodic memory and priming:
1) long retention intervals produce decrements in episodic memory but not in priming
(e.g., Mitchell & Brown, 1988); and 2) switching modality or format has little or no
impact on episodic memory, but causes priming to drop off or disappear altogether (e.g.,
Rajaram & Roediger, 1993). These dissociations contradicted a concept of priming
merely as "a modification o f ht e semantic memory system" (Tulving et al., 1982, p. 341).
Dissociations between episodic and implicit memory are abundant, but what
evidence is there for the distinction between semantic and implicit? One often overlooked
dissociation reported by Dannenbring and Briand (1982) makes a compelling case for the
difference between semantic priming and repetition priming. In semantic priming, one
word is followed by a second related word (e.g., mouse-cheese). In repetition priming the
same word occurs twice (e.g., cheese-cheese). To the extent that a lexical decision
response to the second stimulus produces a faster reaction time, we have priming.
Although both kinds of word sequences evince priming, Dannenbring and Briand varied
the lag between the two trials, and found an interesting divergence between semantic
priming and repetition priming. Their data are plotted in Figure 2, where it is clear that
the magnitude and longevity of semantic priming is entirely different from that of
repetition priming. Corroborating the behavioral priming data, Rugg (1987) found that
event-related brain potentials (ERPs) were very different for semantic priming and word
repetition. He concluded that "semantic priming and word repetition do not have their
effects at a common locus (or loci) within the cognitive system" (p. 123). Similar
electrophysiological repetition effects have been found across adulthood as well, whether
measured by ERPs (Friedman, Hamberger, & Ritter, 1993) or SCRs (Skin Conductance
Responses, Plottffe & Stelmack, 1984).
Subsequent to these findings, the implicit~explicit terminology put forward by Graf
and Schacter (1985) spread across the field like wildfire, and 1986 saw the first aging
experiment on implicit memory published by Light, Singh, and Capps. Although Tulving
used the term "procedural memory" in 1985, he later (1991)used this to refer exclusively
to skills and simple conditioning phenomena. Tulving and Schacter (1990) used the term
"priming," in particular reference to a Perceptual Representation System, which they
proposed is responsible for the perceptual identification and priming of the structural
components of objects and words. Squire (1992) used the term "nondeclarative (irnplicit)"
to include skills and habits, priming, simple classical conditioning, and nonassociative
learning. Throughout the rest of this chapter, I will use the term "implicit memory" as
116 D.B. Mitchell
Repetition Priming vs. Semantic Printing

(Dannenbring & Briand, 1982)
100
8O
A
Ill 60
O9
\
40 \
SAME WORD
7 k RELATED WORD
20
tr \
o
%
-20 -
-40 I I I I I I I I I I I ! I I I I I
9 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16
LAG BETWEEN 1st and 2rid OCCURRENCE
Figure 2. The different effects of lag interval on repetition priming and semantic priming. In repetition
priming, the same word occurs on the first and second trials (e.g,, cheese-cheese); in semantic priming,
the second word is related to the first word (e.g., mouse-cheese; data from Dannenbring & Briand, 1982).
defined by Schacter (1987): "previous experiences facilitate performance on a task that

does not require conscious or intentional recollection of these experiences" (p. 501).
6. TYPES OF TASKS AND PROCESSING PROCEDURES
Roediger and his colleagues have championed the concept of transfer-appropriate

processing, especially as it applies to various implicit tasks. Transfer-appropriate
processing means that "Tests of retention will benefit ... to the extent that the processing
operations at test recapitulate or overlap those engaged during prior learning" (Roediger
& Sriaivas, 1993, p. 21). Within this ~amework, a task is perceptual to the extent that
priming is greatest when the surface form of a stimulus is changed the least from input to
test (e.g., word identification: read a word at input, identify the same word presented
tachistoscopically at test). At the other end of the continuum~ a task is conceptual when
priming occurs with only semantic features overlapping between study and test stimuli
(e.g., category instance generation: read a word at input, generate items from the word's
taxonomic category at test). In order classify a given implicit memory task, we have to
known both its input-output parameters (format, modality, test cue) as well as the amount
of priming it produces. Thus, a task is perceptual if any of the following study-to-test
changes reduce priming: 1) modality (e.g., auditory to visual); 2) format (e.g., word to
picture); 3) perceptual details (e.g., typography, object orientation) (Roediger &

McDermott, 1993; Roediger & Srinivas, 1993; Roediger et al., 1994). The "acid test" is
to find a reverse generation effect, i.e., better priming following exposure to a word
presented by the experimenter than to a word generated by the subject (Roediger, Weldon,
& Challis, 1989). This is exactly what Jacoby (1983) found. He had subjects either read
stimuli (e.g., xxx-COLD) or generate targets semantically (e.g., hot-????). Jacoby found
better priming for read than for generated targets in a subsequent word identification task,
but the opposite effect in a recognition memory task (i.e., better memory for generated
items, the standard "generation effect"). As for conceptual tasks, they are so classified if
priming is unaffected by any of the above physical manipulations, and if priming is
enhanced by semantic processing (e.g., levels of processing, generation effect). Indeed, in
a recent meta-analysis of 38 studies, we found a much larger levels of processing effect in
conceptual (mean = .43) relative to perceptual (mean = .12) implicit tests (due to editorial
prerogative, these magnitude means (deep minus shallow processing)/(deep + shallow
means) were not published in Brown & Mitchell, 1994). Roediger and McDermott (1993)
have classified nine verbal implicit memory tasks (see their Table 1) according to the
criteria above, so their classification scheme was used to organize the aging data in the
tables and figures that follow.
Within perceptual tasks, one other distinction is important. When the study and
test item share a similar morphology (e.g., elephant to ele , as in word stem
completion), the task is said to involve direct priming. Priming can be enhanced even
more, however, when both the display and the response are the same at study and test
(e.g., name a picture, name a picture); this is called repetition priming (Roediger &
McDermott, 1993; Roediger, personal communication, November 1994), which "may thus
be viewed as a special case of direct priming" (Roediger, 1990b, p. 380). The difference
between repetition priming and direct priming is illustrated most dramatically by a finding
from word fragment completion. The standard method is to expose subjects to whole
words (e.g., defenestration) and subsequently test them with fragments (e.g.,
d f n str ti n). Although this approach yields better priming relative to input-test
changes in typography or modality, the greatest priming is achieved by presenting the
identical fragment both times. Thus, d f n str ti on is presented along with a semantic
clue ("the act of throwing a person out of a window") and then the same fragment is used
at test. Gardiner, Dawson, and Sutton (1989) found that a fragment presented at input
produced 41% priming compared to only 28% priming for a whole word presented at
input. Variations on this paradigm are ripe for aging studies.
7. IMPLICIT MEMORY AND AGING
7.1. Issues
The remainder of this chapter will focus on the issue of age-related effects in
implicit memory. This is controversial for two reasons, one theoretical and one empirical.
The theoretical debates dispute whether different memory systems can be distinguished
(e.g., episodic vs. semantic, semantic vs. implicit, three systems or more). For instance, in
Kauslel~s (1994) chapter on "generic (semantic) memory" (parentheses his!), he expresses
1 18 D.B. Mitchell
his skepticism with a subhead in the form of a question: "Separate Memory Systems?"
Others have also expressed reservations about memory systems as a vehicle for explaining
age differences (Craik & Jennings, 1992; Salthouse, 1991).
The major empirical argument focuses on a question articulated by Kausler (1994):
"Is implicit memory truly age insensitive?" (p. 377). This question has come about
because whereas most studies have found that age differences in implicit memory are not
statistically simaificant, there is often a small priming advantage favoring the younger
group. Salthouse (1991) speculated that "some of the failures to find significant age
differences in memory when assessed by implicit procedures may be attributable to low
statistical power" (p. 254). Hultsch, Masson, and Small (1991) argued even more
passionately for statistical power failure, saying that "the failure of previous studies to find
significant age effects on tasks similar to the present stem completion measure be may
related to the power of the design rather than the absence of reliable age differences on
such tasks" (p. P29). They took their argument to its logical extreme, and ran 544
subjects!
7.2. Review of the Data
The implicit memory tasks to be reviewed in detail are limited to those that involve
semantic memory at input and output. The focus is on priming in semantic memory
because of our interest in the interface between these two systems. Therefore, the data
review does not cover implicit memory tasks with no verbal component, such as serial
pattern sequences (Howard & Howard, 1992), 2-dimensional representations of potential
3-dimensional objects (Schacter, Cooper, & Valdiserfi, 1992), or novel visual stimuli (e.g.,
Japanese ideograms presented to American subjects, Wiggs, 1993). Even though these
three studies produced beautiful findings vis-a-vis implicit memory and aging (i.e., no age
differences), there are not enough data yet on the question of priming of novel stimuli (see
Howard, in press, for some detailed coverage). Along a similar line, Howard, Fry, and
Bnme (1991) studied implicit memory for new associations (between familiar words) and
found age differences when there were time limits at study; however, with more study
time, age differences disappeared. But these and related findings are as yet inconclusive,
and will not be discussed further. Pve also excluded studies where subjects were either
unconscious (e.g., Brown, Best, Mitchell, & Haggard, 1992) or intentionally distracted by
the experimenter during input (e.g., Howard & Pulido, 1994). In the studies below,
encoding is assumed to have occurred under "normal" circumstances, i.e., subjects were
paying some attention to the stimuli, usually without expecting any kind of memory test.
The data from 36 studies are presented in three tables and one figure, organized
according to repetition priming, direct priming, and indirect or conceptual or unclassified.
These studies contained 48 separate experiments, which yielded 97 contrast pairs (i.e.,
young vs. old). Over 3,000 subjects are represented with overall mean ages of 24 and 71
in the young and older groups, respectively. Each table indicates the test used, its type
(perceptual or conceptual direct or indirect), the encoding task, and other input or test
conditions where relevant. The data include: the number of subjects, mean ages and mean
priming for young and old, an Old:Young ratio, and Effect Size (ES). Priming was
calculated as follows: When the measure was a proportion, the number of targets attained
Semantic processes in implicit memory. 119
for unstudied items (i.e., baseline) was subtracted fi~om the number of studied targets
completed. When the dependent variable was speed, response times for studied stimuli
were subtracted t~om response times for stimuli not studied. The Old:Young ratio is an
index of the degree of age-related deficit (originaUy devised by F.I.M. Craik): The mean
score of the older group is divided by the mean of the younger group. Thus, a ratio near
1.00 indicates no age difference. As the ratio drops below 1, the age deficit increases, and
as the ratio rises above 1, score one for gray panther memory power. The ES estimate is
the value of r, calculated ~om t-test or F-test (ANOVA) values when available (using the
program by Mullen & Rosenthal, 1985). Of 54 possible statistical age group main effect
tests, only 31 values were reported explicitly. The majority of these occur in Table 2,
dominated by the ubiquitous word stem completion task. Among the studies not reporting
inferential statistics, the modal information was simply that the age difference was not
statistically significant.
The tasks listed in Tables 1-3 and Figure 3 are by now all fairly well known
standard implicit memory tasks. Most of them are described and illustrated in Table 1 of
Roediger and McDermott (1993). It is clear that word stem completion (see Table 2) is
far and away the most popular task, showing up in eight studies. In the standard memory
research literature (i.e., non-aging), the word t~agment completion task has been tested
most extensively, so it's surprising that only one aging study (and the first one at that!) has
employed this task. In the repetition priming tasks (Table 1), the same response was made
at study and at test to the same stimulus (e.g., the word student had to be classified as
animate or inanimate by pressing one of two buttons in the study by Hamberger &
Friedman). In the direct priming tasks (Table 2), the same item was presented at study
and test, but the test version of the stimulus was either modified (e.g., picture fragment,
anagram) or made difficult to perceive via a brief presentation and/or by a mask (e.g.,
auditory or visual word identification). With the exception of the "free-association-to-
category-name" task--which is a bona fide conceptual priming task--the tasks in Table 3
are a motley collection reflecting the diversity in the field and the creativity of our
cognitive colleagues. Finally, Figure 3 is dedicated exclusively to homophone priming.
This special treatment was warranted because this task was not classified by Roediger and
McDermott (1993) and because there are six experiments with 11 contrasts, second only
to word stem completion (29 contrasts). In this task, subjects are first exposed to the
lower frequency version of a homophone in a biasing context (e.g., "Joan of Arc was
burned at the stake"). Later, the implicit task is to spell the homophone (presented
aurally) out of context; priming occurs when subjects produce the biased spelling (e.g.,
stake not steak).
Overall, the majority of the O:Y ratios fall below 1 (61 of 97 comparisons, or
63%). A box score, however, is insensitive to the magnitude of the age difference.
Considering the size of the ratios, the overall mean = 1.053; SD = 0.84, median = .890.
Tested against 1.0, the difference is not siLmificant, t (96) = 0.618, p = .538. However,
some of the larger ratios (e.g., 6.00) are statistical outliers, both by z-scores and by the
non-parametric definition of a ratio exceeding the upper hinge by 1.5 the hinge widths of a
distribution. The ratios from some studies were also dropped as methodological outliers,
120 D.B. Mitchell
Table I
Age Differences in R e p e t i t i o n Priming
Mean Priminq Old:Y Effect

Test/Type/Experiment Exp Encoding Task Young Older Ratio Size
LEXICAL DECISION/P/
K a r a y a n i d i s et al., 1993 1 Lex/Dec--0 lag 129ms 125ms .97 9 ns
N=52; A g e s = 27, 67+ 1 Lex/Dec--4 lag 41ms 54ms 1.32
M o s c o v i t c h , 1982 1 Lex/Dec 63ms 53ms .84 .ns

N= 20; A g e s = young, 70
DEGRADED WORD NAMING/P /

H a s h t r o u d i et al., 1991 2 n a m e - - 5 0 % - - i rep. .05 .04 .80 .09
N = 60; A g e s = 19, 69 2 name--50%--8 rep..I0 .13 1.30
n= I0; A g e = 70 2a n a m e - - 3 7 % - - I rep. .-- .02 .40 .Ii
2a n a m e - - 3 7 % - - 8 rep. .-- .I0 1.00
CLASSIFICATION/P? /
H a m b e r g e r & Friedman, 1 typecase 12ms 02ms .17 . ns
1992; N=36; Ages=25, 70+ 1 animate 29ms 30ms 1.03
SPEEDED READING/P?/
H a s ~ t r o u d i et al., 1991 1 inverted--450ms .i0 .05 .50 .28
N=40; A g e s = 20, 70 la inverted--900ms --- .07 .70 .Ii
la n=20; a g e = 69
M o s c o v i t c h et al., 1986 1 transformed--2hr 6.6s 10.0s 1.52 .ns

N=28; A g e s = 22, 68 1 transformed--2wk 2.8s 3.6s 1.29 .ns
M o s c o v i t c h et al., 1986 2 d e g r a d e d sents. 160ms 470ms 2.94 .ns

N=24; A g e s = 23, 71 2 de9. w o r d p a i r s 090ms 470ms 5.22 .ns
M o s c o v i t c h et al., 1986 3 rand. w o r d p a i r s 220ms 430ms 1.95 .ns

N=24; A g e s = 20, 71 3 single words 070ms 260ms 3.71 .ns
plc~-~. N ~ I N Q / C ? /
M i t c h e l l , 1989 1 name--5 lag 178ms 155ms .87 .09
N= 96; A g e s = 22, 70 1 n a m e - - 2 5 lag 176ms 158ms .89
1 n a m e - - 5 0 lag 151ms 136ms .90
M i t c h e l l et al., 1990 1 name--immed. 134ms 104ms .78 .08

N= 96; A g e s = 20, 70 1 name--I day 079ms 059ms .75
1 name--7 days 072ms 055ms .76
1 n a m e - - i l days 055ms 062ms 1.13
Notes. Test Type: P: Perceptual, C= Conceptual. Priming: ms: msec, s= sec, decimal= proportion.
Otd:Y r a t i o = old mean + young mean. E f f e c t size= s based o n s or ~; for f ~ l , assumed B:.50; .ns: Author(s)
reported only t h a t age d i f f e r e n c e was not s t a t i s t i c a l l y s i g n i f i c a n t . +Additional age groups were tested,
but are not reported here. No age d i f f e r e n c e s were s t a t i s t i c a t t y r e l i a b l e in these studies.
Table 2
Age Differences in D i r e c t Priming
Mean Primina Old:Y Effect

TEST/Type /Experiment Exp Encoding Task Young Older Ratio Size
9
WORD FRAGMENT COMPLETION/P /

L i g h t et al., 1986 1 immediate .17 .12 .71 .15
N= 64; A g e s = 23, 69 1 7 days .09 .08 .89
WORD STEM COMPLETION/P/

C h i a r e l l o & Hoyer, 1988 1 vowel--Omin .52 .26 .50 .35*
N=144; A g e s = f 9 , 68 1 vowel--13min .28 .18 .64
1 vowel--46min .21 .23 i. I0
1 pleasant--Omin .47 .38 .81
1 pleasant--13min .34 .29 .85
1 pleasant--46min .36 .28 .78
D a v i s et al., 1990 2 likability .51 .26 .51 .54*

N=147; A g e s = 2 2 , 74+
D i c k et al., 1989 1 read/trial 1 .21 .26 1.24 - .12

N=48; A g e s = 21, 73 1 generate/trial 1 .24 .22 .92
1 read/trial 2 .43 .39 .91
1 generate/trial 2 .40 .37 .92
1 read/trial 3 .60 .47 .78
1 generate/trial 3 .54 .45 .83
G i b s o n et al., 1993 2 l e t t e r - - s a m e type .16 .12 .75 .ns
N= 88; A g e s = young, 68 2 letter--diff.type .15 .I0 .67

2 syllable--s.type .25 .18 .72
2 syllable--d.type .12 .Ii .92
H u l t s c h et al., 1991 1 lexical d e c i s i o n .16 .Ii .69 .16"

N=544; A g e s = 24, 74+
J a v a & G a r d i n e r , 1991 1 letter .21 .14 .67 .12

N=32; A g e s = 22, 73 1 free-ass .19 .18 .95
N=32; A g e s = 21, 70 2 read .30 .27 .90 .12
2 generate .22 .15 .68
L i g h t & Singh, 1987 1 vowel .24 .17 .71 .16

N=64; A g e s = 24, 68 1 pleasantness .29 .24 .83
N=64; A g e s = 23, 68 2 pleasantness .28 .20 .71 .26
P a r k & Shaw, 1992 1 e-a: 3 l e t t e r s .07 .05 .71 .04

N=287; A g e s = 19, 70 1 e-a: 4 l e t t e r s .ii .09 .82
1 pleasant: 3 let. .08 .07 .88
1 pleasant: 4 let. .12 .12 1.00
122 D.B. Mitchell
Table 2 continued
Age Differences in D i r e c t Priming
M e a n Primina Old:Y Effect

Study Exp Encoding Task Young Older Ratio Size
ANAGRAM SOLUTION/P? /
Java, 1992 1 letters .14 .07 .50 .12
N=32; A g e s = 19, 70 1 free-ass .05 .Ii 2.20
DEGRADED WORD.NAMING*/P/
L i g h t & Singh, 1987 3 vowel (80 %) .ii .09 .82 .20
N=64; A g e s = 21, 69 3 pleasant (80%) .16 .12 .75
R i t c h i e et al., 1993 1 read name .27 .27 1.00 -.05

N=575; A g e s = 65, 95+
V I S U A L W O R D IDENTIFICATION~P~
A b b e n h u i s et al., 1990 2 read/high freq. .14 .07 .50 .58"
N= 22; A g e s = 23, 73 2 read/low freq. .19 .12 .63
L i g h t et al., 1992 1 syllable/visual .18 .15 .83 .16

N= 64; A g e s = 22, 72 1 syllable/aud. .08 .09 1.12
1 pleasant/visual .23 .16 .70
1 pleasant/aud. .12 .08 .67
AUDITORY WORD IDENTIFICATION/P/

L i g h t et al., 1992 2 pleasant/aud. .I0 .12 1.20 .09
N= 64; A g e s = 24, 70 2 pleasant/visual .06 .06 1.00
PICTURE FRAGMENT. I D E N T I F I C A T I O N / P /
Heindel et al., 1990 1 name picture .16 .16 i. 00 .ns

N=24; A g e s = 55, 72
R u s s o & Parkin, 1993 1 name picture .17 .16 .94 .05

N= 96; A g e s = 26, 74
HQMOpHONE SPELLING/P? /
See F i g u r e 3
Notes. Test Type: P= Perceptual, C= Conceptual. Priming: decimal= proportion. Old:Y ratio= old mean +
young mean. Effect size= s based on F or ~; for ~<1, assumed g=.50; .ns= Author(s) reported only that age
difference was not s t a t i s t i c a l l y s i g n i f i c a n t . *Author(s) reported age difference was s t a t i s t i c a l l y
s i g n i f i c a n t at 9<.05 or tower. +Additional age groups were tested, but are not reported here. AThis
degraded naming task d i f f e r e d from the task by the same name in Table 1 only v i s - a - v i s direct vs. r e p e t i t i o n
priming; Hashtroudi et at. (1991) started off the f i r s t t r i a l with a degraded stimulus, whereas Light &
Singh (1987) and Ritchie e t a [ . (1993) presented intact words i n i t i a l l y , and tested with degraded words.
Table 3
Age D i f f e r e n c e s in I n d i r e c t / C o n c e p t u a l Priming and Other Unclassified Tasks
Mean Priminu Old:Y Effect

TEST/Type/Experimen t Exp Encoding Task Young Older Ratio Size
,,
FREE ASSOCIATION TO C A T E G O R Y NAME/I/C/
Light & Albertson, 1989 1 pleasantness .18 .13 .74 .17

N= 64; A g e s = 22, 70
G r o b e r et al., 1992 1 name p i c t u r e .46 .31 .67 .ns

9 N= 68; A g e s = 30, 78
S~.~rrv.Nc~. com~T,~.Tios/i/c/
H a r t m a n & H a s h e r , 1991 1 generate - read .01 .06 6.00 .08

N= 68; A g e s = 20, 66 1 generate + read .12 .06 .50
KNOW J o ' o ~ r r s / o / c /
M A n t y l ~ , 1993
I. N=32; A g e s = 27, 72 1 association .II .I0 .91 .12
2. N=32; A g e s = 26, 70 2 association .13 .19 1.46 -.12
Parkin & Walter, 1992

i. N=40; A g e s = 34, 80 1 intentional .22 .37 1.68 .nr
2. N=90; A g e s = 22, 68+ 2 intentional .39 .42 1.08 .nr
.FALSE FAM]E/D/C?/
D y w a n & J a c o b y , 1990 1 name fame -. Ii* .06 I. +? .nr

N= 48; A g e s = 20, 71
B a r t l e t t et al., 1991 2 face fame -.07 .07 I.+? .nr

N= 39; A g e s = 28, 72
PROCESS DISSOCIATION PROCEDURE/D/P?/
J e n n i n g s & Jacoby, 1993 1 read: priming -.06 .04 i.+? .nr

I. N=72; A g e s = 20, 74 1 read: f a m i l i a r i t y .31 .38 1.23 .nr
2. N=32; A g e s = 19, 70 2 read: priming .51 .53 I. 04 .nr
2 solve: priming .23 .37 1.61 .nr
2 read: f a m i l i a r i t y .62 .58 .94 .22
2 solve: familiar. .62 .66 1 .'06
Notes. Test Type: X= I n d i r e c t , D= D i r e c t ; P= Perceptual, C= Conceptual. Exp= Experiment number; Priming

means are p r o p o r t i o n s . Otd:Y r a t i o = otd mean § young mean. Effect size= s based on f or ~; f o r f ~ l ,
assumed ~=.50; .ns= Author(s) reported onty that age d i f f e r e n c e was not s t a t i s t i c a t t y s i g n i f i c a n t ; .nr=
i n f e r e n t i a t s t a t i s t i c s not reported; *dacoby et at. (1989) obtained priming of .08 when young subjecs were
tested o v e r n i g h t . +Additionat age groups were tested, but are not reported here. No age e f f e c t s were
s t a t i s t i c a t t y r e l i a b t e in these studies.
124 D.B. Mitchell
Homophone Priming Studies

ES
Rose et al., 1986 I
Davis et al., 1990
Howard, 1988, Exp. 1

Howard, Exp. 2, lmm.
Howard, " "2-Day
Howard, Exp. 3, Imm.
Howard, " " 2-Day

M & B, 1990, Imm.
M & B, 1990, 1-Day
M & B, 1990, 7-Day 1 YOUNGER
OLDER
M & B, 1990, 21-Day
J J
-10 0 10 20 30 40 50
PRIMING (%)
Figure 3. Mean priming for young and older adults in homophone spelling or usage tasks from six
studies. ES = effect size (r); *statistically significant age difference; .nr = Inferential statistic not reported;
.ns = author reported only that age difference was not significant. Retention intervals: Imm = immediate;
I, 2, 7, and 21 days. M&B= Mitchell & Brown (1990). The data from Davis et al. (1990) are from their
Experiment 1.
when priming was not observed in one of the age groups.* These ratios are detailed in
Table 4, where the means and medians for each of the four groupings are summarized
Across the entire sample of 97 contrasts, a total of nine ratios were dropped. Table 1 lost three ratios as
extreme statistical oufliers (2.94 and up). Table 3 lost one point as a statistical outlier (6.00) and three
other points because the experiments did not produce the phenomenon in the younger adults. These were
process dissociation procedures, in which remembering an item eliminates the false fame or priming for
that item (Bartlett et al., 1991; Dywan & Jacoby, 1990; Jennings & Jacoby, 1993, Exp. 1). Because
younger subjects' episodic memory was working too well, the phenomenon did not occur in these
conditions, and thus an age comparison is not possible. (In order to get the false fame effect in young
adults, Jacoby, Kelley, Brown, & Jasechko, 1989, delayed the test for 24 hours, when episodic memory
was waning.) Two data points were dropped from the homophone studies (Figure 3) because the older
subjects did not produce priming. The failure to get priming in these studies (Davis et al., 1990, Exp. 1;
Rose et al., 1986) suggests a methodological flaw, since recognition memory in the older groups was
above chance. Davis et al. used Rose et al.'s identical materials, and both studies actually observed
negative priming in the older group (i.e., studied homophones were less likely to spelled in the lower
frequency meaning than unstudied homophones. I am not aware of this problem in any other published
study.). Davis et al. (1990) went so far as to conclude that "the homophone test appears to be poorly
suited for assessment of priming in the elderly" (p. 292). Howard (1988, Exp. 1), however, reported too
much test awareness with the spelling technique, and instead had subjects generate sentences at test in
before and after outliers are removed. Only the data in Table 2 reveal an age-related
deficit reliably different from 1.0. Indeed, only four studies reported statistically
si~ificant main effects for age. All of these are in Table 2, and three of them are based on
word stem completion priming. (Three other si~ificant age main effects occur in
homophone priming, Figure 3, but two of the studies (Davis et al., 1990; Rose et aL,
1986) were unable to get any priming at all in the older adults, and the other, Howard,
1988, Exp. 1, uncovered bona fide episodic contamination. Thus, these are treated
separately.)
Table 4.
Old:Young Ratios from Tables 1-3 a n d F i g u r e 3,
Before and After Removing Outliers
Before Removal After Removal
N Mean Median N Mean Median t
Table 1 24 1.323 .935 21 .946 .890 0.63
Table 2 46 .845 .820 46 .845 .820 3.92*
Table 3 16 1.370 1.020 12 1.077 1.050 0.72
Figure 3 II .875 .690 9 1.070 1.080 0.31
Notes. *~<.01, tested against 1.0.
order to determine which homophonemeaning a subject had in mind. We (Mitchell & Brown, 1990)also
utilized this "usage" procedure, which may account for the difference between the positive and negative
findings.
126 D.B. Mitchell
In the sections that follow, we shall see that the data with the smallest O:Y ratios
(i.e., that suggest the largest age deficit in priming) all come from studies where there is
clear evidence of methodological shortcomings. The shortcomings involved either the
susceptibility of the implicit tasks to contamination by conscious recollection, the number
of stimuli (too few), or the number of repeated trials (too many). These problems are
directly responsible for producing a spurious age difference in priming. Once the
problems are eliminated, the data unequivocally support age invariance in priming.
In Table 1, three-quarters of the data are priming based on response speed
measures. The lowest O:Y ratio (.17) was not an outlier, but the three largest values
(2.94 and up) were. Before removing the outliers, a liberal interpretation is that older
adults show the best priming on implicit tasks that involve identical repetition and/or
response speed measures. However, note that the largest O:Y ratio actually comes from
Table 3 (Hartman & Hasher's unique sentence completion task), whether before or after
removal of outliers (see Table 4). Table 3 contains the greatest number of as yet
unclassified tasks, so we can't make any conclusions here about specific tasks or types of
tasks. The worst age deficit occurs in Table 2, but we will see below that this is not due
the use of direct or perceptual tasks per se.
8. EPISODIC CONTAMINATION IN IMPLICIT M E M O R Y TASKS
One explanation that can account for age differences in implicit memory
performance when they do appear is episodic contamination (cf Howard, in press;
Schacter, Bowers, & Booker, 1989). That is, although subjects are not instructed to use
their recollection as a tool for solving the word or picture puzzle in front of them, they do
so anyway. For example, when we give a subject a word stem, INT , the implicit
instructions only ask that the subject complete the stem with the first word that comes to
mind. If the subject has total anterograde amnesia, then we can be confident that he will
follow the instructions. Otherwise, there is usually no guarantee that the subject will not
notice the relation between the test stems and the words previously seen or heard on the
experimenter's list. When subjects do notice, and then go on to utilize their memory to fill
in the missing letters, then implicit memory is no longer being measured properly, and the
performance is "contaminated" by episodic memory. If this phenomenon would occur
with equal frequency across age groups, and if its occurrence would affect priming scores
equally in different age groups, then contamination would be no more than a nagging
nuisance, and--at least for the purpose of comparing age differences--we could ignore it.
Salthouse (1991) has raised a potential problem with this line of thinking:
"While the suggestion that many i~licit assessments are

contaminated by explicit remembering or episodic processes
preserves the original hypothesis, it does so only at the risk of
introducing an uncomfortable circularity. That is, age
differences are presumed to be present because the tasks include
conscious recollection or the episodic system, but the tasks are
apparently inferred to include conscious recollection of the
episodic system because age differences are present. As when any
hypothetical construct is postulated, some independent means of

establishing its existence seems necessary to minimize reliance
upon assumptions whose validity cannot be verified." (pp. 254-256)
Those independent means are now available, and are reviewed below. We shall
see that 1) younger subjects are more likely to become aware that the implicit task is a
veiled memory test, and 2) younger adults are better at exploiting the study item/test cue
relations, thereby (artificially) boosting their implicit memory performance. Akhough the
explanation for why the young adults do both things is that their episodic memory is
working better, note that the evidence for episodic contamination is n o t based on the
finding of an age difference in priming. We will first review direct evidence for both
points, and then see additional ramifications of this phenomenon reflected in certain data
patterns.
Category Exemplar Priming

(Light & Albertson, 1989)
25
AGE GROUP
20 - l "t'otrNc ~/A OLDER
15
r
7
1o
rr
n
5
UNAWARE AWARE & TRIED

SUBJECTS' AWARENESS OF MEMORY TEST
Figure 4. Priming from category exemplar generation in young and older adults, as a function of whether
subjects were aware of the relation between the test and the previously studied exemplars (Light &
Albertson, 1989). Subjects on the left side of the figure reported being unaware of the relation, whereas
those on the fight side reported both awareness and intentional attempts to generate studied target words
from each category.
When experimenters have asked their subjects whether they noticed a relationship
between the implicit task and previous stimuli, younger subjects are the best detectives.
Light and Albertson (1989) asked their subjects "whether they noticed that they were
generating previously seen list members and whether they deh'berately tried to do so" (p.
489). Although 34% of the young and 12% of the old reported intentionally trying to
128 D.B. Mitchell
produce kems from the study list, fully 87% of the young indicated awareness relative to
54% of the elderly. In Hartman and Hashel~s (1991) study, 54% of their young adults
"became aware of the relationship between the study sentences and the sentence
completion test" (pp. 591-592), compared to only 13% of the older subjects. However,
among their young subjects, awareness was not correlated with priming (sentence-
completion task). Park and Shaw (1992) found awareness in only 10% of their large
sample, about two thirds of which were young. Within their young subjects, awareness
was not correlated with priming. However, when aware subjects were excluded, the
analysis revealed identical priming (.08 each) for young and older adults (cs the means in
Table 2). When Light and Albertson compared the priming performance as a function of
episodic awareness, the age differences varied dramatically. These differences are
illustrated in Figure 4, where it is clear that pure implicit memory performance was age
invariant, in stark contrast to an enormous advantage for the younger subjects who
noticed the relationship between test cues and list items, tried to use that information
explicitly, and succeeded.
Explicit Contamination" Picture Fragment

(Russo & Parkin, 1993)
2.0
EPISODIC STATUS
I NOT RECALLED ~ RECALLED
A
r 1.5
_.1
tlJ
>
iii
_.!
1.0
o
z
tw
12. 0.5
0.0
OLDER ADULTS YOUNGER ADULTS
Figure 5. The relationship between picture fragment identification priming and recall of specific items in
young and older subjects (Russo & Parkin, 1993).
Positive relationships between test awareness and priming have also been reported
by Howard (1988), Grober et aL (1992), Park and Shaw (1992), and Russo and Parkin
(1993). Grober et al. found a statistically nonsi~ificant yet substantial age difference in a
free association task to category names following picture naming (O:Y ratio = .67).
However, in a second experiment, Grober et al. demonstrated 1) that 58% of their young
subjects were test-aware, and 2) those that were aware produced si~ificantly more
targets than test unaware subjects (48% vs. 15%). These contamination effects are not
limited to conceptual priming tasks. Russo and Parkin (1993) found that even in picture
fragment identification--arguably a strongly perceptual task (cs Roediger and Srinivas,
1993) potentially impermeable to episodic contamination-item specific recall was utilized
advantageously by younger adults but not by the older ones. These data are displayed in
Figure 5. (Note that these groups are different from the subjects listed in Table 2; the
latter were n o t given a recall test prior to picture fragment identification.)
Word Fragment Priming

(Light, Singh, & Capps, 1986)
20
YOUNGER
OLDER
IMMED. 7 DAYS
RETENTION INTERVAL
Figure 6. The effect of retention interval on word fragment completion priming in young and older adults
(Light, Singh, & Capps, 1986). Since episodic recognition dropped precipitously (from 70% to 24%)
across the same interval, there was much less opportunity for episodic contamination after 7 days.
Confider another angle. In the studies above, subjects were asked. But we can
see evidence without asking, as follows. If episodic contamination is indeed a major
factor accounting for the ubiquitous small but statistically unreliable age differences in
implicit memory, then the age difference should get smaller as the opportunity for
contamination diminishes. Experiments with multiple retention intervals give us the
proper conditions to test this hypothesis. We know that explicit memory performance
declines systematically over time, for both young and old alike. Therefore, if the
contamination hypothesis is correct, we would expect any age differences in implicit
memory to actually get smaller, even when the explicit memory age differences become
130 D.B. Mitchell
greater. This is exactly the case in studies where both explicit and implicit memory were
tested over long retention intervals. At short intervals, where there was greatest
oppommity for contamination, age differences in implicit performance were largest. At
long intervals, where explicit performance dropped off equally at best or even more for the
older group--implicit performance did not follow this pattern. Thus, Light et at (1986)
reported a much smaller age difference in priming after 1 week than after 1 hour (see
Figure 6). Second, Chiarello and Hoyer (1988) found the largest age difference in word-
stem completion priming on an immediate test, and the smallest difference after a 46-rain
interval, in spite of parallel drops in explicit memory. Finally, Mitchell et al. (1990) found
the smallest age difference-actually favoring the elderly--at a 3-week interval, when the
age difference in explicit memory was greatest. This phenomenon is plotted in Figures 7
and 8 using the O:Y ratio as the index. As can be seen, the ratio exceeded .90 in Chiarello
and Hoyer and flipped above 1.0 in Mitchell et at
Old:Young Memory Ratios

(Chiarello & Hoyer, 1988)
1.0
0.9
0.8
0.7 ,a~ IMPI.ICIT
" "-El- EXPLICIT

0.6- x
%%
0.5- %
E3
0.4-
0.3 I I I I I I I I I I I ! I 1 I ! I ! I I I ! ! I I I I I ! I I ! I I I I I ! ! I I I ! I I !
0 13 46.
TEST DELAY (minutes)
Figure 7. The effect of short retention intervals (minutes) on word stem completion priming, with O:Y
priming ratios as the dependent variable (calculated from Chiarello & Hoyer, 1988). As argued in Figure
6, there was less episodic involvement at longer delays, diminishing the younger adults' advantage.
The flip side of this can be seen when performance is elevated by repeated trials.
Dick et al. (1989) found that word stem priming improved across repeated trials, but at a
higher rate for young subjects than for older subjects. The greater benefit of increases in
conscious recollection for young adults' priming is easy to see in Figure 9. Taken together
with the evidence presented earlier, word stem comletion as a task seems to be particularly
susceptible to contamination. Further evidence for contamination in word stem priming
can be seen in the correlation between the magnitude of priming and the O:Y ratio. (The
contamination hypothesis predicts that age deficits are inflated as priming increases; that
is, priming is abnormally large because of a boost from episodic participation.)
Considering word stem priming alone, the young adults' mean = .235, SD = .123.
Chiarello and Hoyer (1988) and Davis et al. (1990, Exp. 2) each had a priming cell mean
of .52 and .51, respectively. These are statistical outliers, but more to the point, were
associated with two of the lowest O:Y ratios (.50 and .51; see Table 2). With these
numbers in, r = -.39, but the correlation between priming and O:Y drops to .08 without
them One other finding not listed on Table 2 was reported by Hultsch, Hertzog, Small,
McDonald-Miszczalg & Dixon (1992). Even though Hultsch et al. (1991) found a
statistically significant age difference in word stem priming (see Table 2), their 1992 study
Old:Young Memory Ratios

(Mitchell, Brown, & Murphy, 1990)
RATIO
1.2
1.1
1.0
0.9
0.8
0.7
"V
0.6 - I --~- EXPLICIT ~ IMPLICIT [
I I I I I I I I I I I I I I I I I. I I I I I
0.5
0 1 7 21
RETENTION INTERVAL (Days)
Figure 8. The effect of long retention intervals (days and weeks) on picture naming priming again with
O:Y priming ratios as in Figure 7 (calculated from Mitchell, Brown, & Murphy, 1990). The same
rationale mentionedunder Figures 6 and 7 applies here, demonstratingthat with less episodic support, the
age deficit in priming can be eradicated.
did not find an age difference in the same task (ages 65 to 68 and 75 to 78, either cross-
sectionally or longitudinally), and their sample was large (N = 328).
Besides the word stem completion studies, only one other study in Table 2
produced a large and statistically reliable age difference in priming. Although the authors
132 D.B. Mitchell
did not consider contamination, a careful examination reveals some problems. Abbenhuis,
Raajimakers, Raajimakers, and van Woerden (1990, Exp. 2) used individually set
thresholds for tachistoscopic exposures in a word identification study. Three groups of 14
words were presented once, twice, or three times. The tachistoscopic exposure time was
calibrated individually for each subject, starting at 100 msec for young, 200 msec for older
adults, and then working up or down by 20-40 msec increments until a criterion near 40%
correct identifications was reached. The older adults' threshold was much higher than the
young adults' (132 msec and 27 msec, respectively). Even with the longer exposures, the
older group had reliably lower priming performance (9.6% vs. 16.6%; their recognition
memory deficit was also substantial 63% vs. 88%). However, Abbenhuis et aL reported a
hardware problem that may have disproportionately aided the young adults: Their CRT
had a ret~esh cycle of 20 msec, so that words were sometimes exposed for 40 msec, and
thus "some [young] subjects could read the words too well" (p. 580). The same 20-msec
variability around the older adults' mean threshold (132 msec) would not have given them
Word-Stem Priming: Repeated Trials

(Dick, Kean, & Sands, 1989)
60
50
40
30 / f
YOUNG
OLDER
20 I ! I
1 2 3
TRIAL
Figure 9. The effect of repeated trials on word stem priming in young and older subjects (Dick, Kean, &
Sands, 1989). Here it is argued that while repeated exposure enhanced both implicit and episodic
memory, it was the latter that gave the younger subjects the greater boost in priming (cf. Tulving's, 1991,
concept of co-determination).
any comparable advantage. In addition, there was a substantial educational gap between
the two age groups. The young group was composed of university students, with an
average of 16.6 years of education. In contrast, the older adults had less than a high
school education on the average (mean = 11.2 years). Since other studies have found
correlations between education and cognitive functioning, it would not be surprising if this
confound accounted for more of the age difference in priming than aging did. Finally, the
fact that a third of the target items were presented three times each (and then combined
with twice-presented targets for the analysis, unfommately) may have given the young
subjects an edge: That is, three exposures (cs Figure 9, Dick et aL, 1989) may have
pushed many of the targets over into a level of awareness that promoted episodic
contamination in the young adults. Indeed, their recognition performance for this group of
words was near ceiling (mean-92%).
9. AGE INVARIANCE ACROSS THE BOARD?
In spite of the compelling data reviewed above, perhaps not all implicit tasks will
be so robust with respect to age-invariance. Consider an interesting case: When format is
switched between input and test, priming usually suffers. One exception we have found is
Picture Priming
(Steen-Patterson, Jones, Brown, & Mitchell, 1993}
MSEC
150
120
9O 'UT PRIME
II PICTURE
~] WORD
30
YOUNG ADULTS OLDER ADULTS
Figure 10. The effect of changing format (word to picture) in picture naming priming in young and old
adults (Steen-Patterson, Jones, Brown, & Mitchell, 1993). Subjectsread either words or pictures at input,
and always named pictures at test. The words matched the pictures' names.
that words can prime picture naming just as well as pictures do, which we termed "transfer
inappropriate processing" (Brown, Neblett, Jones, & Mitchell, 1991). However, when we
test this paradigm with older adults, there is clear evidence of transfer appropriate
processing. That is, they show less priming from words than from pictures (see Figure 10,
Steen-Patterson, Jones, Brown, & Mitchell, 1993). It's not that words can't prime picture
134 D.B. Mitchell
naming in older adults. Across a very brief interval (1.2 sec), words primed picture
naming m o r e in older adults than in younger adults (see Figure 11, Thomas, Fozard, &
Waugh, 1977). The important thing to note, however, is that even w h e n older adults
show more priming than younger adults, they still show less cross-format priming. (This
is not true, however, for cross-modality priming, as Light et aL, 1992, found similar
patterns across age groups; see Table 2. For some explanations of why older adults
sometimes show greater priming, see Laver & Burke, 1993, and the chapter by Ober and
Shenaut, this volume).
Could the age difference in cross-format priming implicate a perceptual
breakdown? That is, do older adults have greater difficulty perceiving the object, as
opposed to needing more time to retrieve the lexical label? I have argued (Mitchell, 1993)
that picture naming is a conceptual task, but aging data may necessitate a revision of that
claim
Picture Priming
(Thomas, Fozard, & Waugh, 1977)
MSEC
300
270
240
210
180
150
1 INPUT PRIME
I PICTURE
120 WORD
90
60
30
0
YOUNG ADULTS OLDER ADULTS
Figure 11. The effect of changing format (word to picture) in immediate sequential trials in young; and
older adults (Thomas, Fozard, & Waugh, 1977). In Figure 10 (Steen et al., 1993), the study and test trials
were in blocks separated by a few minutes.
10. CONCLUSIONS
We have reviewed data from a slew of studies to answer two questions: 1) Is

implicit memory age-invariant? and 2) Can implicit memory be considered a memory
system separate from semantic memory? The answer to both questions is affirmative,
albeit with different levels of certainty.
Regarding the first question, we examined the claim that age differences in implicit
memory are real but subtle, simply requiring more statistical power for detection. A
casual glance at Tables 1-3 and Figure 3 would appear to support this point of view:
After all, the majority of the studies report slightly better priming for younger subjects,
and a few studies even obtained statistically reliable age differences. But careful scrutiny
revealed that every single one of the studies with a statistically reliable age difference
could be discounted by methodological problems. The most parsimonious explanation for
siLmificant age differences in three word stem completion priming studies (Table 2) was
episodic contamination, and additional procedural problems plagued one word
identification study (Table 2). Two homophone priming studies that found an "age
difference" turned out to have no priming performance at all in the older group. But what
of the majority of studies that were successful in producing viable priming in older
subjects, but still obtained slightly lower magnitudes? Even excluding the problematic
studies, 55 ofthe remaining 88 contrasts (62.5%) have a ratio below 1.0. My sense is that
many of those still have contamination effects. For instance, word stem completion
dominates the sample, where the proportion of ratios falling below 1.0 is 93%! Clearly,
we need a wider sample of implicit tasks in aging research, with greater care taken to
avoid tasks and/or procedures that are likely to be affected by episodic contamination. On
the second question, we do not yet have enough information on the memory systems. It
seems clear, however, that most implicit memory is not really affected adversely by normal
aging, and that in combination with findings from other areas of cognitive research, the
evidence will clearly indicate the presence of at least one memory system dissociated t~om
both semantic and episodic memory. Furthermore, the differential aging effects in implicit
memory are also consistent with the view of process-specific age differences (el Allen,
Madden, and Slane, this volume). Thus, the notion of general slowing is inadequate to
account for the implicit memory phenomena reviewed in this chapter.
11. SUMMARY
A large of number of studies of implicit memory in normally aging adults was

reviewed. The outcomes of 36 studies produced 97 contrasts between young and older
adults, with 18 implicit memory tasks represented. Although some studies obtained
statistically significant age differences in implicit memory tasks, these differences were
virtually eliminated when episodic contamination was taken into account. That is, in a
number of tasks intended to test only implicit memory, in fact subjects used episodic
memory to artificially boost their performance. In every case, the majority of these
subjects were young adults, thus producing a spurious age difference in implicit priming
scores. I concluded that implicit memory performance is not affected by normal aging,
and that implicit memory phenomena can be best understood in the context of multiple
memory systems. This modular approach allows for selective effects of aging on some
systems and processes, leaving other cognitive components intact.
136 D.B. Mitchell
REFERENCES
Abbenhuis, M.A., Raaijmakers, W.G.M., Raaijmakers, J.G.W., & van Woerden, G.J.M.
(1990). Episodic memory in dementia of the Alzheimer type and in normal aging:
Similar impairment in automatic processing. The Quarterly Journal of Experimental
Psychology, 42A, 569-583.
Albert, M.S., Heller, H.S., & Milberg, W. (1988). Changes in naming ability with age.
Baddeley, A. (1992). Working memory. Science, 255, 556-559.
Balota, D.A., Black, S.R., & Cheney, M. (1992). Automatic and attentional priming in
young and older adults: Reevaluation of the two-process model. Journal of
Balota, D.A., & Duchek, J. M. (1988). Age-related differences in lexical access, spreading
Bartlett, J.C., Strater, L., & Fulton, A. (1991). False recency and false fame of faces in
young adulthood and old age. Memory & Cognition, 19, 177-188.
Bowles, N.L., &Poon, L.W. (1985). Aging and retrieval of words in semantic memory.
Brown, A.S., Best, M.Ik, Mitchell, D.B., & Haggard, L.C. (1992). Memory under
anesthesia: Evidence for response suppression. Bulletin of the Psychonomic Society,
30, 244-246.
Brown, A.S., & Mitchell, D.B. (1991). Age differences in retrieval consistency and
response dominance. Journal of Gerontology: Psychological Sciences, 46, 332-339.
Brown, A.S., & Mitchell, D.B. (1994). A reevaluation of semantic versus nonsemantic
processing in implicit memory. Memory & Cognition, 22, 533-541.
Brown, A.S., Neblett, D.R., Jones, T.C., & Mitchell, D.B. (1991). Transfer of processing
in repetition priming: Some inappropriate findings. Journal of Experimental
Burke, D.M., MacKay, D., Worthley, J., & Wade, E. (1991). On the tip of the tongue:
What causes word finding failures in young and old adults? Journal of Memory and
Language, 30, 542-579.
Burke, D.M., & Peters, L. (1986). Word associations in old age: Evidence for
consistency in semantic encoding during adulthood. Psychology and Aging, , 283-
292.
Cerella, J., & Fozard, J.L. (1984). Lexical access and age. Developmental Psychology,
20, 235-243.
Collins, A.M., & Loftus, E.F. (1975). A spreading activation theory of semantic
processing. Psychological Review, 82, 407-428.
Craik, F.I.M. (1986). A functional account of age differences in memory. In F. Klix & H.
Hagendorf (Eds.), Human memory and cognitive capabilities (pp. 409-422).
Am~erdam: Elsevier.
Craik, F.I.M., & Jennings, J.M. (1992). Human memory. In F.I.M. Craik & T.A.
Salthouse (Eds.), The handbook of aging and cognition (pp. 51-110). Hillsdale, NJ:
Erlbaunl
Semanticprocesses in implicit memory 137
Dannenbring, G.L., & Briand, I~ (1982). Semanticpriming and the word repetition effect
in a lexical decision task. Canadian Journal of Psychology, 36, 435-444.
Davis, H.P., Cohen, A., Gandy, M., Colombo, P., Van Dusseldorp, G., Simolke, N.,
Romano, J. (1990). Lexical priming as a function of age. Behavioral Neuroscience,
104, 288-297.
Dywan, J., & Jacoby, L. (1990). Effects of aging on source monitoring: Differences in
susceptibility to false fame. Psychology and Aging, 5, 379-387.
Finley, G.E., & Sharp, T. (1989). Name retrieval by the elderly in the tip-of-the-tongue
paradigm: Demonstrable success in overcoming initial failure. Educational
Friedman, D., Hamberger, M., & Ritter, W. (1993). Event-rdated potentials as indicators
of repetition priming in young and older adults: Amplitude, duration, and scalp
distribution. Psychology and Aging, 8, 120-125.
Gardiner, J.M., Dawson, A.J., & Sutton, E.A. (1989). Specificity and generality of
enhanced priming effects for self-generated study items. American Journal of
Gibson, J.M., Brooks, J.O., Friedman, L., & Yesavage, J.A. (1993). Typography
manipulations can affect priming of word stem completion in older and younger adults.
Grat~ P., & Schacter, D.L. (1985). Implicit and explicit memory for new associations in
normal and amnesic subjects. Journal of Experimental Psychology: Learning,
Grober, E., Gitlin, H.L., Bang, S., & Buschke, H. (1992). Implicit and explicit memory in
young, old, and demented adults. Journal of Clinical and Experimental
Hamberger, M., & Friedman, D. (1992). Event-related potential correlates of repetition
priming and stimulus classification in young, middle-aged, and older adults. Journal of
Hartman, M., & Hasher, L. (1991). Aging and suppression: Memory for previously
relevant information. Psychology and Aging, 6, 587-594.
Hashtroudi, S., Chrosniak, L.D., & Schwartz, B.L. (1991). Effects of aging on priming
and skill learning. Psychology and Aging, 6, 605-615.
Hayrick, L. (1984). When does aging begin? Research on Aging, 6, 99-103.
Hebb, D.O. (1978). On watching myself get old. Psychology Today, 12(6), 15-23.
Heindel, W.C., Salmon, D.P., & Butters, N. (1990). Pictorial priming and cued recall in
Alzheimer's and Huntington's Disease. Brain and Cognition, 13, 282-295.
Heller, 1LB., & Dobbs, A.R. (1993). Age differences in word finding in discourse and
nondiscourse situations. Psychology and Aging, 8, 443-450.
Howard, D.V. (1988). Implicit and explicit assessment of cognitive aging. In M.L. Howe
& C.J. Brainerd (Eds.), Cognitive development m adulthood: Progress in cognitive
development research (pp. 3-37). New York: Springer-Verlag.
Howard, D.V. (in press). The aging of implicit and explicit memory. In Blanchard-Fields,
F., & Hess, T.M. (Eds.), Perspectives on cognitive changes m adulthood and aging.
New York: McGraw-Hill.
138 D.B. Mitchell
Howard, D.V., & Howard, J.H. (1992). Adult age differences in the rate of learning serial
patterns: Evidence from direct and indirect tests. Psychology and Aging, 7, 232-241.
Howard, D.V., & Pulido, A. (1994). Differential effects of age and distraction on
implicit and explicit tests: Mere exposure versus recognition of Turkish words.
Cognitive Aging Conference, Atlanta, April.
Hultsch, D.F., Hertzog, C., Small, B.J., McDonald-Miszczak, L., & Dixon, 1LA. (1992).
Short-term longitudinal change in cognitive performance in later life. Psychology and
Aging, 7, 571-584.
Hultsch, D.F., Masson, M.E.J., & Small, B.J. (1991). Adult age differences in direct and
indirect tests of memory. Journal of Gerontology: Psychological Sciences, 46, P22-
30. [Corrected table, P339.]
Jacoby, L. (1983). Perceptual enhancement: Persistent effects of an experience. Journal
of Experimental Psychology: Learning, Memory, and Cognition, 9, 21-38.
Jacoby, L.L., & Dallas, M. (1981). On the relationship between autobiographical memory
and perceptual learning. Journal of Experimental Psychology: General 110, 306-
340.
Jacoby, L.L., Kelley, C., Brown, J., & Jasechko, J. (1989). Becoming famous overnight:
Limits on the ability to avoid unconscious influences of the past. Journal of
Personality and Social Psychology, 15, 326-338.
James, W. (1890). The principles of psychology (vol. I). London: Macmillan.
Jennings, J.M., & Jacoby, L.L. (1993). Automatic versus intentional uses of memory:
Aging, attention, and control. Psychology and Aging, 8, 283-293.
Karayanidis, F., Andrews, S., Ward, P.B., & McConaghy, N. (1993). Event-related
potentials and repetition priming in young, middle-aged, and elderly normal subjects.
Cognitive Brain Research, 1, 123-134.
Kausler, D.H. (1991). Experimental psychology, cognition, and human aging. New
York: Springer-Verlag.
Kausler, D.H. (1994). Learning and memory in normal aging. San Diego: Academic
Press.
Kolers, P.A. (1976). Reading a year later. Journal of Experimental Psychology: Human
Learning and Memory, 2, 554-565.
Laver, G.D., & Burke, D.M. (1993). Why do semantic priming effects increase in old
age? A meta-analysis. Psychology and Aging, 8, 34-43.
Light, L. L. (1991). Memory and aging: Four hypotheses in search of data. Annual
Review of Psychology, 42, 333-376.
Light, L.L,, & Albertson, S.A. (1989). Direct and indirect tests of memory for category
exemplars in young and older adults. Psychology and Aging, 4, 487-492.
Light, L.L, LaVoie, D., Valencia-Laver, D., & Mead, G. (1992). Direct and indirect
measures of memory for modality in young and older adults. Journal of Experimental
Light, L. L., & Sing,h, A. (1987). Implicit and explicit memory in young and older adults.
Journal of Experimental Psychology: Learning, Memory, and Cognition, 13, 531-
541.
Light, L. L., Singh, A., & Capps, J. L. (1986). Dissociation of memory and awareness in
young and older adults. Journal of Clinical and Experimental Neuropsychology, 8,
62-74.
Lima, S.D., Hale, S., & Myerson, J. (1991). How general is general slowing? Evidence
from the lexical domain. Psychology and Aging, 6, 416-425.
Madden, D.J., Pierce, T.W., & Allen, P.A. (1993). Age-related slowing and the time
course of semantic priming in visual word identification. Psychology and Aging, 8,
490-507.
Mantyl/i, T. (1993). Knowing but not remembering: Adult age differences in recollective
experience. Memory & Cognition, 21, 379-388.
Maylor, E.A. (1990a). Age, blocking, and the tip of the tongue state. British Journal of
Maylor, E.A. (1990b). Recognizing and naming faces: Aging, memory retrieval and the
tip of the tongue state. Journal of Gerontology: Psychological Sciences, 45, P215-
P226.
Milgram, N.W., Head, E., Weiner, E., & Thomas, E. (1994). Cognitive functions and
aging in the dog: Acquisition of nonspatial visual tasks. Behavioral Neuroseience,
108, 57-68.
Mitchell, D.B. (1989). How many memory systems? Evidence from aging. Journal of
Experimental Psychology: Learning, Memory, and Cognition, 15, 31-49.
Mitchell, D.B. (1993). Implicit and explicit memory for pictures: Multiple views across the
lifespan. In P. Graf & M.E.J. Masson (Eds.), Implicit Memory: New Directions in
Cognition, Development, and Neuropsychology (pp. 171-190). Hillsdale, NJ:
Erlbaunl
Mitchell, D.B., & Brown, A.S. (1988). Persistent repetition priming in picture naming
and its dissociation from recognition memory. Journal of Experimental Psychology:
Learning, Memory, and Cognition, 14, 213-222.
Mitchell, D.B., & Brown, A.S. (1990, March). Generation Gaps: Procedural Memory
Anomalies in Time and Aging. Cognitive Aging Conference, Atlanta.
Mitchell, D.B., Brown, A.S., & Murphy, D.1L (1990). Dissociations between procedural
and episodic memory: Effects of time and aging. Psychology and Aging, 5, 264-276.
Mitchell, D.B., Hunt, 1L1L, & Schmitt, F.A. (1986). The generation effect and reality
monitoring: Evidence from dementia and normal aging. Journal of Gerontology, 41,
79-84.
Mitchell, D.B., & Schmitt, F.A. (1994). Multiple memory systems: Normal aging vs.
Alzheimer's disease. Unpublished manuscript.
Morton, J. (1979). Facilitation in word recognition: Experiments causing change in the
logogen model. In P.A. Kolers, M.E. Wrolstad, & H. Bouma (Eds.), Processing of
visible language (vol. 1). New York: Plenun~L
Moscovitch, M. (1982). A neuropsychological approach to perception and memory in
normal and pathological aging. In F.I.M. Craik & S. Trehub (Eds.), Aging and
cognitive processes (pp. 55-78). New York: Plenum Press.
Moscovitch, M., Winocur, G., & McLachlan, D. (1986). Memory as assessed by
recognition and reading time in normal and memory-impaired people with Alzheimel~s
140 D.B. Mitchell
disease and other neurological disorders. Journal of Experimental Psychology:

General, 115, 331-347.
Mullen, B., & Rosenthal, It (1985). BASIC meta-analysis: Procedures and programs.
Hillsdale, NJ: Erlbaum
Myerson, J., Ferraro, F.~, Hale, S., & Lima, S.D. (1992). General slowing in semantic
priming and word recognition. Psychology andAging, 7, 257-270.
Park, D.C., & Shaw, R.J. (1992). Effect of environmental support on inrplieit and explicit
memory in younger and older adults. Psychology and Aging, 7, 632-642.
Parkin, A.J., & Walter, B.M. (1992). Reeolleetive experience, normal aging, and frontal
dysfunction. Psychology and Aging, 7, 290-298.
Perlmutter, M. (1980). An apparent paradox about memory aging. In L.W. Poon, J.L.
Fozard, L.S. Cermak, D. Arenberg, & L.W. Thompson (Eds.), New directions in
memory and aging (pp. 345-353). Hillsdale, NJ: Erlbaum
Perlmutter, M. & Mitchell, D.B. (1982). The appearance and disappearance of age
differences in adult memory. In F.I.M. Craik & S. Trehub (Eds.), Aging and
Cognitive Processes. New York: Plenum Press, 1982.
Plouffe, L., & Stelmack, 1LM. (1984). The electrodermal orienting response and memory:
An analysis of age differences in picture recall. Psychophysiology, 21, 191-198.
Rajaram, S., & Roediger, H.L. (1993). Direct comparison of four implicit memory tests.
Journal of Experimental Psychology: Learning, Memory, and Cognition, 19, 765-
776.
Ritchie, IC, Ledrsert, B., & Touchon, J. (1993). The Eugeria study of cognitive ageing:
Who are the "normal" elderly? International Journal of Geriatric Psychiatry, 8, 969-
977.
Roediger, H.L. II1 (1990a). Implicit memory: Retention without remembering. American
Psychologist, 45, 1043-1056.
Roediger, H.L. HI (1990b). Implicit memory: A commentary. Bulletin of the
Psychonomic Society, 28, 373-380.
Roediger, H.L., Guynn, M.J., & Jones, T.C. (1994). Implicit memory: A tutorial review.
In G. d~Ydewelle, P. Eden, & P. Bertelson (Eds.), International perspectives on
psychological science: The state of the art (vol. 2, pp. 67-94). Hove, England:
Erlbaun~
Roediger, H.L, & McDermott, ICB. (1993). Implicit memory in normal human subjects.
In J. Grafinan & F. Boiler (Eds.), Handbook of neuropsychology (vol. 8, pp. 63-131).
Am~erdam: Elsevier.
Roediger, H.L., & Srinivas, IC (1993). Specificity of operations in perceptual priming. In
P. Graf & M.E.J. Masson (Eds.), Implicit memory: New directions m cognition,
development, and neuropsychology (pp. 17-48).
Roediger, H.L., Weldon, M.S., & Challis, B.H. (1989). Explaining dissociations between
implicit and explicit measures of retention: A processing account. In H.L. Roediger &
F.I.M. Craik (Eds.), Varieties of memory and consciousness: Essays m honour of
Endel Tulving (pp. 3-41). Hillsdale, NJ: Erlbaunl
Rose, T.L., Yes~vage, J.A., Hill, 1LD., & Bower, G.H. (!986). Priming effects and
recognition memory in young and elderly adults. Experimental Aging Research, 12,
31-37.
Rugg, M.D. (1987). Dissociation of semantic priming, word and non-word repetition
effects by event-related potentials. The Quarterly Journal of Experimental
Psychology, 39A, 123-148.
Russo, 1L, & Parkin, A.J. (1993). Age differences in implicit memory: More apparent
than real. Memory & Cognition, 21, 73-80.
Salthouse, T.A. (1991). Theoretical perspectives on cognitive aging. Hillsdale, NJ:
Erlbaunl
Scarborough, D.L., Cortese, C., & Scarborough, H.S. (1977). Frequency and repetition
effects in lexical memory. Journal of Experimental Psychology: Human Perception
and Performance, 3, 1-17.
Schacter, D.L. (1987). Implicit memory: History and current status. Journal of
Schacter, D.L. (1992). Understanding i~licit memory: A cognitive neuroscience
approach. American Psychologist, 47, 559-569.
Schacter, D.L., Bowers, J., & Booker, J. (1989). Intention, awareness, and implicit
memory: The retrieval intentionality criterion. In S. Lewandowsky, J.C. Dtmn, & I~
Kirsuer (Eds.), Implicit memory: Theoretical issues (pp. 47-65). Hillsdale, NJ:
Erlbaunl
Schacter, D.L., Cooper, L.A., & Valdiserri, M. (1992). Implicit and explicit memory for
novel visual objects in older and younger adults. Psychology and Aging, 7, 299-308.
Schulz, 1L (1994). Introduction: Debate on generalized theories of slowing. Journal of
Gerontology: Psychological Sciences, 49, P59.
Skinner, B.F. (1983). Intellectual self-management in old age. American Psychologist,
38, 239-244.
Squire, L.1L (1992). Declarative and nondeclarative memory: Multiple brain systems
supporting learning and memory. Journal ofCogmtive Neuroscience, 4, 232-243.
Srinivas, K., & Roediger, H.L. HI. (1990). Classifying implicit memory tests: Category
association and anagram solution. Journal of Memory and Language, 29, 389-412.
Steen-Patterson, L., Jones, T.C., Brown, A.S., & Mitchell, D.B. (November, 1993).
Cross format priming: Transfer appropriate aging. Presented at The Psychonomic
Society, Washington, DC.
Thomas, J.C., Fozard, J.L., & Waugh, N.C. (1977). Age-related differences in naming
latency. American Journal of Psychology, 90, 499-509.
Tulving, E. (1985). How many memory systems are there? American Psychologist, 40,
385-398.
Tulving, E. (1991). Concepts of human memory. In L.1L Squire, N.M. Weinberger, G.
Lynch, & J.L. McGaugh (Eds.), Memory: Organization and locus of change (pp. 3-
32). New York: Oxford University Press.
Tulving, E., & Schacter, D.L. (1990). Priming and human memory systems. Science,
247, 301-306.
Warrington, E.IC, & Weiskrantz, L. (1968). New method of testing long-term retention
with special reference to amnesic patients. Nature, 217, 972-974.
Weldon, M.S., & Roediger, H.L. (1987). Altering retrieval demands reverses the picture
superiority effect. Memory & Cognition, 15, 269-280.
142 D.B. Mitchell
Wiggs, C.L. (1993). Aging and memory for frequency of occurrence of novel visual
stimuli: Direct and indirect measures. Psychology and Aging, 8, 400-410.
Evidence for task specificity in age-related slowing: A review of speeded picture-

word processing studies*
Paul C. Amrhein
University of New Mexico
In this chapter, studies concerning aging and speeded picture-word processing are
reviewed. As a metric for their evaluation, prevalent hypotheses concerning aging and visual
cognition--borrowed from the psychometric and nonspeeded memory literature--are discussed
as well as current picture-word processing models. Then, as a nominal distinction, studies
employing comparison tasks are evaluated separately from studies employing production tasks.
A "Brinley plot" regression analysis of the condition latencies of these studies reveals that
beyond additive slowing for both task types, proportional slowing occurs only for comparison
tasks in the elderly data. Moreover, this task distinction is also found for strictly lexical
studies, and thus indicates (at least within the modal response latency range) stimulus modality-
independent task specificity for age-related slowing. Directions for future investigations of this
finding are explored.
1. INTRODUCTION
At first glance, this chapter may seem to depart from the general theme of word and
language processing addressed in many of the chapters of this book. However, picture
processing tasks, specifically picture-naming tasks, are currently used in non-aging studies to
study word retrieval (e.g., Levelt, Schriefers, Meyer, Pechmau, Vorberg, & Havinga, 1991;
Wheeldon & Monsell, 1992). In addition, Bowles (1993, 1994) recently employed a picture-
naming task in her study of age differences in lexical retrieval under various priming
conditions. In these studies, necessary assumptions (implicit or explicit) are made concerning
the processes incurred in picture-naming, including picture-stimulus detection, recognition,
semantic information retrieval, name retrieval (usually concerning phonetic information to be
used in a verbal response) as well as verbal response preparation and execution. Such
assumptions, of course, require a commitment to a theoretical model of picture-word
processing. This is not a trivial matter given that there are a number of such models including
those which focus on picture processing at the processing-stage transfer level: Amodal
conceptual with picture-meaning access advantage (Potter & Faulconer, 1975; M.C. Smith &
Magee, 1980; Glaser, 1992), Amodal conceptual with picture-word meaning access
equivalence (Amrhein, 1994; Amrhein & Theios, 1993; Seymour, 1979; Snodgrass, 1980,
1984; Theios & Amrhein, 1989) and Dual Coding (Paivio, 1971, 1986; te Linde, 1982); and at
AUTHOR NOTE: Correspondence concerning this chapter should be directed to: Paul C. Amrhein,
Department of Psychology, University of New Mexico, Albuquerque, NM 8 7 1 3 1 . E-mail:
amrhein@hydra.unm,edu.
144 P. c. Amrhein
least one which addresses picture-processing at the within-stage network level (Bowles, 1993,
1994; Dell & O'Seaghdha, 1991).
Invoking a comparison between word and picture processing is also intuitive, because
both types of processing involve the perceptual recognition of symbols which can be used in
the retrieval of semantic information, and because elements (denoting concrete concepts) from
one symbol system can be readily translated into corresponding dements in the other symbol
system In the least, if pictures are to be used to study word production, we should have a
good idea how pictures are processed, generally, before we use them as tools for the study of
word retrieval via semantic memory. When studying picture and word processing across the
lifespan, these issues become even more important.
I define "picture-word processing" as a set of mental operations performed on pictorial
and verbal stimuli resulting in a production (verbal or graphic) or comparison (categorization,
judgment or matching) response. Common to all tasks are operations concerned with stimulus
perceptual encoding. Specific to production tasks are operations concerned with within-mode
and cross-modal translation of a pictorial or verbal stimulus; retrieval of the modality-specific
representation designated by the output task, and production preparation and execution.
Frequently layered onto the operations for production tasks are additional operations which
occur in picture-word priming paradigms (exhibiting facilitation, inhibition, or Stroop-like
interference effects) such as those concerned with perceptual recognition of the prime stimulus
and concomitant activation or inhibition of related or unrelated stimulus forms (where
relatedness is either phonetic, orthographic, or graphic) or stimulus semantic information (see
e.g., Bajo, 1988; Bowles, 1994; Glaser, 1992; Glaser & Dunkelhofl~ 1984; Glaser & Glaser,
1989; Levelt et al., 1991; McEvoy, 1988). Specific to comparison tasks are operations
concerned with translation of an input stimulus into a conceptual representation for
comparison with a second stimulus or categorical memory structure. Depending o n the model,
this representation may be modality specific (Dual Coding model: pictorial or verbal) or
modality non-specific (Amodal models). Remaining operations consist of the comparison
process itself and finally, (oRen binary) response preparation and execution.
The purpose of this chapter is four-fold: (a) Review the prevailing theoretical
perspectives concerning aging and picture-word processing; (b) Review the current literature
concerning age differences and similarities in speeded picture-word processing; (c) Discuss
how the studies reviewed impact on the current debate concerning the nature of age-related
slowing in speeded tasks: Generalized (e.g., Cerella, 1990, 1994; Myerson & Hale, 1993) or
localized (e.g., Amrhein & Theios, 1993, Geary, Frensch & Wiley, 1993; Allen, Madden,
Weber & Groth, 1993; Fisk & Fisher, 1994; Fisk, Fisher & Rogers, 1992; Perfect, 1994); and
(d) Discuss the future directions of aging and speeded picture-word processing research.
2. LITERATURE REVIEW
2.1. Theoretical Issues
To understand the studies conducted on aging and speeded picture-word processing to

date, it is important to discuss two basic hypotheses that have pervaded that literature: First,
elderly persons should be slower or more errorful than young persons in responding to
pictorial or verbal stimuli and second, this slower or poorer performance should be more
pronounced for pictorial stimuli than verbal stimuli (henceforth "elderly slowing" and "elderly
Evidence for task specificity in age-related slowing 145
spatial deficit" hypotheses, respectively). Somewhat surprisingly, neither hypothesis is based

on the findings of studies which have specifically investigated age and speeded picture-word
processing. Rather, the elderly slowing hypothesis has been generalized from numerous earlier
reaction time studies (see e.g., Botwinick, 1984), and the elderly spatial deficit hypothesis has
been derived from psychometric test results and findings from experiments concerned with
non-speeded episodic memory tasks (otten employing, for example, imagery as a mediator in
memorial encoding). To be discussed later, these hypotheses are at the root of the prevalent
slowing debate on whether this slowing is generalized across stimulus and task domains or is
specific to these domains.
Concerning the elderly spatial deficit hypothesis further: From the psychometric
literature (see Kausler, 1991 for a review), there is indeed evidence that with increasing age,
there is a steeper decline in scores from the Performance scales (which tap spatial abilities) than
in scores on the Verbal Scales of the WAIS-1L In addition, from the episodic memory
literature is the related evidence of an elderly deficit in picture memory (Arenberg, 1978;
Harker & Riege, 1985; Park & Puglisi, 1985; Pezdek, 1983; A.D. Smith, Park, Cherry &
Berkovsky, 1990; Trahan, Larrabee & Levitt, 1986; Tubi & Calev, 1989; Winograd & Simon,
1980; Winograd, A.D. Smith & Simon, 1982) and more specifically in spontaneously
generating images from words, at least in non-speeded, word-learning tasks (e.g., Canestrari,
1968; Hulicka & Grossman, 1967; Treat & Reese, 1976; Winograd & Simon, 1980). In fact,
one researcher concluded a decade ago that "the decline of spatial abilities before verbal
abilities is so well established that it is often referred to as 'the classic aging pattern'" (p. 201,
Halpem, 1984). As was stated before, this conclusion was made primarily on the basis of non-
speeded episodic memory tasks, or in the case of the WAIS-R scales, where response latency
was not explicitly measured or analyzed for the various picture-word processing operations
listed earlier.
A third, much less tested, hypothesis concerns age differences in regard to specific
predictions made by picture-word processing models in the general literature. Such a
hypothesis test is potentially valuable, because such differences can be used to test predictions
of these models as truly "general" cognitive models which cover the lifespan. Conversely,
these models can act as catalysts for the testing of hypotheses concerning aging and picture-
word processing. Of course, sufficient task conditions (e.g., using picture and word stimuli)
must be manipulated for discriminating model tests to be possible; as will be seen later, this has
rarely been the case.
There are two picture-word processing models typically tested in the empirical
literature: the Dual Coding and Amodal (or Common-code) models. These theoretical models
are grounded in the human information processing paradigm and as such make model
predictions in terms of processing stages and latency to carry them out for task completion.
The most venerable of these is the Dual Coding model of Paivio (1971, 1983, 1986). Figure 1
presents a representation of this model denoting relationships among processing stages for a
range of tasks (e.g., comparison, production, etc.). This model posits modality specific
processors which are concerned with the representation and processing of different aspects of
the meaning of a concept. For example, the verbal processor deals with aspects of concept
meaning such as categorical membership, whereas the pictorial processor deals with aspects of
concept meaning such as relative size of the concept when represented as an object.
146 P. C. Amrhein
WORD PICTURE
Early Visual Processing
Surface/ l Surface/
Conceptual Conceptual
Linguistic Pictorial
Processor Processor
Phonetic, Imagistic,
Orthographic Graphic
and and
Semantic Semantic
Codes Codes
Language
Production Binary Response DrawingProduction
(Speech & Writing) System System
Systems
WORD YES/NO PICTURE
Figure 1. Flow diagram representing the general architecture of the Dual Coding model (Paivio, 1971, 1986),
configured for comparison and production tasks.
Of particular relevance here, this model also posits that accessing a verbal code for a
picture stimulus should require less time than accessing an image code for a verbal stimulus
(see Paivio, 1971, 1986; PeUegrino, Rosinski, Chiesi & Siegel, 1977). The apparent logic for
this prediction is based on Paivio's claim that for a non-speeded episodic task, "pictures of
common objects would be remembered even better than concrete words on the assumption that
subjects are more likely to label pictures spontaneously than they are to image to concrete
words" (Paivio, 1986, p. 160). Paraphrasing Amrhein (1994), Paivio's claim suggests two
types of subprocesses for cross-modal translation: spontaneous and less-than-spontaneous. If
we assume that response latency for a given trial is a cumulative function of the latency of each
incurred subprocess weighted by its probability of occurrence (see Falmagne, 1965; Lupker &
Theios, 1975) then the prediction of asymmetric cross-modality transfer latency can be
explained. Specifically, given a cross-modality transfer situation where spontaneous
subprocesses are more likely to function than less-than-spontaneous subprocesses, it is then
expected that the cross-modality transfer latency will be less than for a cross-modality transfer
situation where the less-than-spontaneous subprocesses are more likely to function than the
spontaneous subprocesses.
Figure 2 gives a representation of the Amodal model denoting relationships among
processing stages for a range of tasks (e.g., comparison, production, etc.). As can be seen in
Figure 2, the Amodal model posits modality specific surface processors (and associated
memory stores), one for pictures and one for words. Each surface processor has direct access
to an amodal conceptual meaning processor (and associated memory store). Thus, in contrast
to the Dual Coding model, the amodal modal treats meaning as modality-independent.
The Amodal model has two variants: One variant posits that pictures and words are
recoded into an amodal meaning representation at an equivalent rate (Snodgrass, 1980, 1984;
Seymour, 1973, 1979; Theios & Amrhein, 1989); the second variant posits that pictures are
recoded into an amodal meaning representation at a faster rate than words (e.g., Bajo, 1988;
Glaser, 1992; Potter and Faulconer, 1975; M.C. Smith and Magee, 1980). However, support
for this model has been derived primarily from experiments either involving large pictures and
small words or pictures uncontrolled for featural similarity; confounds which when removed
[i.e., when stimuli subtend 2.2-8.0 ~ of visual angle (Theios & Amrhein, 1989), or disparity
between visual similarity within and between conceptual categories is reduced (Snodgrass &
McCullough, 1986)] result in no such advantage for pictures. With the prediction of the
picture advantage removed, this type of model becomes functionally equivalent to other
amodal models (although this was recently challenged in the literature, see Glaser, 1992).
WORD PICTURE
Early Visual Processing
Surface Surface
Linguistic Abstract
Conceptual Pictorial
Processor Processor
Processor
Phonetic & Imagistic &
Semantic
Orthographic
Codes Graphic
Codes Codes
Language
Production Binary Response Drawing Production
(Speech & Writing) System System
Systems
WORD YES/NO PICTURE
Figure 2. Flowdiagram representing the general architecture of the amodal picture-word processing model of
Theios and Amrhein (1989), configuredfor comparisonand production tasks.
When the stimulus conditions listed above are met, and the concepts denoted by the
stimuli are familiar, then the latency to transfer processing of a stimulus from one surface
processor to the other via an amodal, conceptual processor is equivalent regardless of stimulus
modality, in direct contrast to the Dual Coding prediction of asymmetry concerning cross-
modality transfer. This, of course, impacts on dual stimulus comparison tasks. For example,
148 P. C. Amrhein
semantically comparing a picture to a preceding word stimulus should require the same latency
as semantically comparing a word to a preceding picture stimulus. However, the Dual Coding
model predicts that semantically comparing a picture to a preceding word stimulus should
require more time than semantically comparing a word to a preceding picture stimulus, because
semantic information (e.g., concerning categorical concepts) is lexically based. (However, see
te Linde, 1982, for a qualification of the original theoretical position).
Meeting such stimulus conditions also allows a direct determination of cross-modality
transfer latency. For example, using a picture-naming, word reading task (see e,g., Amrhein,
1994; Cattell, 1886; Fraisse, 1969; Theios & Amrhein, 1989), the latency to access a picture's
meaning and ultimately the phonological code for its name is given by the latency difference
between onset to name the picture and onset to read that picture's name. In the case where
picture-to-word and word-to-picture transfer is assessed (as in a drawing-writing task, see
Amrhein, 1994; Amrhein & Theios, 1993), these transfer latencies should be equivalent, again
in contrast to the asymmetry predictions of the Dual Coding model.
Finally, as can be seen in Figures 1 and 2, there are also common features among the
Dual Coding and Amodal models. Specifically, both models (explicitly or implicitly) posit
stimulus modality-independent processing for early stimulus detection and ultimate response
preparation and execution.
A different, recent model devised to account for priming effects in picture-naming is
the interactive network model of Dell and O'Seaghdha (1991). As a speech production model,
it can be tested using a picture-naming task (Bowles, 1993, 1994). In this model, there are
interconnected network levels concerned with the detection of physical and semantic features
of the stimuli, and phonemic features used in the naming response. This network model allows
for bidirectional excitatory and inhibitory connections within and between levels. Behaviorally,
this means that in stimulus priming paradigms, the influence of the prime stimulus should be
equally distributed across all featural networks involved in processing the subsequent target
stimulus. For example, naming a picture (e.g., "sheep") should activate not only conceptually
related concepts (e.g., "goat"), but also words that are phonemically related to the
conceptually related concept (e.g., "goal"). By simulating the activity of the networks, this
model can make quite precise predictions concerning featural activation flow of picture and
word stimuli over time, thus providing a test of the nature and symmetry (or asymmetry) of
cross-modality transfer situations (i.e., picture-word or word-picture prime-target pairs). To
date, however, only word-picture prime-target pairs have been employed (Bowles, 1993, 1994;
Levelt et al., 1991).
While the information processing and network models may be seen as competitors,
with the former cast as modular and the latter as distributed (Bowles, 1993), a different view is
that they are in fact complementary. That is, given the picture-naming findings of Levelt et al.
(1991), it appears there is evidence for modular processing in picture-naming at a between-
network level, but also distributed processing at a within-network level. For example, contrary
to the prediction given before, Levelt at al. (1991) found that naming a picture activates a
semantically related concept but activation of this related concept does not extend to its
phonemic relatives. Therefore, the best model may be a hybrid one incorporating aspects of
modular and interactive models. Indeed, in an exchange between Dell and O'Seaghdha (1991)
and Levelt, Schriefers, Vorberg, Meyer, Pechman & Havinga (1991), both parties agreed that
"...connectionist models of lexical access...must incorporate a substantial degree of modularity.

One could say that the lexical networks themselves should be relatively 'cold' networks. Their
pattern of performance is to a substantial degree produced by staged structure input from
outside the network proper". (p. 617, Levelt et al., 1991).
In sum, information processing models remain valuable in accounting for between-

processor activity occtm~g in picture-word comparison and cross-modality translation. Once
differences and similarities are noted at this level, then the specifics of processes occurring
within picture and word processors can be assessed using more detailed distributed models.
I would like to add here that in discussing the utility of information processing and
distributed models in revealing the nature of age differences in picture-naming, Bowles (1.993)
misrepresents the utility of the former. Concerning information processing models, she claims
that the model "stages are simply "olack boxes' that represent processing for which the
mechanisms are undefined. Typically, each stage is assumed to complete its processing before
its output is passed on to the next stage." (p. 306). Bowles (1993) adds that attempts at
identifying age effects in these stages (e.g., those concerned with picture-naming) has
produced results which have often "proven ambiguous, because in such models, impairment at
any stage would result in impairment at all subsequent stages, that is a finding of an age effect
on one stage could be due to a problem at any of the preceding stages as well." (p. 307).
Firstly, the general literature is flooded with explanations as to the nature of the
stimulus-transformation operations which occur in these supposedly "undefined" "olack boxes'
(see e.g., Levelt et al., 1991). (Whether such explanations sometimes lack the mathematical
rigor of a connectionist model is, however, a fair criticism) Secondly, only serial models
prescribe that one stage must complete its processing before the next stage begins whereas
parallel and cascade models certainly do not. Moreover, it can be argued that information
processing models of speeded performance can account for serial, parallel and cascade
processing configurations by assessing the additional time each subprocess contributes to
overall performance latency (see Theios & Amrhein, 1989). Finally, as will be reviewed below,
Amrhein and Theios (1993) have demonstrated that age effects can indeed be isolated using a
stage model; in their study, age-related slowing was found for sensory-motor processing but
not for processing concerning cross-modality transfer.
2.2. Studies
In aging and speeded picture-word processing studies, two methodologies have been
primarily employed: Dual stimulation comparison tasks (e.g., Halpem, 1984; Elias &
Kinsbourne, 1974; Mergler & Zandi, 1983; Nebes, 1976) and production tasks (i.e., drawing-
writing: Amrhein & Theios, 1993, and picture-naming: Bowles, 1994; Mitchell, 1989; Pooh &
Fozard, 1978; Thomas, Fozard & Waugh, 1977; see also Goulet, Ska & Kahu, 1994, for a
review). I will review the studies employing comparison tasks first, followed by those
employing production tasks.
Comparison Tasks
Elias and Kinsbourne (1974) presented elderly and young subjects pairs of stimuli
consisting of either arrows or letter bigrams which corresponded to left or fight clock~se and
counter-clockwise rotations [e.g., the concept "counter-clockwise rotation up from the fight"
150 P. C. Amrhein
was represented by the non-verbal stimulus " ~" " or the verbal stimulus "UR" (for Up-
Right)]. The subject's task was to decide whether the members of a given pair (presented
sequentially) were consistent in their representation of directional rotation. Elderly subjects
were slower than the young subjects across the various trial conditions. In addition, a gender
effect was found which interacted with age group. Specifically, male subjects exhibited no
effect for stimulus modality but female subjects exlfibited faster response latencies for verbal
than for pictorial stimuli. Furthermore, elderly females exhibited an amplified version of this
stimulus modality effect relative to young females.
In the Nebes (1976) study, elderly and young subjects decided whether two stimuli,
either a verbal description or a picture, and a following picture (at a variable delay of 0, 1, 2, or
3 seconds) represented the same concept. Nebes found that when there was no delay between
the first and second stimulus, decision latency was slower for verbal description-picture than
picture-picture trials. However, after a delay of one second (and continuing on for delays of 2
and 3 seconds), no latency difference between these two conditions was found, causing Nebes
to conclude that a one second delay provided strfficient time to recode the verbal description
stimulus into a representation (possibly an image) comparable to that of the following picture
stimulus. Overall, elderly subjects were slower than young subjects across condition latencies.
Furthermore, the pattern of results was similar for both age groups.
Mergler and Zandi (1983) presented elderly and young subjects with stimulus triads
consisting of a central "standard" traffic sign (symbolic or verbal) and two different flanking
traffic signs (both verbal or symbolic). Subjects were timed on how quickly they could
determine which flanking traffic sign message matched the standard sign. Overall, elderly
subjects were slower than the young subjects. Further, modality of standard and flanking sign
affected performance such that latency when comparing pictorial flankers to a verbal standard
was less than when comparing verbal flankers to a pictorial standard; an effect which did not
interact with age group.
Finally, a related study using a different set of traffic signs (but with two signs also used
by Mergler & Zandi, 1983) was conducted by Halpem (1984). On a given trial, elderly and
young subjects were presented with a traffic sign message read out loud followed by a slide of
a verbal or symbolic version of that or another traffic sign message. The subjects' task was to
determine whether the message read out loud matched the message conveyed by the traffic
sign in the slide. Overall, elderly subjects were slower than young subjects. But, there was
also an interaction between age group and stimulus modality: Elderly subjects responded faster
to verbal than to symbolic signs, whereas young subjects exhibited no difference between the
two sign modalities.
Theoretical Analysis
In terms of the elderly slowing hypothesis given earlier, all the studies reviewed here
found that elderly were slower than young subjects in overall performance. However, the
findings are mixed concerning the elderly spatial deficit hypothesis. The elderly female subjects
of the Elias and Kinsboume (1974) study and elderly subjects of the Halpem (1984) study
were slower when processing pictorial than verbal stimuli. However, age differences in
processing pictorial vs. verbal stimuli were not found for the male elderly subjects of Elias and
Kinsbourne (1974) (relative to their young male subjects), or for the elderly subjects of Nebes
(1976) and Mergler and Zandi (1983).
Concerning the third hypothesis which deals with whether these data support or refute
general picture-word processing models: Apparent support or refutation for any of these
models is handicapped by methodological weaknesses among these comparison task studies.
In general, none of these experiments manipulated a semantic variable such as category
membership, a variable which has been used often to test picture-word models in the general
literature (e.g., Harris, Morris & Bassett, 1977; Pellegrino et al., 1977; Potter & Faulconer,
1975; Snodgrass & McCullough, 1986; te Linde, 1982). Moreover, Elias and Kinsbourne
(1974) used unfamiliar representations for the concept of angular rotation. In the Nebes
(1976) study, one second was likely too lengthy of a delay to sensitively detect when elderly
and young subjects actually recoded the verbal description stimulus for comparison to the
following picture stimulus. And, Nebes (1976) did not replicate the gender effects reported by
Elias and Kinsboume (1974). Also damaging, the data of Mergler and Zandi (1983) suffer
from a age-differential speed-accuracy tradeoff: Their elderly subjects were more errorful on
their faster condition latencies, but less errorfid on their slower condition latencies; a pattern
which was not exhibited by their young subjects.
Most importantly, none of these studies involved a comprehensive comparison of
within-modality (e.g., word to word or picture to picture) and cross-modality (i.e., word to-
picture or picture to word) access for picture and word stimuli. Such comparisons are critical
to the testing of picture-word processing models. For example, Elias and Kinsboume (1974)
contrasted only within-modality stimulus pairs: nonverbal-nonverbal vs. verbal-verbal, while
Nebes (1976) contrasted only one type of within-modality with one type of cross-modality
stimulus pair: verbal description-picture vs. picture-picture.
Likely because of these weaknesses, the support for the Dual Coding and Amodal
models that can be gleaned from these studies is mixed. For example, the Elias and
Kinsboume (1974) study supports the Amodal model (with no picture processing advantage)
for male, but not for female subjects, whose data refutes all three information processing
models. The Nebes (1976) study indicates that subjects made their comparisons based on the
modality of the first stimulus (verbal description or picture) rather than some common amodal
meaning representation, and thus supports the Dual Coding model (assuming these were
semantically based comparisons). Model support from the Mergler and Zandi (1983) study
simply cannot be determined because they used a simultaneous presentation of mixed-modality
stimuli, thus precluding determination of the processing order of standard and flanking stimuli
(although they claim that the standard stimulus modality dictated modality of the comparison
process). Also, as noted before, their data stiffer from an age-differential speed-accuracy
tradeoff. Finally, the data from the young subjects of the Halpern (1984) study support the
Amodal model (with no picture processing advantage), but the data from their elderly subjects
support the Dual Coding model.
Production Tasks
Picture-Naming
Production tasks include picture-naming, word-reading and picture-drawing, word
writing methodologies. However, with the exception of Amrhein and Theios (1993), all other
production studies have involved only picture-naming (i.e., where onset to name a picture is
the primary dependent variable). In a recent review of aging and picture-naming experiments,
Goulet et al. (1994) discuss 25 studies, of which only three involved a measurement of picture-
152 P. C. Amrh ein
naming latency (Mitchdl, 1989; Poon & Fozard, 1978; Thomas, et al., 1977). The remaining
22 studies assessed only picture-naming accuracy; and of these studies, nine established
accuracy norms for various picture-naming tests. Only the remaining 16 studies actually
investigated the role of aging on picture-naming accuracy. The focus of their review
concerned simply whether these studies had found a si~ificant increase in response latency or
errors for older relative to younger age groups (i.e., the elderly slowing hypothesis)--
sometimes involving a comparison of an older and a younger group, and other times involving
a comparison of age groups across the adult lifespan. Among the 13 empirical studies
assessing naming accuracy, there was little consensus about the presence of such an age group
difference which would indicate aging deficits in picture-naming. While eight of the studies
reported a significant age group increase in naming errors, four reported a nonsi~ificant
difference, and one study reported b e t t e r accuracy for elderly over young picture-naming.
However, as is documented in their review, there is little standardization among the
accuracy studies concerning stimuli, subject health and educational level. Moreover, many of
the studies suffer from low statistical power or incomplete or incorrect statistical analyses. As
if this review isn't disconcerting enough, a problem with accuracy studies, in general, is that
without a co-measure of naming latency, it is difficult to determine whether elderly subjects
were slower because of true age deficits or simply because of a speed accuracy trade-off
The result pattern is similar for the three studies assessing naming latency (and
accuracy). Thomas et al. (1977) reported si~ificantly greater latency for older subjects, while
Mitchell (1989) reported a nonsi~ificant age group difference. Finally, Poon and Fozard
(1978) reported, dependent on stimulus condition, a si~ificant naming latency increase and
decrease for older relative to younger subjects. These three studies are now reviewed in
greater detail below.
Thomas et al. (1977) had subjects ranging in age from 25 to 74 years (organized into
five groups with age ranges of 25-35, 36-45, 46-55, 56-65, or 65-74 years) either name a
picture in isolation (i.e., "Naming" condition), or after a picture was preceded by a word which
on half of the trials represented that picture's name and on the other half of the trials
represented another picture's name (i.e., "Matching" and "Nonmatching" conditions,
respectively). Thomas et al. reasoned that naming latency for the Matching condition would
provide a measure of perceptual and motor subprocesses involved in a Naming condition
response. The additional latency of the Naming condition over that for the Matching condition
would represent the time required for retrieval of a picture's name from semantic memory.
Thomas et al. found similar slowing in naming latency with increases in age for both
conditions; they concluded that much of the age increase in latency for the Naming condition
(especially after uflticient practice) was due to perceptual and motor differences. In other
words, age had a minimal impact on the latency to retrieve the names of the pictures.
Poon and Fozard (1978) also investigated picture-naming by presenting subjects
ranging in age from 18 to 70 years (categorized as young, M d = 20 years; middle age, M d = 50
years; and older, A i d = 65 years) pictures depicting "dated" (circa 1910) and "contemporary"
(circa 1970) objects that were either unique to, or common across these eras. For example, a
sample "unique dated" objects would be a spittoon; a sample "unique contemporary" object
would be a calculator. A sample "common dated" object would be a wood-burning stove; a
sample "common contemporary" object would be the corresponding 1970s version of a stove.
Employing a version of the dual-stimulation paradigm of Thomas et al. (1977), subjects named
each picture either in isolation ["Naming Latency (NL)]", or when it was preceded by the its
name ["Correctly Primed (CP)]" or the name of another picture ["Incorrectly Palmed (IP)"].
As in Thomas et al. (1977), the assumption was that when the preceding picture name matched
that of the to-be-named picture, the picture-naming latency assessed input and output
processes, but not semantic information retrieval processes (the IP condition was used as a
means to reduce fast anticipatory responding potentially induced by the CP condition).
Accordingly, the difference between NL and CP conditions was taken as a latency estimate of
these semantic retrieval processes.
Overall, the results indicated that for all conditions, older, middle and young age
groups did not differ siLmificantly. However, NL latencies were fastest for the older and young
groups (but not for the middle age group) according to datedness of the depicted objects; older
subjects named dated pictures faster than younger subjects, but young subjects named
contemporary pictures faster than older subjects.
Consistent with Thomas et al. (1977), the difference between NL and CP conditions
was equivalent for all age groups, suggesting that semantic retrieval processes do not undergo
slowing with age, a finding that has been reported concerning semantic information retrieval
via word stimuli (e.g., Allen, Madden & Crozier, 1991; Bowles & POOh, 1981, 1985; Cerella
& Fozard, 1984). The slowing that Pooh and Fozard did observe occurred in both NL and CP
conditions prompting the researchers to conclude that such slowing is located in stimulus input
(e.g., picture perception) and response output processes (e.g., response preparation and
execution).
Among other measures, Mitchell (1989) investigated age differences in picture-naming
by manipulating stimulus repetition. Of relevance here is that Mitchell assessed naming latency
of "high and low codability" pictures (i.e., more reliably and less reliably named pictures,
respectively, from Snodgrass & Vanderwart, 1980) as a means of studying age differences in
semantic memory retrieval (responses on first stimulus occurrence) and procedural memory
(changes from first to second occurrence, which varied in position: 5, 25 or 50 items later).
Upon first presentation, picture stimuli were named 61 ms slower by elderly than by young
subjects, a difference which was nonsi~ificant (technically, nonsignificant at their chosen .01
level; it was siLmificant, however, at the .025 level) and this difference did not interact with the
latency difference favoring high over low codability stimuli. Moreover, this difference did not
change with the stimulus' second occurrence regardless of its position in the stimulus ensemble
or its codability.
Finally, an important picture-naming study was recently reported by Bowles (1994).
Bowles used a word-picture priming paradigm in which she manipulated prime condition
[semantically related, semantically unrelated or no-prime (XXXXX)] and target picture onset
asynchrony (SOA). Her manipulation of SOA is particularly noteworthy because the levels
were indexed for individual stimuli and subjects based on latencies for various threshold levels:
subthreshold (0% correct identification), threshold (50% correct identification) and full
threshold (100% correct identification). Bowles found that relative to the no-prime condition,
naming latencies for semantically related and unrelated conditions were elevated at an
(adjusted) SOA of slightly more than 100 ms for both age groups, but that this elevation
persisted longer (about 150 ms) for the elderly than the young subjects. Moreover, by the
longest (adjusted) SOA, both age groups exhibited no-prime and semantically unrelated
latencies that were elevated equivalently above those of the semantically related condition.
This pattern of results indicated that elderly subjects exhibit greater persistence in prime-
induced inhibition. What makes this work exciting is that this finding could be simulated using
154 P. c. Amrhein
a network model with three parameters concerning the relation between the word prime and
picture target: excitation, inhibition, and rate of activation decay.
Only Bowles (1994) and Thomas et al. (1977) found statistically significant evidence
for the elderly slowing hypothesis. Indeed, Goulet et al. (1994) have pointed out that picture-
naming studies investigating aging have often lacked sutticient statistical power to find such a
difference between age groups. Moreover, all of these studies assessed only cross-modality
latency (say word given picture stimulus) without assessing within-modality latency (say word
given word stimulus). Of course, excluding a within-mode condition precludes any test of the
elderly spatial deficit and third hypotheses (concerning predictions of the three information
processing models). By their very nature, these models require more contrastive picture-word
conditions to test their assumptions. As will be discussed below, age differences and
similarities in specific aspects of picture and word processing can be isolated when a complete
set of cross-modality and within-modality conditions are measured. In this regard, however,
the Bowles (1994) study is quite noteworthy because her results were interpreted in terms of a
theoretical (network) model specifically devised to account for aging effects in picture name
retrieval.
There are also other problems with these picture-naming studies. Concerning Pooh and
Fozard (1978), it should be noted that the interaction they reported for age group and picture
datedness also occurred for the CP and IP conditions, indicating that some semantic retrieval
processing also occurred in these conditions. Indeed, it seems unlikely that age group and
datedness factors differentially impacted on stimulus input processes and response output
processes. Also, the target picture stimuli used by Bowles (1994) subtended 14~ of visual
angle, likely requiring saccadic eye movements to be completely viewed. Statistically, this
would result in greater variability in naming latencies; indeed, Bowles did not find a significant
three-way interaction concerning age, SOA and priming condition. Rather, her condition
differences concerning prime stimulus inhibition were found to be siLmificant using "planned"
contrasts comparing the no-prime condition to (collapsed) related and unrelated prime
conditions.
Drawing-Writing
In general picture-naming tasks have a distinct advantage over dual stimulus
comparison tasks used in picture-word processing research (categorization: e.g., Harris,
Morris & Bassett, 1977; Pellegrino et al., 1977; Potter & Faulconer, 1975; Snodgrass &
McCuUough, 1986; te Linde, 1982) and semantic matching (e.g., Elias & Kinsboume, 1974;
Nebes, 1976; Pellegrino, et al., 1977; Theios & Amrhein, 1989, Experiment 2). That is, while
dual stimulus comparison tasks allow inferences concerning encoding and access of verbal and
pictorial representations from semantic memory, these inferences may be influenced by the
effects of decision subprocesses also required for task performance. By contrast, (picture and
word) naming tasks generally provide a more direct manner of assessing representational
access (see e.g., Balota & Chumbley, 1984, 1985; Levelt et al., 1991; Schriefers, Meyer &
Levelt, 1990). For example, in the case of the naming-reading task, the difference between
onset latency to name a picture and onset latency to read that picture's name can be used to
estimate the latency to access the picture's name via semantic memory (Glaser, 1992; Theios &
Amrhein, 1989).
However, because the picture-naming, word-reading task involves only verbal

responses, it allows assessment only of the transfer latency involved in accessing a picture's
name from its picture. Moreover, until recently, the primary mode of comparing picture to
word and word to picture translation has been to contrast picture-naming latencies with image
generation latencies (see Paivio, 1966, 1986; Snodgrass, 1980). Contrasting picture-naming
and image generation latencies is problematic because of non-comparable tasks (Snodgrass,
1980). For example, the picture-naming task typically involves an observable pronunciation
onset response. By contrast, the image generation task cannot involve a comparable
observable response (onset or otherwise). While latency to make a manual ready-to-draw
response is sometimes assessed and followed by the drawing response, latency of drawing
onset is not assessed (see e.g., Paivio, 1966; Paivio, Clark, Digdon & Bolls, 1989).
Furthermore, the introspective nature of the image generation task allows for uncontrolled
subject variability in generation strategies, often leading to conflicting results (Farah &
Kosslyn, 1981; Paivio, 1986).
A more comprehensive production task for the study of picture-word translation would
assess the transfer latency involved in accessing a picture's name from its picture and transfer
latency involved in accessing a picture from its name. Such a task is the drawing-writing task
(Amrhein, 1994; Amrhein & Theios, 1993). In this task, subjects either draw a picture from a
picture or a word stimulus, or write a word from a word or picture stimulus. The dependent
measure is response onset latency, in terms of the time to begin writing a word or drawing a
picture upon stimulus presentation. This drawing-writing task provides a direct assessment of
stimulus encoding and cross-modality transfer subprocesses, independent of stimulus and task
modality. Equations (1-4) below, which are based on the architecture of the model processors
in Figure 2, demonstrate the related input stimulus encoding, cross-modality transfer, output
representation retrieval and production onset latencies incurred by the conditions of the task.
For purposes of comparison, the equations of the naming-reading task are given in (5-6):
(1) Wr,e( ~ , wj ) : t~( ~ ) + t~o + to( wj ).

(2) Write(Pi, Wj) = tE (Pi) + tT (Pi, Wj) + tLo + to (Wj).
(3) Draw(Wi,Pj) : tE (Wi) + tT(Wi,Pj) + tp + to (Pj).
(4) Draw(Pi,Pj) = tE (Pi) + tp + to (Pj).
(5) Name(Pi, Wj) = tE (Pi) + tT(Pi, Wj) + tLp + to (Sj).
(6) Read(Wi, Wj) = tE (Wi) + tLp + to (Sj).
Concerning Equations 1-4, Write(Wi, Wj) and Write(Pi, Wj) represent the total time to
initiate the writing of a word (W~) from, respectively, a corresponding (i.e., same concept)
word (14d) or picture (Pi) stimulus. Draw(I~,Pj) and Draw(Pi,Pj) represent the total time to
initiate the drawing of a picture (Pj) frOm, respectively, a corresponding (i.e., same concept)
word (Wi) or picture (Pi) stimulus. The time to encode a word or picture stimulus into its
corresponding Surface Processor, Linguistic or Pictorial is given by the parameter, tE(Wi) or
tE(Pi). In the case of writing a word from a picture stimulus (2) or drawing a picture from a
word stimulus (3), the additional time to transfer information l~om the Surface Processor
(Pictorial or Linguistic) corresponding to the modality of the input stimulus to the Surface
156 P. c. Amrhein
Processor (Linguistic or Pictorial) corresponding to the modality of the output production--via

the Abstract Conceptual Processor--is given by the transfer parameters, tT(Pi,~) and tT(~,Pj).
The additional latency to retrieve an orthographic code from the Surface Linguistic Processor
corresponding to the word to be written is given by tLo. The additional latency to retrieve a
graphic code from the Surface Pictorial Processor corresponding to the picture to be drawn is
given by tp. Lastly, the additional time to prepare for and initiate a production, either writing a
word or drawing a picture, is given by the parameters, to(~) or to(Pj), corresponding,
respectively, to the processing incurred by the Writing or Drawing Production Systems (see
Figure 2).
Concerning Equations 5-6, Name is the observed response latency to begin to name a
picture. Read is the observed response latency to begin to read a word out loud. The
additional latency to retrieve a phonetic code from the Surface Linguistic Processor
corresponding to the word to be written is given by tLp. Finally, to(S) is the final increment of
time needed to prepare for and initiate a speech response, corresponding to processing
incurred by the Speech Production System (see Figure 2). As mentioned before, Equations 5
and 6 reflect the finding in the literature that naming a picture involves accessing semantic
memory but reading a word does not (Bajo, 1988; Glaser, 1992).
In one of the few earlier investigations of picture-word processing using a drawing
task, Seymour (1974) found that for young subjects, drawing a picture from a picture stimulus
was initiated in much less time than drawing a picture from a sentence stimulus. However,
Seymour used relatively complex sentences (e.g., "The circle is inside the square.") and
pictures (e.g., ~-~ ) in his study. Such stimuli likely produced an overestimate of the transfer
time required to access a cross-modality representation (picture, here). Indeed, this value--
which represents an estimate of tT(W,P) for his stimuli--was 585 ms! In addition, because this
task only involved drawing a picture, only the transfer from a verbal stimulus to a pictorial
representation was assessed.
Of particular interest in the Amrhein and Theios (1993) drawing-writing study was the
comparison--across and within age groups--of the two encoding parameters, tE(W) and rE(P),
and in particular, the two transfer parameters, tT(P,W) and tT(W,P) (see Equations 1-4).
(Henceforth, to represent values averaged over stimulus concepts, subscripts i and j will be
omitted). If elderly individuals have greater difficulty making the cross-modal transfer to
access a picture representation relative to making the cross-modal transfer to access a word
representation (i.e., the elderly spatial deficit hypothesis), then we expected that tx(P,W) <
tx(W,P). While such a finding would also support the Dual Coding model, we expected that
this difference would be greater for the elderly than the young subjects if the Dual Coding
model correctly accounted for task performance, in general. On the other hand, if the elderly
spatial deficit hypothesis does not hold (at least for speeded semantic memory tasks) then
elderly subjects were expected to exhibit a pattern of tx(P, W) and tx(W,P) values consistent
with those of young subjects (i.e., a pattern that either supported one of the Amodal models or
the Dual Coding model).
The results of Amrhein and Theios (1993) revealed support for the elderly slowing
hypothesis. However, both age groups exhibited no differences for picture and word encoding
or in accessing cross-modality representations, thus refuting the elderly spatial deficit
hypothesis. Moreover, their results provide support, from both age groups for the Amodal
picture-word processing model of Theios & Amrhein (1989), and thus refute the Dual Coding
model (Paivio, 1971, 1975, 1983, 1986), and Amodal theories which posit a temporal
advantage for pictures in accessing an amodal, conceptual memory processor and store (e.g.,
Bajo, 1988; Potter & Faulconer, 1975; M.C. Smith & Magee, 1980).
3. PICTURE-WORD PROCESSING AND THE AGE-RELATED SLOWING DEBATE
Few issues have so occupied cognitive aging researchers as the debate concerning the
nature of slowing observed in elderly relative to young subjects in speeded processing tasks
(see e.g., Cerella, 1991a, 1991b; vs. Fisk et al., 1992; Cerella, 1994; Myerson et al., 1994 vs.
Perfect, 1994; Fisk & Fisher, 1994; and Cerella & Hale, 1994 vs. Molenaar & van der Molen,
1994). The primary mode of empirical support for General Slowing Theory has been ~om
meta-analyses based on regressions of elderly on young subject condition latencies (Hale,
Myerson, & Wagstafl~ 1987; Hale, Lima, & Myerson, 1991; Lima, Hale & Myerson, 1991) and
more recently, condition latency differences (e.g., Myerson, Ferraro, Hale, & Lima, 1992).
Not trivially, Perfect (1994) has argued--based on comments by Cerella (e.g., Cerella & Hale,
1994)--that General Slowing theory is "anti-Cognitive Psychology" because it reduces all age-
based performance differences to a mathematical description of changes in neurological
function efficiency, thus removing the need to reference stimulus or task characteristics
(beyond a dimension of "complexity") in order to predict and explain age-related slowing.
Indeed, Cerella and Hale (1994) argue that General Slowing theory (as a one parameter theory)
can account for the increase and later decrease in processing speed from childhood to late
adulthood.
There are two criticisms typically levelled at this theory. One concerns the mode of
analysis typically employed by General Slowing proponents: Meta-analysis using nonlinear or
more often linear regression. For example, Perfect (1994) has argued recently that the results
oft he "Brinley" plot regression approach can misrepresent the underlying task parameters that
determine an age group's overall performance. The other criticism comes directly from studies
using a range of tasks, the data from which either fail to exhibit Age Group X Condition
interactions or which exhibit Age Group X Condition interactions that do not indicate uniform,
proportional slowing in the elderly subjects (e.g., Amrhein & Theios, 1993, Geary, Frensch &
Wiley, 1993; Allen, Madden, Weber & Groth, 1993). Another kind of evidence against
General Slowing Theory are cases where the age-related slowing observed from a meta-
analysis does not indicate the type of proportional slowing typically reported in the meta-
analyses of General Slowing proponents. For example, following up on Myerson et al. (1992),
Laver & Burke (1993) conducted a meta-analysis of a substantially larger set of aging and
lexical priming studies. Contrary to Myerson et al. (1992) who reported proportional slowing
(i.e., the slopes of the best fitting lines approximated 1.5 with negligible or negative intercepts),
Laver and Burke (1993) reported no proportional slowing (i.e., the slopes of the best-fitting
lines approximated 1.0 with a positive intercept). Whereas Myerson et al. (1992) used their
results to argue that semantic priming effects exhibit local evidence for proportional slowing
also found at the global level, Laver and Burke (1993) argued that elderly slowing in these
priming effects was due to sensory slowing which slows semantic processing of the target
stimulus.
Finally, evidence from recta-analyses indicating domain or task specificity concerning
age-related slowing (or lack of it) would also argue against at least a simple single parameter
158 P. c. Amrhein
value account. For example, as will be detailed below, General Slowing theory proponents
have themselves reported that elderly slowing for lexical and nonlexical tasks differs (e.g.,
Lima et al., 1991).
A critical asstunption made by those researchers using the "Brinley plot" regression
approach is that task complexity can be readily defined. But the definition of"task complexity"
itself seems circular (e.g., see Myerson & Hale, 1993). To elaborate, in the a priori
application of this approach, increases in the number of specifiable subprocesses underlying
task performance should increase overall response time. However, it is often is difficult to
specify exactly what these additional processes would be, so the acl hoc application is then
used. In the ad hoc application, greater response latency for a condition (which is hopefully
not compromised by a speed-accuracy tradeoff) is taken as prima facie evidence that that
condition is more "complex" in an information processing sense. Regardless of how
complexity is defined, the General Slowing Theory predicts that elderly subjects will exhibit
proportionally longer response latencies for more "complex" experimental conditions relative
to young subjects.
For each study reviewed, I have listed what appears to be the "complexity" variable(s)
at work in determining response latency.
Comparison tasks: For the Elias and Kinsboume (1974) and Nebes (1976) studies,
there are two factors which relate to task complexity: (a) Concerning the variable of
interstimulus interval, with less time to process preceding the verbal description more time will
be needed to process the following picture or verbal stimulus (a priori); and (b) Depending on
the encoding strategy used (verbal or nonverbal), encoding efficiency will be less, thus more
complex, when there is a stimulus modality-encoding strategy mismatch (ad hoc). Factor (b)
also accounts for the definitions of complexity for Halpem (1984) and Mergler and Zandi
(1983).
Production tasks: For the Mitchell (1989) study, fewer repetitions and longer lags
should increase the complexity of the processing situation [similar to Nebes (1976) but
Mitchell employed a much more sensitive methodology] (a priori). Concerning Bowles
(1994), trials with longer SOA should present a less complex stimulus processing situation,
because the prime stimulus becomes more useful as its identification becomes more probable.
(a priori) Especially at early SOA, where inhibition is expected, this should mean that picture-
naming is most complex because other responses are in competition (a priori). For Thomas et
al. (1977) and Pooh and Fozard (1978), prior correct picture-name information should reduce
the complexity of naming target picture stimulus, by reducing naming uncertainty and
facilitating name retrieval. By contrast, prior incorrect picture-name information should
increase the complexity of naming target picture stimulus by increasing naming uncertainty and
not facilitating name retrieval. Simple naming should be intermediate between these two
conditions in complexity (a priori). Finally, for Amrhein & Theios (1993), increases in
complexity are due to the output task (drawing > writing) (ad hoc) and type of output
representation retrieval (cross-modality > within-modality) (a priori).
To date, meta-analyses have been conducted separately for speeded "lexical" tasks
(e.g., Lima et al., 1991; Laver & Burke, 1993; Myerson et al., 1992) and "non-lexical" tasks
(e.g., Hale et al. 1987; Hale et al., 1991). A common finding of these meta-analyses is that
age-related slowing for nonlexical tasks appears to differ from that of lexical tasks.
Specifically, the slowing for nonlexical tasks has been shown to be nonlinear, and best
accounted for by a power law, whereas the slowing for lexical tasks is linear and best
accounted for by a regression line with a slope of around 1.5. However, according to Lima et
al. (1991), if the response latencies fall within the typical range of 0-3000 ms for both age
groups, a straight line provides a good approximation of the relationship between elderly and
young lexical and nonlexical latencies. For lexical tasks, this line is again expected to have a
slope around 1.5, whereas for nonlexical tasks the line is expected to have a slope around 2.0
(i.e., at least a slope significantly greater than 1.5). I should note, however, that the nonlexical
tasks analyzed to date have represented a mixed bag of stimuli and task types--including
diagrams used in image rotation tasks as well as simple fight displays used in choice reaction
time tasks.
Curiously, while most of the picture-word studies reviewed earlier were published prior
to the appearance of the meta-analyses found in the aging literature (except of course, Amrhein
& Theios, 1993 and Bowles, 1994), none of them were included in those meta-analyses. For
this reason, I conducted a meta-analysis of these studies to determine their contribution to the
debate on age-related slowing. Although, as discussed earlier, there are problems concerning
their methodologies in terms of allowing tests of picture-word processing theories, the data
from these studies still allow a good approximation of more global issues concerning overall
processing speed and age, because both age groups, within a given study, performed these
tasks under the same methodological conditions.
An interesting question is whether data from the picture-word studies will exhibit
different slowing functions according to stimulus modality. Slope distinctions between picture
and word stimuli are complicated because, picture stimuli may be processed ultimately as
words, as in picture-naming or writing from a picture stimulus, and conversely word stimuli
may ultimately be processed as pictures, as in drawing from a word stimulus. In comparison
tasks, it is at least theoretically possible that picture and word stimuli are recoded in opposing
modalities (a view consistent with the Dual Coding model if the comparisons are semantic, and
consistent with all three information processing models if they are not; see Theios & Amrhein,
1989) prior to a decision response. Thus, given that the latencies from picture-word studies
analyzed here all fall within the range stipulated by Lima et al. (1991), I expected that their
best-fitting line would exhibit a slope intermediate to those found for strictly lexical and
nonlexical tasks (i.e., around 1.75). Moreover, this slope should be the same for comparison
and production tasks; speeded lexical tasks such as lexical decision, categorization, judgment,
and naming are all accounted for by a best-fitting line with a slope around 1.5 (Lima et al.,
1991) (but see the following).
Mean latencies from the reviewed studies were obtained either from tables or estimated
from figures presented in the published articles. Table 1 presents the details concerning
number of conditions contributed from each study and their source in each article.
I have excluded from this analysis the data from Mergler and Zandi (1983) and three of
the conditions from Poon and Fozard (1978). The Mergler and Zandi (1983) data represent a
set of extremely long response latencies (range: 1600-2400 ms for young subjects, 3000-5700
160 P. C. Amrhein
ms for elderly subjects), falling outside the 99% confidence interval for all the response
latencies of the comparison-task studies reviewed earlier. These latencies also would have
clearly biased a good linear fit of the data because of their influence as outliers (see also Fisk &
Fisher, 1994 and Hertzog, 1992 on this point). Also, some of these latencies fall outside of the
'able 1
'ASK STUDY CONDITIONS SOURCE
',omparison
Elias and Kinsboume (1974) 12 Figure 2
Nebes (1976) 16 Table 1
Halpem (1984) 16 Table 1
rest-Fitting Line: RTELDERLY= 1.47 RTvotrso + 183.8 ms (r 2 = .757)
~roduction
Picture-Naming
Thomas, Fozard and Waugh (1977) 3 Figure 4
Pooh and Fozard (1978) 3 Figure 1
Mitchell (1989) 14 Table 3
Bowles (1994)
Experiment 1 24 Figure 2
Experiment 2 24 Figure 9
Drawing-Writing
Amrhein and Theios (1993) 4 Figure 3
~est-Fitting Line: RTELDERLY = .82 RTyotrso + 259.6 ms (r 2 = .671)
Word-Naming Studies t~om Lima et al. (1991)

Balota and Duchek (1988) 20 Figure 1
Cerella and Fozard (1984) 4 Table 1
Nebes, Boller and Holland (1986) 3 Table 2
Iest-Fitting Line (Cumulative)" RTELDERLV= .93 RTvotrso + 167.4 ms (r z = .876)
Other Word-Naming Studies

Balota, Black and Cheney (1992)
Experiment 1 12 Table 2
Nebes, Brady and Huff(1989) 2 Table 1
test Fitting Line (Cumulative)" RTELDERLY= .86 RTvotrsG + 222.2 ms (r 2 = . 8 6 4 )
Cote: Studies are ordered as discussed in the text.

range stipulated by Lima et al. (1991) for linear tests. Lastly and most problematic, these data
are suspect because of the presence of an age-differential speed-accuracy tradeoff described
earlier.
The "unique-contemporary", "unique-dated" and "common-dated" condition latencies
of Poon and Fozard (1978) were also excluded because these conditions seemed particularly
confounded with age-group cohort effects (concerning, for exan~le, familiarity of use); effects
which were borne out in their findings reviewed earlier, and which were not present in the
conditions of the other studies. Thus, only the "common-contemporary" condition latencies of
Poon and Fozard (1978) were included in the meta-analysis. Finally, for comparability with the
majority of experiments which assessed only elderly and young age groups, I included only the
youngest and oldest age groups from the age group ranges of Poon and Fozard (1978) and
Thomas et al. (1977).
./(.. j
O9 2000, ....
E
>.,
0
C 1500-
0
__J
c-
J
O I000-
t~
, M
c-
O Tosk Type
500- 9 Comparison
>.,
9 Production
LM
0 ! I ' i' '
0 500 1000 1500 2000
Young Condition Lotency (ms)
Figure 3. Scatter plot and best-fitting lines for comparison and production tasks from the
aging and picture-word processing literature.
The scatter plot of the condition latencies is given in Figure 3. Two distinct patterns
can be seen, one for comparison tasks and one for production tasks. Separate regression lines
were computed for each task type; line equations are given in Table 1. The regression lines for
the comparison and production tasks account for 75.7% and 67.1% of their respective elderly
condition mean variance. The slopes of these two lines differ si~ificantly, t(ll2) = 4.03, p
162 P. C. Amrhein
<.001" The line for comparison tasks has a slope of 1.47 indicating approximately 50% slowing
for the elderly groups, but the line for production tasks has a slope of. 82 indicating no slowing
for elderly over young groups (indeed, a slight increase in elderly-group processing speed).
Finally, the intercepts for comparison and production tasks, 183.8 and 259.6 ms, respectively,
indicate additive slowing for both tasks likely due to slowed sensory-motor processes in the
elderly (see Botwinick, 1984; Cerella, 1990). As an additional indicator that the production
and comparison tasks represent two distinct slowing functions, a regression line fitting
latencies across task (whereRTELDERLY= .74 R T y o ~ + 457.5 ms) accounts for only 16.6% of
the elderly condition mean variance.
In Cerella's (1990) terms, the comparison task regression line represents a case of
"multilayered" slowing where one layer reflects sensory-motor slowing (given by the additive
intercept) and a second layer reflects "computational" slowing (given by the 1.47 slope).
Accordingly, this means that the production task regression line represents a case of single-
layered slowing, due only to sensory-motor slowing; because the slope of this line
approximates 1.0, there is no such computational slowing. Thus, these two regression lines are
problematic for at least a simple version of General Slowing which posits a uniform slowing
proportion (i.e., a slope approximating 1.5) in accounting for elderly slowing in speeded
cognitive tasks. Specifically, comparison tasks appear to be consistent at a global level with
this theory; indeed the slope of 1.47 is quite consistent with that of 1.5 typically found for
strictly lexical tasks (e.g., Lima et al., 1991). However, the production tasks are clearly not
consistent with such an account. Indeed, except for sensory-motor slowing (indicated by the
intercept value of 259.6 ms), there is no further slowing for remaining cognitive processes
concerned with within-mode or cross-modal retrieval of picture or word representations from
picture or word stimuli.
This was also the conclusion reached from the findings of Amrhein and Theios (1993).
In that study, we found that "overall" slowing for a drawing-writing task was a composite
function of subprocesses which exhibit no slowing and those which exhibit additive rather than
proportional (i.e., multiplicative) slowing. Importantly, those subprocesses which exhibit no
age-related slowing are those concerned with semantic memory retrieval a finding consistent
with strictly lexical studies (e.g., Allen, Madden & Crozier, 1991; Bowles & Pooh, 1981,
1985; Cerella & Fozard, 1984.) Coriously, our "overall" slowing was indicated by a slope of
1.56 from the regression line; but because this line was based on only four conditions, this line
was a "false positive" indicator (Fisk & Fisher, 1994) of proportional slowing for elderly
subjects.
In sum, the current meta-analysis indicates that age-related slowing in picture-word
tasks appears to be task specific, even at the global "Brinley" plot level. In addition, the reader
should keep in mind that this meta-analysis (and this type of meta-analysis, in general) ignores
important interactions which indicate differential impact of stimuli and tasks on age group
performance (see also Fisk et al., 1992, for further arguments concerning age differences and
stimulus/task dependent effects in speeded task performance).
Proportional slowing was not found for the picture-word production tasks reviewed in
this chapter, but as stated before, has been found for strictly word-naming tasks (Lima et al.,
1991). Or has it? The naming studies included in the Lima et al. meta-analysis (see Table 1)
were embedded among other non-production tasks (i.e., judgment, comparison and
categorization), thus possibly masking evidence against proportional slowing. To test this
possibility, I included data from the three word-naming studies listed in Lima et al.'s list of
"other lexical tasks" in their Table 4. However, I omitted the delayed pronunciation data of
Balota and Duchek (1988). Because subjects in that task were assessed for word-naming
onset after viewing the stimulus for a variable duration, the time measure did not reflect
stimulus perception as it did in the other studies analyzed earlier. Age-based perceptual
slowing is anticipated to be additive (Botwinick, 1984); its absence in the data would thus
artifactually reduce the additive component in the regression equation.
O0 2000-
E
>,,
o
c 1500--
(D
0
___!
c-
O 1000 -
73
o
c-
O
500 -
~.~t~~
.#, mosk Type
* Comporison
/
I
L.. 9 Production I
&)
-(3 I
Ld Urea et al., 1991) !
o I I i
o 500 1000 1500 2000

Young Condition Lotency ( m s )
Figure 4. Scatter plot and best-fitting lines for comparison tasks and production tasks from
the aging and picture-word processing literature. Also, included in the production task data
are the three word-naming studies from the Lima et al. (1991) meta-analysis.
As can be seen in Figure 4, adding these three word-naming studies to the production
task data set alters the regression line little and actually strengthens the fit. The production
task now has a slope of .93 with an intercept of 167.4 ms, and accounts for 87.6% of the
condition mean variance. Also, the slope of .93 is si~ificantly less than the slope of 1.47 of
the comparison task line, t(139) = 5.22, p < .001. As an additional indicator that the
production and comparison tasks continue to represent two distinct slowing functions, the
regression line fitting latencies across task (where RTELDmLY= 1.05 RTvotmG+ 204.1 ms) now
accounts for only 34.8% of the elderly condition mean variance.
Finally, for sake of completeness, I again recomputed this regression line including two
other word-naming studies in the aging literature (see Table 1). The resulting regression line
now has a slope of. 86 with an intercept of 222.2 ms, and accounts for 86.4% of the condition
mean variance. The slope of .86 for this production task line is again si~ificantly different
from the slope of 1.47 for the comparison task line, t(177) = 6.91, p < .001. And again, as an
164 P. C. Amrhein
additional indicator that the production and comparison tasks continue to represent two
distinct slowing functions, the regression line fitting latencies across task (where RTELDF~LY=
1.13 RTyousG+ 133.4 ms) now accounts for only 45.6% of the elderly condition mean
variance.
The message here is clear: Production tasks involving picture or word stimuli do not
undergo proportional, age-related slowing, whereas comparison tasks do. However, both
tasks share an additive increase likely due to sensory-motor deficits in the elderly subjects (see
Botwinick, 1984). This suggests that lexical and pictorial retrieval are spared in aging (see
also Bowles, 1993; Duchek & Balota, 1993), but aspects of decision processing are not. By
design, all comparison-task studies in Table 1 have a decision component (e.g., responding
'yes' or 'no'). This is also the case for the non-production lexical tasks (i.e., lexical decision,
judgment, categorization, etc.) analyzed by Lima et al. (1991) which also exhibit proportional
slowing. Indeed, if latencies from such tasks were included in the regression analysis of the
picture-word comparison tasks, the goodness of fit (r 2) should increase similarly to that seen
for the regression analysis of the production tasks when the word-naming studies were
included. Thus, at least within the response range analyzed here and by Lima et al. (1991),
stimulus modality (i.e., lexical or nonlexical) does not impact on the nature of age-related
slowing, but task type does.
The reader should also note that in the first recta-analysis, only two within-mode
condition latencies (i.e., from Amrhein & Theios, 1993: draw picture from picture stimulus;
write word from word stimulus) were included among the predominant cross-modality
condition latencies (i.e., from Amrhein & Theios, 1993: draw picture from word stimulus;
write word from picture stimulus; and from the remaining studies: name picture stimulus).
However, even though cross-modality latencies are longer than within-mode latencies, both
condition types should fall on the same regression line because the additional latency increment
for cross-modality transfer has been found to be age-, stimulus modality-, and production task-
independent (Amrhein & Theios, 1993; Amrhein, 1994). This claim is especially supported by
the minimal impact that the latencies of the word-naming (i.e., a within-mode condition)
studies had on the slope of the regression line in the two subsequent recta-analyses.
In stun, this demonstration indicates task specificity in age-related slowing, and thus the
need to explain task performance differences at the underlying process level. In other words, a
fixed proportional slowing account based on overall task performance does not explain aging
effects and non-effects in speeded lexical and nonlexical tasks.
4. FUTURE DIRECTIONS
As was discussed earlier, the drawing-writing task used here offers a balanced solution
to the incompleteness of the traditional naming-reading task. The drawing-writing task can
also be implemented to more comprehensively address issues such as picture-word priming and
Stroop-like interference effects. Only a handful of findings have been reported in the literature
concerning these issues in their relation to aging (and then using only the picture-naming task;
e.g., Bowles, 1994; Mitchell, 1989).
Implicit in Equations 1-6 presented earlier is the assumption that within-modality
conditions (i.e., draw a picture given a picture stimulus, write a word given a word stimulus,
read aloud a word stimulus) do not involve semantic memory access whereas cross-modality
conditions (draw a picture given a word stimulus; write a word given a picture stimulus, name
a picture stimulus) do involve it. Incorporation of a picture-word priming paradigm into the
drawing-writing task would provide a critical test of this assumption.
For young subjects, the evidence from naming tasks suggests support for the Amodal
models, where pictures and word share a common semantic store (e.g., Bajo, 1988).
Specifically, Bajo (1988) found that conceptually related word and picture primes facilitate
naming picture stimuli but not reading word stimuli. Accordingly, if her study was conducted
using the drawing-writing task, the expectation is that conceptually related word and picture
primes would also facilitate writing names for picture stimuli but not writing down word
stimuli. Moreover, if picture-word processing is balanced temporally as it appears to be given
the studies of Amrhein (1994), Amrhein and Theios (1993) and Theios and Amrhein (1989)
then it should also be found that conceptually related word and picture primes facilitate
drawing pictures from word stimuli but not drawing pictures from picture stimuli.
From the aging and word priming literature, it appears that prime facilitation may
actually be greater for elderly over young subjects (although the rate of corresponding
spreading activation may be constant, see Balota & Duchek, 1988, but also see Howard, Shaw
& Heisey, 1986). This difference may be due to slower word encoding and response processes
which give semantic priming mechanisms (i.e., spreading activation) additional time to function
(see e.g., Balota & Duchek, 1988; Bowles & Pooh, 1985; Burke, White & Diaz, 1987;
Howard, McAndrews, & Lasaga, 1981). Indeed, the data of Bowles (1994) suggest that when
age-based perceptual differences among the primes are accounted for, prime facilitation at long
SOAs (500-700 ms) is equivalent for elderly and young subjects.
By using the drawing-writing task, issues of age-related slowing in semantic priming
can be comprehensively addressed by testing pictures and word as primes, target stimuli and
output productions. In addition to determining similarities or differences in semantic priming
effects due to modality of prime, target and output production, the model approach presented
in Equations 1-6, when applied to these data, allows the determination of the specific
subprocesses which are influenced (and to what extent) by subject age, prime-target
relatedness, as well as prime, target, and output production modality (see Amrhein, 1994;
Amrhein & Theios, 1993).
Noticeably absent from the aging and picture-word studies reviewed using comparison
tasks was the manipulation of a semantic variable, notably category membership. This variable
has received substantial attention in the general picture-word processing literature because of
its importance in testing theoretical models (e.g., Harris, Morris & Bassett, 1977; Pellegrino, et
al., 1977; Potter & Faulconer, 1975; Snodgrass & McCuUough, 1986; te Linde, 1982). In a
categorization tasks, where two stimuli are presented for a binary category membership
decision, proportional slowing for elderly subjects should be found that is consistent with that
found for picture-word comparison tasks reviewed here. Moreover, because I am arguing that
it is the decision subprocesses underlying comparison tasks, in general, that exhibit age-related
proportional slowing, the slowing for picture-word categorization tasks should be the same as
that seen for categorization tasks (actually all comparison tasks) using only lexical stimuli.
Finally, one obvious limitation of this (and most other) meta-analyses is that the
production and comparison task data included were from different experiments (i.e., different
subjects appeared across the two task types). A critical test of the task specificity shown here
would be to assess elderly and young groups on a set of stimuli for production and comparison
tasks within-subjects. Indeed, given a stimulus set which satisfies the conditions of size,
featural similarity and familiarity specified by Theios and Amrhein (1989) and Snodgrass and
166 P. C. Amrhein
McCullough (1986), and a comprehensive set of conditions for each task (concerning cross-
modality and within-modality representation retrieval), the three hypotheses (elderly slowing,
elderly spatial deficit, and specific picture-word processing model assumptions) could be tested
collectively.
REFERENCES
Allen, P.A., Madden, D.J. & Crozier, L.C. (1991). Adult age differences in letter-levd and
word-levd processing. Psychology and Aging, 6, 261-272.
Allen, P.A., Madden, D.J., Weber, T.A. & Groth, ICE. (1993). Influence of age and processing
stage on visual word recognition. Psychology and Aging, 8, 274-282.
Amrhein, P.C. (1994). Temporal invariance for picture-word translation: Evidence from
drawing-writing and naming-reading tasks, Memory & Cognition, 22, 442-454.
Amrhein, P.C., & Theios, J. (1993). The time it takes for elderly and young individuals to draw
pictures and write words, Psychology and Aging, 8, 197-205.
Are~berg, D. (1978). Differences and changes with age in the Benton Visual Retention Test.
Bajo, M.T. (1988). Semantic facilitation with pictures and words. Journal of Experimental
Balota, D.A. & Chumbley, J.I. (1984). Are lexical decisions a good measure oflexical access?
The role of word frequency in the neglected decision stage. Journal of Experimental
Balota, D.A. & Chumbley, J.I. (1985). The locus of word-frequency effects in the
pronunciation task: Lexical access and/or production? Journal of Memory and Language,
24, 89-106.
Balota, D.A. & Duchek, J.M. (1988). Age-related differences in lexical access, spreading
Botwinick, J. (1984). Aging and Behavior. New York: Springer.
Bowles, N.L. (1993). Semantic processes that serve picture naming. In J. Cerdla, J. Rybash,
W. Hoyer, & M.L. Commons (Eds.), Adult information processing: Limits on loss. (pp.
303-328). San Diego, CA: Academic Press.
Bowles, N.L. (1994). Age and rate of activation in semantic memory. Psychology and Aging,
9, 414-429.
Bowles, N.L. & Pooh, L.W. (1981). The effect of age on speed oflexical access. Experimental
Aging Research, 7, 417-425.
Bowles, N.L. & Pooh, L.W. (1985). Aging and retrieval of words in semantic memory.
Canestrari, 1LE. (1968). Age changes in acquisition. In G.A. Talland (Ed.), Human aging and
behavior (pp. 169-188). New York: Academic Press.
Cattdl, J.M. (1886). The time it takes to see and name objects. Mind, 11, 63-65.
Cerdla, J. (1990). Aging and information-processing rate. In J. Birren & I~W. Schaie (Eds.),
Handbook of the psychology of aging (pp. 201-221). San Diego: Academic Press.
Cerella, J. (1994). Generalized slowing in Brinley plots. Journal of Gerontology, 49, P65-71.
Cerdla, J., & Fozard, J.L. (1984). Lexical access and age. Developmental Psychology, 20,
235-243.
Evidencefor task specificity in age-relatedslowing 167
Cerella, J., & Hale, S. (1994). The rise and fall of information processing rates over the life
span. Acta Psychologwa, 86, 109-198.
Dell, G.S., & O'Seaghdha, P.G. (1991). Mediated and convergent lexical priming in language
production: A comment on Levelt et al. (1991). Psychological Review, 98, 604-614.
Duchek, J.M., & Balota, D.A. (1993). Sparing activation processes in older adults. In J.
Cerella, J. gybash, W. Hoyer, & M.L. Commons (Eds.), Adult reformat~on processing:
Limits on loss. (pp. 384-406). San Diego, CA: Academic Press.
Elias, M.F., & Kinsbourne, M. (1974). Age and sex differences in the processing of verbal and
nonverbal stimuli. Journal of Gerontology, 29, 162-171.
Falmagne, J.C. (1965). Stochastic models for choice reaction time with application to
experimental results. Journal of Mathematical Psychology, 12, 77-124.
Farah, M.J., & Kosslyn, S.M. (1981). Structure and strategy in image generation. Cognitive
Science, 4, 371-383.
Fisk, A.D., & Fisher, D.L. (1994). Brinley plots and theories of aging: The explicit, muddled,
and implicit debates. Journal of Gerontology, 49, P81-89.
Fisk, A.D., Fisher, D.L. & Rogers, W.A. (1992). General slowing alone cannot explain age-
related search effects: Reply to Cerella (1991). Journal of Experimental Psychology-
General, 121, 73-78.
Fraisse, P. (1969). Why is naming longer than reading? Acta Psychologica, 30, 96-103.
Geary, D., Frensch, P., & Wiley, J. (1993). Simple and complex mental subtraction: Strategy
choice and speed-of-processing differences in younger and older adults. Psychology and
Aging, 8, 242-256.
Glaser, W.1L (1992). Picture naming. Cognition, 42, 61-105.
Glaser, W.1L, & Dungelhofl~ F.-J. (1984). The time course of picture-word interference.
Journal of Experimental Psychology: Human Perception and Performance, 10, 640-654.
Glaser, W.IL, & Glaser, M.O. (1989). Context effects in Stroop-like word and picture
processing. Journal of Experimental Psychology: General, 118, 13-42.
Goulet, P., Ska, B., & Kahn, H.J. (1994). Is there a decline in picture naming with advancing
age? Journal of Speech and Hearing Research, 37, 629-644.
Hale, S., Lima, S.D., & Myerson, J. (1991). General cognitive slowing in the noniexical
domain: An experimental validation. Psychology and Aging, 6, 512-521.
Hale, S., Myerson, J., & Wagstafl~ D. (1987). General slowing of nonverbal information
processing: Evidence for a power law. Journal of Gerontology, 42, 131-136.
Halpem, D.F. (1984). Age differences in response time to verbal and symbolic traffic signs.
Harker, J,O. & Riege, W.H. (1985). Aging and delay effects on recognition of words and
designs. Journal of Gerontology, 40, 601-604.
Harris, P.L., Morris, P.E., & Bassett, E. (1977). Classifying pictures and words: Implications
for the dual-coding hypothesis. Memory & Cognition, 5, 242-246.
Hertzog, C. (1992). Aging, information processing speed, and intelligence. In I~W. Schaie
(Ed.), Annual review of gerontology and germtries (Vol. 11, pp. 55-79). New York:
Springer.
Howard, D.V., Shaw, R,J., & Heisey, J.G. (1986). Aging and the time course of semantic
Hulicka, I.M. & Grossman, J.L. (1967). Age group comparisons for the use of mediators in
paired-associate learning. Journal of Gerontology, 22, 46-51.
168 P.c. Amrhein
Kausler, D.H. (1991). Experimental psychology, cognition, and human aging. New York:
Springer-Verlag.
Kosslyn, S. M. (1980). Image andMind. Cambridge, MA: Harvard University Press.
La Heij, W. (1988). Components of Stroop-like interference in picture naming. Memory &
Cognition, 16, 400-410.
Laver, G.D., & Burke, D.M. (1993). Why do semantic priming effects increase in old age? A
Meta-Analysis. Psychology and Aging, 8, 34-43.
Levelt, W.J.M., Schriefers, H.,Meyer, A.S., Pechman, T., Vorberg, D., & Havinga, J. (1991).
The time course of lexical access in speech production: A study of picture naming.
Levelt, W.J.M., Schriefers, H., Vorberg, D., Meyer, A.S., Pechman, T., & Havinga, J. (1991).
Normal and deviant lexical processing: Reply to Dell and O'Seaghdha (1991).
the lexical domain. Psychology andAging, 6, 416-425.
Lupker, S., & Theios, J. (1975). Tests of two classes of models for choice reaction time.
McCauley, C., Parmalee, C.M., Sperber, 1LD., & Carr, T.H. (1980). Early extraction of
meaning from pictures and its relation to conscious identification. Journal of Experimental
McEvoy, C.L. (1988). Automatic and strategic processes in picture naming. Journal of
Experimental Psychology: Learning, Memory and Cognition, 4, 618-626.
Mergler, N.L., & Zandi, T. (1983). Adult age differences in speed and accuracy of matching
verbal and pictorial signs. Educational Gerontology, 9, 73-85.
Mitchell, D.B. (1989). How many memory systems? Evidence from aging. Journal of
Molenaar, P.C.M., & van der Molen, M.W. (1994). On the discrimination between global and
local trend hypotheses of life-span changes in processing speed. Acta Psychologica, 86,
273-293.
Myerson, J., & Hale, S. (1993). General slowing and age invariance in cognitive processing:
The other side of the coin. In J. Cerella, J. Kybash, W. Hoyer, & M.L. Commons (Eds.),
Adult information processing: Limits on loss. (pp. 115-141). San Diego, CA: Academic
Press.
Myerson, J., Ferraro, F.1L, Hale, S., & Lima, S.D. (1992). General slowing in semantic priming
and word recognition. Psychology and Aging, 7, 257-270.
Myerson, J., Hale, S., Wagstaff~ D., Pooh, L.W., & Smith, G.A. (1990). The information loss
475-487.
Nebes, 1LD. (1976). Verbal-pictorial recoding in the elderly. Journal of Gerontology, 31, 421-
427.
Nebes, 1LD., Boiler, F., & Holland, A. (1986). Use of semantic context by patients with
Alzheimer's Disease. Psychology and Aging, 1, 261-269.
Nebes, 1LD., Brady, C.B., & Hut~ F.J. (1989). Automatic and attentional mechanisms of
semantic priming in Alzheimefs Disease. Journal of Clinical and Experimental
Evidencefor task specificity in age-relatedslowing 169
Paivio, A. (1966). Latency of verbal associations and imagery to noun stimuli as a function of
abstractness and generality. Canadian Journal of Psychology, 20, 378-387.
Paivio, A. (1971). Imagery and verbal processes. New York: Holt, Rinehart & Winston
Paivio, A. (1975). Perceptual comparisons through the mind's eye. Memory & Cognition, 3,
635-648.
Paivio, A. (1983). The empirical case for dual coding. In J.C. Yuille (Ed.), Imagery, memory
and cognition (pp. 307-332). Hillsdale, NJ: Lawrence Erlbaum Associates.
Paivio, A. (1986). Mental representations. New York: Oxford University Press.
Paivio, A., Clark, J.M., Digdon, N., & Bons, T. (1989). Referential processing: Reciprocity
and correlates of naming and imaging. Memory & Cognition, 17, 163-174.
Park, D.C., & Puglisi, J.T. (1985). Older adult's memory for the color of pictures. Journal of
Pellegrino, J.W., gosinski, 1L1L, Chiesi, H.L., & Siegel, A. (1977). Picture-word differences in
decision latency: An analysis of single and dual memory models. Memory & Cognition, 5,
383-396.
Pezdek, I~ (1983). Memory for items and their spatial locations by young and elderly adults.
Poon, L.W., & Fozard, J.L. (1978). Speed of retrieval from long-term memory in relation to
age, familiarity, and datedness of information. Journal of Gerontology, 33, 711-717.
Potter, M.C., & Faulconer, B.A. (1975). Time to understand pictures and words. Nature, 253,
437-438.
Seymour, P.H.K. (1973). A model for reading, naming, and comparison. British Journal of
Seymour, P.H.I~ (1974). Generation of a pictorial code. Memory & Cognition, 2, 224-232.
Seymour, P.H.K. (1979). Human visual cognition. New York: St. Martin's Press.
Smith, A.D., & FuUerton, A.M. (1980). Age differences in episodic and semantic memory:
Implications for language and cognition. In L.W. Poon, J.L. Fozard, L.S. Cermak, D.
Arenberg & L.W. Thompson (Eds.), New directions in memory and aging (pp. 139-155).
Hillsdale, NJ: Lawrence Erlbaum Associates.
Smith, A.D., Park, D.C., Cherry, I~, & Berkowsky, K. (1990). Age differences in memory for
concrete and abstract pictures. Journal of Gerontology: Psychological Sciences, 45, 205-
209.
Smith, M.C., & Magee, L.E. (1980). Tracing the time course of picture-word processing.
Snodgrass, J.G, (1980). Towards a model for picture and word processing. In P. Kolers & M.
Wrolstad (Eds.), Processing of visible language (Vol. 2, pp. 565-584). New York: Plenum
Publishing.
Snodgrass, J.G. (1984). Concepts and their surface representations. Journal of Verbal
Learning and Verbal Behavior, 23, 3-22.
Snodgrass, J.G., & McCullough, B. (1986). The role of visual similarity in picture
categorization. Journal of Experimental Psychology: Learning, Memory, and Cognition,
12, 147-154.
Snodgrass, J.G., & Vanderwart, M. (1980). A standardized set of 260 pictures: Norms for
name agreement, image agreement, familiarity, and visual complexity. Journal of
Experimental Psychology: Human Learning and Memory, 6, 174-215.
170 P.c. Amrhein
Theios, J., & Amrhein, P.C. (1989). Theoretical analysis of the cognitive processing oflexical
and pictorial stimuli: Reading, naming, and visual and conceptual comparisons.
Thomas, J.C., Fozard, J.L., & Waugh, N.C. (1977). Age-related differences in naming latency.
American Journal of Psychology, 90, 499-509.
Trahan, D.E., Larrabee, G.J., & Levin, H.S. (1986). Age-related differences in recognition
memory for pictures. Experimental Aging Research, 12, 147-150.
Treat, N.J., & Reese, H.W. (1976). Age, pacing, and imagery in paired-associated learning.
Tubi, N., & Calev, A. (1989). Verbal and visuospatial recall by younger and older subjects:
Use of matched tasks. Psychology and Aging, 4, 493-495.
Waugh, N.C., Thomas, J.C., & Fozard, J.L. (1978). Retrieval from different memory stores.
Wheeldon, L.I~, & Monsell, S. (1992). The locus of repetition priming of spoken word
production. The Quarterly Journal of Experimental Psychology, 44A, 723-761.
Winograd, E., & Simon, E.W. (1980). Visual memory and imagery. In L.W. Pooh, J.L. Fozard,
L.S. Cermak, D. Arenberg & L.W. Thompson (Eds.), New directions in memory and aging
(pp. 485-506). Hillsdale, NJ: Lawrence Erlbaum Associates.
Winograd, E., Smith, A. & Simon, E.W. (1982). Aging and the picture superiority effect in
recall. Journal of Gerontology, 37, 70-75.
Aging and the distribution of resources in w o r k i n g memory*
Elizabeth A. L. Stine
University of New Hampshire
In a 1988 address at the Cognitive Aging Conference, Tim Salthouse diagnosed the
Cognitive Aging community with a terrible disease, RORAP Syndrome, an acronym for an
insidious pathology in which its victims are compelled to provide vacuous accounts for age-
related changes in cognition, and in particular, to Rely On Resources As a Pseudo-explanation
.(Salthouse, 1988). He helped those of us suffering t~om this syndrome to identify ourselves,
and then went on to provide a 12-step program to help us avoid the circular reasoning that
commonly strikes those afflicted with the syndrome. In this talk and subsequent "pamphlets"
.(Salthouse, 1988a; Salthouse, 1988b; Salthouse, 1988c; Salthouse, 1991a; Salthouse, 1991b;
Salthouse & Babcock, 1991), he has discussed three dominant metaphors used to
conceptualize an age-related reduction in processing resources, alternatively, as a reduction in
processing speed (slowing), a reduction in mental energy (attention), and finally (which brings
us to the topic of this chapter), a reduction in working memory capacity.
The working memory (WM) construct is an intellectual descendent of William James'
.(1890) notion of "primary memory." Essentially, the world is bigger than we are, and our
capacity for constructing knowledge is virtually infinite, but mediating the two is a buffer that
is extremely limited in its capacity for storing and manipulating information. The assumption is
that this buffer varies between as well as within age groups, and that individual differences in
this capacity affect a wide range of cognitive abilities. The years since the diagnosis of
RORAP Syndrome have seen a plethora of studies in which recovering RO1LAPs have worked
to operationalize this elusive construct. So time as a resource is the change in reaction time in
response to conditions varying in complexity; mental energy is the ability to effectively carry
out multiple activities at once (as in divided attention tasks), and working memory capacity is
measured in a loaded span task as number of items that can be recalled after they are encoded
during some concurrent activity. Interestingly, these investigations have revealed an homology
among these metaphors. Salient examples are McDowd and Craik's .(1988) demonstration that
divided attention increases processing time, especially for older adults, Salthouse's numerous
demonstrations (e.g., .Salthouse, 1991a ) that perceptual-motor speed can account for loaded
span performance, Tun et al.'s .(1991) demonstration that older adults with higher spans are
better at divided attention tasks, and data t~om our lab .(Stine & Hindman, 1994) showing that
readers with greater working memory capacity are faster at encoding the idea units from text.
AUTHOR NOTE: Address correspondence to: Department of Psychology, Conant Hall, University of New
Hampshire, Durham, NH 03824, (603) 862-3806. Email: eals@christa.unh.edu This chapter is based on a
presentation at the Fifth Cognitive Aging Conference in Atlanta, GA on April 7, 1994. The research described
from our lab was supportedby grant R29 AG08382 from the National Institute on Aging. I am grateful to Dan
Morrow and Phil Allen for helpful commentson earlier drafts of this manuscript.
172 E.A.L. Stine
So energy is time, and time is space, and space is energy, and I suppose we are left with
if the reader will excuse the expression - - "resources." In fact, Salthouse .(1991b) has
articulated the interchangeable nature of these constructs very nicely:
Although it is convenient to categorize speculations about processing resources in
terms of metaphors of time, space, and energy, these conceptualizations are not
necessarily distinct and independent. Not only are the arguments for the each resource
based on similar reasoning, but the same results are sometimes interpreted as evidence
for different resource conceptualizations .... [T]he different metaphorical interpretations
may be interrelated .... [I]mpairments in divided attention might be a consequence of
slowness in alternating between processes .... reductions in attention would lead to
increase delays between successive processing operations .... Slower rates of
processing have additionally been linked to reduced capacity of working memory ....
(Salthouse, 1991, pp. 346-348)
In this paper, I would like to do two things: First, I'd like to give a brief historical
overview of the working memory concept, considering some recent findings in the cognitive
aging literature implicating the role of WM resources in age-related changes in performance.
Second, I'd like to describe yet another affliction from which those in my lab and elsewhere are
trying to recover, WM UPROAR Syndrome: the Woefully Misbegotten Underestimate of the
Partitioning of Resources to Overcome Age-Related deficiencies. That is, even though many
of the age-related differences we observe may well be explicable in terms of a reduction in
working memory resources, I'd like to argue that there is some flexibility in the system in the
manner in which resources are allocated, and such allocation can go a long way toward
maintaining a high level of performance in later adulthood.
1. AN A C C E L E R A T E D HISTORY OF THE W O R K I N G M E M O R Y CONCEPT AND

ITS APPLICATION TO COGNITIVE A G I N G R E S E A R C H
As noted earlier, WM can be traced back to James' (1890) discussion of primary

memory, our consciousness of the "specious present" (Vol 1, pp. 641-642). With over a
hundred years of rumination and empirical contortions of this concept behind us, James'
description of the ephemeral nature of primary memory and the consequent quest for
coherence is still compelling:
[For] a state o f mind to survive in memory it must have endured f o r a certain
length o f time .... Any state of mind which is shut up to its own moment and fails to
become an object for succeeding states of mind, is as if it belonged to another stream of
thought. Or rather, it belongs physically, not intellectually, to its own stream, forming
a bridge from one segment of it to another, but not being appropriated inwardly by later
segments or appearing as part of the empirical self.... All the intellectual value for us of
a state of mind depends on our after-memory of it .... Only then does it count for us.
(pp. 643-644, italics are James')
That metaphorical single step into the stream of consciousness, that ever-present
"now," was incorporated into experimental psychology as a limited-capacity bottleneck in the
cognitive system Waugh and Norman .(1965) drew heavily on James' presentation in their
seminal paper on primary memory (PM), but specifically conceptualized this bottleneck as a
passive store:
Aging and the distribution of resources in working memory 173
An event in PM has never leit consciousness and is part of the psychological

present .... PM is a faithfifl record of events just perceived .... James believed that PM
extends over a fixed period of time. We propose instead that it encompasses a certain
number of events regardless of the time they take to occur" (pp. 92-93).
Their goal was to distinguish PM from a more enduring secondary memory store. As a
mathematical function relating the number of items recalled to the number of intervening items,
the emphasis was on storage.
This was true even as the model was elaborated into a full multistore model (with a
"short-term store" and a "long-term store") and granted "control processes" that managed the
contents of the short-term store .(Atkinson & Shiffdn, 1968): the primary, short-term memory
was regarded as a number of slots that were continually filled and replaced. The multistore
approach was criticized on many counts, but most saliently on the grounds that capacity and
duration of the short-term store was not, in fact, independent of the effects of knowledge, a
rather fundamental assumption of the model.(Craik & Lockhart, 1972).
In spite of the fact that short-term memory is periodically buried in the literature, the
core conceptualization behind James' primary memory has remained with us. To accommodate
findings that capacity depended on the nature of what was being done with items as well as the
nature of concurrent processing, the passive store was endowed with a capacity for processing
which competed with storage demands and re-christened "working memory." A primary
architect has been Alan Baddeley:
The core of the working memory system consists of a limited capacity 'work
space' which can be divided between storage and processing demands (Baddeley &
Hitch, 1974, pp. 75-76).
[This] model subdivided WM into three components, the Central Executive,
which formed the control centre of the system, was assumed to select and operate
various control processes. It was assumed to have a limited amount of processing
capacity, some of which could be devoted to the short-term storage of information. It
was able to ottload some of the storage demands to subsidiary slave systems, ... the
Articulatory Loop, which was able to maintain verbal material..., and the Visuo-Spatial
Scratch P a d [responsible for] the visualization of spatial material .(Baddeley, 198L, p.
18).
Some theorists now reserve the term "short-term memory" to refer to the acoustically-based
storage buffer needed for the preliminary analysis of language and use the term "working
memory" to refer to the processing component that manipulates information (Graesser, Singer,
& Trabasso, 1994), but the point is that concern with James' "specious present" has largely
shifted from issues of storage properties to issues of processing function.
With Daneman and Carpenter's .(1980) demonstration that the sentence span task
(which requires subjects to process sentences while holding their final words in memory) was
highly predictive of verbal SAT scores, working memory became entrenched in the literature as
the individual difference that made the difference in language processing, in memory, in
problem solving.
The trade-off between processing and storage seems like a potential source of
individual differences in reading comprehension .... The better reader might have more
efficient processes so that he/she effectively would have more capacity for storing and
maintaining information. (p. 451)
174 E.A.L. Stine
Table 1. Studies testing the prediction of an Age by Complexity interaction.
Study Dependent variable Difficulty manipulation AX C?

Babcock Digit and location recall Concurrenttask/ No
& Salthouse (1990) retrieval demands
Crossley Tapping speed Reading normal vs rotated text Yes

& Hiscock (1992) Speech repetition vs. fluency
Maze solving vs. tracking
Salthouse et al. ( 1 9 9 0 ) Accuracyin Number variables tracked No

numerical and spatial Numberof operations
tracking
Salthouse (1992) Reasoning accuracy Number of premises Yes

(omitting Ss at chance)
Salthouse Accuracy in matching Angle of cube Yes

& Skovronek (1992) rotated cube
Tun et al. (1991) Speech recall Concurrent RT No
Tun et al. (1992) RT during speech recall Numberof RT choices No

Propositional density
Wiegerson Digit recall Digit span vs. missing span Yes

& Meertse (1990)
This research was important in expanding the focus of working memory from experimental
work illustrating the general principles of memory in the generic college sophomore to a
consideration of the differences between people who were more or less successful in their
cognitive accompli~qhments.
The divergent approaches of Baddeley and Daneman offer an interesting contrast that
has set the tone for much of the cognitive aging work in this area. In the former, support for
the model is garnered by charting the deterioration of performance as demands for storage and
processing are increased. In the latter, support rests on intercorrelations between a presumed
estimate of working memory and criterial measures of cognitive performance.
Given the success of the working memory construct in explaining cognitive
performance among the young, the extension to aging was obvious. It could be that there is a
reduction in working memory capacity or resources, and that this is responsible for much of
the observed cognitive declines. Consistent with the approaches just outlined, this has been
addressed in two ways. The first examines the differential effects of increasing storage and
processing demands on older adults. If aging brings a decrease in WM resources, then elderly
performance should be disproportionately depressed by these demands (in terms of requiting

more time or engendering lower accuracy). The meta-analytic approaches of Hale, Myerson,
Cerella, and others .(Cerella, 1990; Myerson, Hale, Wagstait~ Poon, & Smith, 1990) have for
the most part supported the resource deficit hypothesis. Manipulations within individual
studies, however, have met with mixed success. Table 1 summarizes several studies which
have addressed this question.
This table is by no means exhaustive but illustrates what I think are some representative
studies in this area. There is about a 50:50 hit rate in obtaining the critical Age X Complexity
interaction. While there have been some interesting post hoc attempts to explain what kind of
difficulty exacerbates age differences and what kind does not, we really do not have a good
theory of the nature of the complexity that taxes elderly resources. For example, the
distinctions between storage capacity and processing capacity, or between structural capacity
and operational capacity, while intuitively appealing, don't seem to account for the difference.
In a recent paper, Salthouse, Babcock, and Shaw (1991) has suggested that what may be
required is a transformation of the essential nature of the representation to a more abstract
form This remains to be tested. In any case, we still don't have a theory of what kind of
empirical difficulty strains the resource capacity of older adults.
Even though the Age by Complexity interaction has been elusive, the individual
differences approach has been somewhat more successful in supporting working memory as a
mediator of age differences in cognitive performance. Table 2 summarizes representative
studies using this approach. Again, this list is not exhaustive but (I hope) fair in showing that
using a wide variety of indices of working memory m measures derived out of the context of
criterial performance, like loaded span or perceptual speed, and measures derived from within
the context of criterial performance, like repeating requests for information and a diverse
set of measures of cognition, age-related variance in performance can often be substantially
accounted for by a decline in working memory resources.
These approaches are, of course, predicated on the assumption that there is an
undifferentiated pool of resources in working memory. Some theorists, however, have argued
otherwise, that is, that working memory is best thought of as a set of distributed capacities of
different modalities. This is exemplified by Monsell .(1984), who after a thorough review of
patterns of selective interference, argued: "The simplest conclusion is that WM is no more (or
less) than a heterogeneous array of independent temporary storage capacities intrinsic to
various subsystems specialized for processing in specific domains" (p. 344). A similar
conclusion was reached by Daneman and Tardif.(1987). Creating a set of working memory
measures, some verbal and some spatial, they also obtained separate measures for processing
performance on these tasks with and without the storage component. Contrary to the notion
that it is a generalized capacity for simultaneous processing and storage that is responsible for
individual differences in cognitive outcome, spatial tasks were more predictive of math SAT,
verbal tasks were more predictive of verbal S A T - and simultaneous storage didn't make a
difference! That is, the processing component of the task was just as predictive of
performance as was the processing plus storage component. They argued:
We think we now have a wider range of measures of working memory capacity
and that the picture suggests the need for abandoning the notion of a "general and
central limitation on information processing .... " .... At the very least, we may have to
posit two separate processors, one for representing and manipulating verbal-symboli
176 E.A.L. Stine
Table 2. Do individual differences in an index of working memory account to some extent for
age-related change in cognition?
Study WM Index Context Criterion
Hultsch et al. Reading span Out Text recall No

(1992) Word recall
Morrow et al. Reading span Out Text recall Yes

(1992)
Salthouse (1992) Reading and Out Reasoning accuracy Yes

computation span
Perceptual speed Out
Recognition of
premises In
Salthouse Line span Out Accuracy in matching Yes

& Skovronek Repeated requests rotated cube
(1992) for infn In
Recognition of infn In
Stine et al. Average loaded span Out Recall of spoken Yes

(1993) narrative
Troyer et al. Executive function Out Word recall Yes

(1994) (concept formation, Visualmemory
flexibility)
Tun et al. (1991) Average loaded span Out Recall of spoken Yes
expository texts
information, and a second for representing and manipulating spatial information" (p.
502).
While there seems to be some consensus that verbal and non-verbal processing
represent different factors in working memory, issues of the structure of WM remain
unresolved. The "distributed capacities" approach is increasingly incorporated into cognitive
aging research. For example, even the generalizability of "general slowing" seems to depend
on whether the task is in the verbal or nonverbal domain .(Hale, Lima, & Myerson, 1991;
Myerson, Ferraro, Hale, & Lima, 1992).
2. W O R K I N G M E M O R Y , L A N G U A G E UNDERSTANDING, AND A G E
One outcome of the distributed capacities notion appears to be that the literature on
working memory has become more insular to particular domains, and it is particularly
flourishing in the domain of language. It is to the role of working memory in language
processing that I now turn. In fact, language provides an interesting laboratory for the study
of working memory. That is, one way to study the effects of straining working memory
resources is to have subjects simultaneously encode word lists, monitor a CRT for pink
elephants, pat their heads, and rub their bellies. Another way is to have them comprehend
language. In so doing, they must access word meanings, create a text-based representation of
propositional content, and construct a broader representation of the situation at the mental
model l e v e l - as when text understanding entails encoding relative spatial arrangement, or the
emotional reaction, predisposition, or goals of characters.
The application of working memory models in language has been a major force in
driving the dominant metaphor from what was once a box of items to what is now more often
conceptualized as the amount of activation available in a knowledge net. Good examples of
this perspective may be found in recent papers by Just and Carpenter .(1992) and by Engle and
colleagues.(1992):
[Our] purpose ... is to present a theoretical integration of the storage and
processing functions of working memory in language comprehension .... In this
framework, capacity can be expressed as the maximum amount of activation available
in working memory to support either of these functions .... We propose that individuals
vary in the amount of activation they have available for meeting the computational and
storage demands of language" (Just & Carpenter, 1992, pp. 123-4).
Working memory consists of those knowledge units that have recently been
activated either from objects in the environment or as a result of productions and are in
various states of loss of activation through either decay or inhibition" (Engle, Cantor, &
Carullo, 1992, p. 990).
This metaphor accommodates a number of interesting findings. For example,
Gernsbacher's .(1991) work showing that high-span subjects are more likely to suppress the
irrelevant meanings of words suggests that high-span readers have more resources because
they are effective in selectively activating appropriate knowledge nodes, i.e., distributing their
resources to relevant information. Just and Carpenter's (1992) work showing that the
impenetrability of syntactic analysis by semantic constraints is not true for high-span readers is
explicable in terms of the greater activation they have available; thus, high-span readers are
better able to simultaneously process the multiple levels of discourse. Finally, there are many
findings with respect to reference and coherence in discourse that suggest that information that
is not currently active in working memory can, nevertheless, be readily accessible under some
circumstances. Notions invoked to explain these findings must abandon simple assumptions
about information either being "in" or "not in"working memory.(Kintsch, 1988); for example,
Sanford and Garrod .(1981) make a distinction between information in an explicit focus in
working memory that is highly active (and can therefore be unambiguously referenced with
pronouns) and information in an implicit focus that is less active but relatively available
because of inferential connections entailed by what is active; Glenberg and colleagues .(1987)
describe "discourse pointers" in which information that is active in working memory points to
(or makes more available) relevant information that is not active; Graesser et al..(1994) have
argued that a "search for meaning" in discourse drives production rules in WM that guide long-
term memory searches; finally, Hintzman .(1986) and O'Brien .(1995) have argued for the
existence of an automatic resonance process invoked by the featural overlap between the active
nodes in working memory and knowledge net.
178 E.A.L. Stine
This shiR in the foundational metaphor for representing the contents of working
memory may prove to be important for understanding age differences in language processing.
Let me give you an example of some recent findings from our lab that would be hard to explain
with the WM "box" metaphor but are well accommodated by the "activated nodes" metaphor
(Hakala, RizeUa, Stine, & O'Brien, in preparation). In this study, young and older adults read
narratives in which a protagonist is initially introduced as having a given predisposition, for
example, Andy is described as being a vegetarian. The narrative line proceeds with filler
material without again making reference to this fact. At a later point, the character performs
an act that is inconsistent with the earlier characterization, for example, Andy eats a five-
course meal including steak tartare, lobster medallions, and raspberries with whipped cream
When readers encounter this information, their reading times increase (rdative to control
passages in which these statements represent no inconsistency, for example, if Andy were
described as being a gourmet), suggesting that they, in fact, reco~ize that (in the context of
this narrative anyway), this action is inconsistent with the nature of the protagonist. Now this
is hard for a traditional WM store model to explain. Since the description about Andy is no
longer "in working memory," readers should be oblivious to the break in coherence. Thus, it
must be that there is some mental model level of representation that is at least in implicit focus
to guide comprehension, enabling the reader to reactivate critical information into explicit
focus so that the inconsistency can be resolved; presumably the increase in reading time in the
inconsistent condition represents the allocation of resources needed to search, reactivate, and
perform the elaborative processing needed to achieve coherence. In support of this notion,
recall is actually higher in the inconsistent condition than in the consistent one .(Albrecht &
O'Brien, 1993).
Two results, however, suggest that the identical increment in reading time for younger
and older adults (i.e., no Age X Consistency interaction) was not sufficient to complete
coherence processing among the old when the pieces of information to be reconciled were far
apart in the text. First, under these conditions, older adults did not show the improvement in
recall that younger adults did .(Hess & Tate, 1991). In addition, there was evidence that the
average older adult did not reactivate into explicit focus the information needed to resolve the
inconsistency. In another experiment relying on a technique developed by Jerry Myers and
colleagues, subjects responded to probes testing the availability of the protagonist's features.
So for example, on target trials subjects verify statements like "Andy is a vegetarian." While
this information was less available aider the filler information for both younger and older adults,
it was only the younger readers who reactivated the target in the face of inconsistency
presumably to resolve it and update the mental model in light of this new information, e.g.,
OK, so Andy is a vegetarian with occasional lapses. Thus, it would appear the average older
adult in our sample did not do this, and to the extent that it is generally the case that elders do
not allocate resources in working memory to such reintegration, this might contribute to age-
related declines in discourse processing. This explanation is consistent with Craik's argument
.(Craik & Jennings, 1992) that elderly adults are less likely to self-initiate processing. This
account has potentially profound implications for understanding the phenomenological
experience of language comprehension in later life. In James' terms, if each segment were
"shut up in its own moment," then the "intellectual value" of the discourse would be
diminished.
This resource allocation hypothesis must be examined in light of the fact that older
adults did indeed slow their reading when faced with the inconsistency. On the average, older
adults were as responsive to the break in global coherence as were the young - - assuming that
the allocation of reading time is equally effective for young and old. But perhaps that
assumption is not warranted. This is an important issue and it is to the interpretation of
reading time allocation that I would like to direct the argument. In a recent study, Hartley,
Stojack, Mushaney, Atmon, and Lee .(1994) have shown across a variety of methods that when
young and older adults are matched on reading time, older adults recall systematically less, and
that that difference increases as more time is allocated (see their Figure 1). In addition,
evidence for cognitive slowing in a variety of tasks (see Salthouse, 1991b)would also suggest
that the allocation of equivalent amounts of time by younger and older readers would not be
expected to yield the same outcome. A Brinley plot of our reading times makes this point.
With a correlation of .98 between the reading times of young and old, and a slope of .94, the
Brinley analysis supported the contention that the older adults were responding j u s t as the
young. There were two things striking about this analysis. First, the extremely high
correlations between the reading times of the young and the old in a domain that does not
involve discrete trials, but rather self-paced reading, is noteworthy since it suggests a
qualitative similarity in resource allocation across a range of text demands.
In addition, the fact that the slope of the fimction relating reading times of the old to
those of the young was just about unity suggests a curious exception to the now familiar
Brinley plot with a slope of about 1.5. In fact, we have collected a lot of reading time data in
my lab lately, and we have been consistently struck by the fact that these Brinley plots never
conform to the predictions of generalized slowing. Note we are not concerned with
discriminating degrees of slowing .(Perfect, 1994); we simply never observe it at all in reading
time! Now it could be that such results suggest a preservation of p r o c e s s i n g - a failure to
find age-related slowing in the domain of language, as suggested by the meta-analysis oflexical
decision times by Laver and Burke .(1993). This seems unlikely if these are the reading times
that ultimately yield recall or comprehension d e f i c i t s - as is the case with the criterion
measure of this task, which were the probe verification times.
A perusal of recent literature on text processing in which reading time was measured
suggests that it is not atypical for the average adult reader to fail to accommodate for cognitive
slowing. For example, Hartley .(1993) found that within-group variability overshadowed
between group differences in reading speed. Similarly Hartley et al. (1994) did not find age
differences in self-paced reading speed. A Brinley analysis of the mean reading times in
different conditions by Hamm and Hasher .(1992) yield a slope of.84 (r=.70). For the domain
experts in Morrow et al..(1992), the Brinley slope is 1.18 (r=.90) and for the domain novices,
it is .84 (r=.82). The one reading time data set for which a Brinley analysis produces a high,
positive slope is that of Connelly, Hasher, and Zacks .(1991) in which the slope was 2.86
(r=.99) (numerical values were estimated from the data in Figure 3). In this study, however,
differences among conditions were created by adding interfering material. The variation in
difficulty in this study then may not reflect natural variation in the demands of reading. In any
case, the bulk of the data suggest that older adults are not particularly slower at reading in
terms of overall reading time nor in terms of how they keep pace with text difficulty. Thus,
there is now considerable evidence for age-constancy in reading time allocation. This age-
equivalence would presumably not accommodate cognitive slowing and is furthermore often
coupled with age differences at retrieval.
Another study from our lab has begun to address more specifically how these resources
are allocated on-line .(Stine, Loveless, & Soederberg, in preparation). Subjects read texts
180 E.A.L. Stine
sector-by-sector (sectors were groups of words that were syntactically well-formed and
presented in response to a button press). Again, we measured reading time (this time for each
sector), and across three conditions the slope ranged from.86 to 1.08. Thus, as in the Hakala
study, the reading times did not conform to the predictions of cognitive slowing. In this case,
however, there was no group age difference in subsequent recall performance. (This appeared
to be an artifact of the higher verbal ability of our elderly group: in a regression analysis of
recall, vocabulary level was a positive predictor and age was a si,~m~ificantnegative predictor.)
Using techniques pioneered by Doffs Aaronson and Karl Haberlandt (of. Aaronson &
Scarborough, 1977; Haberlandt, 1984), we used regression analyses to decompose the reading
times for each subject to reflect the allocation of time to word-level (i.e., length in syllables,
word frequency (f)), text-level (i.e., number of propositions, number of new concepts
introduced, syntactic complexity as measured by Yngve depth), and discourse-level (i.e., serial
position) features. Figure 1 shows the average values of these regression coefficients for
younger and older adults. Overall, the qualitative way in which resources are allocated in
working memory to process these texts is quite similar for young and old: both groups slowed
down for longer words (Syll), informationally dense sectors (Props), the introduction of new
concepts (NewConc), and complex syntax (Yngve); both groups read more quickly when the
words in the sector were more familiar (M log f) and when the sectors were later in the text
(SerPos). In spite of this qualitative similarity, there are some subtle age differences in the
extent to which the two age groups responded to features that represent the formation of a
text-based level of meaning. Older adults allocated less time to process the text-based meaning
of the passage, spending less time per proposition, p<.03, and per new concept, p<.06. Thus,
the regression analysis amplifies the Brinley analysis by showing specifically where the elderly
readers were failing to allocate processing resources. In short, the average elderly reader spent
less effort on constructing the text-based representation.
These results seem to naturally invoke the question of what kind of resource allocation
among the elderly would produce better recall. That is, if the claim is that it is an inappropriate
strategy that contributes to memory deficits, then among those who show better-than-average
recall, we ought to observe some adaptation in strategy. This brings us to our discussion of
WM UPROAR Syndrome: did younger and older adults have to partition their resources
differently so as to be proficient in recall? In order to address this question, we divided
subjects into "good" and "poor" recallers on the basis of a median split of subsequent recall
scores.
Consistent with our earlier discussion, good recallers scored higher on a loaded span
task than poor recallers. Furthermore, there was a multivariate interaction between age and
recall level suggesting that young and old did differentially allocate resources in order to be
good recallers. Younger adults who showed high recall allocated more time to the text-based
features of the text relative to those who showed lower recall, spending more time to process
complex syntax, propositional content, and new concepts. Older adults who were higher in
recall also increased their reading time in response to complex syntax, but otherwise, this age
group did not vary in text-based time allocation as a function of recall level. It was the serial
position effect, a variable thought to reflect the creation and reliance on a global schema
Figure 1. Averageregression coefficients reflectingtime allocated to different demands of text.
.(Haberlandt, 1984), that distinguished the good and poor recallers among the old group. Thus,
while younger adults appeared to use a "bottom-up" strategy (that relied on more thorough
elaboration of the text base) to be good recallers, older adults appeared to use more of a "top-
down" strategy (that relied on building a mental model level of representation early on and
using it to facilitate the encoding of subsequent text) to be good recallers.
We should be somewhat cautious here in noting that these are differences in degree and
not kind. Both "top-down" and "bottom-up" variables were needed to explain variance in
reading time among both younger and older adults. It was only the balance between the two
that shitted differentially for young and old to achieve high levels of recall. The point here is
that what makes an effective reading strategy for text memory is different for younger and
older adults. Also, note that the high recallers for both groups were high in working memory
capacity, so in part it may be that one virtue of having an efficient working memory is the
ability to effectively allocate resources. This is not all there is to it, however. Hierarchical
regression analyses in which recall was predicted from strategy and individual ability variables
showed that these strategy variables accounted for variance in recall apart from working
memory span and vocabulary level; the other point to be made about this analysis was that the
positive effects of vocabulary operated independently of strategy.
A recent study by Morrow and colleagues .(Morrow, Leirer, Andrassy, & Stine, 1994)
also suggests that older adults use different strategies in order to be effective in language
comprehension. In this experiment, subjects read narratives in which a character moved
through a known spatial layout and interacted with objects in this layout. Reading times for
young and old were longer for sentences referring to an object in a location through which the
protagonist had already passed than they were for an object near the location of the
182 E.A.L. Stine
protagonist. So for example, if Art is walking from his office through the subject reception
area into the kitchenette, reading times would be faster for a sentence describing Art
wondering if there is a beer in the refrigerator in the kitchen than they would be for one
describing Art wondering if the calculator is in his desk drawer in his office. These data thus
lend support to the contention that readers use a mental model of the layout to guide narrative
comprehension. But reading time is also affected by explicit mention of the location. Readers
have been found to take longer to comprehend sentences in which the location is n o t
mentioned, presumably because full comprehension involves instantiating the location through
inferential processes. In this study younger adults showed this "mention" effect regardless of
accuracy level on subsequent questions testing comprehension. Older adults who were below
average in comprehension showed a similar mention effect as that of the two young groups.
The older adults who were above average in comprehension were unique in showing a much
larger mention effect about twice that of the other three groups. It is noteworthy that
comprehenders were higher in working memory span than noncomprehenders. It would
appear then that some elders (who on average had greater working memory capacity) were
allocating resources differently to accommodate their capabilities and maintain a high level of
performance.
A third example I wish to note is drawn from a recently completed study from my lab
dealing with the on-line resolution of ambiguity .(Stine, Rub, & Hindman, 1994). For example,
in the course of reading a story about Fredda feeling tense and restless after a long day of
conference talks, you might encounter a sentence that begins, "She decided to look around in
the hotel for a bar .... "Now, at that point in the story, the word "bar" is ambiguous. Is the bar
for which Fredda will look an exercise bar on which she will do stretches or a bar which serves
beverages? It is not until you complete the sentence, e.g., "... so she could hear lively Irish
music until the wee hours of the morning" that you know for sure the meaning of "bar."
Earlier work .(Daneman & Carpenter, 1983) has shown that readers allocate slightly more time
on the first disambiguating word (in this case, "hear") relative to controls in which there is no
ambiguity (for example, if the troublesome word "bar" were replaced with "pub"), and even
more time at the sentence final word (in this case, "morning"). These increments are thought
to reflect processes of initial detection of the ambiguity and subsequent reintegration of the
whole sentence. In our study with elderly adults, high-span elderly were unique in allocating
disproportionately more time to the text as soon as it was disambiguated, and unlike the young
and low-span elderly, disproportionately less time at sentence wrap-up. This was also the
group showing the highest subsequent comprehension scores. Note that the successful elders
did not show evidence of integrating more broadly; rather they appeared to allocate resources
so as to minimize memory load.
Together these studies suggest that perhaps what distinguishes effective language
comprehenders among the elderly (who are often found to be measurable higher in out-of-
context assessments of working memory span) is a difference in resource allocation. In a 1989
chapter Carpenter and Just .(1989) note that high-span young subjects were more responsive to
load in changing their reading strategy. Specifically, they spent less time on lexical access
when a simultaneous memory load was required, suggesting that they accommodated the load
by encoding the meaning of the sentences more superficially. Thus, it may be that high-span
subjects are more flexible in how they use the activation they have. So is that all that span is?
An effective use of resources? Probably not. Engle and colleagues .(1992) in a recent JEP
article consider a number of empirical predictions that fall out of a "resource allocation"
account of working memory among the young and dismiss it on the grounds that the predicted
trade-offs in reading time do not always obtain.
A more compelling account is that there are individual differences in the amount of
activation available to be parceled out m and I reiterate that the high-span elders who are
doing so well on these tasks, are only high-span relative to their age group, and in terms of the
amount of activation available they are most comparable to the low-span young but in
addition, there is flexibility in the way in which these resources are allocated. It may be that
successful cognitive aging entails not having more activation m or a larger capacity working
memory m but rather the effective allocation of the resources that are available.
3. SUMMARY
While limitations in human information processing capacity have long been

acknowledged, the conceptualization of these limitations in experimental psychology has
evolved from that of a passive storage system to one of an activated subset of an extended
knowledge net; rather than a unitary system, this working memory is thought of as a group of
distributed capacities or modules, both dynamic and flexible in their operation. The focus of
the present chapter has been on age differences in how resources are distributed within this
working memory system, particularly in the course of language comprehension.
Brinley analyses of reading times do not appear to conform to the patterns observed in
response times for correct responses in discrete-trial tasks, e.g., lexical decision, mental
rotation. While similar to discrete-trial data in showing an orderly relationship between the
latencies of younger and older adults, the slope of the resulting functions appear to be routinely
close to unity rather than to the 1.5 typically found in discrete-trials tasks in which time is
measured for successful performance. Unlike those of discrete-trial tasks, the Brinley plots of
reading time are not directly relevant to the Slowing Hypothesis. Rather, because any
individual reading time does not necessarily correspond to a unit of successfully completed
processing (e.g., orthographic decoding, lexical access, contextual instantiation,
intraconstituent organization, interconstituent integration, integration with world knowledge,
etc.), Brinley plots of reading times reflect the relative allocation of processing resources by
young and old as text demands increase. The slope of unity suggests a large measure of
similarity between resource allocation strategies of younger and older readers. Subsequent
memory and comprehension performance, however, often reveal age differences favoring the
young, suggesting that age constancy in reading strategy may not be adaptive.
The Brinley approach was augmented by the use of regression analyses of reading time
in which specific cognitive constructs underlying the reading times could be identified. These
analyses substantially supported the Brinley analysis in showing great similarity between how
younger and older readers responded to specific text demands, but also suggested that older
adults were allocating somewhat less time to developing a cognitive representation of the text-
based meaning of the discourse. Older adults who showed high levels of memory performance
were those who used a more top-down approach, thus implicating the importance of
developing new styles of resource distribution to accommodate processing declines.
REFERENCES
Aaronson, D., & Scarborough, H.S. (1977). Performance theories for sentence coding: Some
qualitative models. Journal o f Verbal Learning and Verbal Behavior, 16, 277-303.
184 E.A.L. Stine
Albrecht, J. E., & O'Brien, E. J. (1993). Updating a mental model: Maintaining both local and
global coherence. Journal of Experimental Psychology: Learning, Memory, and
Cognition, 19, 1061-1070.
Atkinson, R. C., & Shiffdn, IL M. (1968). Human memory: A proposed system and its control
processes. In I~ W. Spence & J. T. Spence (Eds.), Advances in the psychology of learning
and motivation New York: Academic Press.
Babcock, 1L U, & Salthouse, T. A. (1990). Effects of increased processing demands on age
differences in working memory. Psychology and Aging, 5, 421-428.
Baddeley, A. D. (1981). The concept of working memory: A view of its current state and
probable future developments. Cognition, 10, 17-27.
Baddeley, A. D., & Hitch, G. (1974). Working memory. In G. H. Bower (Eds.), The
psychology of learning and motivation. London: Academic Press.
Carpenter, P. A., & Just, M. A. (1989). The role of working memory in language
comprehension. In D. Klahr & I~ Kotovsky (Eds.), Complex information processing: The
impact of Herbert A. Simon (pp. 31-68). Hillsdale: Erlbaunl
Cerella, J. (1990). Aging and information processing rate. In J. E. Birren & I~ W. Schaie
(Eds.), Handbook of the psychology of aging (pp. 201-221). New York: Academic Press.
Connelly, S. U, Hasher, L., & Zacks, R. T. (1991). Age and reading: The impact of
distraction. Psychology and Aging, 6, 533-541.
Craik, F. I. M., & Jennings, J. M. (1992). Human memory. In F. I. M. Craik & T. A. Salthouse
(Eds.), The handbook of aging and cognition (pp. 51-110). Hillsdale, NJ: Erlbaum
Craik, F. I. M., & Lockhart, 1L S. (1972). Levels of processing: A framework for memory
research. Journal of Verbal Learning and Verbal Behavior, 11, 671-684.
Crossley, M., & Hiscock, M. (1992). Age-related differences in concurrent-task performance
of normal adults: Evidence for a decline in processing resources. Psychology and Aging, 7,
499-506.
Daneman, M., & Carpenter, P. A. (1980). Individual differences in working memory and
reading. Journal of Verbal Learning and Verbal Behavior, 19, 450-466.
Daneman, M., & Carpenter, P. A, (1983). Individual differences in integrating information
between and within sentences. Journal of Experimental Psychology: Learning, Memory,
Daneman, M., & Tardi~ T. (1987). Working memory and reading skill re-examined. In M.
Coltheart (Eds.), Attention and performance XII: The psychology of reading Hillsdale, N.
J.: Lawrence Erlbaunl
Engle, 1L W., Cantor, J., & Carullo, J. J. (1992). Individual differences in working memory
and comprehension: A test of four hypotheses. Journal of Experimental Psychology:
Human Learning and Memory, 18, 972-992.
Gemsbacher, M. A., & Faust, M. E. (1991). The mechanism of suppression: A component of
general comprehension skill. Journal of Experimental Psychology: Learning, Memory, and
Cognition, 17, 245-262.
Glenberg, A., Meyer, M., & Lindem, I~ (1987). Mental models contribute to foregrounding
during text comprehension. Journal of Memory and Language, 26, 69-83.
Graesser, A. C., Singer, M., & Trabasso, T. (1994). Constructing inferences during narrative
comprehension. Psychological Review, 101, 371-395.
Haberlandt, ~ (1984). Components of sentence and word reading times. In D. E. ~ &. M. A.

Just (Eds.), New methods in reading comprehension research (pp. 219-251). Hillsdale, NJ:
Erlbaunl
Hale, S., Lima, S. D., & Myerson, J. (1991). General cognitive slowing in the nonlexical
domain: An experimental validation. Psychology andAging, 6, 512-521.
Hamm~ V. P., & Hasher, L. (1992). Age and the availability of inferences. Psychology and
Aging, 7, 56-64.
Hartley, J. T. (1993). Aging and prose memory: Tests of the resource-deficit hypothesis.
Hartley, J. T., Stojack, C. C., Mushaney, T. J., Annon, T. A. IC, & Lee, D. W. (1994).
Reading speed and prose memory in older and younger adults. Psychology and Aging, 9,
216-223.
Hess, T. M., & Tate, C. S. (1991). Adult age differences in explanations and memory for
behavioral information. Psychology andAging, 6, 86-92.
Hintzman, D. L. (1986). "Schema abstraction" in a multiple-trace memory model.
Hultsch, D. F., Hertzog, C., Small, B. J., McDonald-Miszczak, L., & Dixon, 1L A. (1992).
Short-term longitudinal change in cognitive performance in later life. Psychology and
Aging, 7, 571-584.
James, W. (1890). The principles of psychology. New York: Holt.
Just, M. A., & Carpenter, P. (1992). A capacity theory of comprehension: Individual
differences in working memory. Psychological Review, 99, 122-149.
Kintsch, W. (1988). The role of knowledge in discourse comprehension: A construction-
integration model. Psychological Review, 95, 163-182.
Laver, G. D., & Burke, D. M. (1993). Why do semantic priming effects increase in old age? A
McDowd, J. M., & Craik, F. I. M. (1988). Effects of aging and task difficulty on divided
attention performance. Journal of Experimental Psychology: Human Perception and
Monsell, S. (1984). Components of working memory underlying verbal skills: A "distributed
capacities" view. In H. Bouma & D. G. Bouwhuis (Eds.), Attention and performance X:
Control of language processes (pp. 327-350). HiUsdale, N. J.: Lawrence Erlbaum
Associates Inc.
Morrow, D. G., Leirer, V., Andrassy, J., & Stine, E. A. L. (1994). The role of age and
working memory capacity in creating situation models from text. Presented at the Fifth
Cognitive Aging Conference, Atlanta, GA.
Morrow, D. G., Leirer, V. O., & Altieri, P. A. (1992). Aging, expertise, and narrative
processing. Psychology and Aging, 7, 376-388.
Myerson, J., Ferraro, F. 1L, Hale, S., & Lima, S. D. (1992). General slowing in semantic
priming and word recognition. Psychology and Aging, 7, 257-290.
Myerson, J., Hale, S., Wagstaff, D., Poon, L. W., & Smith, G. A. (1990). The information loss
475-487.
O'Brien, E. J. (1995). Automatic components of discourse comprehension. In 1~ F. Lorch & E.
J. O'Brien (Eds.), Sources of coherence in reading (pp. 159-176). Hillsdale: Erlbaum
186 E.A.L. Stine
Perfect, T. J. (1994). What can Brinley plots tell us about cognitive aging? Journal of
Salthouse, T. A. (1988a). Initializing the formalization of theories in cognitive aging.
Salthouse, T. A. (1988b). Resource-reduction interpretations of cognitive aging.
Developmental Review, 8, 1-35.
Salthouse, T. A. (1988c). The role of processing resources in cognitive aging. In M. L. Howe
& C. Brainerd (Eds.), Cognitive development m adulthood (pp. 185-239). New York.
Salthouse, T. A. (1991a). Mediation of adult age differences in cognition by reductions in
working memory and speed of processing. Psychological Science, 2, 179-183.
Salthouse, T. A. (199 lb). Theoretical perspectives on cognitive aging. Hillsdale: Erlbaum
Salthouse, T. A. (April, 1988). Coping with too many resources: Usage of processing
resources concepts in cognitive aging. Presented at the Second Cognitive Aging
Conference, Atlanta, GA.
Salthouse, T. A. (1992). Working-memory mediation of adult age differences in integrative
reasoning. Memory and Cognition, 20, 413-423.
Salthouse, T. A., & Babcock, 1L L. (1991). Decomposing adult age differences in working
memory. Developmental Psychology, 27, 763-776.
Salthouse, T. A., Babcock, IL L., Skovronek, E., Mitchell, D., & Palmon, 1L (1990). Age and
experience effects in spatial visualization. Developmental Psychology, 26, 128-136.
Salthouse, T. A., & Skovronek, E. (1992). Within-context assessment of age differences in
working memory. Journal of Gerontology: Psychological Sciences, 47, P110-120.
Sanford, A. J., & Garrod, S. C. (1981). Understanding written language: Explorations of
comprehension beyond the sentence. New York: John Wiley & Sons.
Stine, E. A. L., & Hindman, J. (1994). Age differences in reading time allocation for
propositionally dense sentences. Aging and Cognition, 1, 2-16.
Stine, E. A. L., Lachman, M., & Wingfield, A. (1993). The roles of perceived and actual
control in memory for spoken language. Educational Gerontology, 19, 331-349.
Stine, E. A. L., Loveless, M. IC, & Soederberg, L. M. (in preparation). Resource allocation in
on-line reading by younger and older adults.
Stine, E. A. L., Ruh, J. L., & Hindman, J. (1994). The effects of age and working memory
capacity on reading time allocation and comprehension of ambiguous text. Presented at
the Fifth Cognitive Aging Conference, Atlanta, GA.
Troyer, A. I~, Graves, 1L E., & Cullum, C. M. (1994). Executive functioning as a mediator of
the relationship between age and episodic memory in healthy aging. Aging and Cognition,
1, 45-53.
Tun, P. A., Wingtield, A., & Stine, E. A. L. (1991). Speech processing capacity in younger and
older adults: A dual-task study. Psychology and Aging, 6, 3-9.
Tun, P. A., Wingtield, A., Stine, E. A. L., & Mecsas, C. (1992). Rapid speech processing and
divided attention: Processing rate vs. processing resources as an explanation of age effects.
Waugh, N. C., & Norman, D. A. (1965). Primary memory. Psychological Review, 72, 89-104..
Wiegersma, S., & Meertse, K. (1990). Subjective ordering, working memory, and aging.
Experimental Aging Research, 16, 73-77
N e u r o p s y c h o l o g i c a l Implications of W o r d Recognition Deficits
Marian B. Patterson a and Danielle N. Ripichb
a Alzheimer's Center, University Hospitals of Cleveland, and Case Western Reserve University
b Case Western Reserve University
Defects of word recognition are seen in patients with a variety of neurological

disorders. In this chapter, we focus on one type of word recognition, the reading of single
words. When such deficits are present in patients who previously possessed the ability but
have lost it as the result of an insult to the brain, they provide a sort of natural experiment on
the relationship between brain functioning and behavior. Such studies are in the realm of
neuropsychological research. One of the goals of such research is to reach an understanding of
the nature of the cognitive processes underlying the deficit. It is in mm hoped that an
understanding of these mechanisms may shed light upon the processes involved in reading
among normal individuals.
The relevance ofneuropsychological studies to the study of normal cognitive processes
is explained by Grodzinsky (1990) as that of providing evidence that can sharpen the
development of a theoretical account of a cognitive process by placing constraints on the
theory. That is, a theory of cognitive function must be able to account for a particular kind of
behavior manifested by a brain-damaged patient. Grodzinsky (1990, Chapter l, p. 17) terms
this "the neuropsychological constraint of breakdown- compatibility." As he points out, not all
neurological conditions that affect word recognition and reading are likely to provide relevant
information. For example, a failure of visual word recognition caused by blindness will be
unrelated to the cognitive processes involved in reading beyond initial sensory perception. A
lesion that results in the loss of all language function is likely to shed little light upon the
processing involved in single word reading. But a brain lesion that results in selectively greater
impairment of the ability to read irregular words (that is, words like "sugar" and "laugh" that
cannot be pronounced by sounding them out phonetically) relative to regular words, a
phenomenon seen in the condition called surface dyslexia, may force us to refine a theory of
word recognition to account for this dissociation.
There are two classes of neuropsychological evidence that have particular relevance to
theories about word reading. The first is an isolated breakdown of a word reading in a context
of relatively preserved auditory language comprehension. In focal conditions such as surface
dyslexia, that result in the loss of ability to read certain kinds of words, one may test a
theoretical prediction that particular cognitive deficits may occur. Thus, a theory that accounts
for the pattern of deficits seen in this condition may be used to generate and test hypotheses
about how the condition would affect the processing of some linguistic variable, such as word
frequency. For example, one theoretical view holds that surface dyslexics have lost the ability
to read whole words and must rely on sounding out words phonemically. Thus, they would be
expected to have difficulty reading all irregular words. However, Bub, Chancelliere, and
Ketrtesz, et al (1985), described a patient (MP) with surface dyslexia who was able to read
high frequency, but not low frequency, irregular words. This observation required the authors
188 M.B. Patterson and D.N. Ripich
to elaborate a modified theory of the process of reading to account for the pattern of
disturbances in this case.
Patients with generalized cognitive impairment associated with Alzheimer's disease
(AD) offer another kind of evidence (see Schwartz, 1988 for an extensive discussion). In some
patients with AD, for example, correct pronunciation of words is typically preserved, even at a
point in the disease when the patients are so impaired that they are not able to understand the
meaning of words that they can pronounce. Both classes of evidence, i.e., impaired language
skills resulting from focal lesions and preserved abilities in the face of generalized impairment,
provide tests that can be used to place constraints on theories of language processing.
In this chapter, we offer examples of the neuropsychological evidence that has
contributed to the elaboration of theories of single word reading. A number of relevant themes
have emerged in neuropsychological research on word recognition over the past two decades.
It is beyond the scope of this chapter to address all the various lines of investigation that have
been pursued; however, we will attempt to present salient examples of this research from an
historical perspective and point to areas where further research is needed. First, we will
outline a theoretical framework of cognitive processing that will serve to organize our
presentation of the neuropsychological research on word recognition. Second, we will review
some of the neuropsychological research on deficits that produce linguistically relevant
evidence. Although our major focus will be on specific, acquired deficits of single word
reading, research on patients with generalized impairment and on language comprehension will
be introduced when relevant to a particular point. Third, we will address the relevance of this
neuropsychological evidence to theories of aging.
1. THEORETICAL APPROACHES TO DISTURBANCES OF LANGUAGE

PROCESSING
Attempts to explain the disturbances in language processing experienced by patients

with brain lesions are directed by a theoretical understanding that determines what sort of
evidence will be collected and how it should be interpreted. Thus, in order to evaluate the
neuropsychological literature on word recognition, one must be familiar with the theoretical
models that motivate the studies. In the following pages, we briefly review three related
theoretical models that currently drive much of the neuropsychological research on word
recognition.
1.1. Functional Localization.
Proponents of functional localization theories, such as Benson (1993), Geschwind

(1965), and H6caen and Albert(1978), conceptualize language processing as comprised of a
group of separate activities that are linked to specific neurological structures. Disturbances of
language correspond to lesions in brain areas underlying particular language activities like oral
expression, repetition of spoken words and phrases, and comprehension of written or spoken
language, or to lesions interrupting the connections between the critical language areas. For
example, focal lesions that involve the parietal and temporal areas but not perisylvian
structures may produce a syndrome termed extrasylvian sensory aphasia. These patients have
defective comprehension of both spoken and written language, and their expressive language,
Neuropsychological implications of word recognition deficits 189
though fluent, is marred by paraphasic substitutions. They can, however, repeat words even
though they do not understand them (Benson, 1993, p. 30).
The functional localization approach to understanding language disturbances developed
in clinical settings in which the patterns and dissociations among the general classes of
communication deficits could be used to identify aphasic syndromes. It~ in a given patient, one
ability was impaired while others remained intact, the area of the brain that was damaged was
inferred to contain the neural center (or the transmission pathway) responsible for that
language activity. Thus, studies of functional localization in brain-damaged patients led to a
theory about the organization of the brain structures responsible for language. It proved
possible to predict the general locus of the underlying neurological lesion on the basis of the
pattern of language deficits displayed by the patient.
1.2. Information Processing.
A number of information processing models were proposed in the 1970's and early
1980's to account for disturbances of reading in patients with focal neurological lesions
(Saffran, 1985; Coltheart, Patterson and Marshall, 1980; Marshall and Newcombe, 1973). In
such models, word reading is conceptualized as being comprised of a series of components or
stages through which information passes, be~nning with the initial perceptions and ending with
some behavioral output. The components, or modules, are relatively independent, in that
damage to one does not necessarily affect the operation of another (although it may affect the
input or output of other components). Disruptions may occur with damage to one or more
functional components of the system such as the lexicon of word meanings or the grapheme-
to-phoneme component by which visual stimuli are converted to auditory ones, or to the
connections between component processes (Coltheart, 1985; 1987).
Cognitive neuropsychologists analyze data from both normal and disordered language
to develop and test their models of information processing. For example, acquired dyslexia,
the loss (or partial loss) of previously intact ability to read, is associated with a variety of
reading errors that can be functionally analyzed through a detailed examination of the
components of specific patterns of breakdown. Although the existence of patients in whom a
particular module is damaged while others remain intact implies that the intact and
malfimctioning modules must have separate brain localization, a knowledge of the precise
neurological structures involved is not necessary in studying the cognitive mechanisms
involved in the breakdown. It is the patterns of performance, not the underlying neurological
lesion, that are of theoretical interest. As illustrated earlier, if the component processes and
pathways proposed in a given model cannot explain the pattern of deficits shown by a patient,
then the theoretical model must be modified or abandoned.
1.3. Modular Processing.
The concept of modularity as an approach to explaining the operations of human

perception and language was introduced by Fodor in 1983. He suggested that many of our
mental processes operate as pre-programmed, fixed modules that process a particular,
circumscribed class of perceptual information and the resulting output. Such modules run
automatically and involuntarily. The operation of these modules is planned and regulated by
non-modular central processes that are under voluntary control and are concerned with higher
order cognitive processes like integration of information from various modules, accessing
knowledge and making inferences.
Fodor (1983) originally proposed explicit criteria that define modular processes.
Subsequent authors (see Moscovitch and Umlita, 1990; Grodzinsky, 1990; Ellis and Young,
1988, for example) have suggested that some of these criteria are not essential to demonstrate
that a particular process is modular. For example, although Fodor (1983) listed as two criteria
of modules, that they are innately specified and not assembled, Moscovitch and Umlita (1990)
proposed an expanded definition to account for processes like reading and other skilled, but
highly automatic activities or habits, which, although learned by the organism, otherwise
behave like modules.
These authors, along with Grodzinsky (1990), and Ellis and Young (1988), view as
critical the modular properties of domain specificity, cognitive impenetrability and information
encapsulation. Domain specificity means that a module can process only one type of
information, for example, that concerned with speech or with visual perception of faces.
Cognitive impenetrability and information encapsulation mean that central processes are
neither aware of nor able to influence the contents and operation of the module once it has
been activated, and modules have no access to any information other than that received within
their domains.
Cognitive neuropsychologists have explored the possibilities of modular organization as
it relates to language and central processing. Information processing models are modular
systems, in that the individual processing components are relatively independent of one another
in their operation, and a module can be added, modified, or deleted without affecting the rest
of the information processing system, as long as input to and output from other components of
the system is not changed (Ellis, 1987). However, while these modules may be domain specific,
in that they receive and process only a certain kind of information, they do not necessarily meet
the critical criterion of information encapsulation, in that the extent to which higher cognitive
processes have access to or are able to influence the component processes is not specified in
the models (Vallar, 1991). The additional theoretical constraints posed by Fodor's
conceptualization of modularity been applied in some of the recent neuropsychological
research on language processing.
2. NEUROPSYCHOLOGICAL RESEARCH ON WORD READING DEFICITS.
Acquired deficits of word reading are most often categorized with respect to the nature
of the errors made by brain-damaged patients whose ability to read is impaired, but not entirely
eliminated. The errors that are evaluated comprise broad classes of diverse language activities,
such as reading phonetically regular words, reading non-words, reading single letters, reading
complex paragraphs, and so on. Certain patterns of errors are typically found in conjunction
with one another and, when they disrupt reading comprehension, define clinical syndromes of
acquired dyslexia. We will describe several of these here to give an idea of the nature of the
data that is sought and interpreted in these neuropsychological approaches.
Syndromes of acquired dyslexia have been subject to several various classifications,
depending on the interests and orientation of the investigator. The historical development
leading to modem functional localizationist views has been summarized by H6caen and Albert
(1978), and Benson and Geschwind (1969), among others. Early reports tended to focus on
the relationship of the reading disturbance to associated language deficits, e.g. spontaneous
speech, naming, writing, repetition and comprehension. Two distinct syndromes with different
underlying neuroanatomical correlations have consistently been noted. One, termed ~lexia with
agraphia, was characterized by inability to read or to write, with relatively normal expressive
speech, auditory comprehension and repetition, and associated particularly with lesions in the
angular gyms of the lei%parietal lobe. The second was alexia without agraphia, a syndrome in
which patients who have lost the ability to read are still able to write, although they cannot
read what they themselves have written. Such a pattern of deficits may be explained on the
basis of damage to both the occipital area of the dominant hemisphere and to the splenium of
the corpus callosunl Therefore, visually presented words can neither be processed.by the
damaged dominant occipital area, nor reach, via callosal transmission from the nondominant
hemisphere, the intact dominant hemisphere structures responsible for language comprehension
(H6caen and Albert, 1978).
Grodzinsky (1990) criticizes the functional localization approach for its inability to
explain how loss of a specific neurological structure or structures could account for patterns of
partial loss of function and such subtle patterns of deficits as those seen in some aphasic
patients, who may lose the ability to process certain grammatical characteristics but not others.
In addition, he points out the lack of a theoretical basis for selecting the particular language
activities to be examined. That is, the variables used to classify patients were selected on the
basis of clinical observations of patient behavior rather than an explanatory structure
concerning the mechanisms of language functioning. If a different set of linguistic
characteristics were selected for observation, it is possible that a new set of brain-behavior
correspondences would be apparent, and even that the generally loose fit between underlying
lesion locus and pattern of language deficits might be sharpened.
Thus, Grodzinsky suggests that the localizationist approach is of only limited value in
developing theories of the cognitive mechanisms of language. While the variability in
patterning of deficits shown by aphasic patients might be attributed to variations in the location
and extent of underlying tissue damage, the imprecise nature of clinical pathological
correlations makes it difficult to specify with precision the theoretical implications of a
particular pattern of deficits.
Cognitive neuropsychologists, building on the observation that most patients with
acquired dyslexia are not entirely unable to read, attempted to characterize the reading deficits
according to those aspects of reading that were preserved or damaged. Working from an
information-processing perspective, Marshall and Newcombe in 1973 described cases in which
the patterns of errors suggested three types of reading disturbances: visual dyslexia, a tendency
to confuse visually similar words and letters; semantic dyslexia, a tendency to substitute
visually dissimilar but semantically similar words (e.g. "speak" might be read as "talk"); and
surface dyslexia, in which grapheme-to- phoneme conversion leads to success with regular
words, but errors with irregular words, and in which the rules for spelling are sometimes
violated. These types of errors, as well as errors made by normal readers under some
circumstances, could be explained by positing disruptions at different locations along the
information processing pathways involved in reading. The existence of these distinct dyslexic
syndromes provided support for the hypothesis that reading is accomplished by a two-route
system, in which, when one route to reading is disrupted by disease, alternative, though less
efficient, routes may be available. Thus, reading may be carded out through the direct
recognition of whole words and their association with their meaning or pronunciation, or it
may be carded out by sounding out the word phonetically, as in the example of surface
dyslexia cited earlier. It is assumed that in the normal reader, both routes are available, and are
used whenever needed to achieve most efficient processing.
Subsequent work focused on the detailed analysis of patients with three varieties of
acquired dyslexia: deep dyslexia, phonological dyslexia, and surface dyslexia (Marshall and
Newcombe, 1973; Newcombe and Marshall, 1981; Saffran, 1985, among others). The aim in
these studies was to apply the data supplied by brain damaged patients to test and modify
theories about the cognitive mechanisms that underlie word recognition and reading. These
syndromes and their theoretical implications have been discussed extensively in the literature
(Patterson, Marshall and Coltheart, 1985; Coltheart, et al, 1980). For clarity of exposition, we
give a greatly simplified account. In general terms, these syndromes are characterized with
respect to how well the patient can read aloud regular words vs irregular words and familiar vs
unfamiliar words or non-words; and by the kinds of reading errors they typically make. The
pattern of failures and the nature of the errors allows the investigator to infer the information
processing route being used by the patient and the point (or points) along the pathway at which
the reading process must be disrupted.
2.1. Deep dyslexia.
Patients with deep dyslexia perform best when reading familiar real words, whether
regular or irregular, but they have great difficulty with non-words. Although they typically
make several different kinds of errors, the most interesting from a theoretical point of view are
their "semantic" errors. For example, the word "nephew" might be read as "cousin" or "city"
as "town" (Saffran, 1985; Coltheart, et al, 1980).
Because such patients cannot pronounce non-words, it is thought that they are unable
to perform grapheme to phoneme conversion, and so must rely on a direct reading route, in
which the whole word is directly associated with its meaning. The presence of semantic errors
suggests that, in addition to a failure of the grapheme-phoneme conversion process, the locus
of the breakdown in such patients may involve the semantic store itself or the process by which
word meanings are attached to their names (Coltheart, 1987). Evidence from detailed studies
of individual patients suggests that there is more than one form of deep dyslexia, characterized
by the locus of their semantic errors (Coltheart, 1987; Shallice and Warrington, 1980). For
example, Coltheart (1987) cites studies of patients with deep dyslexia showing that some make
errors that implicate damage to semantic memory, while others make errors that implicate
problems with name-retrieval.
2.2. Phonological dyslexia.
Like patients with deep dyslexia, those with phonological dyslexia are able to read
familiar words, both regular and irregular, better than non-words, but they do not make
semantic errors. Again, careful case studies have revealed that phonological dyslexia may have
several differing bases and is not a unitary syndrome (Coltheart, 1987). Nevertheless, the
ability of such patients to read familiar words aloud at levels close to normal, as well as the
normal reader's ability to read irregular words, provides evidence that the direct "print-to
sound" reading route bypassing grapheme to phoneme conversion can be used both by
phonological dyslexic and normal skilled readers (Saffran, 1985; Ellis and Young, 1988).
2.3. Surface dyslexia.
Patients with surface dyslexia are generally able to read regular words and
pronounceable non-words relatively well. However, when they attempt to read irregular
words they are likely to make errors by pronouncing the words as though they were sounding
them out phonetically, much like a be~nning reader (Saifran, 1987, p.244). Also like
be~nning readers, these patients tend to make '~ errors, pronouncing words on
the basis of grapheme to phoneme conversion rules rather than by their conventional
pronunciations. Studies of such patients have suggested that they accomplish reading via a
"grapheme to phoneme" route.
Like deep and phonological dyslexia, surface dyslexia turns out not to be a unitary
syndrome. Ellis and Young (1988) cite case studies that show that surface dyslexia can result
t~om disruptions at several different points in the flow of information from orthographic input
to phonological output. Some patients appear to have lost the ability to recognize whole
words, some to have lost the ability to attach meaning to the words that they successfully
sound out, and others to have lost the ability to attach word pronunciations to their visual
representations.
While information processing theories have been very useful in the specification of
likely pathways of information flow during reading, Grodzinsky (1990) and Ellis (1987) cite
limitations with such models as an explanatory framework in neuropsychology. To the extent
that such models of language disturbances can be modified to explain the unique patterns of
deficits shown by individual patients, they are essentially descriptive and unconstrained. That
is, as patients are found whose deficits cannot be explained by a particular model hypothesized
processes or subprocesses can readily be added, and one could conceivably build a unique
model to explain the particular pattern displayed by each individual case. In addition,
Grodzinsky (1990) maintains that such models do not require theoretical explanations of the
content or operations of the component systems within the model. He argues that the selection
of data to be gathered is based on its utility in classifying patients who have difficulty with
certain activities, rather than being based on theories of the structure of the mind. Moreover, it
not clear that the kinds of processing components and pathways, e.g. the grapheme-to-
phoneme processing route, currently envisioned in information processing models of reading,
correspond to the way in which the brain is organized. As Ellis (1987, p.401) notes, "...we
would be deluding ourselves if we thought that any actual set of modules we were to propose
today might bear anything more than a passing resemblance to the ultimate 'true' set (assuming
they are discoverable)". Thus, while information processing theories continue to provide the
basis for much productive neuropsychological research on word recognition, further theoretical
elaboration of the structure and content of the component processes is needed.
These attempts during the early 1980's to fimtter specify the mechanisms underlying the
patterns of breakdown in dyslexic syndromes through detailed case studies made it clear once
again that there are great individual differences among patients (Coltheart, 1987; Ellis, 1987).
As a consequence, rather than focusing on the identification of syndromes, cognitive
neuropsychologists have begun to focus on the specification of detailed models and on the
analysis of the behavior of modules within the models and the interactions between the~L Ellis
(1987, p.403) points out that the task of the cognitive neuropsychologist is to examine the
evidence presented by a patient or group of patients, and to ask, "Is this pattern of data
(deficits and intact skills) interpretable within the framework provided by existing theories, or
are modifications called for?"
2.4. Semantic dyslexia.
There is a fourth group of patients whose ability to read aloud is quite good for non-
words, regular words and familiar irregular words, but whose ability to understand the words
they read is severely impaired (Schwartz, Saffran and Matin, 1980; Saffran, 1985). These
include many patients with progressive dementias like Alzheimer's disease, in whom the pattern
of deficits changes over time. On the basis of her review of studies of patients with semantic
dyslexia, Saifran (1985) has suggested that the available routes to single word reading are
abolished one at a time as the disease progresses. The first lost is the route from printed word
to word meaning, and the second the grapheme to phoneme conversion route, so that
eventually the patient is left only with the direct route from visual representation to word
sound, thus leaving the patient with only the ability to pronounce words without understanding
their meaning.
It is both the preservation of reading in the face of generalized cognitive impairment
and the selective and circumscribed nature of cognitive deficits found in patients with varieties
of acquired dyslexia that provide support for the concept of modularity of cognitive functions.
Moscovitch and Umlita (1990, p.8) have stated, "If domain specificity and information
encapsulation are the primary characteristics of modules, then neuropsychological evidence of
double dissociation in patients with focal brain damage and of sparing of function in dementia
are necessary for establishing that a cognitive system (or process) is modular." According to
these authors, the selective nature of the impairment seen in acquired dyslexia, in which the
breakdown of reading can be related to modules that process a specific domain of information,
provides evidence of domain specificity. The preservation of the ability to read aloud despite
generalized cognitive loss provides evidence of information encapsulation, i.e., higher cognitive
processes neither influence nor access the operation of the module.
A second example of modularity in language processing comes from Grodzinsky's
research on agrammatic patients (Grodzinsky, 1990). These are patients with Broca's aphasia
whose language behaviors show a break-down in grammar. Specifically, Grodzinsky examined
evidence for the modular properties of the "syntactic parser", responsible for grammatical
analysis in the comprehension of sentences. His experiments used the analysis on-line
processing of sentence comprehension to examine the nature of the defects of the syntactic
parser among agrammatic aphasic patients. He demonstrated that, while agrammatic patients
may use information from both grammar-based processing and nonlinguistic cognitive
processes to decode a sentence, the grammar-based processing is completed first, and is
independent of input from and inaccessible to higher cognitive processes. That is, it is
cognitively impenetrable and informationally encapsulated. Finally, Grodzinsky suggests that
since the syntactic parser demonstrates modularity even in patients in whom its language
processing is defective, this must also be the case in normal individuals.
It should be noted at this point that, in stressing modular systems, we have not
addressed a fourth model that is gaining currency in the field of cognitive neuropsychology,
that of parallel distributed processing (PDP). PDP models are unlike the modular systems in
that they are highly interactive and involve simultaneous, rather than sequential, processing of
linguistic information on many levels. For an overview of how PDP theorists account for the
cognitive and anatomic concomitants of syndromes of acquired reading disorders, the reader is
referred to Friedman, Ween and Albert (1993).
2.5. Word and language processing deficits associated with cerebral vascular accidents.
A common cause of disturbances of word and language processing in older individuals

is a cerebral vascular acddent (CVA). This term encompasses stroke, usually used to describe
a thrombotic or embolic cerebral infarction, as well as cerebral hemorrhage. Langauge
disorders most commonly result from lesions in the left hemisphere, which is dominant for
language in over 90% of the population. Although in the weeks or months following a CVA
there is often considerable improvement, patients may be left with varying
Sdegrees of permanent impairment that depends on the nature and location of the vascular
lesion. If the lesion is extensive and involves both anterior and posterior language areas, the
result is a global aphasia, the complete loss of receptive and expressive language function. The
localized syndromes described earlier in this chapter, in which a single language function like
reading is selectively involved, however, are produced by more restricted lesions.
Vascular anatomy is an important determinant of where strokes occur, and thus what
aspects of language function are likely to be disrupted. From the perspective of the clinician
who is evaluating a patient, the presence of a set of clinical signs and symptoms leads to the
inferences about location and nature of the underlying neurological lesion (i.e., hemorrhage vs.
infarction; cortical vs. basal ganglia). The relationships between the neurological lesions and
resulting cognitive deficits are complex and often change rapidly during the acute phase and
early recovery stages, however. A brief overview of some of the word and language
processing deficits associated with stroke is presented here. For a more extended discussion of
cerebrovascular disorders in the elderly, the reader is directed to chapters by Daroff and
Conomy (1988), Funkenstein (1988), and Kase and Molar (1984).
Lesions in the distribution of the middle cerebral artery in the left hemisphere of fight
handed individuals produce some of the more frequently seen language disorders. When the
anterior division of the middle cerebral is involved, frontal and anterior parietal structures, may
be damaged, and several varieties of nonfluent aphasia, including agrammatism, or Broca
aphasia result. In these patients, paralysis and sensory loss contralateral to the site of the lesion
are common and other symptoms such as dyspraxia and dysarthria may be present. While the
comprehension of language may be less severely affected than expression, there is generally
some disturbance of spoken and written material as well. CVAs involving the posterior branch
of the middle cerebral artery affect the temporal and parietal lobes, and are more likely to result
in language disorders such as Wernicke's aphasia in which the patient speaks fluently, but may
convey little meaning, and there are frequent paraphasic substitutions of one word for another.
In these patients, comprehension of both spoken and written material is defective. Often there
is a contralateral hemianopia, the loss of vision in the half of the visual field opposite the side
of the lesion.
Other branches of the cerebral arterial system may also produce lesions that disrupt
language function. When the anterior cerebral artery is affected, lesions typically involve
frontal lobes, and may extend into the corpus callosum and subcortical regions such as the
hypothalamus. There may be motor weakness, difficulties with pronunciation or writing, and,
in the acute stage, a disruption of attention and spontaneity of movement. A CVA involving
the posterior cerebral artery usually affects structures in the temporal and occipital lobes, the
splenium of the corpus callosum, as well as the midbrain and thalamus, such lesions produce
visual field defects and somato-sensory deficits. The syndrome of alexia without agraphia
described earlier may result from lesions involving this distribution.
3. RELEVANCE OF NEUROPSYCHOLOGICAL EVIDENCE TO THEORIES OF

AGING
It is difficult to evaluate the effects of age per se on acquired deficits of word

recognition, since the disease processes that produce such focal deficits are likely to differ
between young and elderly patients. For example, while deep dyslexia is often associated with
cerebral vascular accidents among the elderly, strokes are uncommon among younger patients,
in whom the etiology of dyslexia might more likely be tumor or trauma. Effects of aging on
the expression of acquired disturbances of word recognition would consequently be
confounded with type of pathology. Furthermore, it has been argued (Caramazza and
McCloskey, 1988) that because one cannot know the exact extent and location of cerebral
lesions, each patient may have a unique constellation of deficits and thus it is dangerous to
draw conclusion from data based on groups of brain-damaged patients. Thus, many cognitive
neuropsychologists rely on single case studies to tease apart the effects of damage to a
particular cognitive process. Grodzinsky, however, argues that group studies of brain
damaged patients are indeed possible, so long as one clearly identifies the cognitive deficits of
interest and demonstrates that patients under study are able to carry out the demands of the
experimental tasks. In such groups, the age of the patient may still be an important variable.
What are implications of the results from the studies of focal and generalized lesions for
the study of age differences? To the extent that domain specific language processing modules
that are informationally encapsulated and cognitively impenetrable can be demonstrated even in
severely abnormal patients like those with Alzheimer's disease, one may speculate that their
operation might also be resistant to the effects of normal aging. If that is so, then it may be
that some modular processes are resistant to age-related changes like slowing of information
processing. Such a scenario might account for results like those of Allen and associates
(1993), that demonstrated that normal individuals show localized age effects on components of
word recognition as well as generalized slowing. Other such local age effects might be
especially likely to be found among processes like that of the syntactic parser studied by
Grodzinsky, in which modularity can be demonstrated. In this case, one might thus turn to
cognitive neuropsychological studies of brain-damaged individuals to identify theoretically
relevant variables that could be examined in the context of age differences among normals.
In summary, cognitive neuropsychologists, by carrying out detailed studies of the
patterns of success and failure exhibited by both normal and brain damaged individuals, have
developed and refined theories about the cognitive mechanisms underlying single word
reading. The evidence provided by patients with acquired deficits of reading, and in particular,
by patients whose deficits show a pattern of dissociation in which some aspects of reading are
impaired while other aspects are preserved, is especially valuable. They have required that
theories of word recognition be sufficiently defined to allow for the unique pattern of
breakdown. Such cases, although rare, provide evidence that some components of language
processes may be relatively independent of one another, i.e., they satisfy one of the criteria for
modularity.
Neuropsychological implicationsof word recognitiondeficits 197
The three theoretical models that we have discussed - functional localization,

information processing, and modular processing - all handle the dissociations of functional
impairment seen in brain-damaged patients, but they differ in the kind of constraints demanded
by the theory. Fodors' (1983) concept of modularity poses the most rigorous constraints, yet
there are examples in the language abilities of agrammatic aphasics and in the reading
performance of patients with focal lesions and generalized dementia, in which modular
processing appears to occur. The implication with respect to cognitive aging is that such
modular processes may provide a vehicle for examining the effects of aging on complex, but
automatic operations that are cognitively isolated from central processes.
REFERENCES
Allen, P.A., Madden, D.J., Weber, T.A., and Groth, I~E. (1993). Influence of age and
processing stage on visual word recognition. Psychology and aging, 8, 274-282.
Benson, D.F. and Geschwind, N. (1969). The alexias. In P.J. Vinken and G.W. Bruyn (Eds.),
Handbook of Clinical Neurology, Volume 4: Disorders of Speech, Perception, and
Symbolic Behaviour. New York: American Elsevier Publishing Co., Inc.
Benson, D.F. Aphasia. In K.M. Heilman and E. Valenstein (Eds.). Clinical Neuropsychology,
Third Edition. New York: Oxford Universities Press.
Bub, D., Chancelliere, A., and Ketrtesz, A. (1985). Whole-word and analytic translation of
spelling to sound in a non-semantic reader. In I~E. Patterson, J.C. Marshall, and M.
Coltheart (Eds.), Surface dy slexia: Neuropsychological and cognitive studies of
phonological reading. London: Lawrence Erlbaum Associates, Ltd.
Caramazza, A. and McCloskey, M. (1988). The case for single-patient studies. Cognitive
Neuropsychology, 5, 517- 528.
Coltheart, M. (1987). Functional architecture of the Language-processing system In M.
Coltheart, G. Sartori and R. Job (Eds.).The Cognitive Neuropsychology of Language.
Hillsdale, New Jersey:Lawrence Erlbaum Associates, Publisher.
Coltheart, M. Cognitive neuropsychology and the study of reading. In M.I. Posaer and O.S.M.
Marin (Eds.), Mechanisms ofAttention: Attention and Performance XI. Hillsdale, New
Jersey:Lawrence Erlbaum
Coltheart, M., Patterson, K, and Marshall, J.C. (1980). DeepDyslexia. London: Routledge &
Kegan Paul.
Darofl~ 1LB., & Conomy, J.P. (1988). Contributions to contemporary neurology. Boston:
Butterworths.
Ellis, A.W. (1987). Intimations of modularity, or, the modularity of mind: doing cognitive
neuropsychology without syndromes. In M. Coltheart, G. Sartori and 1L Job (Eds.). The
Cognitive Neuropsychology of Language. Hillsdale, New Jersey: Lawrence Erlbaum
Associates, Publisher.
Ellis, A.W. and Young, A.W. (1990). Human Cognitive Neuropsychology. Hillsdale, New
Jersey: Lawrence Erlbaum Associates, Publisher.
Fodor, J.A. (1983). The modularity of mind. Cambridge, Massachusetts: MIT Press.
Friedman, 1LF., Ween, J.E., and Albert, M.L. (1993). Alexia. In I~M. Heilman and E.
Valenstein (Eds.). Clinical Neuropsychology, Third Edition. New York: Oxford
Universities Press.
Funkenstein, H.H. (1988). Cerebrovascular disorders. In M.S. Albert and M.B. Moss (Eds.).
Geriatric Neuropsychology. New York: The Guilford Press.
Geschwind, N. (1965). Disconnexion syndromes in animals and man. Brain, 88, I, 237-294,
585-564.
Grodzinsky, Y. (1990). Theoretical Perspectives on Language Deficits. Cambridge,
Massachusetts: The MIT Press.
Hcaen, H. and Albert, M.L.(1978). Human Neuropsychology. New York: John Wiley & Sons.
Kase, C.S. and Molar, J.P. Cerebrovascular diseases in the elderly: Clinical syndromes. (1984).
In M.L. Albert (Ed.). Clinical Neurology of Aging. New York: Oxford Univeristy Press.
Marshall, J.C. and Newcombe, F. (1973). Patterns of paralexia: a psycholinguistic approach.
Journal of Psycholinguistic Research, 175-199.
Moscovitch, M. and Umlita, C. (1990). Modularity and neuropsychology: modules and central
processes in attention and memory. In M. F. Schwartz (Ed.), Modular Deficits in
Alzheimer-Type Dementia. Cambridge, Massachusetts: The MIT Press.
Newcombe, F. and Marshall, J.C. (1981). On psycholinguistic classification of the acquired
dyslexias. Bulletin of the Orton Society, 31, 29-46.
Patterson, I~E, Marshall, J.C. and Coltheart, M. (1985).Surface dyslexia: Neuropsychological
and cognitive studies of phonological reading. London: Lawrence Erlbaum Associates,
Ltd.
Saffran, E.M. (1985). Acquired dyslexia: implications for models of reading. In G.E.
MacKinnon and T.G. Waller (Eds.), Reading Research: Advances in Theory and Practice,
Vol.4.Orlando: Academic Press.
Schwartz, M.F. (1988). Modular Deficits in Alzheimer-Type Dementia. Cambridge,
Massachusetts: The MIT Press.
Schwartz, M.F., Saffran, E.M. and Matin, O.S. (1980). Fractionating the reading process in
dementia: evidence for word-specific print-to-sound associations. In M. Coltheart and
I~Patterson (Eds.), Deep Dyslexia. London: Routledge & Kegan Paul.
Shallice, T. and Warrington, E. (1980). Single and multiple component central dyslexic
syndromes. In M. Coltheart and I~ Patterson (Eds.), Deep Dyslexia. London: Routledge &
Kegan Paul.
Vallar, G. (1991). Current methodological issues in human neuropsychology. In F. Boller and
J. Grafinan (Eds.), Handbook ofNeuropsychology, Vol. 5, pp 343-378.
Stimulus E n c o d i n g in Alzheimer's Disease: A Multichannel View*
Grover C. Gilmore
Case Western Reserve University, Cleveland, OH 44106-7123
In recent years evidence has accrued to support the hypothesis that Alzheimer's disease
(AD) involves a major visual deficit in the detection of stimuli. This work has led us to
consider the impact which the visual deficit may have on higher order processing. Our
investigations have been conducted within the context of a multichannel model of visual
processing. Within this model it is relevant to consider both quantity and quality aspects of the
visual stimuli which are available to the observer. In the quantity domain we consider the
amplitude and phase of the spatial t~equencies which are present and the time course of the
activation of neural channels by these spatial frequencies. In the qualitative domain we
consider the nature of the information which the observer may extract from the available
spatial frequency information as it flows into the visual system Work will be presented here to
support the contention that the information processing performance of AD patients is
profoundly affected by their sensory deficit. Furthermore, it will be demonstrated that at least
a portion of the AD processing deficit may be ameliorated through a simple stimulus
enhancement procedure. The examination of the visual processing of AD patients provides a
model for the study of individual differences in stimulus encoding and the impact of such
differences on higher levels of information processing.
1. MULTICHANNEL MODEL OF VISUAL PROCESSING
There has been a marked change in the way we conceptualize the early stages of visual
information processing. While it had been common to consider the initial stage to be a passive,
static recording of visual stimulation (e.g. Neisser, 1967; Sperling, 1960), it has now become
reasonable to consider that the flow of visual information which arises in the nervous system is
continuous from the onset of the stimulus. The information needed to detect or identify
various aspects of a stimulus develops at different rates following stimulus onset. Rather than
consider the amount of information which is initially available, we now consider the time
course of the information flow with some data becoming available before others. There is
compelling psychophysical and neurophysical evidence to support this viewpoint (c.f.
Breitmeyer, 1984). There exist multiple, neurological channels that respond to narrow bands
of spatial frequencies (Campbell & Robson, 1968; Wilson, McFarlane, & Phillips, 1983). The
pattern of response characteristics of the channels, including the lag and duration of the
* Acknowledgements: I am indebted to the Alzheimer's disease patients and their caregivers

who have unselfishly given their time and insights to the research study. The research reported
here was supported by NIH grants AG04391, AG08012, AG11549.
200 G. C. Gilmore
response, interchannel inhibition, and phase sensitivity, dictate the development and formation
of visual representations (Sekuler & Blake, 1994).
Two major classes of channels have been described. The terminology which has been
used to name the channels has varied with the discipline and the era of the research.
Psychophycists have used the terms transient and sustained to capture the important
differences in the temporal response characteristics of the neural channels. Physiologists
labeled the channels as Y and X, respectively, while anatomists applied the terms magno and
parvo, respectively, to reflect the differences in the size of the cells. Bassi and Lehmkuhle
(1990) have presented an excellent review of the recent literature in these areas and have
suggested that the phrases M-cell and P-cell pathway be used to refer to the major classes of
cells. These terms will be used for simplicity throughout the remainder of the chapter.
For the purposes of the present discussion, it is important to note the following
differences in the characteristics of the two classes of cells in their response to achromatic
stimuli. The M-cells have relatively large receptive fields and respond best to low spatial
frequencies, while the P-cells have small receptive fields and are more responsive to high
spatial frequencies. It must be kept in mind that within each class there may be multiple neural
channels, each maximally sensitive to a different band of spatial frequencies (Wilson et al,
1983). The M-cells are also more sensitive to low contrast than are the P-cells. The latency
and duration of the neural response also differentiates the cells. M-cells fire on the average
about 15 msec faster than P-cells. The duration of the M-cell response is however transient,
while the P-cells sustain a response throughout the duration of a stimulus (Marroco, 1976).
Finally, the M-cells which are driven by abrupt changes in luminance are sensitive to
movement, while the P-cells respond best to static displays.
If one considers the functional significance of the differences in the cell classes, it is
clear that the P-cells are responsible for visual acuity under high luminance conditions with
static stimuli. The M-cell pathway will respond best to flashing or moving, low to moderate
contrast stimuli. When both classes of cells can respond to the same stimulus, the M-cell
pathway will have a temporal advantage. Considering that much visual information processing
research is conducted with moderate contrast stimuli that are tachistoscopically presented, it
may be conjectured that the M-cell pathway will signal the first information concerning the
stimulus. This early data in the visual system will represent the low spatial frequencies to
which the M-cell pathway is optimally sensitive. It may be argued that phenomena, such as
global precedence (Broadbent, 1977) and perceptual grouping (Rosen, 1994), may reflect the
initial response of the M-cell pathway to the low spatial frequencies of the stimulus display.
Furthermore, since these same low frequencies carry sufficient information to perform letter
(Parish & Sperling, 1991) and face (Peli, Goldstein, Young, Trempe, & Buzney, 1991)
identification tasks and to accurately read text (Legge, Pelli, Rubin, & Schleske, 1985), it is
reasonable that many visual tasks are influenced si~,nificantly by the responsivity of the M-cell
pathway.
2. MULTICHANNEL VISION DEFICITS IN ALZHEIMER'S DISEASE PATIENTS
Memory disorders are the hallmark of A17heimer's disease (AD). However, there is a
growing body of literature to support the hypothesis that A17heimer's disease (AD) also
involves a processing disorder which has its origin in the visual sensory pathways. While
investigators have noted for many years that AD involves serious visual perception
Stimulus encoding in Alzheimer's disease 201
disturbances (e.g. Williams, 1956), it was assumed that these problems were related to higher
cognitive problems and not to sensory deficits. Such an assumption was supported by a
number of facts. First of all, the acuity of AD patients is not notably different from that of
nondemented elderly adults (Katz & Rimmer, 1989). Also, and perhaps critically, the visual
cortex is relatively spared of neurofibrillary tangles (NFT) and plaques, which are the
definitive sign of the disease (Lewis, Campbell, Terry, & Morrison, 1987). Finally, positron
emission tomography of the brains of AD patients has shown that the primary visual cortex is
quite active relative to the low levels of activity in other cortical areas (Benson, 1982).
However, recent postmortem studies have illustrated that there are significant abnormalities in
the visual system of AD patients which may influence the apprehension of visual images.
2.1. Neuropathological Evidence
Hinton, Sadun, Blanks, and Miller (1986) reported that Alzhdmer patients had
widespread axonal degeneration in the optic nerves and a significant reduction in the number of
ganglion cells despite the fact that none of the patients had a history of ophthalmologic disease.
The observation of optic neuropathy (Sadun, 1989) and retinal ganglion call degeneration
related to AD and not to aging has been supported by several investigations (Blanks, Hinton,
Sadun, & Miller, 1989; Blanks, Torigoe, & Blanks, 1991; Blanks, Torigoe, Sper Gauderman,
& Blanks, 1990).
A recent report by Curcio and Drucker (1993), based on a study of only four AD eyes,
has argued that the ganglion cell loss may be attributable to age and not AD. The investigation
found no significant difference between age matched normal elderly adults and AD patients in
the number of ganglion cells. However, it is noteworthy in the latter study that, for three of
the four age matched pairs, the AD eyes did have fewer ganglion cells than the eyes of the
normal dderly. Only one AD patient, the youngest (67 years) in the sample, had more
ganglion cells than his age matched control. In such a small sample the inclusion of this one
pair introduced sufficient variability to negate a finding of a statistically si~ificant difference
between the groups. Thus, despite the conclusion of Curcio and Drucker (1993), it may be
argued that the majority of the AD patients in their study did strffer a loss of ganglion cells not
attributable solely to aging. An observation which is consistent with the above literature on
retinal cell loss in AD.
Hedges and Barbas (1990) reported that retinal abnormalities are detectable in AD
patients. They found that there were clinically detectable abnormalities in a retinal nerve fiber
layer of AD patients in the later stages of the disease. This examination of the living eye
combined with the weight of pathological evidence suggests that AD involves a si~ificant
optic disturbance. Gwen such an abnormality, Hinton et al. (1986) proposed that there may be
an as yet undescribed, characteristic pattern of visual function impairment related to the optic
neuropathy that may potentially aid in the identification of AD.
Hof and Morrison (1990) have demonstrated that, while there is much less NFT formation in
the occipital areas than in the prefrontal and temporal association areas, there is a significant
loss of cells in specific layers of areas 17 and 18. There was a si~ificant reduction of cells in
layer IVB of area 17 and in layer Hid of area 18, which are part of the M-cell pathway. The
cells of these layers have long corticocortical projections to the medial temporal area (MT or
V5). Thus, Hof and Mordson (1990) argue that AD patients may exhibit specific visual
202 G. C. Gilmore
problems linked to poor transmission of visual information to area MT, which is an area of the
cortex specialized for the perception of motion (Zeki et al., 1991).
The observation of a M-cell pathway deficit in AD has been made by other
investigators. Sadun (1989) suggested that the axonal degeneration existed predominantly in
the largest axons of the retina and that it could be observed early in the progression of the
disease. Blanks, Torigoe, and Blanks (1991) reported that, while there is significant cell loss in
both the large and small diameter ganglion cell layer neurons, there is a tendency for a greater
loss among the larger neurons.
It must be noted that there is an inherent difficulty in classifying cells on the basis of
size alone in neuropathological studies. Whitehouse, Hedreen, Jones, and Price (1983), in a
study of cell size in the nucleus basalis of Meynert in patients with AD, observed that a
shrinkage of all neurons may be interpreted inappropriately as a specific loss of large cells, if
size of cell was the sole criterion for classification. Thus, a cautious interpretation of the
majority of the existing neuropathological evidence is that there is a loss or shrinkage of both P
and M ganglion cells in the retina and in the primary cortical layers of the M-cell pathway.
2.2. Psychophysical Investigations
Studies with AD patients of contrast sensitivity and motion detection are consistent
with the neuropathological evidence. To appreciate this work one must note that M-cell and
P-cell channels respond optimally to different types of visual events (Basil & Lemhkule, 1990).
The M-cell channel neurons are optimally responsive to abruptly changing, low spatial
frequency stimuli. The P-cell channel neurons respond best to static, high spatial frequencies.
From the extensive loss of both P (small) and M (large) ganglion layer neurons, it would be
expected that spatial contrast sensitivity would be reduced for all spatial frequencies in AD
patients. Such a result is shown in Figure 1 for a sample of moderately demented AD patients
and an age-matched group of healthy elderly subjects (Gilmore, Thomas, Koss, & Townsend,
1994). This effect has been reported by a number of investigators (Cronin-Golomb et al.,
1991; CAlmore & Levy, 1991; Nissen et al., 1985) in studies of 9 to 25 AD patients. The one
exception in the literature was a study by Schlotterer, Moscovitch, and Crapper-McLachlan
(1984) which examined only six AD patients. Furthermore, in a recent longitudinal
investigation CAlmore and Whitehouse (in press) have demonstrated that the sensitivity of AD
patients to flickered, low frequency stimuli declines more rapidly over a one year period than
does their sensitivity to higher spatial frequencies. This last result is consistent with the
proposition that the M-cell channel neurons become dysfunctional more rapidly than P-cell
channel neurons. Thus, in studies of spatial contrast sensitivity with sufficient statistical
power, through large sample sizes or within-subject longitudinal designs, the evidence is clear
that AD patients experience a significant loss of contrast sensitivity in both the M-cell and the
P-cell channels with the rate of decline being possibly greater for the large cell M channel.
Studies of motion detection have appeared which strongly support the contention by
Hof and Morrison (1990) that AD patients have difficulty with information processed by area
MT. Using a correlated motion paradigm which has been developed to maximally stimulate
the M channel areas of 17 and MT (Newsome & Pare, 1988), Trick and Silverman (1991) and
Gilmore, Wenk, Naylor, and Koss (1994) have reported higher motion thresholds for AD
patients relative to healthy elderly adults. The weakness in responding to spatial changes over
time is likely to be related to the sluggish temporal integration reported by Coyne, Liss, and
3.0
7.5 Hz
"y 2 . 5
t--
oI
"15
t-
O
9 N o r m a l Elderly
o Alzheimer's Patients
~
0
1.0
0.5 . . I I . , I I, i I I . . . I ]
0.5 1 2 4 8
Spatial F r e q u e n c y (cpd)
Figure 1. Contrast sensitivity functions for normal elderly adults and moderately demented Alzheimer's
Disease patients. Spatialfrequencies were presentedat 7.5 Hz counterphase.
Geekler (1984) for AD patients in a visual masking study. The implication of these
psychophysical findings taken in conjunction with the neuropathology evidence of Hof and
Morrison (1990) is that AD patients suffer from a neural degeneration in the M-ceU channel
that disrupts their perception of abrupt temporal events, including simple motion.
RiT~o et al. (1992) recently have argued that the visual deficits of AD patients are due
primarily to dysfunction in the visual association cortices rather than from precortical damage
as hypothesized by Sadun and Bassi (1990). Their conclusion is based on an extensive clinical
neuro-ophthalmological examination of AD patients and healthy elderly adults. While the
patients in their study did have demonstrably lower spatial contrast sensitivity, they also had
normal critical flicker fusion thresholds, pattern visual evoked potentials, and full-field
electroretinograms. Rizzo et al. (1992, p. 98) concluded that there was "no convincing
evidence of damage to the retinocalcarine pathway associated with AD."
While there may not be consensus on the anatomical locus of the visual processing
deficit, the existence of a visual dysfunction, particularly in spatial contrast sensitivity, is well
documented. Furthermore, the pattern of strengths and weaknesses in visual psychophysical
tests reveals an important distinction in the performance of AD patients. AD patients function
at normal levels in tasks which use high contrast or very intense stimuli, such as critical flicker
fusion and evoked potential tasks. The spatial contrast tasks which employ low contrast
stimuli consistently yield poor performance from the patients. The AD visual deficit may
involve particularly those neural processes which respond to low contrast, high temporal
events, while high contrast processing is relatively spared. Thus, the AD patient may benefit
from the presentation of high contrast stimuli in visually loaded tasks, such as reading.
204 G. C. Gilmore
3. IMPACT OF MULTICHANNEL DEFICITS ON HIGHER ORDER PROCESSING
The demonstration of basic visual sensitivity deficits raises a question concerning the
impact of such deficiencies on higher order processing. It may be that the visual impairment
simply slows down the initial encoding and does not impact subsequent stages, as Steinberg
(1967) demonstrated with young observers. Or, the subject with reduced sensitivity may be
forced to deal with a degraded representation that affects the accuracy and efficiency of all
levels of processing. Indeed, Sekuler and Blake (1987) suggested that a "sensory underload"
produced by a prolonged reduction in stimulation may lead to si~ificant reductions in
cognitive activity.
It may be argued that poor contrast sensitivity creates weak visual signals which are
ineffectively processed by the higher order visual areas. Using this reasoning, Regan,
Raymond, Ginsburg, and Murray (1981) argued that a contrast sensing deficit could
masquerade as a higher order disability, such as object recognition. Work with low vision
patients, who typically suffer from a loss of moderate to high spatial frequency contrast
sensitivity (Peli & Peli, 1984), has demonstrated that contrast enhancement can significantly
improve reading (Lawton, 1989), and face recognition (Peli, Goldstein, Young, Trempe, &
Buzney, 1991) performance. Reading disabled children, who suffer a reduction in sensitivity to
low spatial frequencies (Lovegrove, Martin, & Bowling, 1982), also benefit from a
modification of contrast (Williams & LeCluyse, 1990). Recently, the limitations experienced
by elderly adults in reading small or large text has been linked to their relatively minor contrast
sensitivity deficits (Akutsu, Legge, Ross, & Schuebel, 1991).
Difficulty with reading is a common complaint of the AD patient. While patients are
able to read aloud accurately (Crawford, Hart, & Nelson, 1990; Cummings, Houlihan, & Hill,
1986), their reading speed (Nebes, Martin, & Horn, 1984) and comprehension (Cummings et
al., 1986) are poor. Because of the AD patients' good acuity and resistance to interference by
irrelevant cues, such as diagonal lines, it has been assumed that the poor reading performance
of the AD patients is due to a linguistic deficit rather than a visual perception problem
(Cummings et al. 1986). This reasoning is reminiscent of the theoretical stance held by a
number of dyslexia investigators (e.g. Vellutino, 1977).
A new look at the visual deficits of dyslexics has suggested that their reading disorder
may have its roots in the sluggish activity of a weak M-cell channel system (Livingstone,
Rosen, Drislane, & Galaburda, 1991). A similar argument may be made for AD patients who
also may have sluggish M-cell channels. We have suggested that the contrast sensitivity deficit
of the AD patient may have an impact on higher order visual tasks, such as picture naming and
reading (Gilmore, Turner, & Mendez, 1990). The rationale for this argument is that an
elevated threshold for spatial frequencies would result in some stimuli being processed more
slowly or not at all. To determine the direct impact of stimulus contrast on processing speed,
we have conducted several studies in which the contrast of the stimuli was manipulated
explicitly. The extent to which the contrast manipulation modifies or diminates the AD deficit
relative to healthy elderly controls is a measure of the effect of the AD patients' poor contrast
sensitivity on the task.
In the experiments described below the stimuli were created with a PC based image
engineering workstation, which permits the control of stimulus duration, contrast, and spatial
frequency content. The images were transferred to videotape and presented to subjects on a
portable video playback unit. Verbal reaction times were recorded to an accuracy of 1 msec by
synchronizing a tone on the tape with the presentation of the stimulus. The tone started the
reaction time clock and the subject's response stopped the clock. This apparatus has made it
possible to test patients in their home. By following this testing procedure, we place a
minimum burden on the patient and his/her caregiver while we collect reaction time data with
rigorously controlled stimuli.
3.1. Picture Naming
The first study was a dissertation conducted by Turner (1990) in the Perception Lab. It
examined the impact of visual signal strength on picture naming and recognition. Both AD
patients (11=19) and an age-matched group of healthy adults (n=19) were tested. It was
hypothesized that the naming deficit of AD patients can be linked directly to the strength of the
spatial frequency signal in the visual displays. Line drawings of pictures were either enhanced
or degraded by transforming the shape of the magnitude spectrum of the stimuli. The degree
of change was determined by the difference in spatial contrast sensitivity levels documented for
AD patients and healthy elderly adults in Cfilmore, Turner, and Mendez (1990). This is a form
of adaptive filtering.
Patients who have low sensitivity, it was argued, do poorly on these tests because they
do not have sufficient signal present in their systems. Thus, by enhancing the amplitude of the
spatial frequencies we expected to improve the performance of the patients in the naming task.
The major hypothesis was supported partially. As shown in Table 1, the amplitude of the
magnitude spectrum (roughly the contrast of the image) was related to naming latency for high
frequency, that is, common words. An enhanced image led to faster naming times for the AD
patients and degradation resulted in slower times relative to the normal stimulus. This effect
was not present for low frequency, uncommon, words. Thus, the speed with which an
available name (high frequency word) can be accessed was linked to the strength of the visual
signal. This manipulation of the performance level of the AD patients was the first
demonstration that simple visual factors have a direct impact on the performance of a higher
order (picture naming) task.
Table 1.
Picture Naming Reaction Time (msec) For Normal Elderly Adults and Alzheimer's Disease
Patients Presented With Pictures With High Frequency (Common) Or Low Frequency Names
Under Three Levels of Contrast.
Normal Elderly Alzheimer's Disease

Adults Patients
Contrast High Frequency Names
Normal 1305 2259
Enhanced 1254 2048
Degraded 1378 2392
Low Frequency Names
Normal 1368 2393
Enhanced 1427 2651
Degraded 1621 2550
206 G. C. Gilmore
3.2. Letter Naming
The role of contrast and spatial frequencies in the processing of complex stimuli was
investigated fttrther by examining letter identification time (Gilmore, Thomas, K~z, Persanyi,
& Tomsak, 1994). This study was a logical extension of the Turner (1990) study of picture
naming and contrast enhancement. Subjects were shown letters for only 250 msec and were
asked to verbally identify the letters. Reaction time was recorded to an accuracy of 1 reset.
900
Low
800
700
(SO0
500 "I"I I
O
9 400
I I I i I i
E
9OO
Medium
E
9- - 800
i--
t-
O 7OO
O 600
O
-*-' 500
t-
O
"O 400
"+" 900
G)
.._l
High
8OO r, Alzheimer Patients
v Elderly Adults
o Young Adults
700
600
500
400
.9 1.1 1.3 1.6 1.9 No

Filter
Low Pass Filter Cutoff (cplw)
Figure 2. Time to identify single letters by Alzheimer's disease patients, healthy elderly, and young adults
under three conditions of contrast and at five low pass filters. The filters pass the spatial frequencies below the
cycles per letter width (cplw) cutoff. The time to identify letters which were not filtered in each contrast
condition are shown for each subject group. Standard error bars are shown for values greater than 25 msec.
Three groups of subjects were tested, AD patients (n=10), healthy elderly adults (n=10), and
young adults (n=13). The stimuli were shown at a very high contrast (99%) and at two lower
contrasts of 88% and 94%. Pilot testing had shown that AD patients could perform well with
letters that were .48 inches in width and subtended a visual angle of .61 cpd. The spatial
frequency content of the letters was also manipulated with five low pass filters ranging from .9
to 1.9 cycles per letter width (cplw). This range of filters was chosen because good readers
can read text well with only 1.5-2 cplw (Legge, Pelli, Rubin, & Schleske, 1985; Parish &
Sperling, 1991).
In Figure 2 the time for each subject group to read the letters at each contrast and low
pass filter is shown. The most important finding is that under the very high contrast condition
(99%) the AD patients performed identically to the healthy elderly adults. That is, given a
sufficiently strong visual signal the AD patients were able to name the letter in the same time as
the healthy comparison group.
The effect of the low pass filters can be evaluated by comparing the time to identify the
letters in the No Filter condition with the filter conditions. The noteworthy finding is that all of
the subject groups were able to read filtered letters at 1.6 or 1.9 cplw as fast as the intact
letters. This is consistent with the literature on this subject for good readers. Importantly, it
demonstrates that the AD patients, given visual signals of sufficient strength, can perform
comparably to healthy adults.
We believe that this set of findings is quite important for it clearly demonstrates that the AD
patients have the capability to perform simple visual tasks very well. Since AD patients report
reading difficulties with normal text, which has a contrast between 60% and 70%, our results
suggest that AD patients may benefit from a visual intervention strategy which presents reading
material under high contrast conditions.
3.3. Word Reading
A study has been completed which follows the very promising results from the letter
naming study (Gilmore, Groth, & Thomas, 1993). As in the latter study, AD patients,
nondemented elderly adults, and young adults served as subjects. The stimuli were words.
The major manipulations were the contrast of the words and the repetition of a subset of the
stimuli.
In the first experiment each subject viewed the words under three contrast conditions: 69%,
84%, and 98% contrast. In this experiment three groups of subjects were tested, AD patients
(n = 12), healthy elderly adults (n - 12), and young adults (n = 12). Subjects were shown
blocks of words at the three contrast levels for 250 msec and were asked to verbally identify
the words. The main findings from this experiment were that the contrast manipulation had
very little impact on the time for the young observers to read the words, however, the healthy
elderly and the mildly demented AD patients yielded their fastest reaction times in the highest
contrast condition. Furthermore, a subgroup of mildly demented AD patients identified the
words as quickly as the nondemented elderly adults.
In the second experiment the contrast manipulation was expanded to include five
different contrast levels, three oft he levels used in the first experiment (69%, 84%, 98%) and
two lower levels (28%, 51%). Rather than blocking by contrast, as was done in the previous
e~eriment, the words at the five different contrast levels were intermixed. Also, a word
repetition manipulation was added. Each contrast condition involved the presentation of 20
words for identification. Ten of the words in each condition were novel; they were presented
only once in the experiment. Ten specific words were repeated in each of the contrast
conditions; these words were seen a total of five times across the contrast conditions. Finally,
a baseline reaction time measure was collected for each subject.
208 G. C. Gilmore
Repeat Novel
[] Alzheimer Patients, Mild []
9 Healthy Elderly Adults v
9 Young Adults, Filtered 0
9 Young Adults 0
950 ~-~ Repeated Words.

0
m 900
E
850 9T~ = T T T
,
.E
_
I--
800
750
.,~
700
650
600
950 -~ t Novel Words

%-
m 900
E
~ g50
E
o--
I--
go0
I
._.
9"o 750
i"""""" T ....__ T
n,,"
-o 700
t_
0
650
600 I I I I I
28 51 69 84 98
Contrast (%)
Figure 3. Word reading time (ms) and standard error for each subject group in the repeated (A) and novel (B)
word conditions and at the five contrast levels in Experiment 2 of the word reading study. The Alzheimer's
Disease patients were not able to respond to the stimuli presentedin the 28% contrast condition.
Four groups of subjects were tested: AD patients (n = 10), healthy elderly adults (n =
12), young adults (n = 12), and young adults who viewed the stimuli through a neutral density
filter to simulate the reduction in retinal illuminance suffered by elderly adults (n = 12).
As in the first word reading experiment, the contrast manipulation had an effect. At the
three original contrast levels there was no effect on the word reading time of the young adults.
At the lowest contrast level, however, the young adults' word reading time did increase. The
healthy elderly and AD patients showed an increase in word reading time with decreased
contrast. Only four of the 10 AD patients were able to respond to any of the words in the
lowest contrast condition. Two of those four who were able to respond to the lowest contrast
words only responded to those words that were repeated.
As Figure 3 shows, the repetition of the words moderated the contrast effect for all
subject groups. Each subject group was able to read repeated words more rapidly than novel
words but this repetition effect emerged only at lower stimulus contrast levels. These findings
indicate that the repetition of the stimuli produces an immediate familiarity or lexical priming
effect which facilitates the identification of a repeated stimulus. The interaction between
contrast level and stimulus repetition suggests that the locus of the familiarity effect may be at
the level of stimulus encoding. The fact that the AD subjects yielded a strong familiarity effect
demonstrates that their processing components are generally intact at this early stage of
information processing.
A behavioral consequence of the AD patients' poor contrast sensitivity was evident in
their inability to read words at the 28% contrast level. As in Gilmore, Thomas, Klitz, Persanyi,
and Tomsak (1994), the reading time of the AD patients improved with increases in contrast
levels, asymptoting with repeated words at 51% and with novel words at 84% contrast. This
finding demonstrates that a demented group of subjects can improve their reading speed
through a simple visual intervention.
The interaction reported here between contrast level and stimulus repetition supports
earlier proposals that repetition affects the encoding of weak stimuli (den Heyer & Benson,
1988; Norris, 1974). All subjects were able to use the familiarity created by the repetition of
the words to overcome the disadvantage of the lower contrast stimuli. This observation
suggests that the encoding of a familiar (repeated) stimulus can be accomplished with less
distal stimulus information.
The differential effects of stimulus contrast and repetition on the subject groups suggest
encoding speed differences. In the verification model of word processing, variation of stimulus
quality affects feature extraction (Becker, 1976). It is apparent that the older subjects, who
were more affected by the contrast reductions, had slower rates of feature extraction. A
portion of this age related difference may be attributable to sensory factors, such as reduced
retinal illuminance, since the young subjects who viewed the stimuli through a filter had
elevated reading times in the lowest contrast condition.
The AD patients yielded a pattern of performance very similar to the healthy elderly
adults. When differences in baseline reaction time levels are taken into account, it can be
argued that the AD patients had a feature extraction rate that was comparable to healthy
elderly adults at stimulus contrasts of 69% and higher. The lower contrast levels in this
experiment apparently approached or exceeded the thresholds of the patients and severely
disrupted their encoding. It is for this reason that it is noteworthy that the AD patients were
able to perform so well with the repeated words in the 51% condition. These findings suggest
that the automatic encoding processes involved in word recognition remain intact in mildly
demented AD patients and function normally given stimuli of sutticient strength.
4. SIMULATION OF ALZHEIMER-LIKE CONTRAST SENSITIVITY DEFICIT IN

HEALTHY ADULTS
The above studies of letter and word reading have demonstrated that the contrast of the
stimulus can have a marked impact on the performance of all subjects. Furthermore, evidence
was presented to suggest that AD subjects may perform at the levels of healthy elderly adults
when given visual signals of sufficient strength. The extent to which the AD subjects' deficit in
210 G. C. Gilmore
visual processing tasks may be linked to their contrast sensitivity deficiency may be fttrther
examined by simulating the AD contrast sensitivity deficit in healthy adults.
The picture naming speed and accuracy of healthy young and elderly adults was
evaluated in a study as part of a larger investigation of the visual processing capabilities of AD
patients (Cfilmore, Thomas, Koss, & Townsend, 1994). The goal of the study was to simulate
in healthy adults the picture naming behavior of the AD patients by exposing the healthy
subjects to contrast degraded stimuli. The extent to which the Alzheimer simulation affected
the behavior of the healthy subjects would illustrate the potential impact of the AD patients'
contrast sensitivity deficit on a higher order processing task, picture naming.
In confrontation naming tasks, the subject is presented with a stimulus and asked to
provide its name. AD patients perform quite poorly at such tasks (Appel, Kertesz, & Fisman,
1982). There has been considerable debate in the literature as to the relative contribution of
semantic versus perceptual disturbance in producing naming errors. Analysis of the linguistic
content of the naming errors has led a number of investigators to conclude that semantic
processing problems are the major source of the deficit (e.g. Bayles & Kaszniak, 1987;
Bowles, Obler, & Albert, 1987). However, there is evidence that AD patients are quite
sensitive to the perceptual quality of a stimulus. The closer the representation of a stimulus is
to its real world attributes, the better the recognition performance of AD patients. For
example, AD patients do well with real three-dimensional objects and colored pictures but have
difficulty with line drawings (Bisiach, 1966; Kirshner, Webb, & Kelly, 1984).
The literature on confrontation naming was conducted under the then prevailing
assumption that AD did not involve a primary visual deficit. Recent empirical evidence
reviewed above has called that assumption into question. Since AD patients require a stronger
signal to detect an object, it may be argued that the presentation of normal contrast stimuli
results in a weak proximal representation in the visual system of the AD patients. That is, it is
suggested that one source of the naming problems exhibited by AD patients is the weak signal
carried by their visual system.
The extent to which a weak stimulus is responsible for naming errors can be evaluated
by presenting contrast degraded stimuli to nondemented adults. In the present study the
degree of degradation was dictated by the empirically observed contrast sensitivity deficit of
AD patients. It was expected that the degradation of the stimuli to contrast levels purportedly
experienced by AD patients would lead the healthy adults to commit a significantly higher
number of errors. This study is the type of simulation study suggested by Lindenberger and
Baltes (1994) to examine the impact of sensory deficits on higher-order processing.
Enhanced stimuli also were presented to the healthy subjects. If the strength of the normal
stimulus was mdticient for optimum identification speed and accuracy, then the enhancement
should have no impact on performance. However, enhanced contrast may assist subjects with
weaker contrast sensitivity profiles.
Three sets of 32 pictures were drawn from a set developed by Snodgrass and
Vanderwart (1980). The sets were matched for familiarity, name and image agreement, name
frequency, and visual complexity. Three sets of foils for the recognition task were also chosen
from the same database of pictures. A previous study reported equivalent naming times for the
three sets when presented under normal contrast (Turner, 1990).
In the present study the stimulus sets were presented at one of three contrasts: Normal,
Enhanced, or Degraded. The alteration of the contrasts was based on the contrast sensitivity
functions (CSF) measured for normal elderly adults and moderately demented AD patients in a
Figure 4. Sample stimuli in the normal, enhanced, and degraded contrast conditions. The Alzheimer Filter,
based on the contrast sensitivity differences between the subject groups depicted in Figure 1, was used to
transform the contrast of the images.
longitudinal study (Gilmore & Levy, 1991, Gilmore & Whitehouse, in press).
As shown in Figure 1, the AD subjects yielded an overall contrast sensitivity deficit of
.4 log units relative to the healthy sample of subjects. The CSF data from each group were fit
to a fourth-order model, using a five parameter, least-squares fitting algorithm described
previously (Thomas, ~ o r e , & Royer, 1993). Given the model parameters for the two CSF
curves, filters were constructed to compensate for the differences between the CSF functions.
The compensating filters were designed using the ratio of the fit to the normal elderly and the
AD subjects' functions. The filter was then implemented by multiplying the Fourier transform
of a stimulus image and the two-dimensional filter. Finally, the product was inverse
transformed to obtain the filtered image.
The filter, hereafter referred to as the A17:heimer Filter, was applied to the magnitude
spectrum of each image. For images to be presented in the Degraded condition, the spectrtma
of the image was diminished to fit the function representing the CSF of the AD group. For
212 G. C. Gilmore
Enhanced images, the magnitude spectrum was increased to fit the CSF for the nondemented
elderly group.
Sample stimuli are presented in Figure 4. For ease of exposition the same stimulus is
presented here under each of the three Contrast conditions. In the study, the subjects were
exposed to different stimuli in each Contrast condition. Because the original stimuli were line
drawings, the digitized images contained a range of gray levels defining the figure. This range,
of course, was affected by the application of the Alzheimer Filter. In order to evaluate the
contrast of the individual stimuli, the luminance profile of the figure and ground were
determined separately and the average luminance was calculated. Contrast was then defined as
the difference in luminances between the figure and ground over the sum of the figure and
ground luminances. This procedure takes into account the range of luminances present in each
figure. For example the sample stimulus in Figure 4 yielded a contrast of 36% in the Normal
condition, and 99% and 10% in the Enhanced and Degraded conditions, respectively.
The Alzheimer Filter degradation had a major impact on the confrontation naming
performance of both the young and elderly adults. As shown in Table 2, both reaction time,
F(2,68)=68.02, p<.0001, and accuracy of naming, F(2,68)=46.90, p<.0001, were affected by
the contrast degradation. While there was a trend, particularly in the elderly adults, for an
improvement in performance in the Enhanced condition, this effect was not si,~mificant.
Table 2.
Reaction Time (RT, msec) and Accuracy (%) Scores for Healthy Young and Elderly Adults in
the Picture Naming Task. Subjects Viewed Stimuli Which Were Either Normal In Contrast Or
Enhanced Or Degraded.
Young Elderly
Adults Adults
Contrast RT % RT %
Normal 616 93 839 88
Enhanced 618 97 806 94
Degraded 839 84 1273 69
As shown in Table 3, the time taken to recognize a stimulus as "old" also was affected
by the contrast degradation, F(2,68)=159.65, p<.0001. The accuracy of recognition memory
was quite high, except for the elderly adults viewing the Degraded stimuli, F(2,68)=9.06,
p=.0003. While there was a trend for the subjects to recognize enhanced stimuli more quickly,
accuracy of memory was not aided by enhancement.
Table 3.
Reaction Time (RT, msec) And Accuracy (%) Scores For Healthy Young And Elderly Adults
In The Picture Recognition Task. Subjects Viewed Stimuli Which Were Either Normal In
Contrast Or Enhanced Or Degraded.
Young Elderly
Adults Adults
Contrast RT % RT %
Normal 701 99 1009 99
Enhanced 660 98 991 99
Degraded 842 98 1423 87
The purpose of the study was to demonstrate that a contrast reduction comparable to
that sttffered by Alzheimer's disease (AD) patients can lead to a si,~nificant irnpairment in a
confrontation naming task. This goal was achieved in that both healthy young and elderly
adults exhibited marked decrements in both naming time and accuracy when confronted with
degraded stimuli. Furthermore, the memory of the elderly adults also was hampered by the
Alzheimer degraded stimuli. Thus, the results imply that the contrast sensitivity loss of
moderately demented AD patients may play an important role in limiting their performance on
higher order visual processing tasks.
This conclusion is supported by our ongoing investigation of the impact of contrast
enhancement on the naming performance of AD subjects. AD subjects are being presented
with the Normal and Enhanced contrast stimuli of the present study. All of the AD subjects to
date have benefited in their naming accuracy from the use of the enhanced stimuli. Thus, the
application of an adaptive filter to compensate for the sensory deficit of the AD patients may
be an effective tool for improving their visual information processing.
There are several irnplications of the AD simulation study. The first is that the contrast
sensitivity loss experienced by AD patients is important in that it can be linked to a reduction in
the ability to recognize objects. Secondly, the elementary visual deficit may exacerbate the
cognitive problems of the AD patients. Without a strong visual signal, the patients may suffer
a sensory tmderload which limits their interaction with the environment (Sekuler & Blake,
1987). Finally, the findings lead to the suggestion that the application of a compensating filter
may alleviate the encoding burden of the AD patients and thereby irnprove not only their
apprehension of information but also their comprehension and memory of material.
5. CONCLUSIONS
The studies presented above were stimulated by the observation that Alzheimer's
disease patients experience a marked decline in contrast sensitivity. The decline was attributed
to deficiencies of signal processing especially within the M-cell channels. It was speculated
that such an elementary deficit may be linked to the poor performance of the patients on tasks
which involve the visual presentation of stimuli. To evaluate the hypothesis, several pattern
identification experiments were conducted which involved the manipulation of stimulus
contrast. The empirical observations have demonstrated that the performance of AD patients
214 G. C. Gilmore
can be improved by presenting stimuli under very high contrast conditions. Furthermore,
healthy adults presented with stimuli degraded to levels, which simulate the visual condition of
the AD patient, exhibit marked performance decrements. Such findings validate the conclusion
that the contrast sensitivity deficit associated with AD contributes significantly to the patients'
difficulty in interacting with and learning from their visual world.
Generalizations from the latter studies are made cautiously given that isolated letters,
pictures, and words were the reading material and the primary dependent measure was speed.
While in the low vision literature, it has been shown that reading speed is related to
comprehension (Legge, Ross, Luebker, & LaMay, 1989), no such claim may be made in the
AD literature. Indeed, given the severe cognitive disturbances associated with the illness, it
may seem doubtful that an increase in reading speed may bring about a higher level of
comprehension.
This concern may be addressed by noting that AD patients do have intact cognitive
processes that may be underused because of insufficient visual stimulation. Moscovitch,
Wincour, and McLachlan (1986) have determined that AD patients can be shown on tests of
implicit memory, such as speeded reading, to have formed new associations even though their
performance on an explicit test, such as recognition memory, is poor. Grosse, Wilson, and Fox
(1990) have demonstrated that when the encoding of the material is maximized, AD patients
can exhibit normal implicit memory. Furthermore, Nebes (1989) in an extensive review of the
literature has reported that some aspects of semantic memory are spared in AD. Turner
(1990), as reviewed above, demonstrated that the time to generate a high frequency name for a
picture can be improved in AD patients through a modest increase in stimulus contrast. That
is, the time to access a relatively close name in the semantic network was related to the
strength of the visual signal. Thus, it seems reasonable that by improving the quality of stimuli
through contrast enhancement, the patient will be better able to encode and process stimuli at
higher cognitive levels.
To argue that the AD patients' visual deficits may influence their higher order
information processing is consistent with recent theories of information processing and
recognition memory. Murdock (1989) and Theios and Amrhein (1989 a, b) have incorporated
the quality of stimulus encoding in their recognition memory and picture-word processing
models, respectively. Elsewhere in this volume, Amrhein further discusses the impact of
stimulus quality. It is clear that manipulations of physical characteristics of the stimulus, such
as its size or spatial frequency content (Theios & Amrhein, 1989a), or its duration (Lofts,
1974; Yonelinas, Hockley, & Murdock, 1992), can have direct impacts on the processing
speed and accuracy of young subjects whose visual systems are intact. The degradation of a
stimulus may have both a main effect on performance by slowing sensory acquisition and
encoding processes and also an interactive effect with higher order decision and processing
stages (den Heyer & Benson, 1988; Norris, 1984). It is suggested here that if manipulations of
the distal stimuli can influence information processing, then alterations of the proximal stimulus
by a deficient visual system can create similar processing burdens. The marked encoding
deficits of AD subjects in memory tasks (c.f. Nebes, 1992) may be linked to their poor contrast
sensitivities. The work reviewed above on word identification (CAlmore, Groth, and Thomas,
1993) suggests that a portion of the AD patients' encoding deficit which may be remediated
through contrast enhancement is their slow and inaccurate reading of weak visual stimuli.
Specifically, it is argued that differences among individuals in their sensitivity to spatial
frequencies critical for task performance will be reflected in their information processing speed
and accuracy.
The range of research studies conducted in our lab were stimulated by the conceptual
structure of the multichannel model of vision. Conceiving of the visual information processor
as one whose response is driven by the strength of spatial frequency components arising from
the continuous flow of stimulus information has given rise to our questions concerning the
processing behavior of AD patients. It is argued that the weak response of the visual system
can have a ripple effect on higher order processing, conveying poorly encoded representations
for higher order processing. While we are not suggesting that the profound cognitive
disturbances associated with AD are caused by the sensory deficits, we are conjecturing that a
portion of the AD patients' behavioral problems are at least exacerbated by the weak visual
signals. To the extent that a stimulus enhancement procedure ameliorates the behavioral
deficit of the AD patient, we may be able to improve the daily living activities of the patients.
Future work is needed to extend the findings reported here to domains outside of the
laboratory to determine if AD patients can benefit from the enhancement of visual stimuli in
their everyday environment. For example, the use of large print, high contrast material may
enable the AD patients to read more easily. Also, the creation of a high contrast living
environment may permit the demented patient to move about and interact more effectively
(Calkins, 1988). The payoff of this research approach is the ".Improvement of the quality of life
of the AD patient.
REFERENCES
Akutsu, H., Legge, G. E., Ross, J. A., & Schuebel, I~ J. (1991). Psychophysics of reading. X.
Effects of age-related changes in vision. Journal of Gerontology: Psychological Sciences,
46, P325-331.
Appel, J., Kertesz, A., & Fisman, M. (1982). A study of language functioning in Alzheimer
patients. Brain and Language, 17, 73-91.
Bassi, C. J., & Lehmkuhle, S. (1990). Clinical implications of parallel visual pathways. Journal
of the American Optometric Association, 61, 98-110.
Bayles, I~ A., & Kaszniak, A. W. (1987). Communication and cognition m normal aging and
dementia. New York, NY: Little, Brown, & Co.
Becker, C. A. (1976). Allocation of attention during visual word recognition. Journal of
Benson, D. F. (1982). The use of positron emission scanning techniques in the diagnosis of
Alzheimer's disease. In S. Corkin, I~ Davis, J. H. Growdon, E. Usdin, & 1L J. Wurtman
(Eds.), Alzheimer's disease: A report of progress m research (pp 79-82). New York:
Raven Press.
Bisiach, E. (1966). Perceptual factors in the pathogenesis of anomia. Cortex, 2, 90-95.
Blanks, J. C., Hinton, D. 1L, Sadun, A. A., & Miller, C. A. (1989). Retinal ganglion cell
degeneration in Alzheimer's disease. Brain Research, 501, 364-372.
Blanks, J. C., Torigoe, Y., & Blanks, 1L H. I. (1991). Ganglion cell loss in retinal whole-
mounts from patients with Alzheimer's disease. Investigative Ophthalmology & Visual
Science (Supplement), 32, 1230.
216 G.C. Gilmore
Blanks, J. C., Tofigoe, Y., Spee, C., Gauderman, W. J., & Blanks, 1L H. I. (1990). Ganglion
cell loss in macula of patients with Alzheimeds disease. Investigative Ophthalmology and
Visual Science (Supplement), 31, 356.
Bowles, N. L, Obler, L. K., & Albert, M. (1987). Naming errors in healthy aging and dementia
of the Alzheimer type. Cortex, 23, 519-524.
Breitmeyer, B. G. (1984). Visual Masking. Oxford, England: Oxford University Press.
Broadbent, D. E. (1977). The hidden preattentive process. American Psychologist, 32, 109-
118.
Calkins, M. P. (1988). Design for dementia: Planning environments for the elderly and the
confused. Owings Mills, MD: National Health Publishing.
Campbell, F. W., & Robson, J. G. (1968). Application of Fourier analysis to the visibility of
gratings. Journal of Physiology, 197, 551-566.
Coyne, A. C., Liss, L., & Geckler, C. (1984). The relationship between cognitive status and
information processing. Journal of Gerontology, 39, 711-717.
Crawford, J. R., Hart, S., & Nelson, H. E. (1990). Improved detection of cognitive impairment
with the NART: An investigation employing hierarchical discriminant function analysis.
British Journal of Clinical Psychology, 29, 239-241.
Cronin-Golomb, A., Corkin, S., Rizzo, J. F., Cohen, J., Growdon, J. H., & Banks, I~ S.
(1991). Visual dysfunction in Al~eimer's disease: Relation to normal aging. Annals of
Neurology, 29, 41- 52.
Cummings, J. L., Houlihan, J. P., & Hill, M. A. (1986). The pattern of reading deterioration in
dementia of the Al~eimer type: Observations and implications. Brain and Language, 29,
315-323.
Curcio, C. A., & Drucker, D. N. (1993). Retinal ganglion cells in Alzheimer's disease and
aging. Annals of Neurology, 33, 248-257.
den Heyer, I~, & Benson, K. (1988). Constraints on the additive relationship between semantic
priming and word recognition and on the interactive relationship between semantic priming
and stimulus clarity. Canadian Journal of Psychology, 42, 399-413.
Cfilmore, G. C., & Levy, J. (1991). Spatial contrast sensitivity in Alzheimer's Disease: A
comparison of two methods. Optometry and Vision Science, 68, 790-794.
Cfilmore, G. C., Groth, K. E., & Thomas, C. W. (1993). Stimulus contrast and word reading
speed in Alzheimer's disease. Manuscript submitted for publication.
Gilmore, G. C., Thomas, C. W., Klitz, T., Persanyi, M., & Tomsak, R. (1994). Contrast
enhancement eliminates letter identification speed deficits in Alzheimer's disease.
Manuscript submitted for publication.
Gilmore, G. C., Thomas, C. W., Koss, E., & Townsend, L. (1994). Impact of Alzheimer-type
contrast filter on picture naming and memory in healthy adults. Poster presented at the
annual meeting of the Optical Society of America, Dallas.
Gilmore, G. C., Turner, J., & Mendez, M. (1990). Contrast sensitivity and Alzheimer's disease
(Tech. Rep. No. PL90-4). Cleveland, OH: Case Western Reserve University, Perception
Lab.
Gilmore, G. C., Wenk, H., Naylor, L., & Koss, E. (1994). Motion perception and Alzheimer's
disease. Journal of Gerontology: Psychological Sciences, 49, P52-P57.
Cfilmore, G. C., Wenk, H., Naylor, L., & Stuve, T. (1992). Motion perception and aging.
Psychology anclAging, 7, 654-660.
C_filmore, G. C., & Whitehouse, P. (1995). Contrast sensitivity in Alzheimer's disease: A one
year longitudinal analysis. Optometry and Vision Science., 72, 83-91.
Grosse, D. A., Wilson, 1k S., & Fox, J. H. (1990). Preserved word-stem-completion priming
of semantically encoded information in Alzheimer's disease. Psychology and Aging, 5, 304-
306.
Hedges, T. 1L m, & Barbas, N. 1L (1990). Clinical evaluation of the retinal nerve fiber layer in
AlzJaeimer's disease. Investigative Ophthalmology and Visual Science (SupplemenO, 31,
356.
Hinton, D. 1L, Sadun, A. A., Blanks, J. C., & Miller, C. (1986). Optic-nerve degeneration in
Alzheimefs disease. The New England Journal of Medicine, 315, 485-487.
Hof~ P. 1L, & Morrison, J. H. (1990). Quantitative analysis of a vulnerable subset of
pyramidal neurons in Alzheimer's disease: II. Primary and secondary visual cortex. The
Journal of Comparative Neurology, 301, 55 - 64.
Katz, B., & Rimmer, S. (1989). Ophthalmologic manifestations of Alzheimer's disease. Survey
of Ophthalmology, 34, 31-43.
Kirshner, H. S., Webb, W. G., & Kelly, M. P. (1984). The naming disorder of dementia.
Neuropsychologia, 22, 22-30.
Lawton, T. B. (1989). Improved reading performance using individualized compensation filters
for observers with losses in central vision. Ophthalmology, 96, 115-126.
Legge, G. E., Pelli, D. G., Rubin, G. S., Schleske, M. M. (1985). Psychophysics of reading. I.
Normal vision. Vision Research, 25, 239-252.
Legge, G. E., Ross, J. A., Luebker, A., & LaMay, J. M. (1989). Psychophysics of reading.
VIII. The Minnesota Low-Vision Reading Test. Optometry and Vision Science, 66, 843-
853.
Lewis, D. A., Campbell, M. J., Terry, 1L D., & Morrison, J. H. (1987). Laminar and regional
distributions of neurofibrillary tangles and neuritic plaques in Alzheimefs disease: A
quantitative study of visual and auditory cortices. The Journal of Neuorscience, 7, 1799 -
1808.
Lindenberger, U., & BaRes, P. B. (1994). Sensory functioning and intelligence in old age: A
strong connection. Psychology and Aging, 9, 339-355.
Livingstone, M. S., Rosen, G. D., Drislane, F. W., & Galaburda, A. M. (1991). Physiological
and anatomical evidence for a magnocellular defect in developmental dyslexia. Proceedings
of the National Academy of Science USA, 88, 7943-7947.
Loflus, G. Ik (1974), Acquisition of information l~om rapidly presented verbal and nonverbal
stimuE Memory & Cognition, 2, 545-548.
Lovegrove, W., Martin, F., Bowling, A. (1982). Contrast sensitivity functions and specific
reading disability. Neuropsychologia, 20, 309-315.
Marroco, lk T. (1976). Sustained and transient cells in monkey lateral geniculate nucleus:
Conduction velocities and response properties. Journal of Neurophysiology, 39, 340-353.
Moscovitch, M., Winocur, G., & McLachlan, D. (1986). Memory as assessed by recognition
and reading time in normal and memory-impaired people with Alzheimer's disease and
other neurological disorders. Journal of Experimental Psychology: General, 115, 331-347.
Murdock, B. B. (1989). Learning in a distributed memory model. In C. lzawa (Ed.), Current
issues in cognitive processes: The Tulane Floweree symposium on cognition (pp. 69-106).
Hillsdale, N. J.: L. Erlbaunl
218 G.C. Gilmore
Nebes, R. D. (1989). Semantic memory in Alzheimer's disease. Psychology Bulletin, 106, 377-
394.
Nebes, R. D. (1992). Cognitive dysfunction in Alzheimer's disease. In F. I. M. Craik and T. A.
Salthouse (Eds.), The handbook of aging and cognition (pp. 373-446). Hillsdale, N. J.: L.
Erlbaum,
Nebes, R. D., Martin, D. C., & Horn, L. C. (1984). Sparing of semantic memory in
Alzheimer's disease. Journal of Abnormal Psychology, 93, 321-330.
Neisser, U. (1967). Cognitive Psychology. New York, NY: Appleton-Century-Crofts.
Newsome, W. T., & Pare, E. B. (1988). A selective impairment of motion perception
following lesions of the middle temporal visual area. The Journal of Neuroseience, 8,
2201-2211.
Nissen, M. J., Corkin, S., Buonanno, F. S., Growdon, J. H., Wray, S. H., & Bauer, J. (1985).
Spatial vision in Alzheimer's disease. Archives of Neurology, 42, 667-671.
Norris, D. (1984). The effects of ~equeney, repetition and stimulus quality in visual word
recognition. Quarterly Journal of Experimental Psychology, 36A, 507-518.
Parish, D. A., & Sperling, G. (1991). Object spatial fxequencies, retinal spatial l~equencies,
noise, and the efficiency of letter discrimination. Vision Research, 31, 1399-1415.
Peli, E., Goldstein, R. B., Young, G. M., Trempe, C. L., & Buzney, S. M. (1991). Image
enhancement for the visually impaired. Investigative Ophthalmology & Visual Science, 32,
2337-2350.
Regan, D., Raymond, J., Ginsburg, A. P., & Murray, T. J. (1981). Contrast sensitivity, visual
acuity, and the discrimination of Snellen letters in multiple sclerosis. Brain, 104, 333-350.
Rizzo, J. F., Cronin-Golomb, A., Growdon, J. H., Corkin, Z., Rosen, T. J., Sandberg, M. A.,
Chiappa, I~ H., & Lessee S. (1992). Retinocalcarine function in Alzheimer's disease.
Archives of Neurology, 49, 93-101.
Rosen, A. (1994). Perceptual organization, aging, and the multichannel model. Unpublished
doctoral dissertation, Case Western Reserve University, Cleveland, OH.
Sadtm, A. A. (1989). The optic neuropathy of Alzheimer's disease. Metabolic, Pediatric, and
Systemic Ophthalmology, 12, 64-68.
Sadun, A.A., & Bassi, C.J. (1990). Optic nerve damage in Alzheimer's disease.
Ophthalmology, 97, 9-17.
Schlotterer, G., Moscovitch, M., & Crapper-McLacMan, D. (1984). Visual processing
deficits as assessed by spatial t~equency contrast sensitivity and backward masking in
normal aging and Alzheimer's disease. Brain, 107, 309-325.
Sekuler, R., & Blake, 1~ (1987). Sensory underload. Psychology Today, 21, 48-51.
Sekuler, R., & Blake, R. (1994). Perception (3rd). New York, NY: McGraw-Hill.
Snodgrass, J. G., & Vanderwart, M. (1980). A standardized set of 260 pictures: Norms for
name agreement, image agreement, familiarity, and visual complexity. Journal of
Experimental Psychology: Human Learning and Memory, 6, 174-215.
Steinberg, S. (1967). Two operations in character recognition: Some evidence from reaction
time measurements. Perception & Psychophysics, 2, 45-53.
Theios, J. & Amrhein, P. C. (1989a). The role of spatial t~equeney and visual detail in the
recognition of patterns and words. In C. Izawa (Ed.), Current issues in cognitive
processes: The Tulane Floweree symposium on cognition (pp. 389-409). Hillsdale, N. J.:
L. Erlbaun~
Theios, J. & Amrhein, P. C. (1989b). Theoretical analysis of the cognitive processing oflexical
and pictorial stimuli: Reading, naming, and visual and conceptual comparisons.
Thomas, C. W., Cfilmore, G. C., & Royer, F. L. (1993). Models of contrast sensitivity in
human vision. 1EEE Transactions on Systems, Man, and Cybernetics, 23, 857-864.
Trick, G. L., & Silverman, S. E. (1991). Visual sensitivity to motion: Age related changes and
deficits in senile dementia of the Alzheimer type. Neurology, 41, 1437-1440.
Turner, J. A. B. (1990). Visual perception in normal aging and Alzheimer's disease: Influences
on picture naming and recognition. Unpublished doctoral dissertation, Case Western
Reserve University, Cleveland, OH.
Vellutino, F. 1~ (1977). Dyslexia: Theory and Research. Cambridge, MA: MIT Press.
Whitehouse, P. J., Hedreen, J. C., Jones, B. E., & Price, D. L. (1983). A computer analysis of
neuronal size in the nucleus basalis of Meynert in patients with Alzheimer's disease. Annals
of Neurology, 14, 149-150.
Williams, M. (1956). Studies of perception in senile dementia: Cue selection as a ftmction of
intelligence. British Journal of Medical Psychology, 19, 270-279.
Williams, M., & Lecluyse, IC (1990). Perceptual consequences of a temporal processing deficit
in reading disabled children. Journal of the American Optometric Association, 61, 111-
121.
Wilson, H. 1L, McFarlane, D. K., & Phillips, G. C. (1983) Spatial frequency tuning of
orientation selective units estimated by oblique masking. Vision Research, 23, 873-882.
Yonelias, A. P., Hockley, W. E., & Murdock, B. B. (1992). Tests of the list-strength effect in
recognition memory. Journal of Experimental Psychology: Learning, Memory, and
Cognition, 18, 345-355.
Zeki, S., Watson, J. D. G., Lueck, C. J., Friston, K. J., Kennard, C., & Frackowiak, 1L S. J.
(1991). A direct demonstration of the functional specialization in human visual cortex.
The Journal of Neuroscience, 11, 641 - 649.
220 9 1995 Elsevier Science B.V. All rights reserved.
Aging, Alzheimer's Disease, and Word Recognition: A Review of the Recent

Literature*
F. Richard Ferraro
University of North Dakota
1. C H A P T E R G O A I ~
It is the goal of the present chapter to provide the reader with a comprehensive review
of the recent word recognition literature l~om the perspective of normal, healthy aging and the
age-related disorder of Alzheimer's disease (AD). The majority of the chapter will detail visual
word recognition, although I will also discuss, albeit briefly, the growing literature
investigating auditory word recognition in these two populations. My review will examine
theory as well as data, in an effort to paint a picture of what is currently known about word
recognition in aging and AD. I will discuss the specific tasks that have been used primarily to
investigate word recognition (lexical decision, word naming) and the demands associated with
these tasks and how these demands can influence subsequent performance (i.e., Balota,
Ferraro, & Connor, 1991). I will also attempt to provide a look at where the field of aging,
AD, and word recognition is going. Given the sheer number of individuals living longer,
coupled with the increasing number of individuals likely to have AD in our society, this review
should nicely provide the reader with a base on which to evaluate the role of fundamental
aspects of the communication process play in language comprehension.
The first part of this chapter will review the concepts of normal healthy aging and
compare and contrast this with dementia and AD. This review will include demographic
information, operational definitions, incidence and prevalence rates of AD, and the latest, most
up-to-date information pertaining to diagnosis of AD as it relates to normal, healthy aging.
The next section will examine the recent developments pertaining to cognitive
processing (word recognition in particular) in normal healthy older adults, and how this leads
into the study of similar processing in AD individuals. This particular section will also examine
current models of visual and auditory word recognition (i.e., Cohort Model (Grosjean, 1980);
Logogen Model (Morton, 1969), Serial Search Model (Forster, 1971), Interactive-Activation
AUTHOR NOTES: Many of the ideas expressed and discussed in this chapter resulted from my tenure (1989-
1992) as a Postdoctoral Fellow/Research Fellow at the Alzheimer's Disease Research Center (ADRC) at the
Washington University School of Medicine and the Department of Psychology at Washington University in St.
Louis, MO. ADRC Grant P50 AG05681 (L. Berg, PI) and Healthy Aging and Senile Dementia (HASD) Grant
P01 AG03991 (L. Berg, PI, D. A. Balota, Project Leader, F. R. Ferraro, Co-Project Leader), both from the
National Institute on Aging (NIA) graciously supported my research while in St. Louis. Many people at
Washington University deserve thanks: Dave Balota, Leonard Berg, Jan Duchek, Emily LaBarge, Kathy Mann-
Koepke, John Morris, John Stern, and Martha Storandt. I especially thank all the patients and their families I
came into contact with while in St. Louis for allowing me to spend time with them. Finally, I thank Cameron
Camp and all those authors who responded to my request for articles, chapters, and preprints placed on the
Cognitive Aging Electronic Mail Network.
Aging, Alzheimer's disease, and word recognition 221
Model (Rumelhart & McClelland, 1981), Verification Model (Becker, 1976), Connectionist
Model (Seidenberg & McClelland, 1989) and the role they have played and will play in the
future development of understanding the word recognition processing system in healthy aging
and AD.
Finally, this chapter will examine what the future holds regarding word recognition
processing in these two populations. In particular, it will be proposed that this very simply
task (word recognition) be included among other diagnostic tools for studying normal and
deficit lexical information processing.
2. INTRODUCTION
2.1. Normal Healthy Aging
One of the fastest-growing segments of our population are those individuals aged 60 or
older, and this increase can be traced to, among other factors, declining birth rates, reduction
in mortality, and increases in overall life expectancy (e.g., Powell & Whitla, 1994). It is not
surprising, then, to reveal that in 1900, for instance, this group represented only 4% of the
total population. Today, in 1994, this figure is closer to 10-11% of the total population. It is
estimated that this percentage will jump to approximately 12% by the year 2000 and 15-16%
by the year 2020 (Morrison, 1982). C~en the fact that older adults are now also living longer
(into their 80s, 90s, and 100s) and are healthier, in general than in previous decades, the
percentage ofthe population over age 60 will reach nearly 25% by 2040 (Cantor, 1991). ffwe
break down this increase even further, we find that these increases are even more dramatic
when we consider people living to 75 and beyond as well as 85 and beyond. In fact, there is
now evidence supporting the observation that adequacy of cognitive resources (of which word
recognition processes depend heavily) are critical even in individuals who surpass the 100-
year-old mark (Pooh, Martin, Clayton, Messner, Noble, & Johnson, 1992, but see PowelL
1994). Thus, the accumulating evidence suggests that adequate word recognition skills and
associated processing are relevant (and necessary) for people well beyond their 60th birthday.
2.2. Alzheimer's Disease
The fast-increasing aging population not only brings with it more elderly adults, it also
increases the proportion of elderly adults who will acquire and show evidence of a variety of
dementing illnesses (i.e., Costa, Whitfield, & Stewart, 1989). Alzheimer's disease (AD,
Alzheimer, 1907) is but one of nearly 80 forms of dementia, and is typically the most common
form of dementia in the over-60 population. Current estimates reveal that approximately 50%
of all dementia cases are Alzheimer type dementia (Rocca, Amaducci, & Schoenberg, 1986;
Tomlinson, 1982). AD is characterized as a progressive neurological disorder that affects
processing associated with memory, daily-living skills, personality, communication, problem-
solving, and visuospatial ability. The course of the disease is highly variable, with some
patients dieing within 1-2 years of the initial diagnosis while in other individuals the disease
course can last 20-25 years or more (see Martin, 1990, for a discussion regarding sub-groups
in AD individuals). There are no known cures for AD at present, and the only realistic way to
determine if an individual has AD is with a brain autopsy. This is relevant, because the brains
of AD individuals are characterized by microscopic changes involving neurofibrilary tangles,
222 F.R. Ferraro
neuritic plaques, and specific locations of granulovacular degeneration (Berg, McKeel, Miller,
Baty, & Morris, 1993; McKeel, Ball, Price, Smith, Miller, Berg, & Morris, 1993; Price, Davis,
Morris, & White, 1991; Terry & Katzman, 1983; Tomlinson & Henderson, 1976; Wisaiewski
& Merz, 1985).
Currently, approximately 4 million Americans have Alzheimer-type dementia, and
100,000 individuals die from the disease annually (Alzheimefs Disease and Related Disorders
Association, 1987). These figures make Alzheimer's disease the forth leading causes of death
(behind heart disease, cancer, and stroke) in elderly adults aged 60 and over. Specific theories
of possible causes of AD have been advanced (i.e., genetic transmission, neurotoxicity,
infection), although it is likely the case that some combination of these and other causes is
most likely the more accurate picture of causation. This disease is also an economic burden,
not only for families caring for their loved-ones, but also for health-care in general (Max, 1993;
Williamson & Schulz, 1993).
3. DIAGNOSIS OF ALZHEIMER'S DISEASE AND D I F F E I ~ N T I A T I O N FROM

NORMAL AGING
Before one can perform research into "normal" aging, one must be able to operationally
define "normal" aging. Likewise, in addition to defining what is "normal", it is important to
distinguish "normal" from non-normal, or pathological, aging. This distinction becomes
especially important when performing aging research, primarily because of what is known as
age-associated memory impairment. That is, there are certain memory declines that
accompany normal, healthy aging that are typically not the sign of disease. In other words,
age-associated memory impairment would be a component of a "normal" definition of aging.
While many definitions of "normal" abound in the aging literature, a recent chapter by Malec,
Ivnik and Smith (1993, p. 85) provide what appears to be an acceptable and accurate definition
of "normal" aging. A nice feature of their operational definition is the explicit fact that many
elderly adults who are "normal" still have many medical ills (both medical and psychological)
that are a normal aspect of the aging process. Thus, a person can be old and "normal" while
still having a variety of ills that are part of the typical aging process. In this way, a more
representative picture of normal aging can emerge. I will adopt this definition for the present
chapter.
The definition advanced by Malec and colleagues includes several criteria, all of which
must be in evidence by the participant. These criteria include a) no active central nervous
system (CNS) and/or psychiatric disorders that affects cognitive performance, b) no evidence
of cognitive difficulties, c) acceptable levels of psychoactive drug and medication use, such
that cognitive performance is not compromised by the amount and type taken, d) no adverse
cognitive effects associated with any disorder(s) that can affect cognition (i.e., closed-head
injury, substance abuse), and e) no adverse effects on cognitive performance although some
form of chronic illness (i.e., diabetes) may be present as long as the illness does not
compromise performance. In essence, one performs a process of elimination to arrive at a
relatively "normal" subject group, free from any disease or disorder that may adversely affect
cognitive performance. Much the same method is employed when attempting to diagnose
Alzheimer's disease.
Since a brain autopsy is the only conclusive method of determining whether or not a
person has definite Alzheimer's disease, diagnostic criteria for AD have become more
sophisticated over the past 10-15 years (Margolin, 1992). McKhann, Drachman, Folstein,
Katzman, Price, and Stadlan (1984) devised clinical diagnosis criteria, under the auspices of
the National Institute of Neurological and Communicative Disorders and Stroke and the
Alzheimer's Disease and Related Disorders Association (NINCDS-ADRDA), which have
proven to be reliable and accurate with regard to diagnosing individuals with "possible" or
"probable" Alzheimer's disease (i.e., Lopez, Swihart, Becker, geinmuth, Reynolds, gezek, &
Filley, 1990). These criteria also are in agreement with another widely-employed diagnostic
criteria (Diagnostic and Statistical Manual, DSM-1V, American Psychiatric Association, 1994).
Since that time, there has been increased efforts into refining these diagnostic criteria
(Cummings, 1988; Friedland, 1993; Hart & Semple, 1990; Jorm, Fratiglioni, & Winblad, 1993;
Tyn~ 1989). One aspect of these diagnostic criteria that seems dear, however, is a
constellation of impairments that uniquely manifest themselves in Alzheimer disease
individuals. These impairments include any of the following: memory, attention, visuospatial
functioning, problem-solving, calculations, personality, thinking, affect, language, perception,
praxis, awareness of disease, social skills, motor deficits, seizures, and incontinence of urine
and/or stool (Friedland, 1993). Cummings (1988) includes a similar array of deficits and
additionally includes posture, gait, and overall movement deficiencies.
Personally, however, I am most familiar with the Clinical Dementia Rating (CDR) scale
used at the Washington University School of Medicine Alzheimer's Disease Research Center
(ADRC) in St. Louis (Berg, 1984; Berg, 1988; Berg, Miller, Storandt, et al, 1988; Burke,
Miller, Rubin, et al., 1988; Hughes, Berg, Danziger, Coben, & Martin, 1982; Morris, 1993;
Morris, McKeel, Fulling, Torack, & Berg, 1988; McCulla, Coats, Van Fleet, Duchek, Grant, &
Morris, 1989; Storandt, Morris, Rubin, Coben, & Berg, 1992). The CDR stages the severity
of Alzheimer's disease displayed by an individual. In this scale, a CDR of 0 indicates no
dementia; a CDR of.5 indicates very mild or "Questionable" dementia; a CDR of 1.0 indicates
"Mild" dementia; a CDR of 2.0 indicates "Moderate" dementia; a CDR of 3.0 indicates
"Severe" dementia. It consists of a 90-min interview that assesses cognitive ability in areas
including memory, orientation, judgement and problem solving, community affairs, hobbies,
and personal care. The interview is semi-structured and both the patient and the collateral
source (i.e., spouse, child) participate. Interviews are conducted by one of eight board-
certified physicians (four neurologists, four psychiatrists), and are video-taped and
subsequently reviewed by a second physician for reliability. The diagnosis of Alzheimer's
disease by this research team has been excellent, with 145/150 (97%, personal communication,
J. C. Morris, 2/94) individuals diagnosed with SDAT indeed having AD confirmed at a brain
autopsy (Berg, Smith, Morris, et al., 1990; Burke, Miller, Rubin, et al., 1988; Morris, McKeel,
Fulling, Torack, & Berg, 1988; Morris, McKeel, Price, et al., 1988).
4. COGNITIVE PERFORMANCE IN NORMAL AGING AND AD
The last 10-15 years has seen an increase in the amount of research investigating the
basic cognitive processes in both normal elderly adults and individuals with AD (i.e., Light,
1991; Nebes, 1989, 1992; Salthouse, 1982). These processes have ranged from contrast
sensitivity (HuRon, Morris, Elias, & Poston, 1993) to auditory word onset gating (Wingfield,
Aberdeen, & Stine, 1991) to processes including simple and choice reaction time, attention,
and memory (Parasuraman & Nestor, 1993). This increase in research has lead to the
introduction of several new journals, as well, devoted exclusively to aging and Alzheimer's
224 F.R. Ferraro
disease (i.e., Age & Aging; Aging & Cognition; Alzheimer's Disease and Assocmted
Disorders; Archives of Clinical Neuropsychology; Dementia; Developmental
Neuropsychology; Experimental Aging Research; Neuropsychology; Neuropsychology
Review; Journal of Clinical & Experimental Neuropsychology; Journal of Cognitive
Neuroscience; Journal of Neuropsychiatry & Clinical Neurosciences; Psychology & Aging).
Even more important, funding (e.g., National Institute on Aging, NIA), for research into aging
and disease has increased dramatically over this same time period. Given the demographic
statistics discussed earlier, it is not too surprising to see increases in interest for normal and
pathological aging research.
Given the great increase in aging and AD cognitive research over the past 10-15 years,
interest has also been directed at the particular tasks subjects must engage in while these
cognitive processes are being measured. This is especially relevant when studying aging and
demented populations because of the many physical ailments and declines that accompany
these groups. It is quite possible that a cognitive decline noted in a particular experiment is
nothing more than a physical decline. The next section will detail the primary tasks that have
been employed to study word recognition performance. In each case these tasks depend
heavily upon a minimal level of physical dysfunction on the part of the participant. For
instance, in lexical decision both vision (to see the stimulus) and minimal physical strength (to
press one of two possible buttons) are required. The same is true in the naming or
pronunciation task; adequate vision and an adequate vocal apparatus are required for adequate
performance above and beyond whatever the cognitive contribution may be.
4.1. Tasks for Visual Word Recognition: The Naming (Pronunciation) Task and The
Lexical Decision Task
The vast majority of the visual word recognition studies discussed previously have
either employed the lexical decision task or the word naming task. This is of great importance,
especially since these tasks depend highly on cognitive operations that are sometimes
compromised in both older adults and individuals with Alzheimer's disease (i.e., Marcd &
Patterson, 1980). For instance, the lexical decision task, in which the subject must decide
whether or not a string of letters is a real word or not, is not only an identification task but also
a discrimination task. That is, words are usually easier to distinguish and discriminate from
nonwords or pseudowords not only because they are real words but also because words are
usually more familiar to subjects than nonwords (see Balota & Chumbley, 1984; Balota &
Lorch, 1986). Furthermore, the lexical decision task involves the subject making a decision;
deciding whether a string of letters is a word or a nonword. There is compelling evidence that
this process (decision making) is severely compromised in older adults and AD patients,
especially when reaction time is involved (Gordon & Carson, 1990; Madden, Nebes & Allen,
1992; Mahurin & Pirozzolo, 1986; 1993; Pate, Margolin, Friedrich, & Bentley, 1994).
The naming task has also not escaped criticism. Although in the naming task the
subject does not have to make any decisions (i.e., they simply have to name the particular
word), there are still problems with it. For instance, outputing the verbal response can be
especially difficult for older adults or adults with AD. Even more damaging is the fact that
naming deficits are typically a hallmark of AD, especially confrontational naming tests like the
Boston Naming Task (e.g., Appell, Kertesz, & Fisman, 1982; Hufl~ Corkin, & Growdon, 1986;
Martin & Fedio, 1983; Shuttleworth & Huber, 1988; Tipper & Farah, 1994). Furthermore,
deficits in naming are by far one of the most frequently reported disruptions in the language
processing of AD individuals, and there is evidence that the naming task can serve as a good
diagnostic for normal healthy elderly and elderly with mild naming disorders (Bowles, Obler, &
Albert, 1987).
However, despite the numerous confrontational naming deficits in AD, the ability of
AD patients to read words aloud is remarkably well-preserved, especially words that violate
traditional English spelling-to-sound correspondence rules like Yacht and Aisle (Balota &
Ferraro, 1993; Friedman et al, 1992; Seidenberg, Andersen, Kempler, & Jackson, 1994, but
see Patterson, Graham, & Hodges, 1994, for an akemative viewpoint). Although older adults
and individuals with AD can perform the lexical decision task and word naming task, it is
imperative to review the potential difficulties with these tasks nonetheless. This will provide
the researcher with valuable information within which to judge his/her research results.
Finally, any results that are obtained from purely visual processing must also be
tempered by the fact that there are widespread visual deficits in both normal aging (Owsley &
Sloane, 1990) and AD (Cronin-Golomb, Corkin, Kizzo, Cohen, Growdon, & Banks, 1991;
Marin, 1987; Trobe & Butter, 1993). These potential visual information processing deficits
are mentioned so as to make the potential aging or AD researcher more aware of the possible
problems associated with the visual word recognition literature within these subject groups.
The use of the lexical decision task and the naming (pronunciation) task are used
extensively in studies of word recognition in older adults and dementing adults. The results
from these two tasks have accounted for the majority of information that is known about word
recognition processing in aging and age-related disease. The next section will examine in detail
the fundamental aspects of visual word recognition. The majority of studies to be reviewed for
this chapter deal with visual word recognition.
4.2. Visual Word Recognition: Models, Theories, and Applications to Healthy Aging and
Alzheimer's Disease
The particular cognitive process that will be the focus of the remainder of this chapter
is word recognition (visual and auditory). Visual word recognition is central to several higher-
level (i.e., comprehension) processes and has figured significantly in a variety of theories and
models proposed to explain and account for cognitive functioning (Balota, 1993; Becker,
1979; Forster, 1971; Morton, 1969; Rumelhart & McClelland, 1981; Seidenberg &
McCleUand, 1990). A fundamental assumption of many, if not all, of these models is the idea
that the word (in context and out of context) and the processes of word recognition are ideally
suited to convey a variety of information fundamental to cognitive psychology. Many of these
models break down the information processing sequence as being a series of components (i.e.,
encoding, comparing, deciding, responding). Likewise, these models also propose that the
final product of visual word recognition (i.e., i d e n t ~ g a string of letters as a real English
word) is the combination of lower-level (bottom-up), basic visual processing mechanisms
interacting with higher-level (top-down) processes, the result being a single candidate. This
interaction of various bottom-up and top-down processes is time-dependent in that most
literate adults can identify a string of letters as a real English word within 500 ms. Thus, any
breakdowns in either top-down or bottom-up systems, or any time delays, will severely
compromise the word recognition process.
226 F.R. Ferraro
This same fundamental nature of the word recognition process has been a recent
concern of research directed at older adults and individuals with AD (Allen, Madden, Weber,
& Groth, 1993; Balota & Duchek, 1988; Bowles & Pooh, 1981; Cerella & Fozard, 1984;
Ferraro, 1994; Ferraro & Kellas, 1992; Friedman, Ferguson, Robinson, & Sunderland, 1992;
Gilmore, Groth, & Thomas, under review; Madden, 1992; Myerson, Ferraro, Hale, & Lima,
1992; Pirozzolo, Nolan, Kuskowski, Mortimer, & Maletta, 1988). The contribution that these
various studies brings to the models discussed earlier is that they will serve to expand these
models to incorporate the idea of slowing as it relates to the various components fundamental
to these various models. That is, a fundamental theoretical assumption within the cognitive
aging literature is the idea of a generalized slowing function to describe both normal age effects
(young versus old) and disease effects (i.e., old versus AD). Recent papers by Cerella (1985,
1991), Myerson et al. (1992), and Nebes and Brady (1992) highlight this issue.
Several studies in both aging and AD that have examined visual word recognition have
usually couched this process within another process, namely semantic priming. That is, many
of the studies have used word recognition performance as the main dependent measure in
investigating the structure of semantic memory. In the semantic priming task (see Neely, 1991,
for a review) a subject sees a prime (CAT) followed by a target (DOG) and must decide if the
target is a real word or not. On some trials the target is not a real word (BLANT). Also, the
relationship between the prime and the target are manipulated such that on some trials the
prime and target are related (CAT-DOG), unrelated (COUCH-DOG), or neutral (BLANK-
DOG). People are usually faster deciding that DOG is a word when it follows a related prime
than when it follows an unrelated prime (the semantic priming effect; related faster than
unrelated). Likewise, people are usually faster to decide that DOG is a word in comparison to
deciding that BLANT is a nonword or pseudoword (the lexicality effect; word response faster
than nonword or pseudoword response).
Although I will not get into the area of semantic priming and AD (that is being handled
in another chapter in this book by Beth Ober), one important observation that has emerged
from the visual word recognition literature on aging and dementia is the potentially relevant
diagnostic value of the lexicality effect that can be obtained from this literature. By definition,
the lexicality effect is the reaction time (and error rate) difference between subjects' responses
to words and to nonwords or pseudowords. A distinction needs to be made between
pseudowords and nonwords at this time. Pseudowords are typically orthographically legal
strings of letters than has no meaning associated with them. Examples include FLIRP and
BLANT. These letter strings are visually, orthographically, and phonologically similar to real
words, although they have no meaning associated with them. In other words, you would not
find them listed in any dictionary. Similarly, nonwords are (usually) collections of letters that
form a random order. For instance, the collection DSYEGR may be considered a nonword.
That is, it does not look or sound like a real English word. A study reported by Rubenstein,
Garfield, and Millikan (1970) was one of the first to show that these nonwords (i.e., NGTRS)
usually result in a fast NO response when they are part of a lexical decision experiment (see
also Baron & Thurstone, 1973). In a similar vein, it has also been shown that the closer the
pseudoword stimulus approximates a real word (i.e., WlRD), the longer (and more error-
prone) are responses and the greater is the difference between responses to words and
pseudowords (e.g., Coltheart, Davelaar, Jonasson, & Besner, 1977). Henderson (1982) nicely
detailed the lexicality effect literature and concluded that the effect is indeed reliable, especially
when tested under appropriate experimental conditions. Although this may sound patronizing,
it is no small feat these days to obtain appropriate experimental stimuli when performing word
recognition experiments, especially when examining older and demented individuals. In
addition to the time-tested factors ot~ for instance, Concreteness (James, 1975; deGroot, 1989;
Kroll & Mervis, 1986), Polysemy (Jastrzembski, 1981; Kellas, Simpson, & Ferraro, 1988;
Simpson, 1984), Frequency (Balota & Chumbley, 1984), one must also now be aware of
stimulus features such as Familiarity (Gernsbacher, 1984), and Neighborhood Density
(Grainger, 1990). I will now detail the recent literature on visual word recognition in older
adults and individuals with Alzheimer's disease. This review will pay particular attention to the
lexicality effect and its usefulness as a potential clinical diagnostic. It will further be discussed
how results from the lexicality effect across these diverse groups can be used in order to test
the notion that older adults (and individuals with AD) have difficulties inhibiting irrelevant
information. This difficulty in basic inhibitory processing has recently become a hallmark of
both aging and age-related disease (i.e,. Hasher & Zacks, 1988).
Tables 1 and 2 present a summary of those studies that have included word, nonword
(e.g., FGYSWE), and pseudoword (e.g., BLANT) stimuli presented to younger, older and
individuals with Alzheimer's disease. As mentioned above, several of these studies have
couched these particular stimuli within the semantic priming paradigrtt Furthermore, only
those conditions in which a lexicality effect can be obtained are listed. There are actually two
lexicality conditions. The first is the difference between words and pseudowords (PW-W) and
the second is the difference between words and nonwords (NW-W). In this way it will be
possible to examine the influence of approximation to English across these subject groups. As
will be noted in the table, however, only a few studies have included nonwords as part of their
stimuli.
Table 1
Lexicality Effects (PW-W; NW-W) in milliseconds (ms) as a Function of Young Adults and
Elderly Adults
1. Allen, Madden, & Crozier (1991)

- prime was XXXXX, LDT, PWs
- collapsed over number of letters
Very High Medium-High Low Very-Low

Freq. Freq. Freq. Freq.
Young 95 103 118 76

Old 239 196 209 186
228 F.R. Ferraro
2. Allen, Madden, Weber, & Groth (1993)

- LDT, PWs
- case and spacing varied
LDT (Expt. # 1) LDT (Expt. # 2)
Young 181 154

Old 343 251
3. B o w l e s & Pooh (1981)

- 2-word displays, LDT, PWs
- (NW/NW) display used for PWs; avg. of H/H; H/L; L/L for Words
Young = 328
Old - 579
4. B o w l e s & Pooh (1988)

- LDT, PWs
- Related and Unrelated Primes (R+U)/2; THE condition
Young = 199
Old = 139
5. Burke, White, & Diaz (1987)

- LDT, PWs
- prime was either expected or unexpected
410 ms SOA 1550 ms SOA
Young 111 103

Old 11 93
6. Burke & Yee (1984)

- LDT, PWs
- Related/Unrelated Prime; Associated Words; Sentences
Associated Word Whole Sentence Instrument
Young 120 63 91
Old 165 111 94
7. Cohen & Faulkner (1983)

- LDT, PWs
- Context conditions only
High-Probability Low-Probability
Context Context
Young 209 99
Old 234 124
8. Ferraro & Kellas (1992)

- LDT, PWs
- BLANK as neutral prime
target orientation (in degrees)

0 60 120 180
Young 234 300 339 430

Old 743 679 1348 1365
9. Ferraro (1994)
- LDT, PWs, NWs
- BLANK as neutral prime
PW Lexicality
0 30 60 90 120
Young 230 203 156 170 288

Old 395 415 657 612 940
NW Lexicality
0 30 60 90 120
Young 29 -5 -52 -77 -45

Old 215 148 124 194 280
10. Howard (1983)

- LDT, PWs
2 word displays; related + unrelated together collapsed over dominance;
-
PW = NW/NW condition
Young = 38
Old = 30
230 F.R. Ferraro
11. Howard, McAndrews, & Lasaga ( 1 9 8 1 )

- LDT, PWs
- Category + Descriptive (associated/unassociated); NW/NW condition for PWs
Young = 1 3 8
Old = 71
12. Howard, Shaw, & Heisey (1986)

- LDT, PWs
- Words = Associated + Blank + Unassociated; Blank- Nonword for PWs
SOA
150 ms 450 ms 1000 ms
Young 104 71 96
Old 78 111 104
13. Kellas, Simpson, & Ferraro ( 1 9 8 8 )

- LDT, PWs
Young = 1 0 9
Old = 217
14. Madden (1986)

- LDT, PWs
- Neutral was sentence context
Sentence Context
Non-Neutral Neutral
Young 197 186

Old 219 194
15. Madden ( 1 9 8 8 )
- LDT, PWs
- sentence context
Target
Intact Degraded
Neutral Young 133 92

Old 265 211
Non-Neutral Young 109 98

Old 213 226
16. Madden (1989)

- LDT, PWs, 2 expts.
- sentence context
100 m s IS I 1 0 0 0 m s ISI
Neutral Young 66 94
Context Old 109 162
Non-Neutral Young 61 92
Context Old 82 112
17. Madden (1992)

- LDT, PWs
- neutral was BLANK; R/U primes combined
Target
Age Intact BLANK D*e*g*r*a*d*ed
20's 232 225 281

30's 222 184 180
40's 257 276 268
50's 261 225 213
60's 223 249 274
70's 273 258 307
18. M a d d e n & G r e e n e ( 1 9 8 7 )
- LDT, PWs, NWs
Manual LDT Vocal LDT

PWs NWs PWs NWs
Young 85 13 50 4
Old 234 79 163 36
232 F.R. Ferraro
19. Madden, Pierce, & Allen (1993)

- LDT, Naming, PWs
- BLANK as neutral; R + U primes only
Expt. 1
SOA (ms)
128 255 383 510
Young 145 186 167 156

Old 194 193 186 204
Expt. 2
SOA (ms)
71 100 128 156
Young 133 155 156 140

Old 235 262 259 256
Expt. 3
SOA (ms)
100 156 581 708
Young 88 135 138 115

Old 186 218 191 240
Expt. 4
Unrelated only
LDT Manual LDT Vocal
Young 184 156

Old 230 188
Note: Freq. signifies Frequency; H F si~ifies High Frequency; L F signifies Low Frequency;
ISI si~ifies Inter-Stimulus Interval; lex. si~ifies Lexicality Effect; L D T signifies Lexical
Decision Task; ms si~ifies milliseconds; NW si~ifies NonWord; PWs signifies Pseudowords;
R signifies Related; U si~ifies Unrelated; SOA si~ifies Stimulus Onset Asynchrony
Table 2
Lexicality Effect ( P W - W ; N W - W ) in milliseconds (ms) as a function o f Elderly Adults and
A17heimer's Disease Adults
1. Albert & Milberg (1989)

- LDT, P W s
- w o r d = related + unrelated
Old = 327
A D = 701
2. Clark (1980)
- LDTs, P W s
- w o r d s = high-freq. + low-freq.
Y o u n g .......................................... 90
Old ............................................. 170
Mild A D ..................................... 180
M o d e r a t e A D ............................. 150
3. Ferraro & Balota ( 1 9 9 2 )

- LDT, P W s
Y o u n g .......................................... 69
Y o u n g - O l d (< 80 years) ............. 119
Old-Old (> 80 years) .................. 243
Very Mild A D ............................ 235
Mild/Moderate A D ..................... 378
4. Nebes, Brady & H u f f ( 1 9 8 9 )

- LDT, PWs, w o r d = asssociated/unassociated prime
Y o u n g .......................................... 32
Old .............................................. 47
A D ............................................ 238
234 F.R. Ferraro
5. Ober & Shenaut (1988)

- LDT, PWs (misspelled words), NWs (random letters)
- words = high-freq. + low-freq.
PW lex. NW lex.
Old 114 -86

AD 600 -55
Note: AD signifies Alzheimer's disease; Freq. signifies Frequency; HF si~ifies High

Frequency; ISI signifies Inter-Stimulus Interval; lex. si~ifies Lexicality Effect; LDT si~ifies
Lexical Decision Task; LF si~ifies Low Frequency; ms signifies milliseconds; NW si~ifies
NonWord; PWs si~ifies Pseudowords; R si~ifies Related; SOA si~ifies Stimulus Onset
Asynchrony; U si~ifies Unrelated
An immediate observation from Tables 1 and 2 is the fact that across the vast majority
of these studies the pattern of the lexicality effect that emerges is one in which young adults
have a smaller lexicality effect than healthy older adults, who in turn have a smaller lexicality
effect than the demented individuals. Error rates are also consistent with this pattern, with
young adults typically being much more accurate than healthy older adults, who in turn are
typically much more accurate than the individuals with Alzhdmer's disease. Ferraro and Balota
(1993) have interpreted such increases in the lexicality effect as supportive of recent arguments
regarding breakdowns in inhibitory processes in both healthy older adults and individuals with
AD. The reasoning behind this is as follows: Because primes are always words, subjects need
to suppress their word response to the prime items on pseudoword target trials. This
suppression ability (or inability) appears most difficult for older adults (as compared to
younger adults) and for demented individuals (as compared to older non-demented adults),
leading to increases in both response latencies as well as error rates.
This particular interpretation of the lexicality effect results provides converging
evidence regarding other recent investigations into the increasing breakdown in the ability of
these subject groups (old adults, demented adults) in their ability to inhibit partially activated
(but inappropriate) information. Recent investigations with older adults (e.g., Hasher & Zacks,
1988; Hasher, Stoltzfus, Zacks, & Rypma, 1991; McDowd & Oseas-Kreger, 1991; Tipper,
1991) have revealed that this subject group does not appear to inhibit irrelevant information as
much as healthy young adults. Similarly, the same pattern has emerged in investigations that
have tested older adults and demented adults (Balota & Duchek, 1991; Duchek, Balota,
Ferraro, Gernsbacher, Faust, & Conner, 1992; Ferraro, Balota, & Connor, 1991). Since these
various studies have addressed a variety of different tasks, it may be the case that this failure to
inhibit irrelevant information is a general characteristic of both older adults and, to a greater
extent, demented individuals. It is precisely this lexicality effect difference across older and
demented individuals that could potentially be used as a clinical diagnostic marker. This
particular interpretation of the lexicality effect results will be taken up in the Discussion. The
point, however, is that the lexicality effect appears to be a potentially-relevant diagnostic
marker for cognitive declines evident in older and demented populations.
While the majority of word recognition studies performed on older and dementing
adults are visual, several more recent attempts have been made to investigate the role of aging
and AD on auditory word recognition as well. Given the compensatory nature of many older
and dementing adults cognitive perfo~ance, it is important to detail how these populations
perform word recognition processing when the information arrives within a different modality.
Such results have long-term ramifications for how communication processes may be better
served in these populations.
4.3. Auditory Word Recognition: Models, Theories and Applications to Healthy Aging
and Alzheimer's Disease
Like visual word recognition, auditory word recognition also relies on the passage of
time and a variety of bottom-up and top-down processes interacting to arrive at a suitable final
word candidate. However, that is (primarily) where the similarity between these two versions
of word recognition end (although see Johnson, 1992; Johnson & Pugh, 1994, for their Cohort
model of visual word recognition). In general, spoken language typically has more demands
placed upon its comprehension, and these demands would appear to be more compromising for
older and demented adults. For instance, spoken language arrives much faster than written
language, and the listener has much less control over the input, unlike in reading (Wingfidd &
Stine, 1991). There is also the problem of information overload resulting from such rapid
information processing, which would tend to further compromise the elderly and demented
with regard to attentional resource capacity (Kellas, Simpson, & Ferraro, 1988) as well as
short-term working memory capabilities (Morris, Crick, & Craik, 1988).
Various models of auditory word recognition have been advanced with the last 10-15
years, and the similarity to some of the models of visual word recognition detailed earlier is
striking. Grosjean (1980) developed the Cohort model of auditory word recognition, and the
Gating paradigm has been to auditory word recognition what the lexical decision task has been
to visual word recognition. In the Gating paradigm~ participants are presented with short (i.e.,
25-50 ms) segments of individual words and must attempt to decide what word they are
hearing. If the participant is not successful on the initial segment, or gate, successive gates
(usually of 100 ms durations) are presented until the correct response is made. Thus, over real
time, increasing amounts of the stimulus is presented until it is identified. The initial gate
creates what Grosjean termed the Word-Initial Cohort. That is, a relatively large set of
possible word candidates are assembled based on this initial 50 ms segment. With the passage
of time and any additional gating information, this cohort is substantially reduced in size, and
those words the subjects knows that are not consistent with the presented information are
eliminated from the cohort. As additional sensory information is presented, the cohort
eventually has only one member remaining, and the individual then reco~izes the word
(Marslen-Wilson & Welsh, 1978; Tyler, 1984; Wayland, Wingfield, & Goodglass, 1989).
Depending upon the context within which the word is presented, the time needed to identify a
word ranges from approximately 200 ms (in context) to 330 ms (no context) (G-rosjean, 1980;
Marslen-Wilson & Welsh, 1978; Tyler, 1984).
Several researchers in a variety of fields have applied the gating technique to real-time
estimations of word recognition processing and have included children (Elliott, Hammer, &
Evan, 1987; Walley & Metsala, 1990), young adults (Salasoo & Pisoni, 1985), aphasics
(Wingfield, Goodglass, & Smith, 1990), and older adults (Bell, 1989; Craig, 1992; Wingfidd,
236 F.R. Ferraro
Aberdeen, & Stine, 1991; Wingfield & Stine, 1991). As with the visual word recognition
literature detailed earlier, results from auditory word recognition tasks must also be tempered
by the fact that there are specific deficits in the auditory functioning of both elderly adults and
AD individuals (e.g., Sinha, Hollen, Rodriguez, & Miller, 1993). These deficits are noted
simply so that the reader is aware of them With effective screening procedures, results
obtained can still be of theoretical value.
In surveying the auditory word recognition literature, only a handful of studies could be
located that directly tested younger adults and older adults on auditory word recognition (Bell,
1989; Craig, 1992; Elliott, Hammer, & Evan, 1987; Wingfield, Aberdeen, & Stine, 1991;
Wingfield & Stine, 1991). Furthermore, no studies directly testing for auditory word
recognition performance could be located that involved individuals with Alzheimer's disease,
although AD patients have severe dysfunction when attempting to comprehend the speech of
other individuals (Appell, Kertesz, & Fisman, 1982; Kaszniak & Wilson, 1985). Furthermore,
several deficits exist in a variety of auditory fimctions in AD (Cummings & Benson, 1989;
Kurylo, Corkin, Allard, Zatorre, & Growdon, 1993; Margolis, Taylor, & Dunn, 1985).
However, despite the paucity of research within this area with these populations,
important findings have been obtained and it is possible to make for healthy older adults and
specific predictions can, nonetheless, be made for how Alzheimer individuals would likely
perform in such situations.
Of the five research reports that have investigated auditory word recognition in healthy
older adults, the results appeared mixed at best. Bell (1989) found that older adults benefitted
more from semantic context, especially with regard to the word-frequency effect, as compared
to younger adults. He reasoned that the elderly adults' performance may be the result of an
increased reliance on semantic and lexical information as compensation for degraded peripheral
and central encoding (i.e., Stanovich, 1980). In Bell's experiment, young and elderly adults
were compared on auditory word identification performance in noise as a function of target
word frequency, phonemic similarity neighborhood, and degree of semantic context provided
by the carder sentence. Craig (1992) studied real-time isolation monosyllabic word
recognition performance in younger and older individuals. Subjects were asked to listen to
words, guess what they were, and write down their answer as well as indicate (using a 5-point
Likert scale) how confident they were in their decisions. Results revealed that major events in
the real-time understanding process of spoken word identification occurred at a slower rate for
older, as compared to younger, adults. In other words, the older adults were less able to
identify target words at earlier gates and took longer to isolate words, as compared to their
younger counterparts. Craig speculated that this dysfunction could be the result of aging, a
loss of peripheral sensitivity, more central-type auditory differences and changes, or perhaps
due to an interaction of the aging process with both central or peripheral processes. EUiott,
Hammer, and Evan (1987) tested 5-7 year-old children, 17-year-olds, and adults aged 70-85
years on their auditory word identification performance. These authors also had subjects rate
their confidence in their identification performance. In general, teenagers performed better on
the gating task than did the young children and the older adults. Older adults tended to
provide more phonetic guesses than either of the other age groups. Furthermore, both the
teens and the children displayed better performance regarding their average total acceptance
point (i.e., the minimum time of stimulus presentation needed to identify the particular word)
than did the older adults. The conclusion was that although the older adults (presumably) had
greater experience with the words over the course of their lives, this experience was not
stdticient to counterbalance the inherent difficulties in processing briet~ temporally altered

word stimuli.
Wingfield and his colleagues, however, have revealed an opposite pattern to the reports
listed above. In particular, these authors have revealed an age constancy with regard to
auditory word recognition performance (Wingtield, Aberdeen, & Stine, 1991; Wingtield &
Stine, 1991). These authors have found that healthy elderly adults are not compromised in the
least in their ability, compared to younger adults, in auditory word recognition experiments.
Wingfield et al. (1991) presented subjects with 18 sentence contexts (6 high context, 6 low
context, 6 neutral) and the task was to identify the final word (i.e., target) of the sentence.
Each sentence context and target word were presented over headphones at varying (50 ms)
gates. Results revealed the expected main effect of age (young faster than old) and the
expected main effect on context (recall better in high context sentences, followed by low
context, followed by neutral context sentences). However, the age by context interaction was
not simaificant, suggesting that both young and old adults can identify auditorily presented
words with little more than the first half of the word's full acoustic duration. This performance
increased for both groups when the context became more constraining. Thus, healthy elderly
adults can use context effectively in an on-line experimental situation. These results are similar
to those of Kinsbourne (1973), as well as the results offered by Humes, Nelson, and Pisoni
(1991) and Humes, Nelson, Pisoni, and Lively (1993). Thus, it appears that this particular
research area is ripe for further study, especially given the fact that the handful of research
reports examining auditory word recognition in healthy elderly individuals is basically split
down the middle.
The next question concerns how individuals with Alzheimer's disease would perform in
similar auditory word recognition tasks. There is ample evidence that Alzheimer's disease
produces substantial auditory system degeneration (Esiri, Pearson, & Powell, 1986; Sinha,
Hollen, Rodriguez, & Miller, 1993) which can disrupt additional cognitive performance in
these individuals. There is also evidence from a longitudinal study (The Chicago Study) that
auditory comprehension of single words declines quite rapidly over a longitudinal time period
(Kaszniak & Wilson, 1985), suggesting that auditory word recognition performance in a gating
situation would likely evidence a similar sort of pattern.
5. SUMMARY & FUTURE DIRECTIONS
The present chapter has attempted to summarize the recent literature pertaining to
word recognition processes (both visual and auditory) in older adults and individuals with
Alzheimer's disease. The vast majority of the studies reviewed suggests that these very basic
cognitive processes are not totally spared in these individuals (e.g., see also Martin, 1992).
While some breakdowns exist in the sub-processes that influence word recognition
performance (i.e., dysfunctions in visual and auditory functioning for instance), older adults
and adults with AD are very adept at performing these and similar tasks (i.e., Parasuraman &
Nestor, 1993).
It appears very promising to include measures of visual and auditory word recognition
as regular components ofneuropsychological/assessment batteries. Given the paucity of recent
research regarding auditory word recognition, this area appears especially relevant for further
investigation. This enthusiasm stems from the fact that, in the literature reviewed for this
chapter, several studies have shown that the word recognition paradigms can be a very
238 F.R. Ferraro
important and sensitive diagnostic tool. Of course, results from a visual or auditory word
recognition experiment could not be the sole defining criteria for cognitive dysfunction.
However, the diagnostic values of these paradigm~ is related to the diagnostic value reaction
time paradimns (which visual and auditory word recognition fall) have had in both older adults
and adults with a variety of age-related diseases (Gordon & Carson, 1990; Mahurin &
Pirozzolo, 1986; 1993; Muller, Richter, Weisbrod, & Klingberg, 1991). Furthermore, Ferraro
and Balota (1993) and Ferraro and Sturgill (1994) have revealed how the lexicality effect
(difference between pseudowords and words) increases with age and disease status, and how
this fits nicely with the currently-popular theoretical mechanism of a failure to inhibit irrelevant
information in both healthy aging and dementia of the Alzheimer type. Although it could be
said that very little progress has been made in these areas with regard to aging and AD, given
the absololute number of studies reviewed here, it does seem correct in saying that the studies
that have been performed nicely indicate the validity and reliability of these paradimns in
studying basic, elementary cognitive processes in these populations. Future work can only
build on the nice foundation already constructed.
REFERENCES
Albert, M., & Milberg, W. (1989). Semantic processing in patients with Alzheimer's disease.
Brain & Language, 3 7, 163-171.
Allen, P. A., Madden, D. J., & Crozier, L. C. (1991). Adult age differences in letter-level and
word-level processing. Psychology & Aging, 6, 261-271.
Allen, P. A., Madden, D. J., Weber, T. A., & Groth, K. E. (1993). Influence of age and
processing stage on visual word recognition. Psychology & Aging, 8, 274-282.
Alzheimer, A. (1907). A characteristic disease of the cerebral cortex. In I~ Bick, L.
Ammaduci, & G. Pepeu (Eds., & Trans.) (1986). The early story of Alzheimer's disease.
Padua, Italy: Liviana Press.
Alzheimer's Disease & Related Disorders Association (1987). Chicago, IL.
American Psychiatric Association (1994). Diagnostic and statistical manual of mental
disorders (4th Ed.), Washington, DC: American Psychiatric Association.
Appell, J., Kertesz, A., & Fisman, M. (1982). A study of language functioning in Alzheimer
disease. Brain & Language, 17, 73-91.
Balota, D. A. (1993). Visual word recognition: The journey from features to meaning. In M.
A. Gemsbacher (Ed.), Handbook ofpsycholinguistics. NY: Academic Press.
Balota, D. A., & Chumbley, J. I. (1984). Are lexical decisions a good measure of lexical
access? The role of word frequency in the neglected decision stage. Journal of
Experimental Psychology: Human Perception & PerformanJe, 10, 340-357.
Balota, D. A., & Duchek, J. M. (1988). Age-related differences in lexical access, spreading
activation, and simple pronunciation. Psychology & Aging, 3, 84-93.
Balota, D. A., & Duchek, J. M. (1991). Semantic priming effects, lexical repetition effects,
and contextual disambiguation effects in healthy aged individuals and individuals with
senile dementia of the Alzheimer's type. Brain & Language, 40, 181-201.
Balota, D. A., & Ferraro, F. 1~ (1993). A dissociation of frequency and regularity effects in
pronunciation performance across young adults, older adults, and individuals with senile
dementia of the Alzheimer type. Journal ofMemory & Language, 32, 573-592.
Balota, D. A., Ferraro, F. 1L, & Connor, L. T. (1991). On the early influence of meaning in
word recognition: A review of the literature. In P. J. Schwanentlugel (Ed.), The
psychology of word meanings. NJ: LEA.
Balota, D. A., & Lorch, 1~ F. (1986). Depth of automatic spreading activation: Mediated
priming effects in pronunciation but not in lexical decision. Journal of Experimental
Psychology: Learning, Memory, & Cognition, 12, 336-345.
Baron, J., & Thurston, I. (1973). An analysis of the word-superiority effect. Cognitive
Becker, C. A. (1979). Semantic context and word frequency effects in visual word
recognition. Journal of Experimental Psychology: Human Perception & Performance, 5,
252-259.
Bell, T. S. (1989). Age-differences in spoken word identification: Effects oflexical density and
semantic context. Research on Speech Perception Progress Report # 15, 283-302.
Indiana University Speech Research Laboratory.
Berg. L. (1984). Clinical dementia rating. British Journal of Psychiatry, 145, 339.
Berg, L. (1988). Clinical dementia rating (CDR). Psychopharmacology Bulletin, 24, 637-
639.
Berg, L., McKeel, D. W., Miller, J. P., Baty, J., & Morris, J. C. (1993). Neuropathological
indexes of Alzheimer's disease in demented and nondemented people aged 80 years and
older. Archives of Neurology, 50, 349-358.
Berg, L., Miller, J. P., Storandt, M., et al. (1988). Mild senile dementia of the Alzheimer type:
2. Longitudinal assessment. Annals of Neurology, 23, 477-484.
Berg, L., Smith, D. S., Morris, J. C., et al. (1990). Mild senile dementia ofthe Alzheimer type.
3. Longitudinal and cross-sectional assessment. Annals of Neurology, 28, 648-652.
Bowles, N. L., & Poon, L. W. (1981). The effect of age on speed of lexical access.
Bowles, N. L., Obler, L. K., & Albert, M. L. (1987). Naming errors in healthy aging and
dementia of the Alzheimer type. Cortex, 23, 519-524.
Bowles, N. L., & Pooh, L. W. (1988). Age and context effects in lexical decision: An age by
context interaction. ExperimentalAging Research, 14, 201-205.
Burke, D. M., & Yee, P. L. (1984). Semantic priming during sentence processing by young
and older adults. Developmental Psychology, 20, 903-910.
Burke, W. J., Miller, J. P., Rubin, E. H., et al. (1988). Reliability of the Washington University
Clinical Dementia Rating (CDR). Archives of Neurology, 45, 31-32.
Burke, D. M., White, H., & Diaz, D. L. (1987). Semantic priming in young and older adults:
Cantor, M. H. (1991). Family and community: Changing roles in an aging society. The
Gerontologist, 31, 337-346.
83.
Cerella, J. (1990). Aging and information-processing rate. In J. E. Birren & I~ W. Schaie
(Eds.), Handbook of the psychology of aging (pp. 201-221). NY: Academic Press.
Cerella, J., & Fozard, J. L. (1984). Lexical access and age. Developmental Psychology, 20,
235-243.
240 F.R. Ferraro
Clark, E. O. (1980). Semantic and episodic memory impairment in normal and cognitively
impaired elderly adults. In L. K. Obler & M. L. Albert (Eds.), Language and
communication in the elderly. MA: Lexington Books.
Cohen, G., & Faulkner, D. (1983). Word recognition: Age differences in contextual
facilitation effects. British Journal of Psychology, 74, 239-251.
Coltheart, M., Davelaar, E., Jonasson, J., & Besaer, D. (1977). Access to the internal lexicon.
In S. Dornic (Ed.), Attention & performance VI. NJ: LEA.
Costa, P. T., Whitfield, J. 1L, & Stewart, D. (Eds.) (1989). Alzheimer's disease: Abstracts of
the psychological and behavioral literature. Washington, DC" American Psychological
Association.
Craig, C. H. (1992). Effects of aging on time-gated isolated word-reco~iiton performance.
Journal of Speech & Hearing Research, 35, 234-238.
Cronin-Golomb, A., Corkin, S., Rizzo, J. F., Cohen, J., Growdon, J. H., & Banks, K. S.
(1991). Visual dysfunction in Alzheimer's disease: Relation to normal aging. Annals of
Neurology, 29, 41-52.
Cummings, J. L. (1988). Dementia of the Alzheimer type: Challenges of definition and clinical
diagnosis. In H. Whitaker (Ed.), Neuropsychological studies in nonfocal brain damage
(pp. 86-107). NY: Springer.
Cummings, J. L., & Benson, D. F. (1989). Speech and language alterations in dmentia
syndromes. In A. Ardila & F. Ostrosky-Solis (Eds.), Brain organization of language and
cognitive processes (pp. 107-120). NY: Plenunl
DeGroot, A. M. B. (1989). Representational aspects of word imageability and word fxequency
as assessed through word associations. Journal of Experimental Psychology: Learning,
Memory, & Cognition, 15, 824-845.
Duchek, J. M., Balota, D. A., Ferraro, F. 1L, Gemsbacher, M. A., Faust, M. A., & Connor, U
T. (1992). The inhibition of irrelevant information in younger and older adults.
Presented at the XXV International Congress pfPsychology, Brussels, Belgium.
Elliott, U L., Hammer, M. A., & Evan, K. E. (1987). Perception of gated, highly familiar
spoken monosyllabic nouns by children, teenagers, and older adults. Perception &
Psychophysics, 42, 150-157.
Esiri, M. M., Pearson, 1L C. A., & Powell, T. P. S. (1986). The cortex ofthe primary auditory
area in Alzheimer's disease. Brain Research, 366, 385-387.
Ferraro, F. 1L & Balota, D. A. (1993). Semantic and identity priming in senile dementia of
the Alzheimer type (SDAT). Paper presented at the 65th annual meeting of the Midwestern
Psychological Association, Chicago, IL.
Ferraro, F. IL, & Kellas, G. (1992). Age-related changes in the effects of target orientation on
word recognition. Journal of Gerontology: PSYCHOLOGICAL SCIENCES, 47, 279-280.
Ferraro, F. IL, & Stur~ll~ D. (1994). Lexical properties and lexical effects associated with
National Adult Reading Test (NART) stimuli in healthy younger adults and healthy older
adults. Presented at the 5th Biennial Cognitive Aging Conference, Atlanta, GA.
Ferraro, F. 1L (1994). Word unit analysis during visual word recognition in young and elderly
adults. Developmental Neuropsychology, 1O, 13-17.
Forger, K. I. (1976). Accessing the mental lexicon. In 1L J. Wales & E. C. T. Walker (Eds.),
New approaches to language mechanisms. Am~erdam: North-Holland.
Friedland, 1L P. (1993). Alzheimer's disease: Clinical features and differential diagnosis.
Neurology, 43 (Suppl 4), $45-$51.
Aging, Alzheimer'sdisease, and word recognition 241
Friedman, 1L B., Ferguson, S., Robinson, S., & Sunderland, T. (1992). Dissociation of
mechanisms of reading in Alzheimer's disease. Brain & Language, 43, 400-413.
Gemsbacher, M. A. (1984). Resolving 20 years of inconsistent interactions between lexical
familiarity and orthography, concreteness, and polysemy. Journal of Experimental
Cfilmore, G. C., Groth, K. E., & Thomas, C. W. (under review). Stimulus contrast and word
reading in Alzheimefs disease. Submitted for publication.
Gordon, B., & Carson, I~ (1990). The basis of choice reaction time slowing in Alzheimefs
disease. Brain & Cognition, 13, 148-166.
Grainger, J. (1990). Word t~equency and neighborhood ~equency effects in lexical decision
and naming. Journal of Memory & Language, 29, 228-244.
Grosjean, F. (1980). Spoken word recognition processes and the gating paradigag Perception
& Psychophysics, 28, 267-283.
Hart, S., & Semple, J. M. (1990). Neuropsychology and the dementias. London: Taylor &
Francis.
Hasher, L., Stoltzfus, E. 1L, Zacks, 1L T., & gypma, B. (1991). Age and inhibition. Journal
of Experimental Psychology: Learning, Memory & Cognition, 17, 163-169.
Hasher, L., & Zacks, 1L T. (1988). Working memory, comprehension, and aging: A review
and a new view. In G. Bower (Ed.), The psychology of learning and motivation (pp. 193-
225). San Diego, CA: Academic Press.
Henderson, L. (1982). Orthographyand word recognition in reading. NY: Academic Press.
Howard, D. V. (1983). The effects of aging and degree of association on the semantic priming
oflexical decisions. Experimental Aging Research, 9, 145-151.
Howard, D. V., McAndrews, M. P., & Lasaga, M. I. (1981). Semantic priming of lexical
decisions in young and old adults. Journal of Gerontology, 36, 707-714.
Howard, D. V., Shaw, 1L J., & Heisey, J. G. (1986). Aging and the time course of semantic
Hufl~ F. J., Corkin, S., & Growdon, J. (1986). Semantic impairment and anomia in
Alzheimer's disease. Brain & Language, 28, 235-249.
Hughes, C. P., Berg, L., Danziger, W., Coben, L. A., & Martin, ILL. (1982). A new clinical
scale for the staging of dementia. British Journal of Psychiatry, 140, 566-572.
Humes, L. E., Ndson, I~ J., & Pisoni, D. B. (1991). Recognition of synthetic speech by
hearing-impaired elderly listeners. Journal of Speech & Hearing Research, 34, 1180-1184.
Humes, L. E., Nelson, I~ J., Pisoni, D. B., & Lively, S. E. (1993). Effects of age on serial
recall of natural and synthetic speech. Journal of Speech & Hearing Research, 36, 634-
639.
HuRon, J. T., Morris, J. L., Elias, J. W., & Poston, J.N. (1993). Contrast sensitivity
dysfunction in Alzheimer's disease. Neurology, 43, 2328-2330.
James, C. T. (1975). The role of semantic information in lexical decisions. Journal of
Jastrzembski, J. E. (1981). Multiple meanings, number of related meanings, frequency of
occurrence, and the lexicon. Cognitive Psychology, 13, 278-305.
Johnson, N. F. (1992). On the role of cohorts or neighbors in visual word recognition. In 1L
Frost, & L. Katz (Eds.), Orthography, phonology, morphology, & meaning (pp. 147-164).
NY: Elsevier Science Publishers.
242 F.R. Ferraro
Johnson, N. F., & Pugh, I~ 1L (1994). A cohort model of visual word recognition. Cognitive
Psychology 26, 240-330.
Jorm, A. F., Fratiglioni, L., & Winblad, B. (1993). Differential diagnosis in dementia: Principle
components analysis of clinical data t~om a population survey. Archives of Neurology, 50,
72-77.
Kaszniak, A. W., & Wilson, 1L S. (1985). Longitudinal deterioration of language and
cognition in dementia of the Ahheimer's type. Presented as part of a symposium at the
annual meeting of the International Neuropsychological Society, San Diego, CA.
Kellas, G., Simpson, G. B., & Ferraro, F. IL (1988). Aging and performance: A mental
workload analysis. In P. Whitney & 1L Ochsman (Eds.), Psychology & productivity (pp.
35-49). NY: Plenum Publishing Co.
Kinsboume, M. (1973). Age effects on letter span related to rate and sequential dependency.
Knoll, J. F., & Merves, J. S. (1986). Lexical access for concrete and abstract words. Journal
of Experimental Psychology: Learning, Memory, & Cognition, 12, 92-107.
Kurylo, D. D., Corkin, S., AUard, T., Zatorre, 1L J., & Growden, J. H. (1993). Auditory
function in Al~eimer's disease. Neurology, 43, 1893-1899.
Lopez, O. L., Swihart, A. A., Becker, J. T., Reinmuth, O. M., Reynolds, C. F., Rezek, D. L.,
& Daly, F. L. (1990). Reliability of NINCDS-ADRDA clinical criteria for the diagnosis of
Alzheimer's disease. Neurology, 40, 1517-1522.
Madden, D. J., & Greene, H. A. (1987). From retina to response: Contrast sensitivity and
memory retrieval during visual word recognition. Experimental Aging Research, 13, 15-
21.
Madden, D. J. (1986). Adult age differences in visual word recognition: Semantic encoding
and episodic retention. Experimental Aging Research, 12, 71-78.
Madden, D. J. (1988). Adult age differences in the effects of sentence context and stimulus
degradation during visual word recognition. Psychology & Aging, 3, 167-172.
Madden, D. J. (1989). Visual word recognition and age-related slowing. Cognitive
Madden, D. J. (1992). Four to ten milliseconds per year: Age-related slowing of visual word
identification. Journal of Gerontology: PSYCHOLOGICAL SCIENCES, 47, 59-68.
Madden, D., Nebes, R. D., & Allen, P. A. (1992). Cognitive slowing in Alzheimer's disease as
a fimction of task type and response type. Developmental Neuropsychology, 8, 459-471.
Madden, D. J., Pierce, T. W., & Allen, P. A. (1993). Age-related slowing and the time course
of semantic priming in visual word recognition. Psychology & Aging, 8, 490-507.
Mahurin, 1L IC, & Pirozzolo, F. J. (1986). Chronometric analysis: Clinical applications in
aging and dementia. Developmental Neuropsychology, 2, 345-362.
Mahurin, 1L I~, & Pirozzolo, F. J. (1993). Application of Hick's law of response speed in
Alzheimer and Parkinson's disease. Perceptual & Motor Skills, 77, 107-113.
Malec, J. F., Invik, 1L J., & Smith, G. E. (1993). Neuropsychology and normal aging: The
clinician's perspective. In 1L W. Parks, 1L E. Zec, & 1L S. Wilson (Eds.),
Neuropsychology of Ahheimer's disease and other dementias (pp. 81-111). NY: Oxford
University Press.
Marcel, A. J., & Patterson, K. E. (1980). Word recognition and production: Reciprocity in
clinical and normal studies. In J. gequin (Ed.), Attention & performance (Vol 7). NY:
Halstead Press.
Margolin, D. I. (1992). Probing the multiple facets of human intelligence: The cognitive
neuropsychologist as clinician. In D. I. Margolin (Ed.), Cognitive neuropsychology in
clinicalpractice (pp. 18-40). NY: Oxford University Press.
Margolis, 1L B., Taylor, J. M., & Dunn, E. J. (1985). An abbreviated Speech Sounds
Perception Test with a geriatric population. International Journal of Clinical
Matin, O. S. M. (1987). Dementia and visual agnosis. In G. W. Humphreys & M. J. giddoch
(Eds.), Visual object processing (pp. 261-280). NJ: LEA.
Marslen-Wilson, W. D., & Welsh, A. (1978). Processing interactions and lexical access
during word recognition in continuous speech. Cognitive Psychology, 10, 29-63.
Martin, A. (1990). Neuropsychology of Alzhdmer's disease: The case for subgroups. In M. F.
Schwartz (Ed.), Modular deficits in Alzheimer's disease (pp. 143-175). Boston: MIT
Press.
Martin, A. (1992). Semantic knowledge in patients with Alzheimer's disease: Evidence for
degraded representations. In L. Backmann (Ed.), Memory functioning in dementia (pp.
119-134). NY: Elsevier Science Publishers, B. V.
Martin, A., & Fedio, P. (1983). Word production and comprehension in Alzheimer's disease:
The breakdown of semantic knowledge. Brain & Language, 19, 124-141.
Max, W. (1993). The economic impact of Alzheimer's disease. Neurology, 43 (Suppl. 4), S6-
S10.
McClelland, J. L., & Rumelhart, D. E. (1981). An interactive activation model of context
effects in letter perception: Part 1. An account of basic findings. Psychological Review,
88, 375-407.
McCulla, M. M., Coats, M., VanFleet, N., Duchek, J. M., Grant, E., & Morris, J. C. (1989).
Reliability of clinical nurse specialists in the staging of dementia. Archives of Neurology,
46, 1210-1212.
McDowd, J. M., & Oseas-Krueger, D. M. (1991). Aging, inhibition processes, and negative
priming. Journal of Gerontology: PSYCHOLOGICAL SCIENCES, 46, 340-345.
McKeel, D. W., Ball, M. J., Price, J. L., Smith, D. S., Miller, J. P., Berg, L., & Morris, J. C.
(1993). Interlaboratory histopathologic assessment of Al~eimer neuropathology:
Different methodologies yield comparable diagnostic results. Alzheimer's Disease and
Associated Disorders, 7, 136-151.
McKhann, G., Drachman, D., Folstein, M., Katzman, 1L, Price, D., & Stadlan, M. (1984).
Clinical diagnosis of Alzheimer's disease: Report of the NINCDS-ADRDA Group under
the auspices of the Department of Health and Human Service Task Force on Alzheimer's
disease. Neurology, 34, 39-44.
Morris, J. C. (1993). The Clinical Dementia Rating (CDR): Current version and scoring rules.
Neurology, 43, 2412-2414.
Morris, J. C., McKeel, D. W., Fulling, I~, Torack, 1L, & Berg, L. (1988). Validation of the
clinical diagnostic criteria for Alzheimer's disease. Annals of Neurology, 24, 17-22.
Morris, J. C., McKeel, D. W., Price, J. L., et al. (1988). Very mild senile dementia of the
Alzheimer type (SDAT). Neurology, 38, 227.
244 F.R. Ferraro
Morris, 1L G., Gick, M. L., & Craik, F. I. M. (1988). Processing resources and age differences
in working memory. Memory & Cognition, 16, 362-266
Morrison, M. H. (1982). Economics of aging: The future of retirement. NY: Van Nostrand
Reinhold.
Morton, J. (1969). The interaction of information in word recognition. Psychological Review,
76, 165-178.
Muller, G., Richter, R. A., Weisbrod, S., & Klingberg, F. (1991). Reaction time prolongation
in the early stage of presenile onset of Alzheimer's disease. European Archives of
Psychiatry and Clinical Neuropsyehology, 241, 46-48.
Myerson, J., Ferraro, F. R., Hale, S., & Lima, S. D. (1992). General slowing in semantic
priming and word recognition. Psychology & Aging, 7, 257-270.
Nebes, IL D. (1992). Cognitive dysfimetion in Al~eimel~S disease. In F. I. M. Craig & T. A.
Salthouse (Eds.), Handbook of cognition and aging (pp. 373-445). NJ: LEA.
Nebes, 1L D., & Brady, C. B. (1992). Generalized cognitive slowing and severity of dementia
in Alzheimer's disease: Implications for the interpretation of response-time data. Journal
of Clinical and Experimental Neuropsychology, 14, 317-326.
Nebes, 1L D. (1989). Semantic memory in Alzheimer's disease. Psychological Bulletin, 106,
377-394.
Nebes, 1L D., Brady, C. B., & HutZ F. J. (1989). Automatic and attentional mechanisms of
semantic priming in Alzheimer's disease. Journal of Clinical and Experimental
Neely, J. H. (1991). Semantic priming effects in visual word recognition: A selective review of
the current findings and theories. In D. Besner & G. Humphreys (Eds.), Basic processes in
reading: Visual word recognition (pp. 264-336). NJ: LEA.
Ober, B. A., & Shenaut, G. IC (1989). Lexical decision and priming in Alzheimer's disease.
Owsley, C., & Sloane, M. E. (1990). Vision and aging. In. F. Boller & J. Grafman (Eds.),
Handbook ofneuropsychology, Vol. 4. (pp. 229-249). NY: Elsevier Publishers.
Parasuraman, R., & Nestor, P. G. (1993). Preserved cognitive operations in early Alzheimer's
disease. In J. Cerella, J. Rybash, W. J. Hoyer, & M.L. Commons (Eds.), Adult information
processing: Limits on loss (pp. 77-111). NY: Academic Press.
Pate, D. S., Margolin, D. I., Friedrich, F. J., & Bentley, E. E. (1994). Decision-making and
attentional processes in aging and dementia of the Alzheimer's type. Cognitive
Patterson, I~, Graham, N., & Hodges, J. 1L (1994). Reading in Alzheimer's type dementia: A
preserved ability? Neuropsychology, 8, 395-407.
Pirozzolo, F. J., Nolan, B. H., Kushkowski, M., Mortimer, J. A., & Maletta, G. J. (1988).
Latency and accuracy of word recognition in dementia of the Alzheimer type. Alzheimer's
Disease and Associated Disorders, 2, 337-341.
Pooh, L. W., Messner, S., Martin, P., Noble, C. A., Clayton, G. M., & Johnson, M. A. (1992).
The influences of cognitive resources on adaptation and old age. In L. W. Poon (Ed.), The
Georgian centenarian study (pp. 31-46). Amityville, NY: Baywood Publishing.
Powell, A. L. (1994). Senile dementia of extreme aging: A common disorder of centenarians.
Dementia, 5, 106-109.
Powell, D. H., & Whitla, D. K. (1994). Normal cognitive aging: Towards empirical
perspectives. Current Directions in Psychological Science, 3, 27-31.
Price, J. L., Davis, P. B., Morris, J. C., & White, D. L. (1991). The distribution of tangles,
plaques, and related immuniohistochemical markers in healthy aging and Alzheimer's
disease. Neurobiology of Aging, 12, 295-312.
Rocca, W. A., Amaducci, L. A., & Schoenberg, B. S. (1986). Epidemiology of clinically
diagnosed Alzheimer's disease. Annals of Neurology, 19, 415-424.
Rubenstein, H., Garfield, L., & Millikan, J. A. (1970). Homographic entries in the internal
lexicon. Journal of Verbal Learning & Verbal Behavior, 9, 487-494.
Salasoo, A., & Pisoni, D. (1985). Interaction of knowledge sources in spoken word
recognition. Journal of Memory & Language, 24, 210-231.
Salthouse, T. A. (1982). Adult cognition: An experimental psychology of human aging. NY:
Springer.
Seidenberg, M. S., Andersen, E. S., Kempler, D., & Jackson, C. (1994). Do word frequency
and consistency effects dissociate with aging? Paper presented at the 5th Biennial
Cognitive Aging Conference, Atlanta, GA.
Seidenberg, M. S., & McCleHand, J. L. (1989). A distributed, developmental model of word
recognition and naming. Psychological Review, 96, 523-568.
Shuttleworth, E. C., & Huber, S. J. (1988). The naming disorder of dementia of Alzheimer
type. Brain & Language, 34, 222-234.
Simpson, G. B. (1984). Lexical ambiguity and its role in models of word recognition.
Psychological Bulletin, 96, 316-340.
Sinha, U. I~, Hollen, I~ M., Rodriguez, 1L, & Miller, C. A. (1993). Auditory system
degeneration in Alzheimer's disease. Neurology, 43, 779-785.
Stanovich, I~ E. (1980). Toward an interactive-compensatory model of individual differences
Storandt, M., Morris, J. C., Rubin, E. H., Coben, L. A., & Berg, L. (1992). Progression of
senile dementia of the Alzheimer type on a battery of psychometric tests. In L. Backman
(Ed.), Memory functioning in dementia (pp. 207-226). NY: Elsevier Publishers.
Terry, 1L, & Katzman, 1L (1983). Senile dementia of the Alzheimer type: Defining a disease.
In 1L Katzman & R. Terry (Eds.), The neurology of aging (pp. 51-84). Philadelphia, PA:
F. A. Davis Co.
Tipper, S. P. (1991). Less attentional selectivity as a result of declining inhibition in older
adults. Bulletin of the Psychonomic Society, 29, 45-47.
TippeR, L. J., & Farah, M. J. (1994). A computational model of naming in Alzheimer's
disease: Unitary or multiple impairments. Neuropsychology, 8, 3-13.
Trobe, J. D., & Butter, C. M. (1993). A screening test for integrative visual dysfunction in
Al~eimer's disease. Archives of Ophthalmology, 111, 815-818.
Tomlinson, B. E. (1982). Plaques, tangles, and Alzheimer's disease. Psychological Medicine,
12, 449-459.
Tomlinson, B. E., & Henderson, G. (1976). Some quantitative cerebral findings in normal and
demented old people. In R. D. Terry & S. Gershon (Eds.), Neurobiology of aging (pp.
183-204). NY: Raven.
Tyler, L. (1984). Structure of the initial cohort: Evidence form gating. Perception &
Psyehophysics, 36, 417-427.
Tyro, E. L. (1989). Diagnostic assessment in dementia. In Katona, C. (Ed.), Dementia
disorders (pp. 18-43). London: Chapman & Hill.
246 F.R. Ferraro
Walley, A. C., & Metsala, J. L. (1990). The growth of lexical constraints on spOken word
recognition. Perception & Psychophysics, 47, 267-280.
Wayland, S. C., Wingfield, A., & Goodglass, H. (1989). Recognition of isolated words: The
dynamics of cohort reduction. Applied Psycholinguistics, 10, 475-487.
Williamson, G. M., & Schulz, K (1993). Coping with specific stressors in Alzheimer's disease
caregiving. The Gerontologist, 33, 747-753.
Wingtield, A., & Stine, E. A. L. (1991). Expert systems in nature: Spoken language
processing and adult aging. In J. D. Sitmott & J. C. Cavanaugh (Eds.), Bridging
paradigms: Positive development in adulthood and cognitive aging (pp. 237-258). NY:
Preager.
Wingfield, A., Aberdeen, J. S., & Stine, E. A. L. (1991). Word onset gating and linguistic
context in spoken word recognition by young and elderly adults. Journal of Gerontology:
PSYCHOLOGICAL SCIENCES, 46, 127-129.
Wingfield, A., Goodglass, H., & Smith, I~ L. (1990). Effects of word-onset cuing on picture
naming in aphasia: A reconsideration. Brain & Language, 39, 373-390.
Wisniewski, H. M., & Merz, G. S. (1985). Neuropathology ofthe aging brain and dementia of
the Alzheimer's type. In C. M. Gaitz & T. Samorajski (Eds.), Aging 2000: Our health care
destiny: Vol. 1. Biomedical issues (pp. 231-243.). NY: Springer-Verlag.
Semantic priming in Alzheimer's disease: Meta-analysis and theoretical

evaluation*
Beth A. Ober a'b and Gregory I~ Shenaut a'b
aDivision of Human Development, University of California, Davis
bVA Northern California System of Clinics, Pleasant Hill, CA
Whether semantic memory is intact in probable Alzheimer's disease (AD) has been the
subject of lively debate since the mid-1980's. There is ample evidence of impaired
performance on a variety of tasks requiring the access and utilization of semantic knowledge
(world knowledge), including knowledge of words and objects. However, the extent to which
this impaired performance is due to semantic memory deficits per se, as opposed to deficits in
retrieval mechanisms, strategy implementation, deployment of attention, etc., has become the
focus of much theoretical discussion and research activity. (For a thorough review of the
literature on semantic memory and AD, see Nebes, 1989; for an update, see Nebes, 1992).
The lexical priming paradigm has frequently been used as a tool for assessing the
intactness of semantic memory in AD. In this paradigm, the effect of related context on
reaction time (RT) to pronounce a word or recognize a word (in a mixed list of word and
nonword targets) is measured. The reduction in target RT produced by a preceding related
context word (related prime) compared to an unrelated context word (unrelated prime) is
known as the semantic priming effect. The most widely accepted explanation of the semantic
priming effect is as follows: the spread of activation in the semantic memory network, from
the prime's concept node to the related target's concept node, increases the activation level of
the target node, allowing more rapid access, matching, and/or retrieval processes for that node.
If associative connections between related concepts have been weakened or eliminated as a
by-product of the neuropathology associated with AD, then one would expect to see
siL-,nificantly less semantic priming in AD than in elderly normal (EN) subjects. (Unless
otherwise indicated, the term "normal" in this chapter refers to EN subjects.) It~ however, the
semantic representations for the concepts have become degraded in some way by AD, the
result would be either: (1) less-than-normal or even zero priming effect (if the concept nodes
are so badly degraded that spreading activation has a less-than-normal or null effect on level of
activation for that concept node), or (2) greater-than-normal priming (if partially degraded
concept nodes have "moreto gain" via spread of activation).
AUTHORNOTES: Correspondenceconcerningthis chapter shouldbe sent to Beth ~ Ober, HumanDevelopment,

Department of Applied Behavioral Sciences, U.C. Davis, Davis, CA, 95616 or to the following e-mail address:
baober@ucdavis.edu. Portionsof this meta-analysiswerepresentedat the AmericanPsychologicalSocietymeetingin
San Diego, June, 1992. B. A. O ~ s research is supported by the Medical Research Service of the Veterans
Administration and by the National Institute on Aging (Grant #R29-AG10848 to B. A~ O~r and Grant
#P30-AG10129to W.J. Jagust).
248 B.A. Ober and G.K. Shenaut
The results of semantic priming experiments with AD and EN subjects are mixed. A
number of experiments have shown equivalent-to-normal AD priming; investigators in these
cases have generally concluded that semantic memory structures and processes are not
significantly disrupted in AD (e.g., Nebes, Martin, & Horn, 1984; Ober, Shenaut, Jagust, &
Stillman, 1991). However, there are quite a few studies which show si,.~nificantly
greater-than-normal AD priming (i.e., hyperpriming). Investigators do not agree in their
interpretation of these results. Hartman (1991) for example, concluded that attentional
abnormalities which affect the utilization of semantic knowledge are responsible for
greater-than-normal AD priming. Chertkow, Bub, and Seidenberg (1989) on the other hand,
argued that hyperpriming is indicative of degraded representations for those concepts in
semantic memory that are used as primes and targets in the lexical priming paradigms.
In this chapter, we present a critical review of theory and methods relevant to research
on lexical, semantic priming in AD compared to EN subjects. We also present several types of
meta-analyses on the data from all available AD-EN semantic priming experiments.
1. STUDIES S U R V E Y E D
We made every effort to find all of the lexical, semantic priming experiments with AD
and EN subjects, through June 1993. Both PsychINFO and Medline searches were conducted
to supplement our '~oy hand" literature searches. We also sent letters of inquiry to 20 of our
colleagues who had authored or co-authored published papers, or delivered papers at
conferences, on semantic priming (or closely related topics) in AD and normal aging in an
attempt to obtain any new AD priming data sets that we would not otherwise be aware of.
Sixteen of these letters resulted in a written response, and we obtained several "in press" or
unpublished data sets in response to these letters.
A total of 22 AD lexical, semantic priming experiments were available to us for the
meta-analysis. One of these experiments--Ober and Shenaut---(1988), showed pronounced
negative priming for the AD (but not the EN subjects) and was an outlier (more than 2.00 SD
below the mean of all experiments) on each of the four priming effect (PE) measures involving
the AD subjects: PE for AD subjects, the difference in PE between AD and EN subjects,
percent PE (PE divided by unrelated RT x 100) for AD subjects, and the difference in percent
PE between AD and EN subjects. Therefore, this experiment was dropped from further
analyses.* The remaining 21 experiments represent seven different research laboratories, and
include 13 independent samples of AD and EN subjects. Table 1 provides a detailed summary
of these experiments, including information about dementia severity for the AD subjects (all
AD samples were mild-to-moderate in dementia severity, with only the highest fimctioning of
The dropped experiment involved continuous presentation of stimuli, with a low proportion of related pairs, and
would therefore have been categorized as an "automatic" priming experiment, if it had been included in Table 1. The
mean AD PE was -59 ms and the mean EN PE was 25; the difference in PE between the two groups (9 AD subjects
and 15 EN subjects) was significant, but in the opposite direction from the hyperprimingexperiments in our survey.
The AD subjects in this experiment comprised an independent sample from any other studies in the Ober & Shenaut
laboratory, they were mild-to-moderatelyimpaired, with an average score of 110 (out of a maximum score of 144) on
the Mattis DementiaRating Scale (Mattis, 1976). Nebeset al. (1989, Experiment 1, row 1 of Table 1) was a positive
outlier (greater than 2 SD) on absolute AD PE and on the difference in absolute PE between the AD and EN groups;
however, this experiment was well within 2 SD of the mean on percent AD PE and on the difference in percent PE
between the AD and EN groups. Therefore,this experimentwas retained. None of the other experiments in this survey
were outliers on any of the four PE measures.
Table 1
S u m m a r y o f M e t h o d s and Findings for 21 Semantic Priming E x p e r i m e n t C o n d i t i o n s with Alzheimer's Disease ( A D ) and Elderly N o r m a l
( E N ) Subject G r o u p s I
Dementia 3 N uRT PE
Citation 2 Index Paradigm 4 s o A5 R.p. 6 #Tfifls 7 AD EN AD EN AD EN C/A 8
I I
1. N e b e s e t a l . (1989, Exp.1) MMS=20.0 L-pairs 750 .50 15 16 16 1374 > 727 200 > 65
2. Chertkow et al. (1989) MMS=17.5 L-pairs 1502 .50 50 6 10 1255 > 831 141 > 25
3. Margolm (1987/1988) CDR=.95 L-palrs 600 .67 20 11 36 1147 > 698 121 > 21
4. Nebes et al. (1989, Exp.2) MMS=20.0 P-parrs 750 .30 15 16 16 874 > 574 118 > 22
5. Chertkow et al. (1993, Exp. 1) MMS=25.6 L-parrs I186 .33 25 14 22 838 > 595 94 > 29 e3
6. Hartman (1991) MMS=19.5 P-parrs 1397 .67 20 24 31 1065 > 785 74 > 40
7. Chertkow et al. (1993, Exp.2) MMS=25.6 L-parrs 500 .33 25 12 21 845 > 607 66 > 31
8. Chertkow et al. (1994) M2VIS=23.0 L-parrs 1158 .33 25 48 25 773 > 582 54 > 27 ~.~.
9. Albert & Milberg (1989)9 Mattis=120 L-parrs 1500 .50 15 10 10 987 > 622 43 < 40
10. B.a]ota etal. (~1991~) CDR=.92 P-parrs 1119 .63 96 ii
32 32 698 > 556 23 > 11 t,..,.
~
11. Chertkow et al. (1993, Exp.3) MMS=25.6 L-parrs 250 .33 25 14 24 866 > 645 53 = 29
12. Ober et al. (1991 a, Exp.4) MMS=19.3 L-parrs 250 17 20 17 19 773 > 665 51 = 28
13. Ober et al. (1991a, Exp.2) MMS=19.3 L-parrs 250 17 12 17 20 739 > 635 48 = 25
14. Ober et al. (1991a, Exp.1) MMS=19.3 P-pmrs 250 17 12 17 20 692 > 599 26 = 22
t~
15. Ober & Shenaut (1990, Exp. 1) MMS=20.4 L-cont. 1500 .36 48 16 21 529 = 505 26 = 24
16. Ober & Shenaut (1990, Exp.2) MMS=20.3 L-cont. 2000 .36 48 16 15 644 > 523 26 = 25
17. Ober et al. (1991a, Exp.6) MMS=19.3 L-pairs 250 17 40 16 19 754 > 651 24 = 7
18. Nebes et al. (1984) MMS=19.0 P-cont. 2000 .25 20 20 20 709 > 596 22 = 19
19. Ober et al. (1991a, Exp.5) MMS=19.3 P-pairs 250 .17 40 17 19 685 > 600 21 = 12
20. Ober et al. (1991b) MMS=21.7 L-cont. 1500 .27 72 17 17 561 > 463 10 = 18
21. Ober et al. (1991a~ Exp.3) MMS=19.3 P-pairs 250 .17 20 17 20 694 > 616 -1 = 20
bO
N refers to number of subjects, uRT refers to unrelated prime condition RT, PE refers to priming effect, AD
refers to Mzheimer's disease subjects, and EN refers to elderly normal subjects. The >, <, and = signs for uRT
and PE refer to statistical significance (and direction) versus statistical equivalence for the group differences.
: Each citation pertains to one priming experiment. The experiments within each of the following three sets of
citations utilized the same subjects: 1 & 4; 5, 7, & 11; 12-14, 17, 19, & 21. M1 other citations involved
independent subject samples, thus there are 13 independent subject samples among the 21 experiments. Ober
et al. (1991a) refers to Ober, Shenaut, Jagust, and Stillman (1991); Ober et al. (1991b) refers to Ober, Shenaut,
and Nelson-Abbott (1991). These abbreviated citations are used only in this table. The uRT (RT to targets
preceded by unrelated primes) and PE (priming effect) data for citations 12, 17, 19, & 21 (from Ober et al.
1991a) are averaged across the different types of category relations which were included among the related
pairs in those experiments.
3 MMS refers to the Mini-Mental Status exam (Folstein et al., 1975); higher scores indicate higher functioning
(the maximum score is 30); the average MMS scores of 17.5-25.6 are indicative of AD samples with
mild-to-moderate degrees of dementia severity. CDR refers to Clinical Dementia Rating (Hughes et al., 1982);
a higher rating indicates lower functioning (only scores of .5, 1, 2, or 3 are given to each patient and the scale
is not quantitative); the mean ratings which are close to 1.00 for rows 3 and 10 indicate AD samples with mild
dementia. Mattis refers to the Mattis Dementia Rating Scale (Mattis, 1976); higher scores indicate higher
functioning (the maximum score is 144); the mean score of 120 for the AD sample in citation 9 is indicative of
mild dementia.
4 L = lexical decision; P = pronunciation; pairs = a prime-target pair on each trial; cont. = continuous lexical
decision or pronunciation, with each trial consisting of a single item. Experiments in rows 15 and 20 used
single-choice (button press to real words only) lexical decision.
In the experiments summarized in rows 2, 5, 6, 8, & 10, subjects responded to the prime as well as the target;
SOAs were therefore uncontrolled and were longer for the AD compared to EN subjects. In these cases, the
average, experiment-wide SOA was estimated based on either (a) the mean prime RT (across subject groups)
and the blank time between prime response and target presentation or (b) estimated mean prime RT (across
subject groups) using the average RT across related and unrelated targets and the blank time between prime
response and target presentation. For the continuous priming experiment in row 18, the blank time was not
controlled precisely; 2000 ms is an estimated SOA.
6 R.P. stands for relatedness proportion, the proportion of word-word prime-target pairings that are
semantically related. In experiments which included pairs with neutral primes such as XXXXX or BLANK,
these pairs were not counted as word-word pairings. For row 10 (Balota & Duchek) the semantic priming
manipulation was embedded in a word-triplet homograph priming manipulation, with the first two words of
each triplet being the prime and target for semantic priming. Five out of every eight two-word sequences,
(first-second and second-third word sequences) were semantically related. For continuous priming paradigms,
the number of sequences of two related words, divided by the total number of two-word sequences was used as
the relatedness proportion. In the continuous, pronunciation paradigm (row 18), the number of two-word
sequences (the denominator for relatedness proportion) equals the number of trials minus 1. However, in the
continuous, lexical decision paradigms (rows 15, 16, & 20), because 50% of the trials are words and 50% are
nonwords, the number of word-word sequences equals 25% of the number of trials minus 1. For example, in
the experiment of row 20, 15 word-word sequences were included among 60 individually presented stimuli (30
words and 30 nonwords); 4 of these 15 word-word sequences were related pairs, yielding a relatedness
proportion of .27.
7 This column contains the number of data points, per subject, on which the PE is based. Thus, for row 1, there
were 15 related prime-target pairs and 15 unrelated prime-target pairs, with each target word serving as its own
control for priming. In Balota and Duchek (1991; row 10), the 96 total trials were actually comprised of 32
unique trials that were given three times, in separate blocks.
s The experimental conditions labeled "C" met all three of the following criteria for controlled priming
paradigms: pairwise-priming, long SOA (500 ms or greater), and a relatedness proportion of .25 or greater.
M1 other experimental conditions have been labeled "A" for Automatic. See the text for details.
9 This is the only experiment out of the 10 experiments meeting the criteria for controlled processing that did
not show significantly greater priming for AD compared to EN subjects. Mthough Mbert & Milberg report a
significant interaction of group by prime type, with AD patients showing less priming than the controls, their
ANOVA was done on log-transformed RTs. In fact, the PE of the AD group is slightly greater than that of the
EN group.
Semantic priming in Alzheimer's disease 251
the moderate severity individuals included), type of priming paradigm (lexical decision or
pronunciation; pairwise or continuous presentation of primes and targets), SOA, relatedness
proportion, number of trials from which the mean priming effect was calculated, number of AD
and EN subjects, mean RT for the unrelated prime condition for each group, and mean PE for
each group. Each experiment was classified as employing a "controlled" or "automatic"
priming paradigm, on the basis of variables known (at least, for young normal subjects) to
influence the likelihood of attentional priming mechanisms contributing to the overall priming
effect (in addition to automatic spread of activation); this will be discussed in detail later. We
placed all of the "controlled" (C) priming experiments in the upper section of the table and all
of the "automatic" (A) priming experiments in the lower section of the table. Within each of
these sections, we ordered the experiments according to the magnitude of the PE for the AD
subjects, since this is the main variable of interest.
For the 18 experiments with mean Mini-Mental Status scores (MMS; Folstein, Folstein,
& McHugh, 1975) available for the AD samples, the correlation coefficient for MMS and
unrelated RT w a s - . 0 9 , for MMS and PE was .02, and for MMS and the AD-EN difference in
PE was -.02. The lack o f association between dementia severity and the RT-based measures
was most likely due to the limited range of dementia severity represented by these subject
samples.
The priming stimuli used in these experiments were generally described by the authors
as highly associated stimulus-response pairs from published word association norms
(homograph norms in the case of row 10 in Table 1); however, word frequency and
association values for related pairs were not given for many of the experiments. Word
association norms include pairs which are related purely by association (contiguity, part-whole,
object-fimction, etc.) and pairs that are related by semantic category as well as by association.
The related pairs in the six experiment series of Ober et al. (1991) were: highly associated a n d
from the same semantic category (rows 13 and 14 in Table 1); category names and category
instances given by at least 50% of the normative subjects to the given category name (rows 12
and 21); or taken from typicality norms with related pair members belonging to the same
semantic category, but having unknown association values (rows 17 and 19).*
Glosser and Friedman (1991) reported an absence of significant priming for AD subjects on a threshold oral reading
task (in which number of correctly read target words is the dependent measure), when the relationship between the
prime and target is semantic and nonassociative (that is the prime does not elicit the target in word association tasks,
e.g, apple - peach). In contrast, when the relationship was only associative, or both associative and semantic, the AD
patients showed significant priming The authors conclude that the nonsemantic associative network (organized via
co-occurrence in language) is intact in AD, and that previously reported abnormalities in semantic priming for AD
subjects may have been due to abnormalities in the processing of semantic relationships. It should be noted, however,
that the targets preceded by related primes were more likely to be correctly identified in each of the three conditions
(associative nonsemantic, semantic nonassociative, and associative semantic) for AD and control subjects; the priming
effect in the semantic nonassociative condition was the only one out of those six conditions that did not reach
significance for the AD. Furthermore, in the same journal issue containing the Glosser and Friedman study, we
reported on equal-to-normal AD priming for prime-target pairs selected, not from association norms, but from
category norms (O~r et al., 1991; Experiments 3-6). Related pairs were selected on the basis of a
superordinate-subordinate (or visa versa) relationship or an intra-category relationship (in the later case, typicality
norms were used). It is interesting to note that the semantically related pairs which were least likely to have any
associative relationship because they were both atypical members of their r e ~ v e semantic category, showed the
greatest priming effects for the AD patients. The extent to which semantic versus lexical-associative processes are
spared versus impaired in AD is a question clearly in need of further investigation. Of course, the extent to which the
2. DEGRADED CONCEPTS IN SEMANTIC MEMORY
2.1. Evidence for Degraded Concepts
Chertkow et al. (1989, row 2 in Table l) gave the AD patients who participated in their
semantic priming study numerous "off-line" tests of semantic knowledge for 150 picturable
items. They were able to identify sets of items for each AD subject that showed relatively
intact versus relatively degraded semantic knowledge. The interaction between priming effect
and intactness was significant; the degraded items yielded a priming effect that was almost five
times larger than the priming effect for the intact items; further, the unrelated RT was about
35% longer for the degraded items. Chertkow et al. (1989) argued that--although overall
slowing may account for some of the difference in priming effect between the two types of
items and, in turn, between the AD and EN groups--the hyperpriming effect is mainly due to
semantic memory degradation.. They proposed two mechanisms for the effect of degradation
on priming: (1) degraded concepts in semantic memory have more to gain from spreading
activation than do relatively intact concepts (an automatic, pre-lexical mechanism), and (2)
degraded concepts in semantic memory can benefit more from a semantic-matching process
than relatively intact concepts (an attentional, post-lexical mechanism). In one of their more
recent studies, Chertkow and colleagues (Chertkow et al., 1994; row 8 in Table 1) showed
hyperpriming for 48 AD subjects, and then looked at a subgroup of AD subjects with PEs
greater than 60 ms. This subgroup of 20 AD subjects was more anomic and more impaired on
verbal fluency tasks than the remaining 28 AD subjects in the total sample. The authors again
concluded that the AD hyperpriming may be related to degradation of semantic memory.
Martin (Martin & Fedio, 1983; Martin, 1987; Martin, 1992a; Martin, 1992b) has also
proposed semantic memory degradation as an explanation for hyperpriming in AD. Martin has
argued that the nature of the impairments shown by AD patients on tasks requiring the use of
semantic knowledge is more consistent with degraded representations than with either
completely normal representations or total absence of certain representations. Many
researchers have documented that AD patients show deficits on verbal fluency, object naming,
and other tasks dependent on semantic knowledge (for a review, see Nebes, 1989). Martin
described the types of errors that AD patients make in object naming (the errors are often
names o f objects from the same category as the presented object or the name of the category)
and the disparity between AD patients' ability to demonstrate knowledge about superordinate
category membership for given objects and their difficulty in answering questions about
specific attributes of objects. He took these types of findings as evidence that AD patients
suffer from a loss of knowledge at a particular level in the hierarchy of the semantic knowledge
network--the level of basic objects. More specifically, Martin argued that the loss of attribute
knowledge for basic objects causes difficulties in distinguishing same-category, related objects
from one another. In other words, the representation for a particular object (e.g., hammer)
may be underspecified with regard to that object's attributes, but this same representation
would then be overgeneralized in relation to same-category, related objects (e.g., saw,
screwdriver). Martin (1987) and Chertkow et al. (1989) indeed showed that the objects which
a given AD patient cannot name are also the objects for which the patient has the greatest
lexical network is Independent of the semantic network is a question which has not been resolved in the normal
memoryand languageliterature.
difficulty when given attribute knowledge questions. Hodges, Salmon, and Butters (1992)
recently reported a si,~nificant correspondence between individual items on which AD patients
show errors across tests of picture naming, word-picture matching, picture-sorting, etc., and
they too, favored a degraded storage explanation of AD semantic memory deficits.
How does Martin's degraded store hypothesis account for hyperpriming? The idea is
that semantically similar objects share many attributes, and as these attributes are randomly lost
due to the pathology of AD, the representations of these related objects become more similar
to one another. In the case of object naming, this would presumably cause more than one
lexical entry to be activated to the presentation of one of these objects (for example, when the
subject sees a hammer, the conceptual representations for screwdriver, wrench, etc., and their
lexical entries might also be activated). In a semantic priming experiment, this would cause
increased activation of a target concept that was closely related to the prime concept (causing
semantic priming to resemble identity priming).
2.2. Difficulties with the Degraded Store Hypothesis
We believe that there are two major difficulties with the degraded store hypothesis as
an adequate explanation of all available experimental findings on AD priming. The first
difficulty has to do with the fact that si~ificantly greater priming for AD compared to EN
subjects was found for only nine out of the 21 priming experiments in Table 1. Automatic
spread of activation is the process which is supposedly affected by degradation of concepts;
spread of activation occurs in all priming experiments. Therefore, hyperpriming is the
predicted outcome for all priming experiments, at least in the Chertkow et al. (1989)
formulation of the degradation hypothesis. In Martin's formulation of the hypothesis (Martin,
1992a; 1992b) only closely related, same-category concepts will end up becoming so similar in
the process of random attribute loss, that semantic priming will approach identity priming
(Ober et al., 1991, found equal-to-normal identity priming in AD and the absence of an
interaction between identity priming and semantic priming for EN and AD subjects). Thus,
Martin's hypothesis predicts significantly greater AD compared to EN semantic priming in all
experiments utilizing mainly same-category, prime-target, related pairs. However, the six
experiments that used exclusively same-category pairs (e.g., nickel-dime, foot-hand; rows
13-17, & 19 of Table 1), and the two experiments that used exclusively
superordinate-subordinate (e.g., furniture-table) or subordinate-superordinate (e.g., pear-fruit)
related pairs (rows 12 & 21) all showed equal-to-normal AD priming. In contrast, eight out of
nine of the experiments that obtained si~ificantly greater priming for AD compared to EN
subjects (rows 1, 2, 4-8, & 10), used a mixed variety of related prime-target pairs, some
belonging to the same semantic category (e.g., nickel-dime, bus-car), and some not belonging
to the same category (e.g., cook-stove, apple-red). (See Ober et al., 1991, for a detailed
discussion of this issue.)
The second major difficulty with the degraded store hypothesis is the mounting
evidence that AD patients have much more knowledge available in semantic memory than
previously believed. When relatively less effortful assessments of semantic knowledge are used
with AD patients, they often show normal performance, in contrast to showing a deficit when
the same semantic knowledge is tested in a relatively more effortful manner. For example,
Chertkow et al. (1989) asked two-choice questions about objects' attributes or functions, so
that for saw the question was "Do you cut things with it or lift things with it?". The AD
patients were impaired on these questions and had more trouble with questions about the items
that they were unable to name, providing the basis for the authors' suggestion that specific
concepts were degraded. However, when Grober, Buschke, Kawas, and Fuld (1985)
presented subjects with a target concept and a list of words, including some attributes of the
target concept, AD patients were highly accurate in checking off'the attributes that went with a
target, suggesting that the attributes for the concept were available and could be readily
recognized as belonging to that concept. One can argue that the "check-off" task was much
less effortful, and involved little or no decision making, in comparison to the two-choice
question. Another example is a recent study by Bayles, Tomoeda, Kaszniak, and Trosset
(1991) in which impaired performance of AD subjects on tests of object naming, word
definition, and superordinate/coordinate naming was shown to be a direct function of task
difficulty, rather than a function of the particular concepts being tested (the same concepts
were used across tests). This type of finding is also evidence against the degradation of
semantic knowledge. For a detailed review of studies which have been taken as evidence for
or against the disruption of semantic knowledge in AD, the reader is referred to Nebes (1989;
1992).
In summary, although degradation of semantic memory concepts in AD may seem
plausible as an explanation for findings such as intracategory object-naming errors and
intra-individual, same-concept errors across a variety of psycholinguistic tasks, there are
several difficulties with the degradation hypothesis either as an explanation for the
hyperpriming findings in AD or as a general explanatory framework for the difficulties that AD
patients have with tests of semantic knowledge. The degradation hypothesis predicts abnormal
AD priming in all experimental situations (per Chertkow et al., 1989) or at least in
experimental situations with a high percentage of intra-category related pairs (per Martin,
1992a; 1992b). The main difficulty with regard to the broader arena of clinical and
experimental assessments of semantic knowledge is the mounting evidence that AD patients
can perform at or near normal levels when the amount of attention/effort required is minimal
(e.g., when tasks are used that do not require decision making or overt retrieval).
3. GENERALIZED SLOWING AND HYPERPRIMING
3.1. Generalized Slowing in Elderly Normals
Although older normal subjects have longer RTs than younger normal subjects on
virtually any information processing task requiring a speeded response, the age difference
varies widely from one task to another. A number of researchers have argued that the reason
for bigger age differences on some tasks than on others is not because the mental operations
involved in these tasks are differentially affected by aging, but because those tasks with bigger
age differences require more operations and more processing (i.e., are more complex) and are
thus more vulnerable to general cognitive slowing effects (Birren,Woods, & Williams, 1980;
Cerella, Pooh, & Williams, 1980; Myerson, Hale, Wagstaff~ Pooh, & Smith, 1990; Somberg &
Salthouse, 1982). Evidence supporting the general slowing hypothesis for normal aging has
come mainly from meta-analyses integrating data from numerous and varied RT tasks. In these
analyses, the mean latencies of older subjects were plotted as function of the mean latencies of
younger subjects in the same experimental condition; results showed that the older subjects'
RTs could be predicted from the younger subjects' RTs regardless of the nature of the task
used in a particular experimental condition. A recent meta-analysis by Lima, Hale, and

Myerson (1991) showed that a linear function of the form
Older RT = y-intercept + 1.5 (Younger RT)
described the relationship between the two groups' RTs when the experimental conditions
involve lexical tasks, accounting for over 90% of the variance. Lima et al. conducted a
separate meta-analysis with non-lexical tasks; a linear function with a steeper slope,
approximately 2.0, best described the relationships between older and younger groups' RTs
(again, accounting for over 90% of the variance). In sum, there is strong evidence for
age-related generalized slowing of RT, and although the degree of slowing seems to depend on
whether the lexical or non-lexical domain is involved, overall slowing deserves consideration
when one is attempting to interpret differential age effects across experimental conditions (i.e.,
age by condition interactions).
3.2. Generalized Slowing and Semantic Priming in Elderly Normals
Myerson, Ferraro, Hale, and Lima (1992) recently surveyed semantic priming
experiments with younger and older normal subjects. In individual studies, the age by
condition (unrelated versus related prime) interactions were rarely significant; that is, the older
and younger controls exhibited statistically equivalent semantic priming. Some researchers had
predicted that older controls would show less semantic priming than younger controls,
especially in long-SOA, controlled priming paradigms (based on the hypothesis that age-related
deficits occur in attentional but not automatic cognitive processes; Hasher & Zacks, 1979;
Zacks & Hasher, 1988). Contrary to this prediction, the priming effects have tended to be
larger (but not significantly so) for the older compared to younger subjects. Myerson et al.
(1992) found that older adults' mean priming effects were 1.44 times greater than younger
adults' mean priming effects for 22 lexical decision, experiment conditions; this ratio was 1.41
for 12 word pronunciation, experiment conditions. The regression line describing the
relationship between older and younger subjects' data for the lexical decision, related prime
conditions was almost identical to the regression line for the lexical decision, unrelated prime
conditions. Further, there were no significant differences between the slopes or intercepts of
these two lines. These findings were consistent with the hypothesis that the cognitive
processes involved in semantic priming are slowed in the elderly to the same degree as the
cognitive processes involved in lexical RT tasks in general. Furthermore, Myerson et al.
(1992) found no evidence that the young-old difference in semantic priming was dependent on
SOA, for the 17 out of 22 lexical decision experiments with known SOAs (ranging from
0-1500 ms), contrary to the hypothesis of aging effects on controlled but not automatic
priming. SOA is not the sole factor, however, in determining whether a paradigm allows for
controlled priming processes, as discussed in the "Automatic versus Controlled Processes"
section of this chapter. Laver and Burke (1993) recently reported an effect-size meta-analysis
of 49 semantic priming experiment conditions with young normal and EN subjects. They
found semantic priming effects to be reliably greater for EN than young control subjects and
took this finding as support for process-specific, rather than generalized, slowing with normal
aging. (For a discussion of general versus specific age-related influences on performance of
cognitive tasks, and the difficulties in separating these two components, see Salthouse, 1992.)
Thus, it may be the case that processes involved in semantic priming are differentially affected
by aging, in comparison to processes involved in lexical RT tasks in general, but that this
difference can only emerge via the synthesis of data from numerous priming experiments.
3.3. Generalized Slowing in AD
Nebes and Madden (1988) included 37 data sets from experimental conditions run in
their laboratory, representing a variety of lexical and nonlexical tasks, in linear regression
analyses with young control RT as the '"X" variable and either EN or AD RT as the 'N"
variable. They found that AD patients' RTs are 2.2 times slower than those of young normal
subjects, as compared to EN subjects' RTs being 1.5 times slower than young normal
subjects'. Nebes and Brady (1992) conducted the same analysis on 24 new experimental
conditions from their laboratory, and on the combined total of 61 experimental conditions, with
essentially the same results. For the 61 conditions, the linear function which best described the
relationship between EN and young subjects was:
EN RT = -71 ms + 1.36 Young RT; (r 2 = .88)
The linear function which best described the relationship between AD and young subjects was:
AD RT = -260 ms + 2.26 Young RT; (r 2 = .64)
(The linear function for AD and EN subjects was not reported.) When the AD patients were
divided into mildly versus moderately demented subgroups, regression analyses showed the
latter group to have 50% greater slowing than the former group. Nebes and Brady (1992)
concluded that AD subjects show generalized cognitive slowing, which is of a greater
magnitude than that shown in EN subjects, and which increases with severity.
3.4. Generalized Slowing and Semantic Priming in AID
The implications of overall slowing in AD for interpretation of si~ificant interactions

of group (AD versus EN) with experimental condition are the same as those of overall slowing
in EN compared to young normal subjects (Nebes & Brady, 1992; Salthouse, 1985). That is,
in order to conclude that an interaction between group (AD versus EN) and experimental
condition (related versus unrelated prime), such as the "hyperpriming" results for AD patients,
is due to a specific cognitive abnormality in AD, the magnitude of the group difference in
priming effect must be disproportionately larger than the difference in RT (i.e., the generalized
slowing).
In each of the 21 priming experiments in this survey, the RT in the unrelated prime
condition was slower for the AD subjects than for the EN subjects. Furthermore, the
differences in RT between the two groups were much larger for the nine experiments showing
AD hyperpriming (average difference in the unrelated RT condition of 324 ms for rows 1-8 &
10 of Table 1) than for the remaining 12 experiments (average difference in the unrelated RT
condition of 126 ms for rows 9 & 11-21 of Table 1). Can the pattern of priming data be
accounted for by generalized slowing of cognitive processes in AD? One could assume that
the experimental conditions showing AD hyperpriming involve more complex cognitive
processes, and thus longer RTs for both groups of subjects, but with disproportionately longer
RTs in the case of AD subjects. (It is well known that relatively small increments in task
complexity result in large performance decrements for AD subjects; for reviews of the clinical
and experimental literatures, see Hart & Scruple, 1990, and Nebes, 1992, respectively.)
There are two ways in which longer RTs might produce larger PEs. First, it has been
found with young normal subjects that stimulus manipulations which cause longer RTs, such as
degraded or low frequency targets, produce larger PEs, presumably because of extended
processing of the target leading to greater build-up of activation (see Neely, 1991). Similarly,
slower readers tend to produce larger PEs, presumably for the same reason (Stanovich, 1980).
However, it does not seem reasonable to adopt these findings to the explanation of increases
in PE due to generalized slowing, because they depend on extended processing: with
generalized slowing, RTs are longer, but since this is due to across-the-board slowing, there
would not be any extra processing (i.e, no additional spread of activation).
The second way in which longer RTs might lead to larger PEs is based on simple
arithmetic: the ratio of the slowed versus baseline PEs would be the same as the corresponding
RTs. For example, if EN subjects produced 420 ms unrelated RT and 400 ms related RT,
resulting in a 20 ms PE, and if AD subjects were slowed by a factor of two, then they would
produce 840 ms unrelated RT and 800 ms related RT, with a 40 ms priming effect. This is the
only prediction of increased PE which can be made from pure generalized slowing, and it leads
to a testable null hypothesis about the relation between RT and PE in the EN and AD groups.
AD/EN RT and PE ratios.

In order to estimate a general slowing factor, the ratio of the AD unrelated RT to the
EN unrelated RT was determined for each of the 21 experiments and then the mean of these
ratios was calculated. The mean unrelated RT for the AD subjects was 1.32 times that of the
EN subjects; when cases (experiments) were weighted by sample size, the RT ratio was 1.21.
The same procedure for obtaining ratios was carried out for PE; for the unweighted analysis,
the AD/EN PE ratio was 2.28; for the weighted analysis, the PE ratio was 1.99. Thus, AD
subjects show a 20-30% increase in RT, in contrast to a 100-130% increase in PE, compared
to EN subjects.
Brinley plot analyses

Using the method of Brinley (1965) plots were constructed for comparing the two
subject groups to one another on both unrelated and related RT (as was done by Laver &
Burke, 1993; and Myerson et al., 1992 in their meta-analyses of semantic priming experiments
with elderly and young normal subjects). These plots and the best-fit regression lines are
presented in Figure 1.
Regression analyses were conducted in which: (1) EN unrelated RT (uRT) was the
predictor variable for AD uRT, and (2) EN related RT (rRT) was the predictor variable for AD
rRT. We wanted to test the hypothesis that the best-fit regression lines for unrelated versus
related RTs would have si~mfificantly different slopes, which if confirmed, would provide
evidence against overall slowing as an explanation for hyperpriming (uRT was used as a
baseline, since so few of the 21 experiments included a neutral prime).
1400
1200
E
//
v
I- 1000 / -/
rr
L
/./
(1)
E
800
..,.,.
(D
t-
N
< Related .....
,./,~;,v,,"; Unrelated .......
600 1 ~,
400 I /
I/Y i , ~ I
400 600 800 1000 1200 1400
Elderly Normal RT (ms)
Figure 1. Best-fit regression lines for AID related RT as a function of EN related RT (AD rRT = -267 + 1.75 x
EN rRT; r 2 = .66) and for AD unrelated RT as a function of EN unrelated RT (AD uRT = -487 + 2.12 x EN
uRT; r~ = .69). Each of the solid lines connect the unrelated and related RTs for one of the 21 experiments;
the unrelated RT data point is at the right end and the related RT data point is at the left end, for each of these
solid lines.
The following regression equations w e r e obtained:
A D uKT = -487 + 2.12 x E N uKT; r 2 = .69

A D rRT = -267 + 1.75 x E N rRT, r 2 = .66
The difference b e t w e e n the slopes o f these two linear functions is nonsignificant ( F < 1). The
correlation o f A D uKT with E N ul{T is sigmificant, r = 83, p <.001. The correlation o f A D
rRT with E N r R T is also si,~lnificant, r =. 81, p <.001.
Difficulties with Brmley-plot analyses of generalized slowing

The absence o f slope differences for the best-fitting linear functions for E N - A D R T
plots o f u R T vs r R T is consistent with an overall slowing explanation o f hyperpriming. That
is, the failure to reject the null hypothesis o f equal slopes can be taken as an indication that A D
subjects are slowed to the same extent on rRT trials as on u R T trials, and that overall slowing
can account for their increased priming effects. H o w e v e r , Fisk, Fisher, and Rogers, (1992)
Semanticpriming in Alzheimer's disease 259
have shown that the Brinley-plot regression analysis of young versus older subjects' RTs can
be insensitive to significant differences in age effects across tasks (e.g., tasks varying in the
response required, the amount of automatic vs. controlled processing involved, etc.). Indeed,
Laver and Burke (1993) came to the same conclusion after their effect-size meta-analysis
showed a reliably greater priming effect for older adults than younger adults, even though the
linear regression of young-older RT plots showed no differences in slope (or intercept)
between the uRT and rRT best-fit lines.* This concern that Brinley plots may be
overestimating overall, or task-independent slowing, together with our observation of larger
AD-EN priming effect differences in experiments that seem to meet criteria for controlled
processing (discussed in detail in the next major section of this paper) led us to look more
carefully at the relationships between RT and PE for each subject group.
Within-group correlations of uRT and PE

The within-group correlation o f u R T and PE is highly significant (r = .89, p < .001) for
the AD group, but only approaches significance (r = .41, p = .07) for the EN group. Indeed,
these two correlation coefficients are si~ificantly different from one another (z = 2.96, p <
.01). This suggests that AD subjects' PEs are much more closely tied to their RTs, than is the
case for EN subjects. As a follow-up to this finding and as a means of broadening our
perspective on the issue of overall slowing and its role in hyperpriming, we conducted
regression analyses in which the within-group priming functions (PE as a function of uRT)
were obtained and compared.
Within-group PE functions and slope comparison

We tested the hypothesis that the rate with which PE increases as uRT increases is
significantly greater for the AD compared to EN subjects, i.e., that the slope of the AD priming
function would be significantly greater than the slope of the EN priming function. Again, since
most of the experiments did not include a neutral priming condition, the RT for the unrelated
prime condition is used as the baseline (uRT). This priming function analysis yielded the
following regression equations:
PE = - 1 1 0 + 0.204 x uRT (AD Subjects); r 2 --- .79

PE = -10 + 0.0572 x uRT (EN Subjects); r 2 = . 17
The slopes of these two functions are significantly different (F(1,38) = 7.83, p < .01),
indicating that, as underlying RT increases, the AD groups' PEs increased more than those of
the EN groups. Figure 2 is a graphical representation of the priming functions of the two
groups.
Effect-size meta-analyses as well as meta-analyses utilizing the actual summary statistics from the individual
experiments (e.g, mean RT, mean percent correct, etc.) have become increasingly common in the cognitive aging
literature. According to Glass, McGaw, and Smith (1981) effect-size analyses are necessary when one wants to
compare findings across studies which have used different scales of measurement and/or different methods for
reporting findings. However, Glass et al. (p. 93) also state that the findings from different studies can be expressed
directly when the scales and methods of reporting findings are the same across the studies. In these cases, standard
methods of data analysis may be applied to the between-study data (Glass et al., p. 153) as in the present analyses of
uRT, rRT, and PE.
Some investigators have suggested that an appropriate method of controlling for

changes in baseline RT is to use percent-PE (PPE), or (PE x 100)/uRT, as the dependent
variable. We repeated the above regression analysis with PPE, still using uRT as the baseline
(predictor) variable, with the following results:
PPE = -4.70 + 0.0136 x uRT (AD Subjects); r 2 = .55

PPE = 2.64 + 0.0024 x uRT (EN Subjects); r 2 = .02
(Note that because of the change in scale, the regression coefficients of the PE and PPE
functions are not directly comparable.) The difference between the slopes of these functions
approaches significance (F(1,38) = 3.05, p = .085). The correlation o f u R T and PPE is highly
siLnaificant (r = .74, p < .001) for the AD subjects, but not significant (r = .13) for the EN
subjects.
If the hyperpriming found in many of the AD priming studies was simply due to
generalized slowing, then the slopes of the PE functions for the AD and EN groups plotted in
Figure 2 should be statistically equivalent. In fact, the slopes for the two groups are
significantly different, which is evidence against overall slowing as the sole explanation of
hyperpriming in AD. Furthermore, when an alternate measure of priming, based on the
percentage decrease in RT with a related prime, is used as the predicted variable in regression,
the difference in slope for the AD versus EN group approaches si~ificance.*
Generalized slowing and the null hypothesis

It is important to note that the generalized slowing hypothesis is essentially an
acceptance of the null hypothesis with regard to a difference in semantic priming processes
between AD and EN groups. Although the slopes of the AD-EN uRT regression line and of
the AD-EN rRT regression line are statistically equivalent, there are several other
multi-experiment analyses reported here (involving the within-groups relationship of uRT to
PE) and findings by other researchers (described above) regarding the insensitivity of the
between-groups Brinley plots, which cause us to consider alternative accounts (to overall
slowing) for the often-obtained hyperpriming in AD.
Normal PE in other neuropsychological populations with slowing of RT

Another type of evidence against the overall slowing explanation of AD hyperpriming
was provided by Chertkow, et al. (1994; row 8 in Table 1). They described 20 AD subjects
(out of a total sample of 48 AD subjects) who showed priming effects greater than 60 ms; 8 of
these 20 AD subjects, however, had unrelated RTs in the same range as the normal controls
(between about 500 and 750 ms; the remaining 12 of these 20 AD subjects had unrelated RTs
greater than 900 ms, substantially above those of any normal controls). This indicates that
* Because the findings from experiments with small sample sizes may be less reliable than those from experiments
with relatively larger sample sizes, we also conducted weighted regression analyses for the within-group priming
functions, with sample size as the weightingfactor. The F-tests of slope differenceswith the weightedanalysesyielded
F(1, 38) = 8.10, p < .01, for the PE function, and F(1, 38) = 2.90, p = .09 for the PPE function, paralleling the
findings for the unweightedanalyses. Whenthe correlationcoefficientsfor the relationship ofuRT to PE were derived
via weighted regressionanalyses, with sample sizes as weights, r = .88 for the AD group, and r =.46 for the EN group;
the weighted correlation coefficientsfor the relationship of uRT to PPE yieldedr =.73 for the AD group, and r =. 18
for the EN group, again parallelingthe findings for the unweightedanalyses.
200
AD
EN .......
, At
E 150
v
o A
M-,,-
m 100 A
r
o._
J A
E
"= 50 a ...................................ii
e . . e ~ t ~ a E
9 / e a
i ! I I il
600 800 1000 1200 1400

Unrelated RT (ms)
Figure 2. Plot of the within-group, uRT-PE functions with best-fit regression lines for the AD subjects (A, a,
solid line) and EN subjects (E, e, dotted line), capital letters indicate that the AD subjects showed significantly
more priming than the EN subjects in the experiment from which the groups' means were obtained.
Regression accounts for 79% of the variance in the case of the AD subjects, whereas regression accounts for
only 17% of the variance in the case of the EN subjects, the slopes of the AD and EN regression functions are
significantly different. See text for details.
significant overall slowing is not a prerequisite ofhyperpriming. In a related vein, Chertkow et

al. (in press) reported the absence of greater-than-young-normal priming effects on a long
SOA, lexical decision task, for normal elderly, depressed elderly, and Parkinsons's disease
subject groups. Thus several subject populations who are flowed relative to young normals
did not exhibit increased priming effects as a "byproduct" of the flowing; this can also be taken
as evidence against a generalized slowing explanation ofhyperpriming in AD.
Summary: Overall slowing

Although the between-groups analyses of uRT and rRT data are consistem with an
overall flowing explanation of the AD priming findings, these Brinley-plot-based analyses
overestimate overall slowing (i.e., mask task-specific slowing). Furthermore, the following
findings constitute converging evidence against overall flowing as a complete explanation of
AD hyperpriming: (1) the significant difference in the rate at which the AD as compared to EN
PE increases as a function of uKT (i.e., the si~ificant difference in the slopes for the AD
versus EN priming functions for the 21 experiments in the survey), (2) the marginally
si~ificant difference in the rate at which the AD as compared to EN percent PE increased as a
function ofuRT, (3) the finding ofhyperpriming in a subgroup of AD patients who show equal
raw RTs to elderly control subjects (Chertkow et al., 1994), and (4) the absence of increased
semantic priming in patient populations other than AD who, nonetheless, showed
greater-than-normal raw RTs (Chertkow et al., in press). Of course, AD subjects always have
slower RTs than EN subjects, and overall slowing most likely played a significant role in the
outcomes of these priming experiments. However, it seems that overall slowing alone does
not provide an adequate explanation, and that we should begin the search for possible task- or
process-specific slowing factors associated with AD, which also play an important role in
semantic priming experiments.
4. A U T O M A T I C VERSUS C O N T R O L L E D PROCESSES
4.1. A Multi-process Model of Semantic Priming
Automatic spread o f activation

Automatic cognitive processes are those that occur without intention, conscious
awareness, or the use of conscious, attentional resources; in contrast, controlled cognitive
processes utilize conscious, attentional resources, are intentional, and are not concealed from
conscious awareness (Hasher & Zacks, 1979; Posner & Snyder, 1975; Shiffdn & Schneider,
1977; Bargh, 1992). The dual-process model of semantic priming (Neely, 1977; Posner &
Snyder, 1975) incorporates the automatic process of spreading activation and the controlled
process of expectancy. The concept of spreading activation was introduced by Collins and
Quillian (1969) and Collins and Loffus (1975). The activation of the semantic memory node
corresponding to a presented word (i.e., the prime) was assumed to spread to the nodes of
semantically and/or associatively related words (i.e., including the related target), thus
decreasing the time required for activation levels in these related nodes to exceed a critical
threshold for recognition. Since spreading activation is fast and automatic, it is expected to
operate in all priming experiments.
Expectancy
Expectancy involves the use of the prime to generate an expectancy set, consisting of
potential targets which are related to the prime; targets which are in this set are processed
more quickly than those which are not (Becker, 1980). Expectancy is assumed to be relatively
slow acting and under strategic control, hence, a controlled process. At short SOAs (such as
250 ms), normal subjects are thought to show only automatic priming effects; that is, the
amount of priming is not affected by the proportion of related pairs in the list or the
instructional set, and there is no interference effect for unrelated as compared to neutral
primes. In contrast, normal subjects begin to show attentional, strategy-driven effects when
SOAs are 400-500 ms and longer. The priming effect increases with the proportion of related
pairs in the list, can be influenced by instructions, and can involve interference from unrelated
compared to neutral primes at these longer SOAs (de Groot, 1984; den Heyer, Briand, &
Dannenbring, 1983; Neely, 1977; but of. McLeod & Walley, 1989).* It is important to note
* Only nine of the 21 experiments (rows 3, 5-8, 11, 15, 16, 20 in Table 1) included a neutral prime condition (the
word '~olank"or a nonword as prime) which potentially enables the separation of the overall priming effect into two
components: facilitation(RT in the neutral condition - RT in the related condition) and inhibition (RT in the unrelated
condition - RT in the neutral condition). Furthermore,in five of these nine experiments(rows 3, 6, 15, 16, and 20) the
mean RT for the neutral condition did not fall betweenthe mean RTs for the related and unrelated conditions for both
that expectancy is a pre-lexical mechanism (i.e., it precedes lexical access of the target), which
can speed up or slow down access of the target concept depending on what the subject is led
to expect. Expectancy-based priming should be equally likely to occur in pronunciation and
lexical decision paradimns. However, a pairwise paradigm, with temporally paired primes and
.targets, would greatly increase the chances for expectancy-based priming in comparison to a
continuous paradignl Furthermore, pairwise priming paradi,~mns with relatively high
proportions of related pairs and relatively long SOAs (i.e., greater than 500 ms) would be
especially prone to expectancy-based priming. In a continuous priming paradigm, subjects are
given single-item trials, with word pronunciation or lexical decision required on each trial;
there is no designation of prime-target pairings for the subjects. For evidence that several
controlled processes, including expectancy, do not play a role in the semantic priming effects
obtained with single-presentation (continuous) priming paradi,~mns, the reader is referred to
Fischler (1977) and Shelton and Martin (1992).
Semantic matching
Another controlled priming process which has recently received attention in the normal
literature is semantic matching. This process, unlike expectancy, is post-lexical and occurs
only in lexical decision (not pronunciation) tasks. Furthermore, semantic matching is
apparently precluded by the use of a continuous priming paradigm, as was used in the lexical
decision experiments summarized in rows 15, 16 and 20 of Table 1 (McNamara & Altarriba,
1988; Shelton & Martin, 1992). As described by Neely and Keefe (1989) between the time
that lexical access of the target has occurred and a word/nonword decision is made for the
target, subjects can use information about the relatedness of the prime-target pair to decrease
the RT for a correct '%vord" decision as well as to decrease the RT for a correct "nonword"
decision. If the subject mentally looks back to the prime and notices that there is a semantic
relationship between the prime and target, then the subject will be biased to a '%vord" response.
On the other hand, any lexical nodes activated by seeing a nonword target will rarely be
semantically related to the prime word with which it is paired. (See de Groot, 1984, for a
similar description of a post-lexical mechanism which she calls "post-lexical coherence
checking.") If the nonword ratio (proportionof all unrelated prime-target pairs in which the
target is a nonword) is high, subjects will be biased toward a nonword response. Neely, Keefe,
and Ross (1989) showed that previous studies of the effects of increasing the relatedness
proportion on lexical decision have actually confounded relatedness proportion and nonword
ratio; they increase together if the probabilities of word versus nonword targets remain the
same. When relatedness proportion and nonword ratio are orthogonally manipulated some
differential effects (e.g., for type of prime-target relationship) are obtained. The important
point for present purposes is that increases in relatedness proportion, with concomitant
increases in nonword ratio, can ma~ify the priming effect through post-lexical semantic
matching.
the AIDand controlgroups, precludingthe use of the neutral conditionas a baselinefrom which to measurefacilitation
and inhibition for both groups. (There are numerous difficulties associated with choosing the appropriate neutral
prime condition for a particular priming paradigm; an overviewof these difficulties is provided in Neely, 1991; see
also Jonides and Mack, 1984).
4.2. Controlled Processes and Hyperpriming
The current set of 21 experiments includes nine experiments which yielded significantly
greater priming in the AD compared to EN group (rows 1-8, & 10). Each one of these
AD-hyperpriming experiments involved a pairwise, long-SOA priming paradigm~ with
relatedness proportions from .33 to .67 (see the third, fourth, and fifth columns in Table 1);
hence, they meet the criteria for controlled priming paradigms. There is only one "controlled"
experiment (row 9; see Table note 9 for further explanation) in which AD hyperpriming was
not obtained. There are 11 experiments (lower section of Table 1) that did n o t show
si~ificantly increased priming for AD compared to EN subjects; all are either short-SOA,
pairwise paradimns, or continuous paradigms. The controlled process of expectancy is
precluded in all 11 of the experiments, since expectancy is induced via pairwise presentation,
an awareness of the presence of related pairs, and an SOA of at least 400 ms. Among the
short-SOA, pain~se priming experiments in the lower section of Table 1, six had a relatedness
proportion of only .17 (rows 12-14, 17, 19, & 21) and one had a relatedness proportion of.33
(row 11). Among the continuous priming experiments in the lower section of Table 1, the
relatedness proportion varied between .25 and .36 (rows 15, 16, 18, & 20). Therefore, even if
a high relatedness proportion (e.g., .50 or greater) could induce expectancy effects in a
continuous paradigm (and we have no data on this at present), this is not a concern for the
current set of continuous priming experiments. It seems reasonable to conclude that
expectancy-based priming was absent or occurred only minimally among the 11 experiments in
the bottom section of Table.I, all of which do n o t show AD hyperpriming.
The second controlled process, semantic matching, would be expected to play a
si~ificant role (in addition to expectancy) in the six pain~ise, lexical decision experiments
(rows 1, 2, 3, 5, 7, 8) which are included among the nine AD-hyperpriming studies in Table 1.
Semantic matching might also occur in the four, pain~ise, lexical decision experiments in the
lower section of Table 1 (rows 11, 12, 13, & 17), but to a lesser extent than in the lexical
decision experiments in the upper section of Table 1, because the relatedness proportion in
three out of four of these experiments (rows 12, 13, & 17) is quite low (.17). The remainder
of the lexical decision experiments in the bottom half of Table 1 involved continuous priming,
which precludes the use of semantic matching. Thus, an additional attentional process,
semantic matching, is much more likely to be contributing to the priming effects in the upper
section as compared to the lower section of Table 1. It is interesting to note that, although
none of the AD-EN priming differences among the 11 experiments in the bottom half of Table
1 were reliable, it is the four pain~ise lexical decision experiments (rows 11, 12, 13, & 17) that
show the largest AD-EN priming differences among these 11 experiments. This is suggestive
of some "seepage" of semantic matching in these experiments, which we have classified as
"automatic" based on the short (250 ms) SOA, which precludes the use of expectancy. In sum,
the pattern of findings suggests that hyperpriming may be the result of AD abnormalities in the
deployment of expectancy and/or semantic matching during the semantic priming trials.
It is important to note that the three Chertkow et al. (1993) experiments summarized in
rows 5, 7, and 11 of Table 1 were all conducted on the same groups of AD and EN subjects,
with the same stimuli and general procedure. In their Experiment 1, Chertkow et al. had
subjects make a lexical decision to the prime, and then there was a 500 ms interval from the
onset of the response to the prime to the onset of the target. In their Experiment 2, Chertkow
et al.'s subjects did not respond to the prime and the SOA was controlled, at 500 ms. Both
Experiments 1 and 2 showed significantly greater priming in the AD compared to EN subjects.

In contrast, in their Experiment 3, Chertkow et al. obtained AD priming which was not
statistically different from EN priming; this experiment is in the "automatic" section of Table 1,
because the SOA was 250 ms (the primes were not responded to), which is not enough time
for expectancy to be utilized. This switch for the same group of AD subjects, from normal
semantic priming to hyperpriming (a 25% PE increase), due solely to an increase in SOA from
250 to 500 ms, provides converging evidence (to the pattern seen across the entire set of
experiments, involving multiple AD subject groups) that the finding ofAD hyperpriming versus
normal priming depends on the automatic versus controlled nature of the experimental
paradigm,
4.3. Controlled-Process-Specific Slowing
The pattern of AD hyperpriming versus equal-to-normal priming across the set of 21

experiments provides strong evidence that automatic semantic priming processes function
normally in AD, and that hyperpriming occurs only in paradigms which encourage controlled
priming processes. Furthermore, increased priming for the AD subjects was strongly
associated with increased uRT. Could there be process-specific slowing for the AD subjects
occurring only in those semantic priming experiments which encourage controlled processes,
and which resulted in RTs which are disproportionately slower than those for the experiments
which involved relatively more "automatic" semantic priming paradigms? The reader is
referred to Figure 3, which illustrates the disproportionate slowing of RT and the
disproportionate PE for the AD subjects in the controlled experiments. The PE (i.e., the
difference between the rRT and the uRT) is 94 ms for the AD in the controlled experiments, as
compared to PEs of 28, 31, and 21 ms for AD automatic, EN controlled, and EN automatic,
respectively.
We conducted a meta-analysis, using ANOVA, in which the following were treated as
independent variables: AD versus EN, controlled versus automatic paradigm, and uRT vs.
rRT. When all 21 experiments were included in the ANOVA, each of the main effects, all of
the two-way interactions (group x paradigm, group x relatedness, and paradigm x relatedness),
and the three-way interaction (group x paradigm x relatedness) were si,~mfificant at either the
.01 or .001 level. The main effects reflect the slowing with AD, the slowing with controlled
paradigms, and the slowing of RT when the prime-target pair is unrelated compared to related.
The two-way interactions reflect the greater difference in RT slowing for controlled compared
to automatic paradigms in the case of AD compared to EN subjects, the greater difference in
priming effect for the AD compared to EN subjects, and the larger priming effect in controlled
compared to automatic paradigm~q. The three-way interaction, F(1, 19) - 14.11, p < .01,
which is the most important effect for present purposes, reflects the disproportionate increase
in priming effect for AD subjects in the controlled condition. We are fully aware that utilizing
"experiments as subjects" in an ANOVA is not a usual practice; however, we conducted the
analysis for the same reasons that we and other investigators have used "experiments as
subjects" in correlation and regression analyses. The essence of meta-analysis is to allow for
the discovery of something meaningful and important that is not apparent in the individual
studies. Indeed, Glass, McGaw, and Smith (1981) state that "...any of the methods of
statistical analysis that have proved to be useful in extracting meaning from data are potentially
useful in meta-analysis. One's attitude toward the data may be exploratory ... or confirmatory,
1100
Alzheimer Patients
1000
E
= l
P-
r
900
0
,=,=,,,,,
o
w 800
I:::
m
Elderly Controls
ID
700
600
Controlled Automatic Controlled Automatic
Figure 3. Graphical representation of the mean unrelated RT (U) and mean related RT (R) for each subject
group, in each of the two kinds of experiments (Controlled and Automatic).
descriptive or inferential; it does not matter" (p. 153). We are using the results of the
ANOVA, particularly the F-ratio for the three-way interaction, in a confirmatory manner and
as a way of demonstrating that the disproportionate increase in PE for the AD subjects in
controlled conditions is reliable. The 21 priming studies involved only 13 independent subject
samples (see footnote 1 from Table 1 for details). In order to eliminate inter-experiment
dependencies, we separately analyzed each of the 36 sets of 13 independent studies. In every
one of the 36 ANOVAs all effects, including the critical three-way interaction (group x
paradigm x relatedness), were significant at the .05 level or better.
5. GENERAL SUMMARY AND CONCLUSIONS
We have utilized data from 21 semantic priming experiments in an evaluation of each of

three explanations for the often-obtained AD hyperpriming. The degraded semantic memory
hypothesis predicts that all semantic priming experiments will show hyperpriming in AD, which
is dearly not the case (only 9 out of 21 experiments showed significantly greater-than-normal
AD priming). Further, many of the previously reported deficits on semantic memory tasks
have been called into question as evidence accumulates that semantic memories are indeed
available in AD, but not always accessible, due to attentional demands of the particular task
being used. The generalized slowing hypothesis states that hyperpriming is due to RT slowing
and that the AD compared to EN increases in PE are proportionate to the AD compared to EN
increases in RT across experiments. The Brinley-plot regression analyses were consistent with
a generalized slowing explanation; however, when priming effects were predicted from
unrelated RT, in separate regressions for AD and EN groups, the slope of the AD priming
function was significantly larger than the slope of the EN priming fimction. Thus, the semantic
priming effect is disproportionately increased with increasing unrelated RT in AD compared to
EN subject groups. The hypothesis that hyperpriming is attentionally-based, being found only
in experimental paradigms which allow for controlled priming processes (such as expectancy
and semantic matching) is strongly supported by the pattern of findings. Nine out of 10
experiments with pain~ise presentation, medium-to-long SOAs, and relatively high proportions
of related prime-target pairs yield AD hyperpriming. In contrast, none of the 11 priming
experiments which utilize either short-SOA, pain~ise paradigms or continuous priming
paradigms (and are therefore classified as "automatic") show AD hyperpriming. Chertkow et
al. (1993) have even shown normal priming and hyperpriming in the same group of AD
subjects, with the same stimuli, in a lexical decision task, by simply increasing the SOA from
250 ms to 500 ms or more. An ANOVA on the mean RTs for the 21 experiments showed a
significant three-way interaction of group x paradigm x prime relatedness, due to the fact that
the AD subjects show disproportionate priming effects compared to the EN subjects only for
controlled priming paradig~ns. (This finding was also obtained when only experiments utilizing
independent groups of subjects were included.)
In sum, we have concluded that AD hyperpriming is attentionally-based, coinciding
with the disproportionate slowing in RT that occurs in controlled, but not automatic priming
paradigms. In this regard, one could think of AD hyperpriming as the result of
controlled-process-specific slowing. The two strategies involved in controlled priming,
expectancy and semantic matching, require the explicit utilization of information in semantic
memory, which is much more difficult (and would therefore cause much more slowing) than
implicit utilization of semantic memory (as occurs in "pure" automatic semantic priming) for
AD individuals. The more specific mechanism(s) for attentionally-based AD hyperpriming
await discovery. One candidate mechanism is an increase in the inhibitory (as opposed to the
facilitatory) component of the priming effect. (See Chertkow et al., 1993, for preliminary
evidence that AD patients show larger inhibitory effects than control subjects in controlled
priming paradigms.) Generalized slowing does, of course, occur in AD and this can readily be
observed in these 21 RT experiments. The fact that there is a trend for AD priming effects to
be larger than the EN priming effects even in the automatic priming experiments of Table 1 is
likely due to generalized slowing; atter all, the AD groups' RTs are significantly slower than
the EN groups' RTs in 20 out of the 21 experiments. Furthermore, partial degradation in the
semantic memory network is likely in AD, although more so in the later compared to earlier
stages of the disease. Generalized slowing and degraded memory stores could each be
contributing somewhat to hyperpriming in the 9 out of 21 experiments in which it occurs.
However, neither of these mechanisms alone or together can account for the pattern of
significantly greater-than-normal versus statistically equal-to-normal AD priming across the 21
experiments, whereas controlled-process-specific slowing can. The single most important
message from our chapter is that researchers should not draw conclusions about the intactness
of semantic memory structures and/or processes in AD based on results obtained with semantic
priming (or other semantic knowledge-dependent) paradigms that allow controlled

(attentional) processes to play a major role in performance on the task.
REFERENCES
Albert, M., & Milberg, W. (1989). Semantic processing in patients with Alzheimer's disease.
Brain and Language, 3 7, 163-171.
and contextual disambiguation effects in healthy aged individuals with senile dementia of
the Alzheimer type. Brain and Language, 40, 181-201.
Bargh, J. A. (1992). The ecology of automaticity: Toward establishing the conditions needed
to produce automatic processing effects. American Journal of Psychology, 105, 181-199.
Bayles, K. A., Tomoeda, C. K., Kaszniak, A. W., & Trosset, M. W. (1991). Alzheimer's
disease effects on semantic memory: Loss of structure or impaired processing? Journal of
Cognitive Neuroscienee, 3, 166-181.
Becker, C. A. (1980). Semantic context effects in visual word recognition: An analysis of
semantic strategies. Memory and Cognition, 8, 493-512
Birren, J. E., Woods, A. M., & Williams, M. V. (1980). Behavioral slowing with age:
Causes, organization, and consequences. In L. W. Poon (Ed.), Aging in the 1980s:
Psychological issues (pp. 293-308). Washington, DC: American Psychological
Association.
Brinley, J. F. (1965). Cognitive sets, speed and accuracy ofperformance in the elderly. In A.
T. Welford & J. E. Birren (Eds.), Behavior, aging and the nervous system (pp. 114-149).
Springfield, IL: Charles C. Thomas.
Cerella, J., Pooh, L. W., & Williams, D. M. (1980). Age and the complexity hypothesis. In
L. W. Poon (Ed.), Aging in the 1980s: Psychological issues (pp. 332-340). Washington,
DC: American Psychological Association.
Chertkow, H., Bub, D., Bermnan, H., Bruemmer, A., Merling, A, & Rothfleisch, J. (1994).
Increased semantic priming in patients with dementia of the Alzheimer's type. Journal of
Clinical and Experimental Neuropsychology, 16, 608-622.
Chertkow, H., Bub, D., Bergman, H., D'Antono, B., Whitehead, V., & Rothfleisch, J. (in
press). Semantic priming on lexical decision in young and elderly normals, elderly
depressed, and Parkinson's disease patients: Absence of effects of slowing. Psychology
and A ging.
Chertkow, H., Bub, D., Bergman, H., Whitehead, V., Merling, A, & Zahirney, G. (1993).
Semantic priming and prime-target lag in patients with dementia of the Alzheimer's type.
(Manuscript submitted for publication).
Chertkow, H., Bub, D., & Seidenberg, M. (1989). Priming and semantic memory loss in
Alzheimer's disease. Brain and Language, 36, 420-446
Collins, A. M., & Quillian, M. R. (1969). Retrieval time from semantic memory. Journal of
Verbal Learning and Verbal Behavior, 8, 240-247.
Collins, A. M., &. Lottus, E. F. (1975). A spreading-activation theory of semantic
processing. Psychological Review, 82, 407-428
de Groot, A. M. B. (1984). Primed lexical decision: Combined effects of proportion of
related prime-target pairs and the stimulus-onset asynchrony of prime and target.
Quarterly Journal of Experimental Psychology, 96, 29-44
den Heyer, K,, Briand, I~, & Dannenbring, G. L. (1983). Strategic factors in a
lexical-decision task: Evidence for automatic and attention-driven processes. Memory &
Cognition, 11, 374-381
Fischler, I. (1977). Associative facilitation without expectancy in a lexical decision task.
Fisk, A. D., Fisher, D. L., & Rogers, W. A. (1992). General slowing alone cannot explain
age-related search effects: Reply to Cerella (1991). Journal of Experimental Psychology:
General, 121, 73-78.
Folstein, M. F., Folstein, S. E, & McHugh, P. 1L (1975). Mini-Mental State: A practical
method for grading the cognitive state of patients for the clinician. Journal of Psychiatric
Research, 12, 189-198.
Glass, G. V., McGaw, B., & Smith, M. L. (1981). Meta-analysis in social research. Beverly
Hills: Sage Publications.
Glosser, G., & Friedman, 1L B. (1991). Lexical but not semantic priming in Alzheimer's
Disease. Psychology and Aging, 6, 522-527.
Grober, E., Buschke, H., Kawas, C., & Fuld, P. (1985). Impaired ranking of semantic
attributes in dementia. Brain and Language, 26, 276-286
Hart, S., & Semple, J. M. (1990). Neuropsychology and the dementias. London: Taylor &
Francis.
Hartman, M. (1991). The use of semantic knowledge in Alzheimer's disease: Evidence for
impairments of attention. Neuropsychologia, 29, 213-228.
Hasher, L., & Zacks, 1L T. (1979). Automatic and effortful processes in memory. Journal of
Experimental Psychology: General, 108, 356-388.
Hodges, J. 1L, Salmon, D. P, & Butters, N. (1992). Semantic memory impairment in
Alzheimer's disease: Failure of access or degraded knowledge? Neuropsychologia, 30,
301-314.
Hughes, C. P., Berg, L., Danziger, W. L., Coben, L. A., & Martin, 1L L. (1982). A new
clinical scale for the staging of dementia. British Journal of Psychiatry, 140, 566-572.
Jonides, J., & Mack, R. (1984). On the cost and benefit of cost and benefit. Psychological
Bulletin, 96, 29-44
Laver, G. D., & Burke, D. M. (1993). Why do semantic priming effects increase in old age?
A meta-analysis. Psychology and Aging, 8, 34-43.
Lima, S. D., Hale, S., & Myerson, J. (1991). How general is general slowing? Evidence from
Margolin, D. I. (1988). Lexical priming by pictures and words in aging, stroke and dementia
(Doctoral dissertation, University of Oregon, Portland, 1987). Dissertation Abstracts
International, 49, 1416B.
Martin, A. (1987). Representation of semantic and spatial knowledge in Alzheimer's patients:
Implications for models of preserved learning in amnesia. Journal of Clinical and
Experimental Neuropsychology, 9, 191-224.
Martin, A. (1992a). Degraded knowledge representations in patients with Alzheimer's
disease: Implications for models of semantic and repetition priming. In L. 1L Squire & N.
Butters (Eds.), Neuropsychology ofMemory 2nd ed. (pp. 220-232). New York: Guilford
Press.
270 B.A. Oberand G.K. Shenaut
Martin, A. (1992b). Semantic knowledge in patients with Alzheimer's disease: Evidence for
degraded representations. In L. Backman (Ed.), Memory functioning in dementia (pp.
119-134). Amsterdam: North-Holland Press.
Martin, A.., & Fedio, P. (1983). Word production and comprehension in Alzheimer's disease:
The breakdown of semantic knowledge. Brain and Language, 19, 124-141.
Mattis, S. (1976). Mental status examination for organic mental syndrome in the elderly
patient. In L. Bellack & T. Katasu (Eds.), Geriatric Psychiatry: A Handbook for
Psychiatrists and Primary Care Physicians (pp. 77-121). New York: Grime and Straton.
McLeod, B. E., & Walley, 1L E. (1989). Early interference in a priming task with brief
masked targets. Canadian dournal of Psychology, 43, 444-470
McNamara, T. P., Altarriba, J. (1988). Depth of spreading activation revisited" Semantic
mediated priming occurs in lexical decisions. Journal of Memory and Language, 27,
545-559.
Myerson, J., Ferraro, F. 1L, Hale, S., & Lima, S. D. (1992). General slowing in semantic
Myerson, J., Hale, S., Wagstafl~ D., Pooh, L. W., & Smith, G. A. (1990). The
information-loss model: A mathematical theory of age-related cognitive slowing.
Nebes, 1L D. (1989). Semantic memory in Alzheimer's disease. Psychological Bulletin, 106,
377-394
Nebes, 1L D. (1992). Cognitive dysfunction in Alzheimer's disease. In F. I. M. Craik & T. A.
Salthouse (Eds.), The handbook of aging and cognition (pp. 373-446). Hillsdale, NJ:
Erlbaum~
Nebes, 1L D., & Brady, C. B. (1992). Generalized cognitive slowing and severity of dementia
in Alzheimer's disease: Implications for the interpretation of response-time data. Journal
Nebes, 1L D., Brady, C. B., & Hufl~ F. J. (1989). Automatic and attentional mechanisms of
semantic priming in Alzheimer's disease, dournal of Clinical and Experimental
Nebes, 1L D., & Madden, D. J. (1988). Different patterns of cognitive slowing produced by
Alzheimer's disease and normal aging. Psychology and Aging, 3, 102-104
Nebes, 1L D., Martin, D. C., & Horn, L. C. (1984). Sparing of semantic memory in
Alzheimer's Disease. Journal of Abnormal Psychology, 93, 321-330
Neely, J. H. (1991). Semantic priming effects in visual word recognition: A selective review
of current findings and theories. In D. Besner & G. W. Humphreys (Eds.), Basic processes
in reading: Visual word recognition (pp. 264-336). Hillsdale, N. J.: Lawrence Erlbaum~
Neely, J. H. (1977). Semantic priming and retrieval from lexical memory: Roles of
inhibitionless spreading activation and limited-capacity attention. Journal of Experimental
Neely, J. H., & Keefe, D. E. (1989). Semantic context effects on visual word processing: A
hybrid prospective/retrospective processing theory. In G. H. Bower (Ed.), The psychology
of learning and motivation: Advances in research and theory, Vol.23 (pp. 207-248).
New York: Academic Press,.
Neely, J. H., Keefe, D. E, & Ross, I~ L. (1989). Semantic priming in the lexical decision
task: Roles of prospective prime-generated expectancies and retrospective semantic
matching. Journal of Experimental Psychology: Learning, Memory, and Cognition, 15,

1003-1019.
Ober, B. A., & Shenaut, G. I~ (1988). Lexical decision and priming in Alzheimer's disease.
Ober, B. A., & Shenaut, G. K. (1990, November). Semantic priming in Alzheimer's disease
and normal aging: Single- and dual-choice lexical decision. Paper presented at the meeting
of the Psychonomic Society, New Orleans.
Ober, B. A., Shenaut, G. K, Jagust, W. J, & Stillman, R. C. (1991). Automatic semantic
priming with varying types of category relationships in Alzheimer's disease and normal
aging. Psychology and Aging, 6, 647-660.
Ober, B. A., Shenaut, G. K., & Nelson-Abbott, R. A. (1991, August). Semantic priming and
word frequency in normal and abnormal aging. Paper presented at the meeting of the
American Psychological Association, San Francisco.
Posner, M. I., & Snyder, C. 1~ R. (1975). Attention and cognitive control. In R. L. Solso
(Ed.), Information processing and cognition: The Loyola symposium. Hillsdale, N.J.:
Erlbaun~
Salthouse, T. A. (1992). Shifting levels of analysis in the investigation of cognitive aging.
Human Development, 3, 321-342.
Salthouse, T. A. (1985). Age-related changes in basic cognitive processes. In M. Storandt &
G. 1~ VandenBos (Eds.), Handbook of the psychology of aging, 2nd ed.(pp. 400-426).
New York: Van Nostrand Reinhold.
Shelton, J. 1~, & Martin, R. C. (1992). How semantic is automatic semantic priming?
Journal of Experimental Psychology: Learning, Memory, and Cognition, 18, 1191-1210.
Shiffrin, 1~ M., & Schneider, W. (1977). Controlled and automatic human information
processing: II. Perceptual learning, automatic attending, and a general theory.
Somberg, B. L., & Salthouse, T. A. (1982). Divided attention abilities in young and old
adults. Journal of Experimental Psychology: Human Perception and Performance, 8,
651-663
Stanovich, I~ E. (1980). Toward an interactive-compensatory model of individual differences
Zacks, R. T., & Hasher, L. (1988). Capacity theory of inferences. In L. L. Light & D. M.
Burke (Eds.), Language, memory, and aging. New York: Cambridge University Press.
Indirect memory tests in Alzheimer' s Disease*
Marilyn Hartman and Matthew L. Pirnot
University of North Carolina at Chapel Hill
1. INDIRECT MEMORY TESTS AND ALZHEIMER'S DISEASE
Anterograde amnesia is one of the core symptoms of Alzheimer's disease (AD) (e.g.,
Corkin, 1982; Kaszniak, 1986) and one of the major criteria for the clinical diagnosis of the
illness (McKhatm, Drachman, Folstein, Katzman, Price, & Stadlan, 1984). Frequently it is the
first symptom to appear and precedes the development of full-blown dementia, in which all
cognitive functions are progressively affected. This inability to learn new information is
traditionally assessed by direct or explicit memory tests that require the patient to intentionally
retrieve information that has recently been presented (e.g., free recall, cued recall, and
recognition memory tests). The topic of this chapter, however, is the performance of AD
patients on a different type of memory test, called an indirect or implicit memory test
(Richardson-Klavehn & Bjork, 1988; Roediger & McDermott, 1993; Schacter, 1987). This
type of test does not explicitly ask the patient to reflect on prior exposure to studied material
but rather measures learning through the facilitation in performance that occurs with repeated
exposure to a set of stimuli.
Indirect memory tests have attracted the attention of cognitive psychologists and
neuropsychologists studying memory, because of the frequent finding that patients with
circumscribed amnesia perform at normal levels on indirect memory tests, while at the same
time show severe deficits on direct memory tests (Schacter, 1987; Squire, 1992; Torres & Raz,
1994). Although both sets of terms, direct/indirect and implicit/explicit, appear commonly in
the literature, we will use the terms indirect and direct in this review. This choice can be
ascribed to our preference for the more descriptive terms, whose only denotation is to
distinguish between tests that differ in the type of instruction to subjects. Thus, subjects on
direct memory tests are always instructed to think back to an earlier experience and recollect
the information that was learned, whereas subjects in an indirect memory test are told either to
perform as efficiently as possible a given task that is seemingly unrelated to the studied
material, or else simply to practice a skill. If they are informed at all of the fact that some
stimuli will reappear in the test, they are also told to proceed with the test without intentionally
trying to use that knowledge. They are never told that the purpose of the test is to assess
learning. Thus, the essential difference between direct and indirect memory tests is that the
instructions differ with regard to the intentionality of retrieval. We have chosen to avoid the
more ambiguous terms implicit and explicit memory tests, because they are frequently used to
refer to both the type of test instructions, as well as to two sets of theoretically independent
memory processes, each with its own neurological substrate. Describing the tests as indirect
Please address correspondence to: Marilyn Hartman, Psychology Department, Davie Hall CB # 3270,
University of North Carolina, Chapel Hill, NC 27599-3270. E-mail: Marilyn_Hartman@ unc.edu
Indirect memory tests in Alzheimer's disease 273
or direct memory tests removes the theoretical implications associated with the terms implicit
and explicit.
Although the literature on indirect memory tests in AD is still relatively small and very
new -- with one exception, all the studies discussed here have been published since 1986 -- it
is worth a close examination in order to establish what is known and what is not known, and to
take stock of theoretical and methodological issues raised by the literature to date. In addition
to adding to our knowledge of Alzheimer's disease, an understanding of indirect memory test
performance in AD can be used to learn more about the cognitive processes that subserve the
tests as well. Because only some indirect memory tests exhibit impairments in AD, a
consideration of the obtained patterns may also provide data concerning the differences among
them and reveal clues regarding their neuroanatomical substrate.
Although the question addressed by all of the studies to be discussed here is whether
AD patients have intact indirect memory test performance, some clarification of this question is
in order. Because severely demented AD patients are generally unable to perform the indirect
memory tests (and thus are unable to participate in the research), it seems reasonable to infer
that they also have si~ificant impairments. Similarly, because healthy older adults show
normal or near normal performance on indirect memory tests (see Light, 1991, for a recent
review), it is clear that AD affects memory on this type of test, above and beyond the effects of
normal aging. Thus, we can confidently conclude at the outset that Alzheimer's disease does
produce deficits on indirect memory tests and that the severity of those deficits is positively
correlated with the severity of dementia. Therefore, the more interesting question, and the one
to be addressed by the research literature, is whether performance on indirect memory tests is
impaired in mildly to moderately demented patients with AD, and, more specifically, whether
all types of indirect memory tests are equally affected at this stage of the illness. Patients
included in these studies have severe impairments on direct memory tests, and are be~nning to
develop cognitive dysfunction in other domains as well. Until recently, however, little was
known about their ability to demonstrate memory on indirect tests.
Because of the focus of this volume on verbal processes, we first discuss indirect tests
involving language tasks; following this, we will also consider the non-verbal tests, because
they help clarify some of the patterns of preserved and impaired abilities on the verbal tests. In
addition, we will also briefly compare the performance of AD patients to patients with
Huntington's and Parkinson's Disease, as well as to patients with pure amnesia, in order to
illustrate the unique effects of Alzheimer's disease.
2. I N D I R E C T M E M O R Y TESTS USING VERBAL S T I M U L I
In the prototypical study of language-based indirect memory, subjects are first

presented a list of stimuli. After a brief delay, an apparently unrelated task is administered,
which contains a mixture of the previously encoded stimuli and new stimuli. This second task
is the indirect memory test, and usually requires subjects to name, identify, or generate words.
Memory is measured by the tendency of subjects to read or identify words more efficiently, or
generate words more frequently, as a function of prior exposure to those words. This effect is
called repetition priming, and is generally calculated as the difference in performance for
repeated items as compared to new items, which serve as a baseline measure. Alternatively, all
items are presented on two or more occasions, and memory is assessed in terms of the
improvement from initial performance. In addition to these absolute indices of memory,
274 M. Hartman and M.L. Pirnot
learning can also be calculated in terms of the change in performance for repeated items as a
proportion of initial or baseline performance. Because baseline performance may differ among
subjects and between groups of subjects, the proportional scores provide additional
information about the size of memory effects. For instance, subjects whose overall
performance is slow must show larger absolute priming effects to reflect the same relative
benefit as subjects who respond more quickly. Both absolute priming effects and proportional
scores will be discussed in this review as appropriate.
The first studies to be considered are those that showed no evidence of impairment in
the AD patients. These included the following indirect memory tests: simple naming of word
triads (Balota & Duchek, 1991), lexical decisions (Chenery, Ingram, & Murdoch, 1994; Ober
& Shenaut, 1988; Ober, Shenaut, Jagust, & Stillman, 1991), reading text (Monti, Gabriel/,
Wilson, & Reminger, 1994), and reading of mirror-reversed words (Grober, Ausubel,
Sliwinski, & Gordon, 1992). In all of these, repetition priming was measured by the reduction
in response times with repeated exposure to the critical stimuli. Not only was there no
indication of impairment in the AD patients, but in two of the six studies there was slightly (but
not significantly) more repetition priming in the AD patients than the elderly controls (Balota
& Duchek, 1991; Ober & Shenaut, 1988). In one study in which the same words were
repeated for six consecutive blocks of trials, AD patients showed continuous irnprovement for
all six blocks, whereas healthy subjects demonstrated little change in reaction times atter the
second block (Chenery et al., 1994). Furthermore, in the studies in which priming effects
could be examined in relation to baseline performance, there were no group differences in the
proportional priming scores. This latter finding is of importance, because overall reaction
times were significantly slower in the AD patients. Thus, a consideration of proportional
scores in the present case confirms the absence of impairment in AD patients.
2.1. Studies of Perceptual Identification
There are also three studies that examine repetition priming in a perceptual
identification task (Abbenhuis, Raaijmakers, Raaijmakers, & van Woerden, 1990; Keane,
Gabrieli, Fennema, Growdon, & Corkin, 1991; Keane, Gabrieli, Growdon, & Corkin, 1994).
In this indirect memory test, subjects attempt to identify words that are presented for very brief
durations. The measure of learning is the reduction in exposure time required to identify
previously presented words. In two of the studies, subjects viewed real words (Abbenhuis et
al., 1990; Keane et al., 1991); in the third pronounceable non-words were used (Keane et al.,
1994). None of the three studies found significant differences between patients and normals,
although under some conditions there was numerically more repetition priming in AD than
elderly controls. One study, however, reported near si~ificant reductions in the proportional
priming scores for AD subjects (Keane et al., 1991, Experiment 3). This conflicting result is
difficult to interpret, because it is not clear whether absolute or proportional repetition priming
is the most appropriate measure. AD patients needed longer exposure times in order to
identify new words in all studies; these large and significant differences in baseline performance
make conclusions ambiguous when absolute and proportional measures of priming point to
different conclusions.
In addition to the possibly inconsistent results across studies, another characteristic of
the data raises additional questions. In examining the performance of the healthy subjects, one
notes that the repetition priming effects, although statistically significant, were very small. For
instance, the mean effect sizes ranged from 8.7 msec to 12 msec in older adults in Keane et al.
(1991); they had a mean of 4.9 msec in Keane et al. (1994). Thus, even ifAD patients were
impaired, the numerical differences between the patients and normals would likely be very
small, and one would need a relatively large number of subjects to detect a real difference. In
the three studies cited, the samples had 10 to 12 subjects in each group. Exactly how large a
sample is needed is difficult to determine, because standard errors are generally not reported,
but clearly a larger number of subjects is needed before any definitive conclusions can be
reached about the perceptual identification test.
2.2. Studies of Stem Completion and Anagram Solution
Studies using the stem completion task by far outnumber studies with any other single
indirect memory test. In this procedure, subjects are given the initial letters of new and
previously studied words and told to complete each word stem with the first word that comes
to mind. Of the ten studies using this indirect test, eight showed impairments in AD (Bondi &
Kaszniak, 1991; Bondi, Kaszniak, Rapcsak, & Butters, 1993; Gabrieli, Keane, Stanger,
Kjelgaard, Corkin, & Growdon, in press; Heindel, Salmon, Shults, Walicke, & Butters, 1989;
Keane et al., 1991; Randolph, 1991; Salmon, Shimamura, Butters, & Smith, 1988; Shimamura,
Salmon, Squire, & Butters, 1987). In addition to the two studies using stem completion tests
that showed no impairment (Grosse, Wilson, & Fox, 1990; Partridge, Knight, & Feehan,
1990), there is one study using anagrams that also found intact performance in AD patients
(Perfect, Dowries, de Mornay Davies, & Wilson, 1992); however, in the latter the control
group was recruited from a geriatric day care center and not screened for cognitive
dysfunction. A potentially important observation regarding the two stem completion studies
that did not find group differences (Grosse et al., 1990; Partridge et al., 1990) is that both used
encoding tasks that required more extensive semantic processing than the studies that did
report impairment in the AD patients. In the studies showing intact performance, subjects
either generated the target words from sentence contexts or else were required to define the
words. In contrast, studies finding impairments in Alzheimer's patients had subjects read
isolated words at the time of encoding plus either make semantic or orthographic judgments
about each word (Gabrieli et al., in press), or, more commonly, rate each word in terms of how
much they liked it. If greater elaborative (e.g., semantic) processing requirements at encoding
indeed lead to normal repetition priming, this would suggest that AD patients do not access
meaning spontaneously, and consequently are more dependent on explicit situational demands
to support semantic processing. Possible reasons to expect such a pattern will be discussed
below. Nevertheless, it appears that under the more typical testing conditions, AD patients are
consistently impaired. This is particularly noteworthy in that baseline performance was
generally equivalent for AD and control subjects, that is, the two groups of subjects were
equally likely to produce the new words as completions of the stems in the baseline condition.
Before reaching final conclusions about the stem completion test, it is important to
confider whether the results can be explained by the differential use of intentional memory
strategies by subjects whose performance is better, that is, whether healthy older subjects used
deliberate retrieval of studied words to a greater extent than did AD patients. Even though
subjects are generally not informed of the relationship between an indirect memory test and
previously studied words, some subjects may realize the connection on their own, if they notice
that some of their responses are identical on the two tasks. After becoming aware, some may
alter their strategies for c a ~ g out the indirect test, and begin to intentionally use the earlier
set of words as responses on the indirect memory test (Bowers & Schacter, 1990; Jacoby,
1991; Richardson-Klavehn, Lee, Joubran, & Bjork, 1994).
One piece of evidence against contamination of the healthy older adults' performance is
the finding by Gabrieli et al. (in press) that the stem completion test showed no effect of the
levels of processing encoding manipulation in either the healthy or AD group. Unlike the
direct test results, which showed significantly better memory for words studied with the
semantic than the orthographic encoding task, there was no such effect on stem completion
priming. Furthermore, the repetition priming effects of AD patients in this experiment were
smaller than those of amnesic patients, who showed priming similar to normals. Because one
would not expect significant use of intentional retrieval strategies by amnesic patients, one can
probably rule out this explanation as an account for the observed impairment in AD patients in
this study. Nevertheless, a cautionary note is in order, as most studies did not assess the
degree of intentional strategy use. In addition, one study reported a siL-mificant correlation
between repetition priming and direct memory performance (Randolph, 1991). This
association may represent a causal link, with intentional retrieval contaminating the indirect test
results, but it may also reflect the fact that priming and explicit memory test performance are
both directly related to dementia severity (Gabrieli et al., in press; Keane et al., 1994).
2.3. Studies Using Free Associations
There are three studies that utilized a free association paradigm (Brandt, Spencer,
McSorley, & Folstein, 1988; Hut~ Mack, Mahlmann, & Greenberg, 1988; Salmon et al.,
1988). In two of them, subjects first studied word pairs (e.g., BIRD - WING), and then were
shown the first word in each pair and asked to generate a single association. Repetition
priming was indicated by the tendency to provide the previously studied associates as
responses. In the third study, subjects studied a list of single words, then were asked to give
associations for a second list of words, half of which were semantically related to the
previously studied list. All three studies showed significant impairments of repetition priming
in AD;* in fact AD patients did not produce significant repetition priming at all in two of them-
As with the stem completion studies, there were no group differences in baseline flee
associations. The question of intentional retrieval is similar to that discussed in relation to stem
completion tasks, however, and there is no evidence presented that addresses this issue. In the
Bran& et al. study (1988), nevertheless, performance on an unrelated free recall test was
uncorrelated with repetition priming of the associates.
2.4. Summary of Verbal Repetition Priming Studies
One of the first conclusions to draw from this review is that one cannot generalize
about indirect memory tests, as patients with AD showed impairment on some but not all the
repetition priming tasks described. When repetition priming was measured by word naming,
lexical decisions, mirror reading, and prose reading, no impairments were found. On tests
requiring stem completions or the generation of semantic associates, however, there was
Although the study by Huff et al. (1988) compared AD patients to left hemisphere stroke patients, one can
reasonably assume that if the AD patients were impaired in comparison to the stroke patients, their
performance was reduced relative to healthy individuals as well.
unequivocal impairment.* More ambiguous are the results from the perceptual identification
test, and further research is needed to clarify the ability of AD patients to demonstrate learning
on this task.
Overall, although all these memory tests involved verbal stimuli and indirect memory
instructions, there were clear differences among them in the effects of AD patients. Before
considering possible explanations of the obtained pattern, however, it will prove useful to
review the pattern of preserved and impaired performance on indirect memory tests that are
non-verbal in nature. This may provide further clarification of the findings from the verbal
repetition priming tasks.
3. INDIRECT M E M O R Y TESTS INVOLVING NON-VERBAL STIMULI AND

SKILLS
The studies that fall in this category include a variety of tasks: adaptation to prisms,
biasing of weight judgments, tracing of geometric figures viewed in mirror image, a pursuit
rotor task, identification of perceptually degraded pictures, reading of mirror-reflected words,
and a serial reaction time task. Only some of these provide evidence of intact functioning in
the patients with AD, and as with the verbal tasks, we will discuss these first.
In one study AD and healthy control subjects showed equal ability to adjust to prisms
that displaced objects in the visual field (Paulsen, Butters, Salmon, Heindel, & Swenson,
1993). Both groups of subjects showed equivalent initial response to the prisms (i.e.,
equivalent baseline performance), and an equivalent rate of adaptation over time. A study that
examined the extent to which judgments of heaviness were influenced by prior exposure to a
set of heavy or fight weights also showed no impairment in the patient group (Heindel, Salmon,
& Butters, 1991). Both healthy and AD subjects who lit~ed fighter weights in the first phase of
the experiment rated a second set of weights as heavier than subjects who lifted a set of heavier
weights first.
Subjects tested with a mirror-tracing task (Gabrieli, Corkin, Mickel, & Growdon,
1993) were required to reach around a barrier and trace a geometric pattern while viewing the
mirror reflection of their drawing hand. AD patients showed a normal degree of improvement
with practice, and when generalization of the skill was assessed with a second design, they
showed equivalent transfer of their newly acquired skill.
There are four studies examining pursuit rotor performance, none reporting significant
impairment in AD patients (Bondi & Kaszniak, 1991; Bondi et al., 1993; Eslinger & Damasio,
1986; Heindel et al., 1989). In this task, subjects had to keep a hand-held stylus in contact
with a small metal target on a rotating disk; performance was measured in terms of the amount
of time the subject maintained contact between the stylus and the target. Although initial,
baseline performance in the AD patients was significantly worse than that of the healthy
elderly, the two groups improved by the same amount when given practice across several
sessions. In the one study that individually adjusted the speed of the disk to equate initial
performance for AD and the control group (Heindel et al., 1989), the performance of AD and
controls was essentially identical. The only contrary evidence appeared in one study that
showed a trend towards worse performance in AD patients during the training sessions
(Eslinger & Damasio, 1986). There was also a si~ificant decline in performance for AD
*There is also one unpublished report of impairment on a category exemplar generation test (Monti, Gabrieli,
Reminger, Grosse, & Wilson, 1992).
patients after a 20 minute post-training delay, whereas the control subjects actually improved
after the rest period. Unfortunately, this study included only eight AD patients, so that
replication of their findings regarding forgetting of the skill would be desirable.
All of the above studies that examine skill acquisition are thought to involve learning of
motor programming skills in response to sensory experience. Thus, patients with Alzheimer's
disease appear to adapt normally to novel sensory stimuli, and acquire performance skills that
reflect this learning.
In contrast to the lack of impairment in AD patients in the learning of motor programs,
there are a number of studies showing impaired performance on indirect memory tests
involving other types of non-verbal stimuli. These include repetition priming of identification
of perceptually degraded pictures (Bondi & Kaszniak, 1991; Bondi et al., 1993; Corkin, 1982;
Gabrieli et al., in press; Heindel, Salmon, & Butters, 1990) and learning in a serial reaction
time task (Ferraro, Balota, & Connor, 1993; Knopman, 1991; Knopman & Nissen, 1987). We
also include here the learning of a primarily non-verbal skill that involves verbal stimuli, a task
involving mirror reading of words (Grober et al., 1992).
In the studies using fragmented pictures, facilitation is measured by reductions in the
amount of perceptual information needed to identify a picture on the second exposure. In four
of the five studies, AD patients showed significantly reduced repetition priming (Bondi &
Kaszniak, 1991; Bondi et al., 1993; Corkin, 1982; Heindel et al., 1990). In the two studies
where it can be determined, there was no repetition priming at all in AD subjects (Bondi &
Kaszniak, 1991; Bondi et al., 1993). In the single study reporting no significant differences
between AD and control subjects, repetition priming was numerically less in AD patients than
healthy individuals, and when priming was calculated in terms of proportional facilitation rather
than absolute differences, AD patients showed considerably less learning (Gabrieli et al., in
press). Because AD patients have sigmificant impairments in their baseline ability to identify
perceptually degraded pictures, the examination of proportional scores is important. Heindel
and his colleagues (Heindel et al., 1990) equated baseline performance by calculating the
amount of perceptual degradation necessary for each subject to perform at a specified level,
then corrected priming scores for individual differences in baseline performance, and still found
clear-cut impairment among the AD patients. In a different type of perceptual learning task,
AD patients' ability to read mirror-reversed words did not improve across trials unless the
identical stimuli were repeated (Grober et al., 1992).
Another test used to assess memory indirectly is the serial reaction time task (SRT)
developed by Nissen and her colleagues (Nissen & Bullemer, 1987). On each trial in this task
subjects are presented with a light on a computer screen at one of four locations and required
to respond by pressing a key corresponding to that location. Subjects are presented with five
blocks of 100 trials. The first four blocks consist of a ten-trial repeating sequence; the last
block switches to a pseudo-random order. Subjects are not informed about the presence of a
repeating sequence. Learning of the embedded sequence is reflected in a decrease in reaction
times across the first four blocks followed by an increase in the fitth block when the pattern
disappears.
The SRT task has been used in three studies of Alzheimer's disease, with mixed results.
In one, AD patients showed reduced learning (Knopman, 1991). In a second study (Knopman
& Nissen, 1987), there was a non-significant trend towards reduced learning in the AD
subjects (p<. 10) and 9 of the 28 patients (32%) did not show any learning. In the third study
(Ferraro et al., 1993), the mildly demented but not the very mildly demented AD patients were
significantly irnpaired. The only study to examine forgetting in this task (Knopman, 1991)
reported that all subjects who showed robust learning during the initial training session retained
the learning over a one-week delay.
In considering the siL_mificance of the conflicting results on the SRT task, one again
notes that there are large group differences in baseline reaction times. In fact, if one examines
learning as a function of baseline speed of responding, even the study that reports no
significant differences between AD and healthy elderly reveals large reductions in the amount
of learning (Knopman & Nissen, 1987): the healthy subjects showed approximately 40%
improvement across the first four blocks of trials as compared to 10% for the patients. Overall
then, it appears that learning of a motor-spatial sequence is not intact in Alzheimer's patients,
except for those in the very earliest stages oft he disorder.
In summary, one notes that as with verbal indirect tests of memory, the results are not
consistent across all types of tasks. In trying to make sense of the observed pattern, there
appear to be two general categories of tests. The first category includes tasks requiring the
learning of new motor programs in response to novel sensory experience; overall these show
no memory decrement in patients with AD (although see Eslinger & Damasio, 1986, for a
possible exception). The second category is harder to describe but includes picture fragment
identification, mirror reading skill~ and the SRT test. There is no obvious common
denominator for these tasks. Possible explanations of the impairments will be considered
below.
4. EXPLANATIONS OF THE PATTERN OF INTACT AND IMPAIRED ABILITIES
Up to this point we have seen that for both language-based and non-verbal indirect
memory tests, there are some tasks that show impairments in the early and middle stages of
Alzheimer's disease and others that appear intact. We now turn to possible explanations of this
pattern and consider the plausibility of each.
4.1. Neuroanatomical Explanations
One type of explanation emphasizes the neuroanatomical correlates of cognitive

processes. According to this hypothesis, different indirect memory tests involve different
regions of the brain, and tests showing intact abilities depend on regions that are relatively
intact in AD. For example, it is well known that the neuropathological changes that are
pathognomonic of the disease are most strongly concentrated in the association areas of the
cerebral hemispheres and the limbic system (Damasio, van Hoesen, & Hyman, 1990). Within
the cerebral hemispheres, areas close to the primary sensory areas are less affected than areas
farther away, and the primary sensory cortices are essentially free of pathology (Damasio et al.,
1990). In addition, cortical and subcortical structures that make up the motor system,
including the basal ganglia and cerebellum, are relatively spared (Damasio et al., 1990).
One consequence of the distribution ofneuropathological changes in AD is that indirect
memory tests with a strong motor component should show intact performance in AD patients,
because of the preservation of the motor systenx This is consistent with the evidence reported
above, and with studies of indirect memory comparing patients with AD to patients with
known basal ganglia dysfimction. The latter include patients with Huntington's disease (HD)
and Parkinson's disease (PD). In direct contrast to the pattern of results for individuals with
AD, HD patients and non-demented PD patients show impairments primarily on tasks

requiring the learning of motor skills. For instance, HD patients do poorly on tests involving
the biasing of weight judgments (Heindel et al., 1991) and visuo-motor adaptation to prisms
(Paulsen et al., 1993), tasks on which AD patients perform normally. As a group, HD patients
are also impaired on the SRT task, although as with the AD patients, there is considerable
heterogeneity of ability level (Knopman & Nissen, 1991). Non-demented PD patients also
show impaired SRT performance similar to that of mildly demented AD (Ferraro et al., 1993).
On the pursuit rotor test, patients with HD show less learning than the AD patients, although
non-demented individuals with PD have normal performance on this task (Bondi & Kaszniak,
1991; Heindel et al., 1989).
Unlike the findings with motor-based indirect memory tests, patients with basal ganglia
disease demonstrate intact learning on other types of indirect memory tests on which AD
patients do poorly. Thus, patients with HD and non-demented patients with PD show normal
repetition priming for fragmented picture identification, stem completion and flee association
tests (Bondi & Kaszniak, 1991; Heindel et al., 1989, 1990; Randolph, 1991; Salmon et al.,
1988; Shimamura et al., 1987), although one study found HD patients to be mildly impaired
(but still better than AD patients) on the stem completion task (Salmon et al., 1988).
Overall these results represent a double dissociation between performance on motor
and non-motor indirect memory tests, with AD patients showing more impairment on non-
motor tasks, and PD and HD patients showing greater deficits on the motor tasks.
Examination of the patients with PD who are also demented provides corroborating evidence.
Unlike non-demented PD patients, these patients are globally impaired on all types of indirect
memory tests. They have difficulty both on motor tasks such as the rotor pursuit, as well as
language-based tests such as stem completion (Heindel et al., 1989). Thus, these patients
appear to have deficits attributable to both subcortical and cortical damage. This is not
surprising since some of these patients are known to have Alzheimer's type brain pathology in
addition to the striatal damage associated with Parkinson's disease (Heindel et al., 1989).
Thus, this set of studies suggests that part of the reason AD patients perform at normal levels
on some indirect memory tests is due to the relative sparing in these patients of the basal
ganglia and the subcortical motor system. In contrast their impairments appear localizable to
the cerebral hemispheres. The only apparent exception to this pattern is the finding of
impairment on the SRT task for patients with both subcortical and cortical disease. The most
likely explanation of this anomaly is that the test is not a pure motor task, but rather involves
motor learning as well as higher level cognitive processes.
The conclusion that the neuroanatomical locus of learning on non-motor indirect tests
lies in the neocortex is also reinforced by studies contrasting the performance of AD patients
with amnesics who have brain damage that affects the limbic system in a relatively selective
manner. AD patients and amnesics have equally severe deficits on direct memory tests, but the
latter do not have the wide range of cognitive deficits seen in dementia. Numerous studies
have reported indirect memory test performance in the normal range in amnesics (see Schacter,
Chiu, & Ochsner, 1993, for a recent review), and compared to AD patients, amnesics had
higher levels of repetition priming on the stem completion task (Bondi et al., 1993; Gabrieli et
al., in press; Salmon et al., 1988; Shimamura et al., 1987). The only conflicting result among
the studies directly comparing amnesia and AD (Bondi et al., 1993; Gabrieli et al., 1993;
Gabrieli et al., in press; Salmon et al., 1988; Shimamura et al., 1987) involved the picture
identification test, which showed impaired performance for the amnesic as well as the AD
Indirect memory tests in AIzheimer's disease 281
patients (Bondi et al., 1993; Gabrieli et al., in press). Several previous studies, however, have
reported intact performance of amnesic patients on this task (Corkin, 1982; Warrington &
Weiskrantz, 1968), so that the reasons for this discrepancy are not clear.
Although studies of amnesics provide indirect evidence of the irrelevance of the limbic
system to indirect memory tests, recent studies using imaging techniques offer very preliminary
but nevertheless inconsistent evidence. One study using Positron Emission Tomography (PET)
indicated that the stem completion task involves posterior right hemisphere cortical areas
(Squire, Ojemann, Miezin, Petersen, Videen, & Raichle, 1992). In contrast, a study utilizing
MRI images of patients with a wide range of amnesic disorders (including AD) suggested that
the limbic system and striatum are both involved in the perceptual identification test (Jemigan
& Ostergaard, 1993). Thus, despite the behavioral evidence from amnesic patients, the
identification of indirect memory tests with particular brain regions dearly requires further
clarification.
In summary, comparisons of patients with AD to other neurological populations
suggests strongly a dissociation between motor learning tasks and non-motor tasks. Based on
the contrast with HD and PD patients, it appears that the intact performance of AD patients on
motor tasks can probably be attributed to intact functioning of the basal ganglia and its
connections to cortical motor structures. Furthermore, amnesic patients with primarily limbic
system damage tend to perform normally both on indirect memory tests that show intact
performance in AD as well as those tests that show impairments. Thus, dysfunction of the
limbic system in AD does not appear a udficient explanation for the pattern of findings among
AD patients, although we still have relatively little direct evidence for the localization of
indirect memory test effects within the cerebral hemispheres. Tentatively, however, we can
conclude that dysfunction within the neocortex plays a role in the indirect memory test
impairments observed in AD patients.
4.2. The Role Of Intentional Retrieval On Indirect Memory Tests
A consideration of our knowledge of the neuroanatomy of indirect memory tests still

leaves the following phenomenon to explain: Why do AD patients do well on indirect memory
tests involving reading of isolated words and text or lexical decisions, but poorly on picture
fragment identification, stem completions, word associations, and the SRT task? One
possibility that surfaces frequently in the literature is that in the studies showing impairments,
healthy control subjects used intentional retrieval strategies that would not have been available
to the AD patients. Although mentioned previously in connection with the stem completion
test, perhaps it is appropriate to consider this argument again, in light of the full set of findings.
Numerous researchers have emphasized that giving subjects instructions to respond as quickly
and efficiently as possible does not guarantee that subjects follow these instructions (Bowers &
Schacter, 1990; Jacoby, 1991). In support of this concern, it has been found that subjects
frequently become aware of the relationship between the study and test procedures (Bowers &
Schacter, 1990; Richardson-Klavelm et al., 1994), and a significant proportion of them then
adopt an intentional retrieval strategy (Richardson-Klavehn et al., 1994).
Despite the possibility that explicit recollection contaminates the performance of
subjects, and particularly of healthy subjects flee of memory problems, one must be cautious
about invoking this explanation without clear evidence. First, the evidence for the confounding
of unintentional and intentional retrieval in young adults usually reveals relatively small effects
(e.g., Richardson-Klavetm et al., 1994; Toth, Reingold, & Jacoby, 1994). These tend to be
much smaller than the observed differences between AD and healthy control subjects. Second,
this explanation does not explain the patients' total absence of repetition priming in two studies
of word associations (Huff et al., 1988; Salmon et al., 1988). Third, as pointed out above, the
results of Gabrieli et al. (in press) show a large discrepancy between the priming effects of AD
and amnesic patients. Fourth, at the present time there is no evidence that conscious memory
processes are used more frequently in some indirect tests than others. Therefore, although it is
plausible that a portion of the repetition priming produced by healthy subjects is due to the
influence of consciously-employed memory processes, there is very little direct evidence for it.
The only data to support such a confounding is the finding in two studies that used the picture
identification test and compared amnesics to AD patients. Contrary to prediction, repetition
priming was reduced in the amnesics relative to control subjects (Bondi et al., 1993; Gabrieli et
al., in press), suggesting that in this particular instance the healthy subjects may have carried
out the task using explicit retrieval strategies. This was confirmed by Gabrieli et al.'s (in press)
informal observations of those subjects' behavior. Thus, further work is needed to rule out this
explanation in all cases in which impairment is observed. Consistent assessment of subjects'
use of strategies would provide a useful adjunct in evaluating such results.
4.3. The Transfer Appropriate Processing Model
Another possible explanation of the indirect memory test results utilizes the transfer
appropriate processing theory (TAP) developed by Roediger, Blaxton, and their colleagues
(Blaxton, 1989; Roediger & Blaxton, 1987). This framework holds that both indirect and
direct memory tests vary in their relative reliance on perceptual or conceptual processing.
Extensive research utilizing this theory has verified this distinction and demonstrated that some
indirect memory tests, such as perceptual identification, stem completion, word fragment
completion, and picture fragment identification tests, are primarily perceptually driven, whereas
tests such as category exemplar production, general knowledge tests, and word association
tests are primarily conceptually-driven.
The main characteristics of conceptually-driven tests are their insensitivity to changes in
the sensory characteristics of the stimuli from study to test, and the fact that they are strongly
affected by the use of semantically-based strategies. Unlike the conceptually-based tests,
perceptually-driven tests produce reduced repetition priming when the appearance of the
stimuli is altered from study to test, and show little or no effect of encoding manipulations that
emphasize elaborative, semantic processing over perceptual processing. Although most
research with amnesic patients has utilized indirect memory tests that are classified as
perceptually-driven, there are examples of intact performance in amnesics on conceptually-
based indirect tests as well (e.g., Graf, Shimamura, & Squire, 1985). There is only one study
showing deficits of amnesic patients on conceptually-driven indirect memory tests (Blaxton,
1992), and this result may be restricted to amnesic patients who have a combination of limbic
system and nearby temporal association cortex damage.
Applying this framework to the literature with AD patients can account for much of the
observed pattern of performance. For instance, one might hypothesize that AD patients have
greater difficulty with conceptual than perceptual processing, as Gabrieli, Keane, and their
colleagues have argued (Gabrieli et al., in press; Keane et al., 1991). This is certainly plausible,
given the relative sparing of sensory compared to association cortex, and the presence of
language deficits (to be discussed in more detail below). Thus, one might predict intact
performance on reading and lexical decision tasks, and impaired performance on word
association tests, and indeed this corresponds to the findings. One would also expect normal
performance on the perceptual identification test, but as argued above, the data with AD
patients are ambiguous on this count.
One problem with this account, however, is the clearly impaired performance of AD
patients on the stem completion and picture fragment identification tests. There is no obvious
explanation in terms of the TAP framework. Let us consider in particular the stem completion
test, because it has been studied more extensively than the pictorial task. The findings with
young healthy adults overwhelmingly suggest that this type of repetition priming is dependent
largely on perceptual learning (see Roediger & McDermott, in press, for a discussion). Thus,
one would expect intact performance by AD patients. To the extent that conceptual
processing contributes, its effect is small; for example, levels of processing manipulations have
little or no effect on stem completion performance (Challis & Brodbeck, 1992). When there is
a conceptual contribution to stem completion, it may involve contamination by intentional
retrieval strategies adopted by some subjects (Tothet al., 1994), and we have argued above
that this is unlikely to fully account for the deficit in AD patients. Overall the evidence
suggests no effect of elaborative rehearsal on the stem completion test, and therefore no strong
conceptual component as defined by the TAP framework. This creates problems for
interpreting the findings from AD patients in terms of the distinction between perceptually-
driven and conceptually-driven tests, and one must look elsewhere to understand these
patients' difficulty with the stem completion test.
Before leaving the TAP framework, it is important to note another feature of the theory
that may be useful in explaining the pattern of findings in AD. In addition to the distinction
between perceptual and conceptual processing, the theory also incorporates the idea that
memory depends in part on the degree of similarity in processing between study and test tasks
(e.g., encoding specificity; Morris, Bransford, & Franks, 1977). This principle makes the
prediction that the more similar the demands of study and test tasks, the better the subjects'
performance. One might then ask whether studies showing deficits in AD patients involve
memory tests that are dissimilar to the study tasks. If patients with AD have impaired memory
processes, greater support or cueing may be essential to induce intact performance.
Viewing the set of intact and impaired performances from this perspective is
informative, but still not completely satisfactory. The naming, lexical decision, and reading
tasks all involve identical procedures on first and second viewing of the stimuli, thus showing a
high level of encoding specificity. It is on these tests that the AD patients show intact
performance. In contrast, when the processing demands of the tasks change from study to
test, impairment is generally observed, as seen in the patients' poor performance on stem
completion and word association tests. Nevertheless, not all results conform to this pattern.
Impairments are found on the picture identification and the SRT tests, where the tasks remain
constant throughout the experiment. Thus, study-test similarity explains the majority of
findings with AD patients, but can not easily account for all of the results.
4.4. Non-Memory Contributions To Indirect Memory Tests
Although an examination of memory processes is clearly useful, a consideration of

impairments in non-memory functions may also be important in understanding the performance
of AD patients on indirect memory tests. If the demented patients were less able than the
control subjects to carry out the tasks used in this research, this may have had an impact on
indirect measures of learning. Two types of evidence are relevant in evaluating this possibility.
First, an examination of impairments in baseline conditions on the various indirect memory
tests may yield clues about the non-memory characteristics of performance. Second, some
studies report relationships between indirect memory test performance and general cognitive
functioning; a survey of these findings may also be informative.
If one examines first the studies of verbal repetition priming, there is consistent
evidence of impairment in initial or baseline conditions when the dependent measure is speed of
response. Thus, the AD patients were overall slower in their reaction times in almost every
case (Balota & Duchek, 1991; Grober et al., 1992; Ober & Shenaut, 1988; Ober et al., 1991),
the only exception being reading times for prose passages (Monti et al., 1994). Even in the
latter case, it is not likely that AD patients had completely normal reading skills. Given the
generalized nature of their cognitive dysfunction, with its concomitant attentional and language
difficulties, comprehension was likely impaired.
In direct contrast to the above-mentioned studies, the experiments that were not
speeded and did not measure reaction time, namely those using stem completion and word
association tests, generally reported equivalent baseline performance in the patient and control
groups. Patients also made very few omission errors. For instance, Shimamura et al. (1987)
found that AD subjects were able to provide completions for essentially all of the stems (98%).
In one exception, Randolph (1991) found an increased number of omission errors in AD
patients. The overall similarity between AD and healthy elderly performance is perhaps
unexpected; yet it is bolstered also by the finding that the types of responses made to the word
stems were similar in both groups of subjects (Gabrieli et al., in press): AD and healthy
subjects responded equally often with the most common completion.
Despite this evidence for equivalent baseline responding, we still have only incomplete
information about the integrity of patients' ability, because stem completion and flee
association performance was untimed. It is likely that, as with the studies of naming and
lexical decision judgments, the patients with AD were significantly slower to respond. Thus,
equivalent baselines demonstrate that the organization of semantic knowledge is normal, but
do not preclude the possibility that access to these stored representations has been affected by
their illness. If lexical-semantic information indeed is less available, priming might be affected
as well. Thus, at the present time, it seems unwarranted to assume that all the cognitive
processes necessary for completing word stems and generating free associations are normal in
AD patients.
As for the non-verbal tests of indirect memory, it again seems that the speeded tests
generally showed impairments in patients with AD in baseline conditions. AD patients had
si~ificantly more difficulty than their healthy counterparts in the pursuit rotor task (Bondi &
Kaszniak, 1991; Bondi et al., 1993; Eslinger & Damasio, 1986; Heindel et al., 1989), and had
slower reaction times on the SRT task (Ferraro et al., 1993; Knopman, 1991; Knopman &
Nissen, 1987). Accuracy on the SRT tasks was affected as well; in the two studies reporting
these data, AD patients made more errors (Ferraro et al., 1993; Knopman & Nissen, 1987).
The only exception here was the absence of impairment in the ability of AD patients to trace a
reflected pattern in their first attempts (Gabrieli et al., 1993), although the AD patients showed
numerically worse performance and the sample was small (N = 9). In contrast, tests that were
not speeded produced similar baseline performance for AD patients and healthy elderly: Both
groups were equally able to judge relative heaviness of weights (Heindel et al., 1990), and their
visuo-motor abilities were equally affected by wearing prisms (Paulsen et al., 1993). In sum,
AD patients are impaired on speeded motor tasks, but appear to show normal visuo-motor
responses to stimuli. The latter tasks may represent the only category of indirect memory tests
in which baseline performance is normal in AD.
If AD patients are impaired in carrying out the majority of tasks used in indirect
memory tests, it then becomes important to consider the relationship between that impairment
and the degree of implicit learning. There are two possible relationships. The first is that
difficulty with the task might lead to a greater benefit for the patient group of prior exposure to
the stimuli (see Stanovich & West, 1979, for an analogous example involving semantic priming
in a comparison of poor and good readers). If the tasks are relatively easier for the healthy
individuals, they may rely less on previous recent experience to boost their performance.
There does not appear to be any evidence for this hypothesis, however, from the studies of AD
patients. In none of the studies did the patients show si~ificantly greater repetition priming
than controls, and there were no reported positive correlations between memory effects and
baseline performance. Nevertheless, it may be premature to rule out this possibility, because a
recent study using a relatively large sample of patients (N = 30) showed that priming in the
perceptual identification test was negatively correlated (r = -.39) with baseline performance
(Jemigan & Ostergaard, 1993). Perhaps the failure to find a similar result with AD patients
may be due to the limits of our data at the present time. Perhaps a wider range of ability and
larger samples would reveal similar relationships for AD patients.
The alternative hypothesis is that baseline performance is negatively correlated with the
benefit of repeated exposure. This hypothesis states that subjects experiencing difficulty in
carrying out a task may also be unable to retrieve recently obtained information relevant to that
task. Although almost no studies reported correlations between baseline performance and
repetition priming, a number of them examined relationships between indirect memory test
performance and the severity of cognitive dysfimction. In the verbal domain, no relationship
was found between dementia severity and improvement in reading text (Monti et al., 1994),
repetition priming for mirror-reversed words (Grober et al., 1992) or perceptual identification
repetition priming (Keane et al., 1991), but si~ificant correlations have been found for stem
completion repetition priming (Gabrieli et al., in press; Keane et al., 1991; Shimamura et al.,
1987). On the flee association test, Huff et al. (1988) reported that patients with greater
anomia and decreased ability to make semantic judgments showed less priming. In the non-
verbal domain, prism adaptation and pursuit rotor performance were not associated with
severity of dementia or initial performance (Heindel et al., 1989; Paulsen et al., 1993). On the
SRT task, inconsistent relationships have been reported. Ferraro et al. (1993) found si~ificant
correlations for mildly demented, but not for the very mildly demented patients, whereas
Knopman and Nissen (1987) showed si~ificant relationships with several tests of non-verbal
reasoning, but no relationship to overall dementia severity.
Given the small samples used for these analyses, the results are surprisingly consistent.
Overall, si~ificant relationships between indirect memory test performance and other
measures of cognitive dysfunction were found for tests that produced reduced learning in AD,
and no relationship for tests showing intact performance. Thus, on indirect memory tests that
give evidence of impairment in AD, there appears to be systematic variability in the patients'
performance, such that patients with greater cognitive impairment produce reduced learning.
If reduced learning on indirect memory measures is correlated with independent

measures of cognitive dysfunction, it remains still to determine the nature of this rdationship.
One possibility is that impairments in basic cognitive capacities such as attention mediate both
the memory and non-memory performances in AD patients (see also Ober & Shenaut, this
volume, for an extended discussion of the impact of attentional deficits on semantic priming).
Although one might then expect all indirect memory tests to be equally affected, this may not
be the case if different tasks vary in their reliance on these basic skills. The alternative is that
deficits in skills specifically needed for a particular task are responsible for the resulting deficit
on the indirect memory test. With regard to verbal skills, one would predict a relationship
between measures of language functioning and repetition priming, and indeed a significant
corrdation between verbal fluency and naming ability and levels of repetition priming has been
reported (Huff et al., 1988).
In examining more closely the relationship between reduced language functioning and
repetition priming in AD patients, it is informative to consider which language abilities are
most impaired in the early and middle stages of the disease. Overall, numerous findings in the
literature indicate intact semantic priming and normal selection of words under highly
constrained semantic contexts, but impaired ability to produce words when there are fewer
constraints, for example in response to category labels, pictures, or incomplete sentences with
many possible endings (see Nebes, 1989, for a review). The conclusions of numerous
researchers has been that this deficit is due to a combination of difficulty in word retrieval and
a loss of semantic knowledge (e.g., Butters, Granholm, Salmon, & Grant, 1987; Chertkow &
Bub, 1990; Chertkow, Bub, & Seidenberg, 1989; Huff, Corkin, & Growdon, 1986; Huff et al.,
1988; Martin, 1987; Randolph, Braun, Goldberg, & Chase, 1993). Furthermore, even when
AD patients have the requisite semantic knowledge, they may be inefficient or inconsistent in
the utilization of that knowledge (Hartman, 1991; Ober & Shenaut, this volume).
The pattern of intact performance under high-constraint situations and poor
performance under low-constraint situations on language tests also appears evident when
examining repetition priming performance in AD patients. The studies showing no impairment
all involved highly constrained reading tasks, whereas the impaired stem completion and word
association tests both had relatively low constraints on responding. It is also interesting to
note that stem completion repetition priming was intact in the two studiesusing an encoding
task that strongly directed semantic processing (Grosse et al., 1990; Partridge et al., 1990).
Other evidence that impairments in semantic processing may be present is the finding of
reduced impact of a levels of processing manipulation on direct memory tests (Gabrieli et al., in
press), a phenomenon that other researchers have noted as well (e.g., Corkin, 1982; Martin,
Cox, Brouwers, & Fedio, 1985). In conclusion, it seems likely that reduced repetition priming
on verbal tasks occurs in AD patients whenever the indirect memory tests tap underlying
language processing abilities that are impaired, and that this occurs despite the finding of
equivalent baseline performance on these tests. As noted above, however, one can not yet rule
out the possibility that more generalized cognitive dysfunction is responsible for the observed
pattern of deficits.
The conclusion that the integrity of language skills is an important factor in verbal
repetition priming also has implications for further refinement of the TAP theory. Based on the
work with Alzheimer's patients, it appears that the tasks traditionally grouped together as
perceptually-driven include at least two types of tasks. In one, repetition priming is largely
dependent on perceptual skills that rely on sensory input, as originally hypothesized (e.g.,
naming, lexical decision, and reading tasks). A second type involves learning based more
strongly on lexical knowledge and access (e.g., the stem completion test). This latter type still
remains distinct from conceptually-driven repetition priming tasks, which can be distinguished
by their sensitivity to elaborative semantic processing in healthy adults.
Regarding the study-test similarity feature of the TAP framework, it should be noted
that the tests with highly constrained language functioning (and showing no impairment in AD)
are also those that use identical tasks on first and second presentations of the stimuli. Thus, it
will remain for further research to ascertain which factor is more important. One way to test
this with the stem completion task, for instance, would be to determine whether the reduced
priming in AD patients is due to the stem completion task p e r se, and/or to an impairment in
the inability to profit from prior experience. One could compare stem completions when they
follow either the study of complete words, as has been done in previous studies, or the
generation of words from stems.
As noted earlier, several sets of findings were not readily explained by the TAP
framework. These involved the picture identification test, and possibly the verbal perceptual
identification test and the SRT task. The first of these is generally considered perceptually-
driven and thus might be expected to show no impairment. Nevertheless, subjects are
instructed to search their lexicon for the name for each picture. It now seems likely that this
requirement may account for the reduced priming in the AD patients. In the test for perceptual
identification of degraded words, it is not yet clear whether AD patients are impaired;
however, this test neither involved a change of procedure from study to test, nor required
lexical search. Thus, one would expect further research to confirm the absence of an
irnpairment on this test. As for the SRT task, the results from previous research is mixed, but
the task itself does not involve any changes in procedure during training. On this basis, one
would expect no impairment. Therefore, it appears that a different type of explanation is
needed to explain the impairments on this test. Research with healthy younger adults indicates
that the associative learning necessary to learn the sequence is sensitive to the amount of
available attention, with dual task procedures reducing the amount of learning on the original
version of the task (Nissen & Bullemer, 1987; but see Frensch & Miner, 1994, for a discussion
of other versions of the SRT task). Increased time between stimuli also reduces the degree of
learning (Frensch & Miner, 1994). Given the increased response times of the AD patients on
the SRT task and independent evidence for reduced attentional capacity in AD patients
(Baddeley, Logic, Bressi, Della Sala, & Spinnler, 1986), it seems likely that these account for
the impairments found in AD.
5. SUMMARY AND CONCLUSIONS
In conclusion, although the research on indirect memory tests with AD patients is still
relatively new, the results to date are fairly coherent. It appears that multiple mechanisms are
involved. First, the ability to learn new motor programs appears intact in AD; this can
probably be attributed to the relative sparing of subcortical structures and cortical motor areas
known to be important for this type of learning. Second, impairments in lexical retrieval and
reduced semantic knowledge (or access to that knowledge), along with intact perceptually-
based word skills, may account for the observed findings with verbal repetition priming tests.
Third, limits on attention may impact some indirect memory tests. These limits may explain
difficulties on the SRT task and may also contribute to the obtained pattern of impairments on
the verbal tasks. Although this proposed list of mechanisms does not provide a parsimonious
explanation of indirect memory test performance in AD, there is relatively strong independent
evidence for each mechanism in the research literature. Given that AD involves multiple
cognitive deficits, the necessity for complex explanations is plausible.
5.1. Implications For Understanding Cognitive Deficits In AD
The study of indirect memory tests in AD has provided evidence of non-uniform

deterioration of abilities in this domain. It is also clear that this is in part due to the non-
uniform deterioration of non-memory abilities, and this research has underlined the importance
of interactions between cognitive domains that occur with this disease. In order to better
understand these complex relationships, and by so doing, to understand the changes in indirect
memory test performance, research should continue to focus on the relationship between
indirect memory tests and impairments in the cognitive processes necessary to carry out these
tests. One way to do this is to obtain on-line performance measures on both the study and test
tasks, using variables such as number of errors of commission and omission, types of
responses, and response latencies. All of these may also provide baselines for measuring
improvements with repeated exposure to critical stimuli, so that multiple measures of learning
may be obtained. This information can help distinguish between a pure memory deficit and a
memory deficit resulting from impairments in the non-memory components of the tasks.
In addition to using more complete assessments in baseline conditions, it may prove
useful to consider the relationship between baseline scores and indirect measures of learning.
Here one needs to consider the methodological issue of how to measure learning. The
traditional method of examining absolute irnprovement (i.e., the difference from baseline)
contains the implicit assumption that there is no relationship between memory effects and
baseline performance. As pointed out above, this may be an unwarranted assumption (see also
Snodgrass, Corwin, & Feenan, 1990, for further discussion of this issue). The evidence that
baseline performance is sometimes correlated negatively to repetition priming scores (Jernigan
& Ostergaard, 1993; Snodgrass et al., 1993) gives further reason for concern. As a
consequence, it may be useful to consistently examine the data both in absolute and
proportional terms, and to interpret the results in relation to other evidence concerning the
integrity of non-memory components of performance.
A further recommendation for future research is that each study include independent
measures that tap the same non-memory cognitive processes that are required for the indirect
memory test. Correlations between indirect memory test performance and these tests can
provide further information about the source of impairments on the memory tests. Such data
may prove more useful than correlations with the less sensitive global measures of dementia
severity. *
* Although global measures of dementia have sometimes been found to correlate with repetition priming test
performance, they are still relatively insensitive tests. For instance, they do not predict performance on indirect
memory tests for dementias of different etiologies. As was noted by Shimamura et al. (1987), the overall
severity of dementia was similar for AD and liD patients who showed very different patterns of performance on
indirect memory tests.
5.2. Implications For Theories Of Indirect Memory Tests
Although the implications of this research for evaluating and developing theories of
memory have been discussed throughout this review, several final points remain. First, the
data are clear concerning the need for differentiating various types of indirect memory tests.
Although the existence of a distinct group of tests that is largely dependent on motor learning
is relatively clear-cut, further work is required in order to draw the boundaries between the
different types of verbally-based repetition priming tasks, and between verbal and pictorial
versions of similar tasks. In addition, the role of attention on complex tasks has yet to be
investigated with AD patients.
The literature on AD patients also speaks to the neuroanatomical basis of both motor
learning and repetition priming of discrete stimuli. With regard to the latter, numerous
researchers have suggested that the area of neocortex necessary for initial processing of a
particular type of stimuli is also the area that accrues an advantage from re-processing the
same stimuli (e.g., Fuster, 1984; Torres & Raz, 1994). Although the data on this matter are
not completely consistent (e.g., Jernigan & Ostergaard, 1993), the bulk of the evidence
supports such a conceptualization (see Torres & Raz, 1994, for a recent review). Thus,
perceptually-driven indirect tests may be primarily dependent on visual cortex in the right
hemisphere (Squire et al., 1992), whereas conceptually-driven indirect tests may involve
association cortex dedicated to language functioning. As Gabrieli et al. (in press) have noted,
the dissociations between AD patients' performance on stem completion and other repetition
priming tests may further help distinguish the structures involved in each of these tests.
Nevertheless, this type of work is still in its infancy. It is our feeling that research with AD
patients provides an exciting opportunity to contribute to the work that is currently being
conducted with healthy and amnesic populations, and offers potentially unique data that can
not be obtained from other types of subjects.
REFERENCES
Abbenhuis, M. A., Raaijmakers, W. G. M., Raaijmakers, J. G. W., & van Woerden, J. G. M.

(1990). Episodic memory in dementia of the Alzheimer type and in normal ageing: Similar
impairment in automatic processing. Quarterly Journal of Experimental Psychology, 42A,
569-583.
Baddeley, A., Logie, 1L, Bressi, S., Della Sala, S., & Spinnler, H. (1986). Dementia and
working memory. Quarterly Journal of Experimental Psychology, 38A, 603-618.
and contextual disambiguation effects in healthy aged individuals and individuals with
Senile Dementia of the Alzheimer Type. Brain and Language, 40, 181-201.
Blaxton, T. A. (1989). Investigating dissociations among memory measures: Support for a
transfer appropriate processing framework. Journal of Experimental Psychology:
Learning, Memory, and Cognition, 15, 657-668.
Blaxton, T. A. (1992). Dissociations among memory measures in memory-impaired subjects:
Evidence for a processing account of memory. Memory and Cognition, 20, 549-562.
Bondi, M. W. &. Kaszniak, A. W. (1991). Implicit and explicit memory in Alzheimer's disease
and Parkinson' s disease. Journal of Clinical and Experimental Neuropsychology, 13, 339-
358.
Bondi, M. W., Kaszniak, A. W., Rapcsak, S. Z., & Butters, M. A. (1993). Implicit and
explicit memory following anterior communicating artery aneurysm rupture. Brain and
Cognition, 22, 213-229.
Bowers, J. S., & Schacter, D. L. (1990). Implicit memory and test awareness. Journal of
Bran&, J., Spencer, M., McSorley, P., & Folstein, M. F. (1988). Semantic activation and
implicit memory in Alzheimer's disease. Alzheimer's Disease and Associated Disorders, 2,
112-119.
Butters, N., Granholm, E., Salmon, D. P. & Grant, I. (1987). Episodic and semantic memory:
A comparison of amnesic and demented patients. Journal of Clinical and Experimental
Challis, B. H., & Brodbeck, D. R. (1992). Level of processing affects priming in word
fragment completion. Journal of Experimental Psychology: Learning, Memory, and
Cognition, 18, 595-607.
Chenery, H. J., Ingram, J. C. L., & Murdoch, B. E. (1994). The effect of repeated prime-
target presentation in manipulating attention-induced priming in persons with Dementia of
the Alzheimer's Type. Brain and Cognition, 25, 108-127.
Chertkow, H., & Bub, D. (1990). Semantic memory loss in dementia of Alzheimer's type.
Brain, 13, 397-417.
Chertkow, H., Bub, D., & Seidenberg, M. (1989). Priming and semantic memory loss in
Alzheimer's Disease. Brain and Language, 36, 420-446.
Corkin, S. (1982). Some relationships between global amnesias and the memory impairments
in Al~eimer's Disease. In S. Corkin, K. L. David, J. H. Growdon, & E. Usdin (Eds.),
Alzheimer's disease: A report of progress in research (pp. 149-164). New York: Raven
Press.
Damasio, A. R., Van Hoesen, G. W., & Human, B. T. (1990). Reflections on the selectivity of
neuropathological changes in Alzheimer's Disease. In M. F. Schwartz (Ed.), Modular
deficits in Alzheimer-type dementia (pp. 83-100). Cambridge, MA: MIT Press.
Eslinger, P. J., & Damasio, A. R. (1986). Preserved motor learning in Alzheimer's disease:
Implications for anatomy and behavior. Journal of Neuroscience, 6, 3006-3009.
Ferraro, F. R., Balota, D. A., & Connor, L. T. (1993). Implicit memory and the formation of
new associations in nondemented Parkinson's disease individuals and individuals with senile
dementia of the Alzheimer type: A serial reaction time (SRT) investigation. Brain and
Cognition, 21, 163-180.
Frensch, P. A., & Miner, C. S. (1994). Effects of presentation rate and individual differences
in short-term capacity on an indirect measure of serial learning. Memory and Cognition,
22, 95-110.
Fuster, J. M. (1984). The cortical substrate of memory. In L. 1L Squire, & N. Butters (Eds.),
Neuropsychology of Memory (pp. 279-286). New York: Cafilford Press.
Gabrieli, J. D. E., Corkin, S., Mickel, S. F., & Growdon, J. H. (1993). Intact acquisition and
long-term retention of mirror-tracing skill in Alzheimer's disease and in global amnesia.
Behavioral Neuroscience, 107, 899-910.
Gabrieli, J. D. E., Keane, M. M., Stanger, B. Z., Kjelgaard, M. M., Corkin, S., & Growdon, J.
H. (in press). Dissociations among structural-perceptual, lexical-semantic, and event-fact
memory systems in Alzheimer, amnesic, and normal subjects. Cortex.
Indirect memory tests in AIzheimer's disease 291
Grat~ P., Shimamura, A. P., & Squire, L. R. (1985). Priming across modalities and priming
across category levels: Extending the domain of preserved function in amnesia. Journal of
Grober, E., Ausubel, 1L, Sliwinski, M., & Gordon, B. (1992). Skill learning and repetition
priming in Alzheimer's disease. Neuropsychologia, 30, 849-858.
Grosse, D. A., Wilson, 1L S. & Fox, J. H. (1990). Preserved word-stem-completion priming
of semantically encoded information in Alzheimer's disease. Psychology and Aging, 5,
304-306.
Hartman, M. (1991). The use of semantic knowledge in Alzheimer's disease: Evidence for
impairments of attention. Neuropsychologia, 29, 213-228.
Heindel, W. C., Salmon, D. P., & Butters, N. (1990). Pictorial priming and cued recall in
Alzheimer's and Huntington's disease. Brain and Cognition, 13, 282-295.
Heindel, W. C., Salmon, D. P. & Butters, N. (1991). The biasing of weight judgments in
Alzheimer's and Huntington's Disease: A priming or programming phenomenon? Journal
Heindel, W. C., Salmon, D. P., Shults, C. W., Walioke, P. A., & Butters, N. (1989).
Neuropsychological evidence for multiple memory systems: A comparison of Alzheimer's,
Huntington's, and Parkinson's disease patients. Journal of Neuroscience, 9, 582-587.
Hufl~ F. J., Corkin, S., & Growdon, J. H. (1986). Semantic impairment and anomia in
Alzheimer's disease. Brain and Language, 28, 235-249.
Hutt~ F. J., Mack, L., Mahlmann, J., & Greenberg, S. (1988). A comparison of lexical-
semantic impairments in left hemisphere stroke and Alzheimer's Disease. Brain and
Language, 34, 262-278.
Jacoby, L. L., (1991). A process dissociation framework: Separating automatic and
intentional uses of memory. Journal of Memory and Language, 30, 513-541.
Jernigan, T. L., & Ostergaard, A. L. (1993). Word priming and recognition memory are both
affected by mesial temporal lobe damage. Neuropsychology, 7, 14-26.
Kaszniak, A. W. (1986). The neuropsychology of dementia. In I. Grant, & D. M. Adams
(Eds.), Neuropsychological assessment of neuropsychiatric disorders (pp. 172-220). New
York: Oxford University Press.
Keane, M. M., Gabrieli, J. D. E., Fennema, A. C., Growdon, J. H., & Corkin, S. (1991).
Evidence for a dissociation between perceptual and conceptual priming in Alzheimer's
disease. Behavioral Neuroscience, 105, 326-342.
Keane, M. M., Gabrieli, J. D. E., Growdon, J. H., & Corkin, S. (1994). Priming in perceptual
identification of pseudowords is normal in Alzheimer's disease. Neuropsychologia, 32,
343-356.
Knopman, D. (1991). Long-term retention of implicitly acquired learning in patients with
Alzheimer's Disease. Journal of Clinical and Experimental Neuropsychology, 13, 880-
894.
Knopman, D. S., & Nissen, M. J. (1987). Implicit learning in patients with probable
Alzheimer's disease. Neurology, 37, 784-788.
Knopman, D. S., & Nissen, M. J. (1991). Procedural learning is impaired in Huntington's
disease: Evidence from the serial reaction-time task. Neuropsychologia, 29, 245-254.
292 M. Hartmanand M.L. Pirnot
Martin, A. (1987). Representation of semantic and spatial knowledge in Alzheimer's patients:

Implications for models of preserved learning in amnesia. Journal of Clinical and
Experimental Neuropsychology, 9, 191-224.
Martin, A., Cox, C., Brouwers, P., & Fedio, P. (1985). A note on different patterns of
irnpaired and preserved cognitive abilities and their relation to episodic memory deficits in
Alzheimer's patients. Brain and Language, 26, 181-185.
McKann, G., Drachman, D., Folstein, M., Katzman, IL, Price, D., & Stadlan, E. M. (1984).
Clinical diagnosis of Alzheimer's disease: Report of the NINCDS-ADRDA Work group
under the auspices of Department of Health and Human Services Task Force on
Alzheimer's Disease. Neurology, 34, 939-944.
Monti, L. A., Crabrieli, J. D. E., Reminger, S. L., Grosse, D. A., & Wilson, 1L S. (1992). A
specifically conceptual priming deficit in patients with Alzheimer's disease. Society for
Neuroscienee Abstraets, 18, 734.
Monti, L. A., Gabrieli, J. D. E., Wilson, 1L S., & Reminger, S. L. (1994). Intact text-specific
implicit memory in patients with Alzheimer's disease. Psychology and Aging, 9, 64-71.
Morris, C. D., Bransford, J. D., & Franks, J. J. (1977). Levels of processing versus transfer
appropriate processing. Journal of Verbal Learning and Verbal Behavior, 16, 519-533.
Nebes, R. D. (1989). Semantic memory in Alzheimer's disease. Psychological Bulletin, 106,
377-394.
Nissen, M. J., & Bullemer, P. (1987). Attentional requirements of learning: Evidence from
performance measures. Cognitive Psychology, 19, 1-32.
Ober, B. A., & Shenaut, G. K. (1988). Lexical decision and priming in Alzheimer's Disease.
Neuropsyehologia, 26, 273-286.
Ober, B. A., Shenaut, G. K., Jagust, W. J., & Stillman, R. C. (1991). Automatic semantic
priming with various category relations in Alzheimer's Disease and normal aging.
Partridge, F. M., Knight, R. G., & Feehan, M. (1990). Direct and indirect memory
performance in patients with senile dementia. Psychological Medicine, 20, 111-118.
Paulsen, J. S., Butters, N., Salmon, D. P., Heindel, W. C., & Swenson, M. R. (1993). Prism
adaptation in Alzheimer's and Huntington's disease. Neuropsyehology, 7, 73-81.
Perfect, T. J., Dowries, J. J., De Momay Davies, P., & Wilson, K. (1992). Preserved implicit
memory for lexical information in Alzheimer's disease. Perceptual and Motor Skills, 74,
747-754.
Randolph, C. (1991). Implicit, explicit, and semantic memory functions in AlzJaeimer's disease
and Huntington's disease. Journal of Clinical and Experimental Neuropsyehology, 13,
479-494.
Randolph, C., Braun, A. R., Goldberg, T. E., & Chase, T. N. (1993). Semantic fluency in
Alzheimer's, Parkinson's, and Huntington's disease: Dissociation of storage and retrieval
failures. Neuropsychology, 7, 82-88.
Richardson-Klavehn, A., & Bjork, 1~ A. (1988). Measures of memory. Annual Review of
Richardson-Klavehn, A., Lee, M. G., Joubran, Ik, & Bjork, R. A. (1994). Intention and
awareness in perceptual identification priming. Memory and Cognition, 22, 293-312.
Roediger, H. L. HI, & Blaxton, T. A. (1987). Retrieval modes produce dissociations in
memory for surface information. In D. S. Gorfein & 1~ R. Hoffman (Eds.), Memory and
learning." The Ebbinghaus Centennial Conference (pp. 349-379). Hillsdale, NJ: Erlbaunl
Indirect memorytests in Alzheimer's disease 293
Roediger, H. L. HI, & McDermott, I~ B. (1993). Implicit memory in normal human subjects.
In H. Spinnler & F. Boiler (Eds.), Handbook ofNeuropsychology. Am~erdam: Elsevier.
Salmon, D. P., Shimamura, A. P., Butters, N., & Smith, S. (1988). Lexical and semantic
priming deficits in patients with Alzheimer's disease. Journal of Clinical and Experimental
Schacter, D. L. (1987). Implicit memory: history and current status. Journal of Experimental
Schacter, D. L., Chiu, P., & Ochsner, K. N. (1993). Implicit memory: A selective review.
Annual Review of Neuroseience, 16, 159-182.
Shimamura, A. P., Salmon, D. P., Squire, L. R., & Butters, N. (1987). Memory dysfunction
and word priming in dementia and amnesia. Behavioral Neuroseienee, 101, 347-351.
Snodgrass, J. G., Corwin, J., & Keenan, K. (February, 1990). The pragmaties of measuring
priming: Applications to normal and abnormal memory. Paper presented at the meeting
of the Imemational Neuropsychological Society, Orlando, FL.
Squire, L. 1L (1992). Memory and the hippocampus: a synthesis from findings with rats,
monkeys, and humans. Psychological Review, 99, 195-231.
Squire, L. IL, Ojemann, J. G., Miezin, F. M., Petersen, S. E., Videen, T. O., & Raiehle, M. E.
(1992). Activation of the hippoeampus in normal humans: A functional anatomical study
of memory. Proceedings of the National Academy of Sciences, 89, 1837-1841.
Stanovich, K. E., & West, R. F. (1979). Mechanisms of sentence context effects in reading:
Automatic activation and conscious attention. Memory and Cognition, 7, 77-85.
Tortes, I. J., & Raz, N. (1994). Toward the neural basis of verbal priming: A cognitive-
neuropsychological synthesis. Neuropsyehology Review, 4, 1-30.
Toth, F. P., Reingold, E. M., & Jacoby, L. L. (1994). Toward a redefinition of implicit
memory: Process dissociations following elaborative processing and self-generation.
Warrington, E. K., & Weiskrantz, L. (1968). New method of testing long-term retention with
special reference to amnesic patients. Nature, 217, 972-974.
1995 Elsevier Science B.V.
Using Event-Related Brain Potentials to Draw Inferences About Human

I n f o r m a t i o n Processing*
K. Richard Ridderinkhof' and Theodore R. Bashore b
aDepartment of Psychology, University of Amsterdam, The Netherlands
bDepartment of Psychology, University of Northern Colorado, Greeley, CO
1. E V E N T - R E L A T E D BRAIN POTENTIALS: A BRIEF I N T R O D U C T I O N
The spontaneous electrical activity of the human brain consists of field potentials
produced by postsynaptic discharges in large populations of neurons. This activity can be
measured indirectly at the scalp in the form of the electroencephalogram or EEG (Berger,
1929). The EEG represents continuous variation in on-going brain electrical activity.
Presentation of a stimulus can elicit a transient change in this otherwise continuous EEG
activity. That is, a stimulus may produce positive or negative shifts in the voltage of the EEG
that are distinguishable from its continuously varying background activity. In an early
observation of this phenomenon Davis (1939) identified a large negative deflection in the EEG
approximately 100 msec after delivery of intense auditory stimulation. This negative change in
the electrical polarity of the EEG was of brief duration and was time-locked to the presentation
of the auditory stimulus. Patterns of positive and negative deflections in the on-going EEG
that are elicited in response to the presentation of a stimulus, like the negative shift observed
by Davis (1939), and/or to the execution of a movement are referred to as event-related brain
potentials or ERPs. ERPs reflect changes in electrophysiological activity generated in the
brain when an individual receives some form of sensory stimulation or initiates a pre-
designated movement. The ERP is manifest as a series of voltage-time fluctuations that are
most apparent in time-locked averages of EEG activity derived from repeated occurrences of
the critical event. Each deflection in the series, identified separately as components of the
ERP, represents synchronous changes in the electrical activity of large populations of neurons.
The latency and amplitude of these components may be sensitive to variations in the physical
characteristics of the eliciting stimulus, in the nature of the intended movement, or in the
processing demands of a task.
These signals are recorded using electrodes that are affixed to the scalp at a number of
different sites. The locations of these sites have been standardized so that brain electrical
ACKNOWLEDGEMENTS: This chapter was written while the second author was a Visiting Professor at the
University of Amsterdam, supported by the Experimental Psychology Graduate School (EPOS) in the
Netherlands, by a visitor's grant from NWO (The Dutch Organization for Pure Research), and by a stipend
from the Research Corporation at the University of Northern Colorado. In addition, preparation of this chapter
was supported in part by grants to the second author from the National Institute on Aging (AG04581,
AG12263), and by a research grant from NWO to M.W. van der Molen that provides postdoctoral support for
the first author.
Using event-related brain potentials 295
activity recorded in different clinics or laboratories throughout the world can be compared
(Jasper, 1958). This system, known as the 10-20 or International Electrode Placement System,
has established the sites shown in Figure 1 as standard. In most ERP studies, recording
electrodes are placed along what are called the midline scalp sites, Fz, Cz, and Pz. When
visual stimuli are used, a fourth midline site is used, Oz. This site was not identified as a
standard site in the original 10-20 system_ However, it has become such a commonly used
Nasion
A1 A2
Ini~}n
Figure 1. A two-dimensional projection plane of the head showing all standard electrode placements according
to the International Electrode (10-20) Placement System (Jasper, 1958). A1 and A2 are reference electrodes,
usually attached to the earlobes. The lengths of the lines connecting the nasion (bone above the nose) to inion
(bone at base of the posterior mid-cranium) and A1 to A2 are measured; electrode positions are determined
relative to the points that divide these lines into segmentsof 10, 20, 20, 20, 20, and 10% of the line length. The
frequently used electrode placement Oz is not part of the official 10-20 system.
location in visual ERP studies that it is for all practical purposes a standard placement site.
Scalp sites are labeled on the basis of the area of cortex they overlie. Thus, Fz refers to the
frontal, Cz to the central, Pz to the parietal, and Oz to the occipital midline site, respectively.
Sites located to the left and the fight of midline are also labeled by number; odd numbers on
the left and even numbers on the fight. For example, the site P3 refers to a location overlying
left parietal cortex. The placement system is referred to as the 10-20 system because the
relative locations of many critical electrode sites are 10% or 20% of certain measured distances
over the scalp. With advancing EEG technology, such as those involved in brain mapping and
deriving the sources for signals measured at the scalp, the number of electrode sites has grown.
296 K.R. Ridderinkhof and Th.R. Bashore
As many as 128 electrodes may be used in some applications. All of these configurations are
derivatives, however, of the 10-20 system
1.1. Deriving ERPs from the Background EEG: Signal Analysis
Event-related brain potentials are embedded in a background of random EEG activity

that is typically many times larger than the components of the ERP. As a result, the
components of interest may not be accurately distinguished from the background EEG
following presentation of a single stimulus. The background EEG, composed of many signals
of different frequencies, is considered noise in the context of ERP research. Noise is assumed
to comprise those constituent frequencies of the EEG that do not respond systematically to a
stimulus or are not time-locked to the execution of a movement. This assumption provides the
rationale for averaging EEG recordings over large numbers of trials: The non-systematic noise
contributions average to zero, whereas the systematic (event- or movement-related) voltage-
time changes remain unaffected. Thus, the time-locked activity is assumed to emerge from the
diminished background noise. It should be noted, however, that because of inter-trial
variability in the latency of ERP components, components in the average ERP may be flatter
and broader in appearance than those evident in the single-trial ERP. The properties of the
ERP that are of interest to investigators in the field, peak amplitude and latency, are usually
derived from the average ERP. Since the average ERP may be influenced by single-trial
variation in both the amplitudes and latencies of its constituent components, estimates of both
the amplitude and latency of a given component may not represent the single-trial distributions
precisely. However, if the signal-to-noise ratio is large enough to allow reliable single-trial
estimates of these properties of the ERP, means can be calculated from values measured on
individual trials. Estimates derived in this manner are more representative of the underlying
distribution than are estimates taken from the average ERP, particularly for components that
have relatively large single trial variability (Smulders, Kenemans, & Kok, 1994). A variety of
procedures have been used to estimate the latency and amplitude of ERP components, both
from single-trial and averaged signals, and are well-documented in the literature (e.g., Donchin
& Heflley, 1978). Of particular concern for single-trial analysis is selection of a digital filter to
reduce the contribution of noise to the morphology of the ERP (e.g., Farwell, Martinerie,
Bashore, Rapp, & Goddard, 1993). Discussion of these techniques falls beyond the scope of
this chapter; however, it is important to be mindful of the fact that any factor effect on the
parameters of ERP components may be influenced by the procedures used to extract signal
from noise in the ERP and to derive the estimates of peak amplitude and latency.
1.2. Taxonomy of Components of the Event-Related Brain Potential
Presentation of a stimulus elicits an ERP composed of components that have been

categorized into two broad types, exogenous and endogenous (Donchin, Ritter, & McCallum,
1978). Exogenous components usually occur within the first 150 msec after stimulus onset
and may even be detected as early as several milliseconds after stimulation. These components
are thought to reflect the obligatory activation of neuroanatomical structures in the stimulated
primary sensory pathways, and are commonly referred to as sensory evoked potentials. The
latencies and amplitudes of exogenous components vary with changes in the physical
properties of the eliciting stimulus (e.g., an increase in stimulus intensity is typically associated
with increases in component amplitudes). The distribution of an ERP component over the
scalp is determined by its relative amplitude at the various electrode sites where it is measured.
Exogenous components have characteristic distributions across scalp electrode sites (i.e., scalp
topographies) that vary with the source of stimulation. Their amplitudes are usually maximal
at scalp sites overlying the primary cortical receptive area in which the stimulation was
delivered. For example, visual stimulation elicits exogenous components that typically have
their largest amplitudes at occipital electrode sites. A fundamentally irnportant property of
exogenous components is that their appearance on the scalp is not dependent on the subject
doing anything other than passively receiving the stimulus; indeed, they can be evoked in a
comatose patient, provided that the primary sensory pathways of the patient are intact.
In contrast to exogenous components, endogenous components are manifest in the
ERP only under conditions in which the subject is actively engaged in a task that requires
processing a stimulus and making some decision about it. These components of the ERP are
relatively insensitive to variations in the physical characteristics of a stimulus. Rather, the
latency and amplitude of endogenous components are sensitive to variations in the stimulus and
response processing demands imposed by a task. Thus, the same stimulus may elicit an
endogenous component under some task conditions, but not under others. By contrast,
presentation of a stimulus will always elicit exogenous components, irrespective of the
processing demands associated with it. The sensitivity of endogenous components to
variations in task demands suggests that changes in their latency and amplitude may reveal
temporal and structural aspects of specific information processing activities. This property
renders these ERP components suitable to augment traditional performance measures (such as
reaction time and accuracy) in studying human information processing. Unlike the exogenous
components of the ERP, endogenous components are usually thought to occur at latencies
longer than 150 msec following stimulus presentation, and to have reasonably invariant scalp
topographies across different stimulus modalities. For instance, the amplitude of the P300
component of the ERP is maximal at the midline scalp site, Pz, regardless of the type of
stimulation (somatosensory, auditory, visual) delivered.
It should be noted that cognitive influences have been identified on what had once been
thought to be exclusively exogenous components of the ERP. These influences are evident on
components within the latency range spanning 30 to 200 msec following stimulation. The
properties of these components can be influenced not only by changes in the physical
characteristics of the eliciting stimulus, but also by variations in the stimulus processing
requirements induced by the task. Recent evidence t~om selective attention tasks, for example,
has demonstrated that the demand to attend differentially to one stimulus over others is
associated with increases in component amplitude as early as 30 to 80 msec following delivery
of the critical stimulus in the somatosensory (e.g., Desmedt, Huy, & Bourguet, 1983), auditory
(e.g., Guterman, Josiassen, & Bashore, 1992), and visual (e.g., Kenemans, Kok, & Smulders,
1993) modalities.
1.3. Nomenclature
By convention, ERP components are labeled on the basis of their electrical polarity
(positive or negative) and the average latency at which they attain their maximum amplitude in
young adults. For example, the P100 component refers to a positive change in electrical
polarity that achieves its peak amplitude about 100 msec following delivery of a visual
stimulus, whereas the N200 component has a negative polarity that is largest about 200 msec
post-stimulus. These potentials are largest at posterior scalp sites. They are thought to index
facilitation of processing for attended visual features, as they are larger in magnitude for stimuli
presented foveaUy (or in the focus of attention) compared to those presented peripherally
(Mangtm, Hillyard, & Luck, 1993). The latency element of these naming conventions,
particularly for endogenous components, is approximate. The actual observed latency for a
component may vary as a function of variations in experimental factors such as stimulus
modality, stimulus complexity, stimulus-response compatibility, response complexity, or
subject age. For example, the P300 component has been observed to peak at about 250 msec
in 15-year-olds when they make a simple discrimination between two tones (Bashore, 1990), at
620 msec in 5-year-olds when they process a multielement visual stimulus array (Ridderinkhof
& van der Molen, in press), and at 800-900 msec in 70-year-olds when they perform a difficult
visual discrimination (Bashore, 1990). Given this latency variability, the component is
identified as a P300 on the basis of its electrical polarity, scalp distribution, and responsivity to
experimental manipulations. Alternatively, ERP components may be labeled according to their
polarity and the order in which they appear in the time series. Following this convention, the
P100 is called P1 and the N200 is called N2. Many investigators follow this naming
convention.
Some components are not named according to either of these conventions. Rather,
they are named on the bails of their putative functional si~ificance. These components are
typically endogenous in nature. Among these components are those that are associated with
preparation and initiation of movements. They are often referred to as movement-related
potentials (Kutas & Donchin, 1980). For example, a slow, ramp-like negative component
always develops between the presentation of a warning stimulus and an imperative stimulus
that calls for some decision. This component was thought by the investigators who discovered
it, Walter, Cooper, Aldridge, McCallum, and Winter (1964), to reflect the development of a
subject's expectancy for the imperative stimulus during the preparatory interval; hence, it was
named the contingent negative variation or CNV. At about the same time, Kornhuber and
Deecke (1965) reported another slow, ramp-like negative-going component that began to
develop between 1000 to 1500 msec prior to execution of an tmsignaled voluntary movement
and reached its maximum amplitude at about the time the movement was made (e.g., flexion of
the index finger). They labeled this component the Bereitschaflspotential (i.e., readiness
potential), reflecting its presumed association with preparation to execute a movement. The
CNV and readiness potential are now thought to share a functional relationship. Work by
Rohrbaugh and colleagues suggests that the CNV may be composed of two constituents: an
early portion that is an exogenous response to the warning stimulus and a late portion,
proximal to the onset of the stimulus, that resembles the readiness potential (Rohrbaugh &
Gaillard, 1983). The heritage of the recently identified lateralized readiness potential (LRP),
described at more length below, can be traced to both the CNV and the Bereitsehaflspotential.
It has proven to be a highly sensitive measure of preferential response preparation (e.g., Coles,
1989).
Other components of the ERP are obtained by subtracting the ERP time series of one
experimental condition from that of another condition, and then identifying si~ificant
deflections in the resulting difference wave. These components are usually endogenous and
most commonly are named on the bails of their presumed functional si~ificance. An example
of such a component is the processing negativity. Identification of this component followed
from the classic research ofHaider, Spong, and Lindsley (1969). They reported a difference in
amplitude of the N100-P200 complex of ERPs to attended as compared to unattended auditory
stimuli. Subsequent research led, in turn, to use of the subtraction method to reveal the
processing negativity (denoted as Nd). The Nd is a long-lasting negative deflection apparent in
the difference wave that results when the ERP to stimuli that are not monitored is subtracted
from the ERP to stimuli that are monitored (for review, see Naatanen, 1975). Another
example of a component that is identified by subtraction is the selection negativity (SN), a
negative deflection in the difference wave generated by subtracting the ERP to the relevant
level of a stimulus dimension from the ERP to the irrelevant level (for review, see Hansen &
Hillyard, 1983). Yet another example of a component revealed through subtraction is the
mismatch negativity (MMN). It is a negative component with a frontocentral scalp distribution
and a variable latency than can be as early as 140 msec after stimulus presentation (for review,
see Naatanen, 1985). The MMN is observed when a subject detects a deviant stimulus (i.e., a
mismatch) in a rapid monotonous sequence of visual or auditory stimuli, both when they are
attended or unattended. It is exposed by subtracting the ERP to the standard stimulus from the
ERP to the deviant stimulus. All of these negative components are thought to reflect the
activation of selective attention mechanisms.
2. ILLUSTRATIVE EXAMPLES OF SOME RELEVANT ERP COMPONENTS
EEG recordings during task performance provide valuable tools for cognitive
psychophysiologists and neuroscientists to draw inferences about the structural and temporal
aspects of human information processing. Rather than providing an exhaustive review of this
research, our presentation is restricted to examples of components that are discussed in the
chapters that follow in this portion of the book (for extensive reviews, see Callaway, Tueting,
& Koslow, 1978; Coles, Donchin, & Porges, 1986; van der Molen, Bashore, Halliday, &
Callaway, 1991; Woods, 1990). In these chapters, a number of different ERP components are
discussed that have been studied to deepen our insight into the effects of aging on language
processing, remembering, and mental processing speed. In this section, we illustrate the use of
some of these ERP components (the N400, Dm, P300, and LRP), and briefly discuss issues
pertinent to deriving and interpreting the functional significance of these components.
2.1. The N400 Component and Semantic Processing
The role of linguistic context in establishing semantic expectancies on the properties of

the ERP has been investigated extensively since the pioneering studies of Kutas and Hillyard
(1980a,b). These expectancies can have a significant influence on the recognition of semantic
units like words and sentences. Kutas and Hillyard (1980a,b) first investigated the role of
contextual factors in linguistic processing by comparing the ERPs elicited by confirmations and
disconfirmations of semantic expectancies as sentences were being read. Some of the
sentences were completed by semantically anomalous words that rendered them nonsensical,
whereas others were completed by semantically-appropriate words. The ERP seen in response
to a semantically-inappropriate word was characterized by the emergence of a negative
component between 300 and 600 msec atter presentation of the closing word that was not
evident when the closing word was semantically-appropriate. Kutas and Hillyard named this
negative component the N400. It is larger in amplitude over posterior than anterior scalp sites,
and is slightly larger over right than left hemisphere sites. The N400 has also been observed in
response to semantic anomalies associated with spoken words, reinforcing the view that it is
sensitive to violations of semantic expectations in the lingt~ic domain (e.g., McCallum,
Farmer, & Pocock, 1984). This inference has also been supported by the failure of pure
grammatical anomalies (e.g., Kutas & Hillyard, 1983) or anomalies in non-linguistic stimulus
sequences such as melodies (e.g., Besson & Macar, 1987) to elicit an N400. Other research
has indicated that an N400 can be invoked in a wide variety of linguistic conditions, provided
that the target word is not predictable from a preceding linguistic context (for reviews, see
Fischler & Raney, 1991; Kutas & van Petten, 1994).
The overall meaning of a sentence may not be the only crucial factor in generating an
N400. For instance, Fischler and co-workers (Fischler, Bloom, Childers, Roucos, & Perry,
1983; Fischler, Childers, Achariyapaopan, & Perry, 1985) showed that false statements formed
by category associates (e.g., 'robins are sparrows') elicited N400s that were larger than those
CRY~,
A
THE PIZZA WAS TOO HOT TO ] ~DRINK.,
- ! ~ "EAT "'"'"
5#V
I
_1.1 I t i i I I t i !
0 400 msee
BEST COMPLETIONS
............. RELATED ANOMALIES
. . . . UNRELATED ANOMALIES
Figure 2. An example of the N400 recorded from the midline parietal electrode site, Pz. It is largest in
amplitude to anomalous sentence completions, smallest to expected sentence completions, and intermediate to
anomalous completions that were related to the expected completion. (Data from Kutas, Lindamood, &
Hillyard, 1984; adapted from Kutas & van Petten, 1988).
seen in response to true statements (e.g., 'robins are birds') but smaller than those generated by
false statements composed of unassociated categories (e.g., 'robins are wagons'). In a similar
vein, consider an illustration of this point used in a review by Kutas, Lindawood, abd HiUyard
(1984). As can be seen in Figure 2, completion of a sentence with a semantically anomalous
word yielded a smaller N400 when the word was related to the expected completion (e.g., 'the
pizza was too hot to drink') than when it was unrelated (e.g., ~he pizza was too hot to cry').
Maturation and aging have also been demonstrated to affect the properties of the
N400. For example, peak latencies of the N400 shorten from age 6 to 9 (Friedman, Putnam, &
Sutton, 1990), are prolonged in middle-aged (with a concomitant amplitude reduction; Gunter,
Jackson, & Mulder, 1992) and elderly (Harbin, Marsh, & Harvey, 1984) adults as compared to
young adults, and the amplitude of the N400 has been observed to be larger in normal children
than that in their reading-disabled peers (Stelmack & Miles, 1990). The chapter by King and
Kutas in this volume reviews the sensitivity of linguistic processing to the effects of older age
as revealed in the N400.
2.2. The Dm Component and Memory Performance
Presentation of stimuli that must be remembered for successful completion of later

memory tasks elicits ERPs with components whose amplitudes are predictive of subsequent
recall or recognition of those stimuli. These ERPs are characterized in general by a positivity
that is larger than that seen in the ERPs invoked by stimuli that are not remembered,
particularly between 400 to 800 msec after stimulus presentation (e.g., Besson & Kutas, 1983;
Friedman & Sutton, 1987; Neville, Kutas, Chesney, & Schmidt, 1986; Paller, McCarthy, &
Wood, 1988). This enhanced positivity is revealed in a difference wave that is called the Dm
(Paller, Kutas, & Mayes, 1987) because of its relation to subsequent memory performance
(D=difference; m=memory).
The Dm has been shown to be influenced in explicit memory tasks (such as recall and
recognition) but not in implicit memory tasks (such as stem completion), supporting the notion
that different aspects of memory are assessed by these two types of tasks (Paller, 1990). This
component has also been shown to be influenced to a larger extent in an intentional memory
task (in which voluntary learning occurs, as in learning a foreign language) than in an incidental
memory task (in which unintentional learning of material from a secondary task occurs;
Muente, Heinze, Scholz, & Kuenkel, 1988). In addition, variations in the Dm are more closely
associated with recollection of words repeated from an imagery task (requiring size
estimations) than of words repeated from an orthographic task (requiring letter counting; Paller
& Kutas, 1992).
Besson and Kutas (1993) investigated the differential effects of context repetition and
terminal word repetition on Dm in a task requiring cued-recall of sentences. As revealed in
Figure 3, context repetition effects on Dm were manifest, whereas terminal word repetition
effects were small or absent. The finding that the amplitude of the Dm varied across different
repetition conditions suggested to Besson and Kutas that the memory-related process
subserving cued-recall performance is not all-or-none, but rather is engaged in a graded
fashion.
The Dm is not manifest in all tasks that require explicit memory operations, however.
Failures to elicit a Dm may occur in tasks that use complex semantic associations, such as
mnemonic aids (e.g., Karis, Fabiani, & Donchin, 1984). Furthermore, in order to obtain a
reliable Din, large numbers of test items are required to ensure sufficient signal-to-noise ratios
(Paller, Kutas, Shimamura, & Squire, 1987). It should be noted that the relation between
modulations of Dm and of other components in overlapping latency intervals is not well-
articulated (see discussions in Besson, Kutas, & van Petten, 1992; Friedman, 1990; Rugg,
Furda, & Lorist, 1988). Nevertheless, even though the extent to which Dm effects result from
modulations of other ERP components is not yet determined, useful inferences about memory
processes may be drawn from observations of experimental effects on its properties (Pallet &
Kutas, 1992). The chapter by Friedman and Fabiani in this volume provides such an example
in its discussion of the effects of aging on memory processes revealed in the Dnl
CDNTE~T EFFECT WORD EFFECT DIFFERENCEWAVE
CENTRAL
PARIETAL /~ .~
TEMPORAL
R. ANTERIOR
TEMPORAL ~ ~.~.~,,,~.~
L. P O S T E R I O R ~
TEMPORAL
TEMPORAL
"1
.......... ,~,"
; " ~ ' ~0' 0 400 800 0 400 ~0mn
Same Context ~ S4um WoPd _ _ Context

.._ OJff Context ... Diff llm'd .__ 1lord
Figure 3. An illustration of the Dm. The effects on subsequent cued-recall performance of context repetition
and terminal word repetition are shown in the left and middle panels, respectively; the fight panel shows Dms
for context (solid lines) and terminal word (dashed lines) repetition effects. Repeating the sentence context did
elicit a Dm, whereas repeating the terminal word of the sentence had little or no effect. (Reprinted from Besson
and Kutas, 1993.)
2.3. The P300 Component and Stimulus Processing
The P300 or P3 is perhaps the most widely studied ERP component in the cognitive
psychophysiological literature (for reviews, see Coles, Gratton, & Fabiani, 1990; Donchin,
Karis, Bashore, Coles, & Gratton, 1986), due in part to the ease with which it can be measured
and detected in the background EEG, even on individual trials. The P300 was first reported in
two papers by Sutton and colleagues in the mid-1960s (for a review, see Bashore & van der
Molen, 1991). Sutton, Braren, Zubin, and John (1965) discovered that a large positive
component in the ERP was elicited when subjects had to guess the occurrence of an
unpredictable auditory stimulus. Their work also revealed that the amplitude of this positive
component, labeled by Sutton et al. as the P300, was negatively correlated with the probability
of occurrence of the target stimulus; the lower its probability, the larger the amplitude of the
P300. Sutton, Tueting, Zubin, and John (1967) showed further that the same external stimulus
events could invoke P300s of different amplitudes, depending on the type of responses subjects
were instructed to make to these stimuli. Moreover, they demonstrated that the P300 could be
generated in the absence of a stimulus event, if that absence was relevant to successful
performance of the task.
Following these initial observations, the sensitivity of the amplitude of the P300 to
variations in target probability, in subjective probability, and in the utility of the cliciting
stimulus event have been demonstrated repeatedly (e.g., Duncan-Johnson & Donchin, 1977;
Johnson & Donchin, 1978). In the paradigmatic procedure for cliciting a P300, the oddball
task, two stimuli or stimulus categories are presented in a pscudorandom sequence, such that
the probability of occurrence of one stimulus is far less than that of the other stimulus. The
subject's task is to count or to press a button to the rare stimulus (or stimulus class) and to
refrain from counting or pressing a button to the frequent stimulus (or stimulus class).
Typically, the rare stimulus event elicits a larger P300 than does the frequent stimulus event
(for review, see Johnson, 1988). IntereStingly, P300 amplitude to an informative pre-cue has
been demonstrated to be predictive of later use of the information conveyed by that pre-cue.
That is, if the P300 to a valid pro-cue is large, the response to the subsequent imperative
stimulus will be faster on average than when the pro-cue is small (Gratton ct al., 1990).
The peak latency of the P300 has been demonstrated to be sensitive to the duration of
stimulus processing and relatively insensitive to the duration of response processing. Thus,
P300 latency has been reported to vary systematically as a function of experimental
manipulations that influence stimulus evaluation processes, but to vary far less systematically to
manipulations that affect response selection and execution (e.g., Coles, Gchring, Gratton, &
Donchin, 1992; Kutas, McCarthy, & Donchin, 1977; Magliero, Bashore, Coles, & Donchin,
1984; McCarthy & Donchin, 1981; Ragot, 1990).
It should be noted that there is some controversy over the precise functional
significance of the P300, particularly as it relates to the amplitude of this component (for
critical reviews, see Pdtchard, 198 l; Verlcger, 1988). It is reasonable to assert, however, that
there is general agreement that the predominant influence on the peak latency of the P300 is
provided by variations in stimulus processing demands.
The putative differential sensitivity of P300 latency to variations in stimulus processing
demands has justified the use of this measure in conjunction with reaction time to study mental
chronomctry (i.e., the structure and timing of mental processing). One example of this type of
research is the effort to evaluate the continuous flow model of Erikscn and Schultz (1979; see,
for example, Coles, Gratton, Bashore, Erikscn, & Donchin, 1985; Riddcrinkhof & van dcr
Molcn, in press a; Smid, Mulder, & Muldcr, 1990). This model assumes that partial stimulus
information is transmitted continuously to the response output system as the stimulus is being
processed. In the paradigmatic (Erikscn) task, subjects are presented a multiclcmcnt horizontal
stimulus array that consists of a center stimulus flanked on each side by two other stimulus
elements. Typically, there are two stimuli (e.g., H, S), one signaling a left response (e.g., H)
and the other a right response (e.g,. S). Hankers are either identical with the center stimulus
(i.e., HHHHH, a congruent array) or are the alternative stimulus (i.e., SSHSS, an incongruent
array). That is, the flankers either signal the same or the opposite response as the center
stimulus. Subjects are instructed to ignore the flankers and to make their response output
decision exclusively on the basis of the identity of the center stimulus. Despite these
2O
~ 0 "7-" ~
"N
~
"~
-20
' Cmgtuem
I
9 Im:ongtuem " ' Neutral
i
2O"1
"~;>" 0 [ . . . . . . . .
.20
'~ o . . . . . . .
.20 J
!
2O
-1000 -500 0 500 1000

Time from Stimulus Presentation (ms)
Figure 4. An example of the P300 recorded from the midline parietal electrode site, Pz, for young adults (upper
panel), 10 to 12 year-olds, (2nd panel), 7 to 9 year-olds (3rd panel), and 5 to 6 year-olds (lower panel). Positive
changes in electrical polarity are shown as upward deflections in this figure. It is equally common to show
positive values as downward deflections. It can be seen that the latency of the P300 (the prominent positive
deflection) progressively decreases with age. P300 latency is delayed further in incongruent conditions (where
flanking stimulus elements signify the response opposite to that designated by the target stimulus element)
compared to congruent and neutral conditions; this effect does not vary as a function of age. (Reprinted from
Ridderinkhof & van der Molen, in press b).
instructions, the reaction times of subjects are always lengthened by the presentation of an
incongruent array. Moreover, P300 latency is also prolonged by this array. This finding
suggests that flankers conveying competing information induce a perceptual conflictthat delays
analysis of the target stimulus. Interestingly,Ridderinkhof and van dcr Molcn (in press b) have
demonstrated that the magnitude of this perceptual conflict may not be sensitive to early
maturational influences. They found that the presence of incongruent flankers induced
comparable increases in P300 latency among young children and young adults (see Figure 4).
However, as discussed in the chapter by Bashore and Smulders in this volume, this
comparability may not extend to later life--P300 latency is prolonged more in the older adults
than in young adults by this perceptual conflict (Zeef& Kok, 1993).
2.4. The LRP and Partial Response Activation
As we described earlier, the readiness potential is a negative-going movement-related

brain potential that precedes voluntary movements. When unilateral movements of the hand
are made (e.g., digit flexion, squeezing, writing), the readiness potential is largest at electrode
sites overlying cortical motor areas contralateral to the hand involved in the movement (e.g.,
Bashore, McCarthy, Heffley, Clapman, & Donchin, 1982; Kutas & Donchin, 1977, 1980;
Rohrbaugh & Gaillard, 1983). Coles and co-workers introduced a subtraction and averaging
procedure to isolate the lateralized readiness potential (LRP) from the readiness potential
(Coles, 1989; Gratton, Coles, Sirevaag, Eriksen, & Donchin, 1988; note that an equivalent
measure was developed independently by deJong, Wierda, Mulder, & Mulder, 1988). The
LRP is derived by Coles and colleagues by subtracting the potential recorded from the scalp
site overlying primary motor cortex ipsilateral to the responding hand from the potential
contralateral to the responding hand, averaged across trials for left- and fight-hand responses
separately; the signals for the lett- and fight-hand responses are then averaged together so that
any lateral asymmetries that are not related to the executed response are cancelled. Derived in
this manner, the LRP associated with activation of the correct response is negative-going.
Alternative, but functionally equivalent, methods for deriving the LRP have been used by
deJong et al. (1988) and Osman, Bashore, Coles, Donchin, and Meyer (1992). Considerable
evidence from neurophysiological studies involving both humans and animals supports the
assertion that the LRP is produced by activation of the motor system and is generated primarily
in primary motor cortex (e.g., Coles, Gratton, & Donchin, 1988; Deecke, 1987; Miller &
Hackley, 1992; Miller, Riehle, & Requin, 1992; Requin, 1985). The LRP has proven to be a
reliable and highly sensitive real-time index of preferential motor preparation in a variety of
experimental tasks.
To illustrate, the LRP can be used to provide insight into the nature of interference
processes by examining the central activation of partial responses that do not reach the
criterion for peripheral activation, as indexed by EMG activity and/or overt manual responses
(e.g., Gratton et al., 1988). The LRP, like the P300, has been utilized to study the influence of
flanking stimuli on processing the central stimulus in the Eriksen task. Flanking stimuli
associated with the response opposite to that signaled by the center stimulus (i.e., incongruent
stimulus arrays) have been observed to elicit a slight and transient positive-going deflection of
the LRP very early in the reaction process (preceding the major negative deflection that
indicates activation of the correct response). This early deflection suggests that the incorrect
response was activated at the central level without producing any consequent peripheral
muscle activation or overt movement (e.g., Gratton et al., 1988; Riddexffldaof & van der
Molen, in press a; Smid et al., 1990). Thus, stimulus arrays that contain conflicting
information may induce partial activation of the inappropriate response on some proportion of
trials. Although this activation of the incorrect response by the flanking stimuli is, as a rule,
not sufficient to exceed the response execution threshold, it competes with activation of the
correct response by the center stimulus. This presumed competitive process is thought to be
reflected in the initial lateralization of the LRP contralateral to the incorrect response side on
trials in which incongruent stimulus information is presented. This pattern of activation of the
LRP is evident in Figure 5, which provides a typical example of a transient partial activation of
the incorrect response channel, followed by the complete activation of the correct response
when the flanking stimuli are arrows that point in the direction opposite to that of the center
arrow.
[ - - Congruent : - Incongruent .... :-NeutrM I
o
0
E
.<
J I *'i * * I I * * J I , ** I I , ,, i i , **
-1000 -500 0 500 1000
Time from Stimulus Presentation (ms)
Figure 5. A representative example of the LRP recorded from young adults. Again, positive changes in
electrical polarity are shown as upward deflections. The predominant negative deflection represents the
activation of the correct response. Its onset latency is delayed in incongruent conditions (when flanking
stimulus elements signify the response opposite to that designated by the target stimulus element) compared to
congruent and neutral conditions. In addition, it is preceded by a small positive deflection that is most clearly
manifest in incongruent conditions and is thought to reflect the transient activation of the incorrect response.
(Reprinted from Ridderinkhof & van der Molen, in press a).
Partial activation of the incorrect response channel can be determined not only in
stimulus-locked LRPs, like those shown in Figure 5, but also in response-locked LRPs. The
latter is determined by averaging back a few hundred milliseconds from the execution of the
overt response (e.g., a button press). The former is determined by averaging from the onset of
the stimulus to the execution of the overt response. To some extent, morphological aspects of
the LRP may be obscured in the stimulus-locked LRP because of between-trial variation in the
time required for stimulus identification processes. This variability may be sttfficient to blur the
early structural properties of the LRP. However, the time from the activation of the response
output system to the execution of the overt response is probably more constant; thus, the
morphological integrity of the LRP may be better preserved in the response-locked LRP (see
Osman & Moore, 1993). Note that the pattern of factor effects on the LRP should not differ
as a function of the averaging procedure. Rather, effects should simply be more robust, under
some conditions, for the response-locked LRP.
Inferences about the onset and time-course of central response system activation are
drawn from the onset latency, slope, and amplitude of the LRP. Several procedures for
estimating these parameters have been proposed, none of which is accepted as standard. Since
the LRP was discovered less than a decade ago, measurement techniques are still being
explored. Nevertheless, the use of this component as an index of response system activation
has become quite popular among cognitive psychophysiologists, and its use is rapidly
expanding. In recent years, LRPs have been utilized to study the inhibition of subthreshold
central response channel activation in the stop-signal paradigm (deJong, Coles, Logan, &
Using event-related brainpotentials 307
Gratton, 1990; deJong, Coles, & Logan, in press); to determine the extent of response system
activation when overt responses are absent in go/nogo tasks (Miller & Hackley, 1992; Osman
et al., 1992); to investigate the effects of preparatory cues on the speed of central processing
(deJong et al., 1988; Gratton et al., 1990; Gehdng, Gratton, Coles, & Donchin, 1992); to
analyze the inhibition of immediate arousal (Ridderinkhof~ van der Molen, & Miller, 1992); to
examine the relationship between visual search and response selection (Staid, Lamain,
Hogeboom, Mulder, & Mulder, 1991); and, as we discussed earlier, to assess response
competition effects in the Eriksen task (Gratton et al., 1988; Ridderinkhof & van der Molen, in
press a; Smid et al., 1990). In addition, the effects of older age on the properties of the LRP
have been investigated (Zeef & Kok, 1993; see the chapter by Bashore & Smulders in this
volume for a discussion of this work).
3. SUMMARY AND CONCLUSIONS
Changes in brain electrical activity in relation to sensory or motor events can be

recorded from the human scalp in the form of the ERP. The latencies and amplitudes of
endogenous components of the ERP are sensitive to variations in the stimulus and response
processing demands imposed upon the cognitive system, while being relatively insensitive to
changes in the physical properties of the eliciting stimulus events. The latter influence the
exogenous components of the ERP. In this chapter, we provided readers who are unfamiliar
with this area of research with a basic introduction to the field. Our attention has been on
endogenous components of the ERP, particularly those discussed in the chapters in this section
of the book. The N400 component was shown to be an index for disconfirmation of semantic
expectancies: Semantically anomalous, but not appropriate, sentence completions produce an
N400. Variations in the amplitude of the Dm were shown to be predictive of the extent to
which stimuli are memorized: The Dm to stimuli that are remembered in subsequent recall or
recognition tasks is larger than the Dm to stimuli that are not remembered. The parameters of
the P300 were shown to index different aspects of information processing: The amplitude of
P300 varies with the probability and relevance of a stimulus event, whereas the latency of P300
is very sensitive to changes in stimulus processing demands while it is much less sensitive to
variations in response processing demands. Finally, the LRP was shown to be a reliable and
highly sensitive real-time index of preferential central response system activation that can
reveal this activation in the absence of any overt movement.
REFERENCES
Bashore, T. 1L (1990). Stimulus-Response compatibility viewed from a cognitive

psychophysiological perspective. In R. W. Proctor & T. G. Reeve (Eds. ),
Stimulus-Response compatibility. Amsterdam, the Netherlands: Elsevier Science
Publishers.
Bashore, T.1L, McCarthy, G., Hettley, E.F., Clapman, 1LC., and Donchin, E. (1982) Is
handwriting posture associated with differences in motor control? An analysis of
asymmetries in the readiness potential. Neuropsychologica, 20, 327-346.
Bashore, T. R., & van tier Molen, M. W. (1991). Discovery of the P300: A tribute. Biological
308 K.R. Ridderinkhofand Th.R. Bashore
Berger, H. (1929). Ueber das Elektrenkphalogramm des Menschen. Archiv fuer Psychiatrie
und Nervenkrankheiten, 87, 527-570.
Besson, M., & Kutas, M. (1993). The many facets of repetition: A cued-recall and
event-related potential analysis of repeating words in same versus different sentence
contexts. Journal of Experimental Psychology: Learning, Memory, and Cognition, 19,
1115-1133.
Besson, M., & Kutas, M., & van Petten, C. (1992). An event-related potential (ERP) analysis
of semantic congruity and repetition effects in sentences. Journal of Cognitive
Neuroseienee, 4, 132-149.
Besson, M., & Macar, F. (1987). An event-related brain potential analysis of ineongnfity in
music and other nonlinguistic contexts. Psyehophysiology, 24, 14-25.
Callaway, E., Tueting, P., & Koslow, S. (Eds. )(1978). Event-related brain potentials in man.
New York: Academic Press.
Coles, M. G. H. (1989). Modem mind-brain reading: Psychophysiology, physiology, and
cognition. Psyehophysiology, 26, 251-269.
Coles, M.G.H, Donchin, E., and Porges, S.W. (Eds.)(1986). Psyehophysiology: Systems,
Processes, and Applications, New York: Guilford Press.
Coles, M. G. H., Gehring, W. J., Gratton, G., & Donchin, E. (1992). Response activation and
verification: A psychophysiological analysis. In G. E. Stelmach & J. Requin (Eds.),
Tutorials in Motor Behavior H. Amgerdam, the Netherlands: Elsevier Science.
Coles, M. G. H., Gratton, G., Bashore, T. IL, Eriksen, C. W., & Donchin, E. (1985). A
psychophysiological investigation of the continuous flow model of human information
11, 529-553.
Coles, M. G. H., Gratton, G., & Donchin, E. (1988). Detecting early communication: Using
measures of movement-related potentials to illuminate human information processing.
Biological Psychology, 26, 69-89.
Coles, M. G. H., Gratton, G., & Fabiani, M. (1990). Event-related brain potentials. In J. T.
Cacioppo & L. G. Tassinary (Eds.), Principles ofpsyehophysiology: Physical, social, and
inferential elements (pp.413-455). Cambridge, MA: Cambridge University Press.
Davis, P. A. (1939). Effects of acoustic stimuli on the waking human brain. Journal of
Neurophysiology, 2, 494-499.
Deecke, L. (1987). Bereitschattspotential as an indicator of movement preparation in
supplementary motor area and motor cortex. In G. Bock, M. O'Conner, & J. Marsh (Eds.),
Motor Areas of the Cerebral Cortex (pp. 231-245). New York: Wiley.
deJong, 1L, Coles, M. G. H., & Logan, G. L. (in press). Strategies and mechanisms in
nonselective and selective inhibitory motor control. Journal of Experimental Psychology:
Human Perception and Performance.
deJong, 1L, Coles, M. G. H., Logan, G. L., & G-ratton, G. (1990). In search of the point ofno
return: The control of response processes. Journal of Experimental Psychology: Human
deJong, 1L, Wierda, M., Mulder, G., & Mulder, L. J. M. (1988). Use of partial information in
responding. Journal of Experimental Psychology: Human Perception and Performance,
14, 682-692.
Desmedt, J. E., Huy, N. T., & Bourguet, M. (1983). The cognitive P40, N60, and P100
components of somatosensory evoked potenitals and the earliest signs of sensory
processing in man. Electroencephalography and Clinical Neurophysiology: Evoked

Potentials, 56, 572-582.
Donchin, E., & Heffley, E. (1978). Multivariate analysis of event-related potential data: A
tutorial review. In D. Otto (Ed.), Multidisciplinary perspectives in event-related brain
potentials research. Washington, DC: US government printing office.
Donchin, E., Karis, D., Bashore, T.R., Coles, M.G.H., and Gratton, G. (1986). Cognitive
psychophysiology and human information processing. In Psychophysiology: Systems,
Processes, and Applications (pp. 244-267), M.G.H. Coles, E. Donchin, and S.W. Porges
(Eds.), New York: Gtfilford Press.
Donchin, E., Ritter, W., & McCallum, W.C. (1978). Cognitive psychophysiology: The
endogenous components of the ERP. In E. Callaway, P. Tueting, & S. Koslow (Eds.),
Event-related brain potentials in man. New York: Academic Press.
Dtmcan-Jolmson, C. C., & Donchin, E. (1977). On quantifying surprise: The variation of
event-related potentials with subjective probability. Psychophysiology, 14, 456-467.
Eriksen, C.W., & Schultz, D.W. (1979). Information processing in visual search: A
continuous flow conception and experimental results. Perception and Psychophysics, 25,
249-263.
Farwell, L.A., Martinerie, J.M., Bashore, T.1L, Rapp, P.E., & Goddard, P.H. (1993). Optimal
digital filters for long latency components of the event-related brain potential.
Psychophysiology, 30, 306-315.
Fischler, I., Bloom, P. A., Childers, D. G., Roucos, S. E., & Perry, N. W. (1983). Brain
potentials related to stages of sentence verification. Psychophysiology, 20, 400-409.
Fischler, I., Childers, D. G., Achariyapaopan, T., & Perry, N. W. (1985). Brain potentials
during sentence verification: Automatic aspects of comprehension. Biological Psychology,
21, 83-106.
FiscMer, I., & Raney, G. (1991). Language by eye: Behavioral and psychophysiological
approaches to reading. In, J. R. Jennings, & M. G. H. Coles, (Eds.), Handbook of
cognitive psychophysiology (pp. 511-575). New York: Wiley.
Friedman, D. (1990). ERPs during continuous recognition memory for words. Biological
Friedman, D., Putnam, L., & Sutton, S. (1990). Longitudinal and cross-sectional comparisons
of young children's cognitive ERPs and behavior in a picture-matching task: Preliminary
findings. International Journal of Psychophysiology, 8, 213-221.
Friedman, D., & Sutton, S. (1987). Event-related potential during continuous recognition
memory. Electroencephalography and Clinical Neurophysiology, 40 (Suppl.), 316-231.
Gehring, W. J., Gratton, G., Coles, M. G. H., & Donchin, E. (1992). Probability effects on
stimulus evaluation and response processes. Journal of Experimental Psychology: Human
Gratton, G., Bosco, C. M., Kramer, A. F., Coles, M. G. H., Wickens, C. D., & Donchin, E.
(1990). Event-related brain potentials as indices of information extraction and response
priming. Electroencephalography and Clinical Neurophysiology, 75, 419-432.
Gratton, G., Coles, M. G. H., & Donchin, E. (1992). Optimizing the use of information:
Strategic control of activation of responses. Journal of Experimental Psychology:
General, 121, 480-506.
Gratton, G., Coles, M. G. H., Sirevaag, E. J., Eriksen, C. W., & Donchin, E. (1988). Pre- and
poststimulus activation of response channels: A psychophysiological analysis. Journal of
Gunter, T. C., Jackson, J. L., & Mulder, G. (1992). An electrophysiological study of semantic
processing in young and middle-aged academics. Psychophysiology, 29, 38-54.
Guterman, Y., Josiassen, 1LC., & Bashore, T.1L (1992). Attentional influence on the P50
component of the auditory event-related brain potential. International Journal of
Haider, M., Spong, P., & Lindsley, D. B. (1964). Attention, vigilance, and cortical evoked
potentials in humans. Science, 145, 180-182.
Hansen, J. C., & Hillyard, S. A. (1983). Selective attention to multidimensional auditory
stimuli. Journal of Experimental Psychology: Human Perception and Performance, 9,
1-19.
Harbin, T. J., Marsh, G. lk, & Harvey, M. T. (1984). Differences in the late components of the
event-related potential due to age and to semantic and nonsemantic tasks.
Electroencephalography and Clinical Neurophysiology: Evoked Potentials, 59, 489-496.
Jasper, H.H. (1958). The ten-twenty electrode system of the International Federation.
Johnson, R. (1988). The amplitude of the P300 component of the event-related potential:
Review and synthesis. In P. K. Ackles, J. R. Jennings, and M. G. H. Coles (Eds.),
Advances in psychophysiology, vol. 3 (pp. 69-138). Greenwich, CT: JAI Press.
Johnson, 1L, & Donchin, E. (1978). On how p300 amplitude varies with the utility of the
eliciting stimuli. Eleetroeneephalography and Clinical Neurophysiology, 44, 424-437.
Karis, D., Fabiani, M., & Donchin, E. (1984). "P300" and memory: Individual differences in
the von Restorff effect. Cognitive Psychology, 16, 177-216.
Kenemans, J. L., Kok, A., & Smulders, F. T. Y. (1993). Event-related potentials to
conjunctions of spatial frequency and orientation as a function of stimulus parameters and
response requirements. Electroeneephalography and Clinical Neurophysiology: Evoked
Potentials, 88, 51-63.
Kornhuber, H. H., & Deecke, L. (1965). Himpotentialnderungen bei Willkurbewegungen und
passiven Bewegungen des Menschen: Bereitschafispotential und reafferente Potentiale.
Pfluger's Arehive fr die gesammelte Psyehologie, 184, 1-17.
Kutas, M., & Donchin, E. (1977). The effects of handedness, of responding hand, and of
response force on the contralateral dominance of the readiness potential. In J. Desmedt
(Ed.), Attention, voluntary contraction, and event-related cerebral potentials (pp.
189-210). Basel, Switzerland: Karger.
Kutas, M., & Donchin, E. (1980). Preparation to respond as manifested by movement-related
brain potentials. Brain Research, 202, 95-115.
Kutas, M., & Hillyard, S. A. (1980a). Event-related brain potentials to semantically
inappropriate and surprisingly large words. Biological Psychology, 11, 99-116.
Kutas, M., & Hillyard, S. A. (1980b). Reading senseless sentences: Brain potentials reflect
semantic incongruity. Science, 207, 203-205.
Kutas, M., & Hillyard, S. A. (1983). Event-related brain potentials to grammatical errors and
semantic anomalies. Memory and Cognition, 11, 539-550.
Kutas, M., Lindamood, T., & HiUyard, S. A. (1984). Word expectancy and event-related brain
potentials during sentence processing. In S. Komblum & J. Requin (Eds.), Preparatory
states and processes (pp. 217-238). Hillsdale, NJ: Erlbaum Press.
Kutas, M., McCarthy, G., & Donchin, E. (1977). Augmenting mental chronometry: The P300
as a measure of stimulus evaluation time. Science, 197, 792-795.
Kutas, M., & van Petten, C. (1988). Event-related brain potential studies of language. In P. I~
Ackles, J. 1L Jennings, and M. G. H. Coles (Eds.), Advances in psychophysiology, vol. 3.
Greenwich, CT: JAI Press.
Kutas, M., & van Petten, C. (1994). Psycholinguistics electrified. In M. A. Gemsbacher (Ed.),
Handbook of Psycholmguistics. San Diego: Academic Press.
Magliero, A., Bashore, T. R., Coles, M. G. H., & Donchin, E. (1984). On the dependence of
P300 latency on stimulus evaluation processes. Psychophysiology, 21, 171-186.
Mangun, G. Ik, Hillyard, S. A., & Lucj, S. J. (1993). Electrocortical substrates of visual
selective attention. In D. E. Meyer & S. Komblum (Eds.), Attention and performance XIV:
Synergies in experimental psychology, artificial intelligence, and cognitive neuroscience
(pp. 219-243). Cambridge, MA: MIT Press.
McCallum, W, C., Farmer, S. F., & Pocock, P. V. (1984). The effects of physical and semantic
incongruities of auditory event-related potentials. Eleetroeneephalography and Clinical
Neurophysiology: Evoked Potentials, 59, 477-488.
Miller, J. O., & Hackley, S. A. (1992). Electrophysiological evidence for temporal overlap
among contingent mental processes. Journal of Experimental Psychology: General, 121,
195-209.
Miller, J. O., Riehle, A., & Requin, J. (1992). Effects of preliminary perceptual output on
neuronal activity of the primary motor cortex. Journal of Experimental Psychology:
Muente, T. F., Heinze, H. J., Scholz, M., & Kuenkel, H. (1988). Effects of a eholinergie
nootropie (WEB 1881 FU) on event-related potentials recorded in incidental and
intentional memory tasks. Neuropsychobiology, 19, 158-168.
Naatanen, Ik (1975). Selective attention and evoked potentials in humans--a critical review.
Neville, H., Kutas, M., Chesney, G., & Sehmidt, A. L. (1986). Event-related brain potentials
during initial encoding and recognition memory of congruous and incongruous words.
Journal of Memory and Language, 25, 75-92.
Osman, A., Bashore, T. R., Coles, M. G. H., Donehin, E., & Meyer, D. E. (1992). On the
transmission of partial information: Inferences from movement-related brain potentials.
Osman, A., & Moore, I~ (1993). The locus of dual-task interference: Psychological refractory
effects on movement-related brain potentials. Journal of Experimental Psychology:
Pallet, K. A. (1990). Recall and stem-completion priming have different eleetrophysiologieal
correlates and are modified differentially by directed forgetting. Journal of Experimental
Pallet, K. A., & Kutas, M. (1992). Brain potentials during memory retrieval provide
neurophysiologieal support for the distinction between conscious recollection and priming.
Journal of Cognitive Neuroscienee, 4, 375-391.
Paller, K. A., Kutas, M., & Mayes, A. Ik (1987). Neural correlates of encoding in an incidental
learning paradimn. Electroencephalography and Clinical Neurophysiology, 67, 360-371.
Paller, K. A., Kutas, M., Shimamura, A. P., & Squire, L. R. (1987). Brain responses to
concrete and abstract words reflect processes that correlate with later performance on a
test of stem-completion priming. Electroencephalography and Clinical Neurophysiology,
40 (Suppl.), 360-365.
Pallet, K. A., McCarthy, G., & Wood, C. C. (1988). Erps predictive of subsequent recall and
recognition performance. Biological Psychology, 26, 269-276.
Pritchard, W. S. (1981). Psychophysiology of P300: A review. Psychological Bulletin, 89,
506-540. Ragot, R. (1990). Cerebral evoked potentials: Early indices of compatibility
effects. In 1~ W. Proctor & T. G. Reeve (Eds.), Stimulus-Response Compatibility: An
Integrated Perspective (pp. 141-162). Am~erdam, the Netherlands: Elsevier Science.
Requin, J. (1985). Looking forward to moving soon: Ante factum selective processes in motor
control. In O. Matin & M. I. Posner (Eds.), Attention and Performance X/(pp. 147-167).
Hillsdale, NJ: Erlbaum
Ridderinkhof~ I~ lk, & van der Molen, M. W. (in press a). When global information and local
information collide: A brain-potential analysis of the locus of interference effects.
Biological Psychology.
Ridderinkhof~ K. 1~, & van der Molen, M. W. (in press b). A psychophysiological analysis of
developmental differences in the ability to resist interference. Child Development.
Ridderinkhof~ K. R., van der Molen, M. W., & Miller, J. O. (1992). Inhibition of immediate
arousal. Psychophysiology, 29, 59 (abstract).
Rohrbaugh, J. W., & Galliard, A. W. I~, (1983). Sensory and motor aspects of the contingent
negative variation. In A. W. K. Gaillard and W. Ritter, (Eds.), Tutorials in event-related
potential research: Endogenous components (pp. 269-310). Amsterdam, the Netherlands:
North-Holland.
Rugg, M. D., Furda, J., & Lorist, M. (1988). The effects of task on the modulation of
event-related potentials by word repetition. Psychophysiology, 25, 55-63.
Smid, H. G. O. M., Lamain, W., Hogeboom, M. M., Mulder, G., & Mulder, L. J. M. (1991).
Psychophysiologi-cal evidence for continuous information transmission between visual
search and response processes. Journal of Experimental Psychology: Human Perception
and Performance, 17, 696-714.
Staid, H. G. O. M., Mulder, G., & Mulder, L. J. M. (1990). Selective response activation can
begin before stimulus recognition is complete: A psychophysiological and error analysis of
continuous flow. Acta Psychologica, 74, 169-201.
Smulders, F. T. Y., Kenemans, J. L., & Kok, A. (1994). A comparison of different methods for
estimating single-trial P300 latencies. Electroencephalography and Clinical
Stelmack, R. M., & Miles, J. (1990). The effect of picture priming on event-related potentials
of normal and disabled readers during a word recognition memory task.. Journal of
Clinical and Experimental Neuropsychology, 12, 887-903.
Sutton, S., Braren, M., Zubin, J., and John, E.R. (1965). Evoked-potential correlates of
stimulus uncertainty. Science, 150, 1187-1188.
Sutton, S., Tueting, P., Zubin, J., and John, E.1L (1965). Information delivery and the sensory
evoked potential Science, 155, 1436-1439. Verleger, 1L (1988). Event-related potentials
and cognition: A critique of the context-updating hypothesis and an alternative

interpretation of P3. Behavioral and Brain Sciences, 11, 343-356.
van der Molen, M. W., Bashore, T. 1L, Halliday, 1L, & Callaway, E. (1991).
Chronopsychophysiology: Mental chronometry augmented by psychophysiological time
markers. In J. 1L Jennings & M. G. H. Coles (Eds.), Handbook of cognitive
psychophysiology: Central and autonomic nervous system approaches (pp. 9-178).
Chichester, GB: Wiley.
Walter, W. (3., Cooper, 1L, Aldridge, V. J., McCallum, W. C., & Winter, A. C. (1964).
Contingent negative variation: An electric sign of sensorimotor association and expectancy
in the human brain. Nature, 203, 380-384.
Woods, D. L. (1990). The physiological basis of selective attention: Implications of
event-related potential studies. In J. W. Rohrbaugh, 1L Parasuraman, & 1L Johnson, Jr.
(Eds.), Event-related brain potentials. New York: Oxford University Press.
Zeef~ E.J., & Kok, A. (1993). Age-related differences in the timing of stimulus and response
processes during visual selective attention: Performance and psychophysiological analyses.
Do the waves begin to waver? ERP studies of language processing in the elderly*
Jonathan W. King a and Marta Kutas b
~Department of Cognitive Science
bDepartments of Cognitive Science and Neurosciences, University of California, San Diego, La

Jolla, California 92093-0515
Until recently, relatively little research had been done on language processing in the
elderly except as an adjunct to the study of aphasia. While language processes have usually
been thought to be fairly impervious to aging, it is now clear that some changes do occur.
Moreover, many of these changes otten seem to be directly linked to changes in the efficacy of
other cognitive operations affected by aging, particularly the active suppression of irrelevant
information and the management of working memory (WM). One way to study these changes
is to record event-related brain potentials (ERPs) from the scalp in response to individual
words either in isolation or in sentences. In this chapter, we review what is currently known
about the electrophysiology of visual language processing in the elderly, and what ERPs may
tell us about age-related changes in the neural substrates of reading.
1. ERP STUDIES OF LANGUAGE PROCESSING IN THE ELDERLY
Much more research has been focused on the development of language abilities from
infancy through early adulthood than on changes in language processing over the next forty to
fifty years. Notable exceptions are the investigations of individuals with gross disturbances in
language caused by certain strokes or dementing illnesses and their age-matched controls. But
a variety of subtle changes do occur within the normal aging brain that can have noticeable
effects on language processing. In this chapter, we review some of our results from studies of
language processing in presumably normal, healthy, elderly individuals employing event-related
brain potentials (ERPs). ERPs provide us with a brain imaging technique that is well-suited for
mapping processes at the psychological and linguistic level onto the relevant neural physiology.
They are especially useful for our purposes because they provide a potentially continuous
measure of on-going brain activity which is sensitive to many of the same variables that
influence language processing. Further, ERPs give us a useful measure that does not require
* Acknowledgements: Work reported in this paper was supported by NICHD grant I-ID22614 and by NIA
grant AG08313 to Marta Kutas. Jonathan W. King was also supported in part by a McDonnell-Pew
postdoctoral fellowship from the San Diego Center for Cognitive Neuroscience, funds from the Center for
Human Information Processing (CHIP; NIMH grant MH-14268), and funds from the Center for Research on
Language (CRL; NIH grant T32 DC00041-01) at UCSD. We also thank Jill Weckerly for her helpful
comments on earlier versions of this work and Heather Mclsaac for data collection.
Do the waves begin to waver ? 315
overt behavior, which allows us to study subject groups who are severely affected in their
ability to generate motor responses. This feature also allows us to note aging-related
differences in central brain processing that might be suggested by, but difficult to pin down
from reaction time (RT) data alone, given the well-known increase in the mean and variance of
RTs in the elderly.* ERPs also allow us to study age-related differences in language processing
due to other changes in aspects of cognition such as the flexible deployment of cognitive
resources, and the maintenance of items in working memory. These processes can also be
studied using RT-based methodologies, but they involve processes that do not always lead to
the generation of RTs in natural situations. Before we discuss the ERP evidence, we will
address these processes from a psychological and a neurological perspective in aim with the
aim of providing a background against which to interpret the ERP data.
1.1. Working Memory
One cognitive process crucial to language processing is working memory, a resource

long known to be degraded by aging (see, e.g., Wechsler, 1987, for aging norms).
Contemporary views of working memory have emphasized that what had previously been seen
as a unitary phenomenon can be analyzed into a system of component subprocesses whose
existence has been inferred from the performance of human subjects (e.g., Baddeley, 1986),
and verified experimentally in non-human primates (Goldman-Kakic, 1987). Virtually all
models of working memory now respect Baddeley's broad distinction between "central"
executive and "peripheral," often modality-specific maintenance, processes. Other commonly
recognized features of most working memory models include its decay function as well as
subprocesses responsible for item-to-item activation and inhibition, and the introduction of
items into WM (e.g., Anderson, 1983; Just & Carpenter, 1992).
The efficiency of the working memory system does seem to decline with age (Cohen,
1988) and this decline affects more than just the ability to remember phone numbers. Kemper
(1988) in particular has discussed how this decline has an impact on the parsing and production
of complex syntactic structures in the elderly, but the specific mechanisms for these age-related
changes remain unknown. General slowing of cognitive processes, by virtue of increasing the
exposure of memories to potential decay, undoubtedly does contribute to the decline of WM.
Recent results, however, also suggest a prominent role for the reduced efficiency of inhibitory
processes that serve to "clarify" the contents of working memory via the active removal of
irrelevant information (Hasher, Stoltzfus, Zacks, & Rypma, 1991). This inhibitory deficit
hypothesis contrasts with views that attribute age-related changes in working memory capacity
(WMC) to a failure in one or several other processes such as the active maintenance of items in
WM, the focussing of attention to specific items in WM, or the introduction of new items to
WM.
One specific and linguistically relevant prediction that follows from the inhibitory
deficit hypothesis is that normal elderly individuals would experience difficulties at establishing
coreference between a pronoun and the noun phrase it refers to, especially in the presence of
multiple possible (even if implausible) candidate noun phrases. This has been observed
experimentally (Light & Capps, 1986), and is apparently not due simply to an increase in the
Note that this is not to suggestthat age-relatedslowingis unimportant, since it is an undeniablycritical aspectof the
data to be explained(Salthouse, 1985). Instead,we are arguingthat ERPs can help clarifythe patterns of results seen
in RT data in cases where statisticalor practicalconsiderationsmakethe latter moredifficultto interpret.
316 J. w. King and M. Kutas
absolute amount of information that must be maintained in WM (Zelinski, 1988). These kinds
of effects, however, are not unique to the elderly but rather to individuals of any age with
reduced WMC. For instance, Gemsbacher and Faust (1991) found that among young adults,
less skilled comprehenders were less efficient and slower in their suppression of contextually
inappropriate meanings of ambiguous words (once activated) than were more skilled
comprehenders.
1.2. Suppression and Long Term Memory
An important corollary of the inhibitory deficit hypothesis of aging in working memory

is that a similar inefficiency in inhibitory influences in long term memory (LTM) can account
for some of the notable problems the elderly have with retrieving long term memories.
Particularly relevant in the language production domain are the word-finding difficulties the
elderly oRen experience. At first glance, such difficulties may seem to be a simple failure of
item retrieval, but the most exasperating part of such '~ip of the tongue" experiences is how
much information actually is retrieved, not how little, and how many incorrect labels are
generated (Brown, 1991). Note that this situation is asymmetric in the sense that normal
elderly subjects do not fail to retrieve meaning given a verbal label. This phenomenon might be
taken as a particular form of the so-called "fan effect", whereby it becomes more difficult to
verify a specific property (here, a name) in a semantic network memory as the number of
properties connected with a node increases (see Anderson, 1983; Cohen, 1990). Control over
this type of search appears to be related to the ability to inhibit irrelevant (but automatically
activated) memory nodes. Thus, according to the inhibitory deficit hypothesis the elderly,
with deficient inhibitory resources to guide their search through memory, should show
increased fan effects, as appears to be the case (Gerard, Zacks, Hasher, & Radvansky, 1991).
A closely related phenomenon that also helps to reveal the nature of such inhibitory
factors is negative priming which refers to a reliable decrement in performance (speed and/or
accuracy) to stimuli that were recently ignored relative to stimuli that are occtm~g for the first
time in that particular experimental situation (e.g., Tipper, 1985). Conditions leading to
negative priming are in exact opposition to the usual state of affairs, wherein stimulus analysis
is faster and more accurate upon a second presentation (i.e. with repetition) compared to the
analysis of new (unrepeated) stimuli. The relevant variable thus seems to be whether the item
on its first occurrence is attended or ignored; upon repetition, the former leads to facilitation,
the latter to inhibition. Insofar as inhibitory processes are deficient in normal elderly, they
should show little or no negative priming in such paradigms, and this also appears to be true in
some situations (Tipper, 1991; Hasher et al., 1991).
1.3. Effects on Linguistic Processes
In the preceding paragraphs we have summarized the evidence that some of the more
profound effects of aging on psychological and language processes seem to stem from changes
in inhibitory processes and their effects on memory. Now we turn briefly to two equally
important issues in the psycholinguistic literature, namely lexical access and the processing of
syntactic structure. All other things being equal, older individuals appear to have si~ificantly
larger vocabularies than younger ones (Botwinick, 1984), but there is little evidence that the
receptive comprehension of words in the elderly is compromised by this fact, although the
Do the waves begin to waver? 317
elderly do show some word-finding difficulties as alluded to above. In fact, while older adults
are systematically slower at performing tasks involving lexical information (such as lexical
decision, word naming, and category membership), a recent analysis by Lima, Hale, and
Myerson (1991) suggests that elderly RTs are systematically slower still in task situations
relying on non-lexical tasks (such as mental rotation, four-choice reaction time, and Steinberg
memory scanning). Given the fact that the elderly have larger vocabularies than the young, it
would be interesting if lexical processing is relatively spared, although more research on this
point is needed. When we move into the realm of syntactic processing, however, the
deleterious effects of aging seem relatively greater. Even as the stack-based, strictly serial
models of parsing that assumed a critical role of WM have fallen into disfavor in favor of more
parallel approaches (e.g. Gibson, 1990; Just & Carpenter, 1992), the vital contribution that
working memory capacity makes to syntactic processing has been more frequently
acknowledged. This, in turn, explains why the production and processing of syntactic
structures making the heaviest demands on WMC might deteriorate with age.
The results of a series of investigations using a variety of techniques have led Kemper
and her colleagues to conclude that the elderly (especially those older than 75) are
disproportionately impaired in their comprehension, production, imitation ot~ and memory for
certain syntactic constructions (Kemper, 1986a; Kemper, 1986b; Kemper, 1987a; Kemper,
19870; Kemper, Rash, Kynette, & Norman 1990; Kynette & Kemper, 1986). Specifically,
these so called "old-old" subjects have trouble with sentences which include embeddings using
that and wh- clauses as sentential subjects, particularly those that produce let~-branching
structures such as The reporter who the senator attacked admitted the error; we will return to
these constructions later. By contrast, the processing consequences of sheer length have been
found to be less important, although not as negligible as they ot~en seem to be in young
subjects. Still, it is important to note that the elderly do not exhibit more difficulties with basic
'Who did what to whom" questions than the young, unless certain WM-dependent roadblocks
are thrown into the path of a syntactic analysis. Richer world knowledge and a larger
vocabulary might ameliorate these effects to some extent (as would be suggested by the expert
memory literature, e.g. Chase & Simon, 1973), but they cannot eliminate them
1.4. A Brief Review of ERPs
To a cognitive neuroscientist, knowledge of the relevant cognitive psychology is only a

first step in understanding how the language comprehension system really works, and where
aging has its most notable effects. For many questions, real insights will come only when the
relevant mental processes are reliably associated with specific features of neurophysiological
measures. Neurophysiological evidence can not only supply additional constraints to currently
popular theories, just as any new data could, but also provide information of a special nature.
Specifically, ERPs are a measure of brain activity~the ultimate generator of mentation and
cognition that can therefore be discussed in both physiological and psychological terms.
Because the ERP is a sign of one level of explanation (neural) and responsive to variables that
influence another (behavioral), cataloguing its behavior under controlled conditions will by
necessity lead to more explicit psychological theories.
The neurophysiology of cognitive processing in humans can be investigated by the
noninvasive recording of event-related brain potential activity from the scalp. ERPs (also
known as evoked potentials) are small voltage fluctuations in the electroencephalogram (EEG)
318 J.W. King and M. Kutas
that are time-locked to sensory, motor, or cognitive events; ERPs, then, reflect patterned
neural activity associated with informational transactions in the brain. While electrical fields
measured at the scalp are remote measures of the brain's activity, they constitute one of the
few techniques available for recording the dynamic patterns of neuro-electric activity
associated with specific cognitive acts and linguistic processes (Callaway et al., 1978; Renault,
Kutas, Coles, & Gaillard, 1989; Munte, Heinze, & Mangun, 1993). Used as converging
operations with behavioral measures, ERP results can assist in classifying perceptual, cognitive
and linguistic processes (reviewed in Naatanen & Michie, 1979; Hillyard & Picton, 1987).
ERPs have several characteristics that make them especially well-suited for addressing
issues central to cognition in the elderly. The first is the multidimensional nature of the ERP
waveform, which can vary along a number of dimensions; specifically, latency, amplitude, and
distribution across the scalp. Thus, in principle, ERPs can reflect not just quantitative
fluctuations in some process but the activity of qualitatively different processing events as well.
Furthermore, the ERP is both a continuous and a real-time measure that is a record of the
brain's processing* over periods that are co-extensive with language- or memory-related
operations. They provide not only information about processing before RTs could possibly be
measured, but also information about processes that bridge the time between multiple
responses when they occur. Importantly, reliable ERP effects can be obtained even in the
absence of any additional task over and above the natural one of reading or listening, an
advantage which becomes even more obvious when working with various patient populations.
Basic research into the brain mechanisms that generate scalp potentials continues to
this day, but it is generally agreed that most of the electrical activity observed in faster ERPs
(i.e. those lasting less than a second) represents the summation of the excitatory and inhibitory
postsynaptic potentials on thousands of large pyramidal neurons aligned perpendicularly to the
cortical surface (Nunez, 1981). These groups of neurons can be modelled by an equivalent
dipole, whose activity is volume-conducted from its point of origin to recording electrodes on
the scalp with no propagation delay, but with substantial smearing of the observed field due to
the manner of conduction even in the ideal case ** (Nunez, 1981). It is frequently assumed
(and may actually be the case) that an increase in firing rate within a cortical region will be
associated with a cortical surface negativity (Speckmann, Caspers, & Elger, 1984), but
whether this appears on the scalp as a negative or positive potential depends crucially on the
orientation of the equivalent dipole in the activated region of cortex relative to the recording
electrode. We raise these issues both because readers might otherwise leap to unwarranted
conclusions about the localization of specific generators, and because we sometimes do
speculate about localizations, albeit mindful of constraints provided by converging evidence
from lesion studies, animal models, or other functional imaging modalities. As detailed below,
we do have some a priori notions about what structures in the central nervous system are
more or less affected by aging, and how these changes affect the cognitive processes described
above.
* It is however,importantto note that not all brain processes are reflectedat the scalp surface. For limits see Allison et
al. (1986).
**A point source on cortex abuttingthe brain case "spreads"to a circular region2.5 cm in diameterat the scalp, even
neglectingthe unavoidableconductionof activitythrough the scalp towardseither a current source or sink, depending
on which direction current is flowing,
1.5. The Aging Brain and its ERPs
It is well-documented that the relative amounts of white and grey matter change over
the course of the lifespan, with adulthood being marked by a slow but steady reduction in grey
matter (Henderson, Tomlinson, & Gibson, 1980). Exactly what portion of this shrinkage is
due to cell loss (suggested by Henderson et al., 1980) versus the shrinkage of neurons and
their dendritic processes is still a subject of debate. However, both processes contribute to this
net shrinkage and thus can have potentially serious consequences on the quality of information
processing that is possible. A combination of basic neuroscience and better imaging techniques
has recently afforded an estimate of changes over the lifespan in neuron counts in specific brain
areas from changes in brain volume, which in turn were estimated from MRI scans (Jemigan,
Press, & Hesselink, 1990). Averaging over the entire cerebrum, Pfefferbaum r al. (1994)
estimated that grey matter volume (and thus cell counts) peaked around age 5, when a grey
matter volume of about 130 mL and a white matter volume of only 50 mL rattle around inside
an 1175 mL brain case. At age 20, they found that the volume of cerebral grey matter had
already decreased by over 15% to -110 mL, but white matter had increased by 40% to 70 mL.
At age 65, grey matter was down another 18% to only about 90 mL, but white matter was
unchanged; overall brain volume was maintained by an increase in cerebrospinal fluid.
Jemigan et al. (1991) observed similar effects of aging in a smaller sample of subjects,
and also assessed grey matter volume losses for different cortical and subcortical regions.
While they found highly reliable losses in almost every sector examined, they noted distinctly
greater losses in the caudate nucleus and slightly greater losses in the medial/basal temporal
lobes than in other cortical regions. An interesting aspect of these results is that both of these
regions have been shown to be crucial for various memory functions, with the medial temporal
lobe system involved in explicit memory formation (Zola-Morgan & Squire, 1993), parts of the
basal and inferior temporal lobe in visual working memory (Miller & Desimonr 1994), and the
caudate apparently critical for implicit motor learning and the organization of motor behavior
generally (Graybiel et al., 1994). Both the temporal lobe structures and the caudate are also
strongly interconnected with prefrontal areas believed to play a quite prominent role in
working memory systems (Goldman-Rakic, 1987). Further, both regions have heavy
dopaminergic irmervations (reviewed by Berger, 1992), and prefrontal areas at least are known
to suffer notable reductions in dopaminergic activity with normal aging (reviewed by Fuster,
1989).
While the functional significance of dopamine in the CNS is still hotly debated, it does
appear to be vital to the performance of at least some working memory tasks (Sawaguchi &
Goldman-Rakic, 1991), and is hypothesized to play a modulatory role in many areas of primate
cortex. Normal aging also is associated with changes in the striatum, (McGeer, McGcer &
Suzuki, 1977), and, in particular, reductions in the width of the pars compacta of the substantia
nigra, which consists almost solely of dopaminergic neurons; these reductions are significantly
correlated with reaction time slowing in a skilled movement task (Pujol et al., 1992). Further,
recently it has been noted that the striatum is involved not only with the direct generation of
motor behavior, but with other "central" cognitive processes as well (e.g., Eslinger & Grattan,
1993). In summary, aging appears to affect precisely the areas where one would expect
damage to be associated with a slowing in motor preparation, declarative memory deficits, and
possibly a decrease in working memory capacity.
1.6. Methodological Issues and Recording Parameters
In the data discussed throughout the remainder of this chapter, all subjects were
right-handed, neurologically normal, native English monolingual speakers. Young subjects
were UCSD students between 18 and 27 years of age who participated either for course credit
or $5.00 per hour. Our elderly subjects were generally recruited from continuing education
classes or volunteers at the Veterans Administration Hospital and also were paid $5-$7 per
hour for their participation. Overall, our elderly subjects had an average education level of
about 15 years. Unless otherwise specified, all ERPs were responses to words embedded in
normal English sentences between 8 and 18 words in length that were presented to subjects
one word at a time in the center of a CRT while their EEG was being recorded. Words were
presented for a duration of 200 msec once every 500 or 550 msec. Specific task instructions
varied from experiment to experiment, but in all cases comprehending the sentences presented
was essential to task performance. Particular electrode montages also varied from study to
study, but all included electrodes over lateral sites at standard 10-20 sites (F7, FS, T5, T6, O1,
O2), and some non-standard pairs approximately over Broca's area, Wernicke's Area, (and
their right hemisphere homologs) and primary auditory cortex (referred to below as Central).
All subjects' EEG was amplified using a time constant of about 8 seconds, and digitized on-line
with a sampling rate of 250 Hz.; eye movement and blink artifacts were rejected off-line prior
to averaging. In the discussion below, the differences we discuss are all statistically reliable at
the .05 level and, unless explicitly noted, the differences in amplitudes or latencies mentioned
are based on measures taken from individual subject's average ERPs (see Ridderinkhof &
Bashore, this volume, for a discussion of this and other relevant details of ERP methodology).
Further details of the general experimental procedure are given in King and Kutas (in press).
As suggested in the Introduction, we generally expect ERP effects to match the timecourse of
the relevant processing, so where WMC usage is of interest, both short (1 second) and longer
(up to 5 second) epochs were prepared from the raw data.
2. SINGLE WORD DATA
2.1. Early Potentials to Words
One of the convenient features of the ERP is that it is possible to examine the integrity,
nature, and timecourse of some of the different processes that are critical for successful
language comprehension. For instance, although all perceptual processes may not be complete
within a hundred milliseconds, there is some relatively early (in both a physiological and
psychological sense) selective processing that has an impact on the extent to which items are at
the focus of attention. Those items that are at the focus of attention are privileged relative to
those that are not, and are therefore processed more efficiently. This facilitated processing,
especially for certain sensory and physical features, is indexed by some of the early components
of the ERP such as the posterior P1 (peaking around 100 msec) and the posterior N1 (peaking
between 170 and 190 msec). The amplitudes of both the P1 and N1 can be changed by a
variety of important factors; thus, N1 and P1 amplitudes are hrger to items that are presented
centrally rather than in the visual periphery, larger to items in the spotlight of attention than to
those outside of its purview, and larger to items physically closer to those attended (provided
PARSING STUDY REPETITION STUDY
N1
/\ / \
L. Frontol
N 170 N400 /
R. Occipitol
~ P1 P2
' ' ': ' , V -
m
I I I I I 3 II~V
I I I I I
0 100 200 300 400 0 100 200 300 400 msec
Young Elderly
Figure 1. Grand average ERPs from two independent studies at two electrode sites for Young and Elderly
subjects. NegativeVoltages are plotted up in this and all other figures.
they share attentionally selected features) than to those which are farther away (Mangun,
Hillyard & Luck, 1993).
Figure 1 shows ERPs to open class words (e.g. nouns, verbs, adjectives, adverbs) from
two of our studies, with the waveforms for young and elderly subjects overlapped to highlight
the effects of aging. Over posterior sites, as shown here for the fight occipital electrode,
young and elderly subjects generate very similar N1 (N170-190) components, suggesting that
the early visual and attentional processing, or at least the first 200 msec of such processing, is
similar in the two groups. (Note that P] amplitudes are routinely more variable in our data, so
the apparent difference between the young and elderly subjects in the Repetition study is
neither a reliable nor necessarily meaningful difference.) As can be seen in Figure 2, these early
potentials at the posterior electrode sites distinguish good from poor comprehenders of written
text, be they younger or older adults. In both groups the amplitude of the posterior P1-N1-P2
complex is somewhat larger in the poor than good comprehenders, even though the Posterior
Temporal N1 of the elderly is drastically reduced in size relative to that of the younger
subjects. The posterior P1-N1-P2 complex has been linked to changes in resource allocation
(in the form of shifting attention). In general, N1 and sometimes P2 amplitudes are greater for
attended stimuli than for unattended stimuli in both the auditory (Hillyard, 1985) and visual
(Mangun & Hillyard, 199 l) modalities, and the former appears to index the amount of
YOUNG ELDERLY
Good Poor Good Poor
Frontal
Anterior -~ v ~ ,~. ~--^k/,~.~^.- ~ J~-,~,-

Temporal
kI
Central .,, ~! v -
,.t
Posterior
Temporal
II
Occipitol _ A/L _A ^ _ A
1
, , - , -
020044)0
,,,,,
0 200 44)0
,,,,,
0 2O0 400
I,,,,
0 200
"t
4O0
Figure 2. Grand average ERPs to all open class words for Good and Poor comprehenders defined by a median
split in both Young(n=24) and Elderly (n=l 8) subjects, with the left and fight hemispheres overlapped.
resources devoted to processing a channel that can be selected on the basis of spatial
information. A rationale for such an attentional difference can be found in the now
conventional hypothesis that poorer comprehenders may allocate more attentional resources to
lower level processes than good comprehenders (Hunt, Lunneborg, & Lewis, 1978; Perfetti &
Lesgold, 1977). A somewhat similar N1-P2 effect has been seen previously in a dual task
study carried out by Raney (1993), in which subjects simultaneously read (or re-read) a series
of texts while also responding to randomly occurring auditory tones in a secondary task.
Raney found that the N1-P2 evoked by the tones increased in amplitude when subjects were
re-reading a prose passage, consistent with the view that more cognitive resources were
available to be devoted to the secondary tone detection task as the processing demands of
word recognition decreased.
Contrasting with the equivalent posterior N Is in the young and elderly, are those at the
more anterior electrode site, where the elderly subjects show markedly larger NIs than do the
young subjects in both studies (see Figure 1). Little is known about the fimctional significance
of the anterior N1 although, like the posterior N1, it is sensitive to manipulations of
visuo-spatial selective attention and has been associated with a posterior dipole in various
modelling attempts (Hillyard, personal communication).
Following the N1 at fronto-central sites, young subjects display atypical P2 (180-210)
component; by contrast, this component is barely evident in the average data from the elderly
subjects. In younger adults, the P2 is one of the more robust of the early components of the
ERP. Nonetheless, relatively little is known about factors that influence its latency and
amplitude. In general, pictures have been shown to yield larger frontal (or vertex) P2s than
words, and non-words yield smaller P2s than words (Jeffreys & Tukmachi, 1992). Puce and
her colleagues (1994; see also Allison et al., 1994) have suggested that at least one generator
of the scalp P2 to faces and words comes from discrete locations on the fusiform gyrus on the
bottom of the temporal lobe.* These data were based on investigations of patients with
implanted subdural electrode arrays who were being evaluated for surgery to relieve
intractable epilepsy.
Anatomically, the fusiform gyms lies between structures of the medial temporal lobe
and other more laterally placed visual association cortices. This region is therefore likely to be
within the zone noted by Jernigan and her colleagues as a region of more sharply decreased
grey matter volume with aging. However, localizing one of possibly several neural generators
to a region whose function is unknown is not particularly revealing about the functional
significance of the P2 or the fact it is so drastically reduced in the elderly. One of the few
studies addressing cognitive effects on the P2 was performed by Chapman, McCrary, and
Chapman (1978; also 1981), who reported that the amplitude of a P2-1ike component with a
peak latency of 250 msec was related to the successful storage of a stimulus into short term
memory in a memory probe task. Such a result would be consistent with a hypothesis of
reduced efficiency of working memory in the elderly, but further work is needed to determine
whether this was a P2 effect or perhaps an early member of the P3 family (Verleger, 1988;
Donchin & Coles, 1988). Following the early sensory components (e.g., P1, N1 and P2) are
the later, so-called endogenous components of which the N400 is one example. The fight
occipital site data in Figure 1 clearly show that young and elderly subjects also differ in the
prominence of their N400 components; this difference will be discussed in greater detail later in
the chapter.
In standard behavioral psycholinguistics, it requires a certain degree of experimental
finesse to detect differences between items belonging to distinct lexical classes, although such
differences do obviously have some importance in the syntactic processing of sentences. (By
lexical classes, we refer to the difference between content words such as nouns, verbs,
adjectives and ly-adverbs and function words such as articles, prepositions, pronouns, and
conjunctions; the difference between the so-called "Open Class" items and their "Closed Class"
cousins.) It is therefore noteworthy that the difference between Open and Closed Class items
is quite clear in their ERPs, as can be seen in Figure 3.
At frontal sites, the ERPs to closed class words in both the young and the elderly
subjects are characterized by a broad, lefi-lateralized negativity relative to open class words;
this difference begins as early as 100 msec post stimulus-onset and continues throughout the
* On the cortical surface, this potential is a negativity,but the polarity reverses within the brain and would yield a
broadly distributedpositivityon the scalp.
YOUNG ELDERLY
A ..
~/" ^ .- =f~ ..f
Frontol
Centr(]l r~
N400
_ A .. _A ..
Occipital
r -'llY- -7
]-
I , I , I I i I , I I , I, L I I I , I
0 2~ 400 0 2~ 400 0 200 400 0 2~ 400mI~
Closed Class Words Open Class Words
Figure 3. Panel A shows grand average ERPs at 6 electrode sites for Open Class and Closed Class words
obtained from Young (n=24) and Elderly (n=l 8) subjects, with the N400 effect shaded for both groups and the
peak of the Lexical Processing Negativity(L,PN)indicated for the Young subjects.
epoch. The functional sL~ificance of this greater negativity for closed class words is
uncertain, although it has been previously noted (e.g. Neville et al., 1992). We have suggested
that it might index the expectancy that a content word will occur soon, given that function
words in English generally introduce new phrasal units (Kutas & King, in press). If this is the
case, ~ e n one could expect to see the resolution of this frontal negativity generated by
function words as a frontal positivity on the following content words. Figure 3 shows that
there is in fact a late, lateralized frontal positivity to open class words whose amplitude is
nearly equal to the amplitude of the negativity seen to closed class words. Thus, the pattern
we observe in the data is consistent with our expectancy generation and resolution idea, and
there appear to be few differences between young and elderly subjects in this pattern.
Whereas the slow potential ERPs from the young and elderly subjects are remarkably
similar over frontal sites, the fine structure of the waveform over these sites differs markedly
between young and old subjects. The ERPs of the young, but not those of the elderly, exhibit
a negative peak at about 280 msec for closed class words and at about 300 msec for open class
words. We have found that the peak latency of this negativity, dubbed the Lexical Processing
Negativity (LPN) is highly sensitive to the length and frequency of daily usage of the words
that elicit them (King & Kutas, submitted). For grand mean data, the proportion of variance in
LPN latency explained by length and frequency is 86%, even at the single subject level these
two factors explain about 44% of the variance in latencies when we consider the median
percentage of variance explained in our set of individual subjects. Neville et al. (1992)
suggested this component (referred to as the N280) was reliably evoked only by closed class
words and thus indexed the operation of lexical class-specific syntactic processes, but our data
suggest this is not the case. First, our data demonstrate that the LPN (aka N280) is not unique
to closed class words. Second, it seems rather unlikely that the LPN indexes some obligatory
syntactic process since it is virtually absent in elderly subjects, who, nonetheless,
comprehended these sentences and appreciated their syntactic structure as well as did younger
subjects. Of course, it is possible that the elderly did generate LPNs but with far less
synchrony and phase-locking than the younger adults. As an alternative, we have hypothesized
that the LPN reflects activity in premotor cortex related to the control of gaze and suppression
of voluntary eye movements in our experimental situation (King & Kutas, submitted). In
standard ERP experiments, subjects are requested not to move their eyes or blink during the
duration of the sentence presentation; honoring this request is quite effortful. Indeed, our
elderly subjects are typically less able to abide by these instructions, and may thus be exercising
less control over their gaze in the sense that they are less able to inhibit automatic eye
movements in this experimental situation. This observation is also consistent with the
inhibitory deficit hypothesis of aging-related cognitive change.
2.2. N400 and Integration As We "Know" It.
Thus far we have glossed over one of the most obvious differences between the ERPs
to content and function words, and between the ERPs to content words in younger and older
subjects, namely the large negative wave especially prominent over fight posterior scalp. This
was done, in part, so that we could consider the earlier and possibly less meaning-driven
components of the ERP to words before turning to issues of deeper processing and
information integration over the course of the sentence. Both Figures 1 and 3 show that, for
young subjects, the ERP to every word in a word-by-word rapid serial visual presentation
format is characterized by a negativity, that peaks between 350 and 450 msec (N400), is larger
posteriorly than anteriorly, and is slightly larger over right than left hemisphere rites (for a
review see Kutas & Van Petten, 1994). Content or open class word EKPs contain larger
N400s than do function or closed class word ERPs. In fact, at the be~nning of a sentence the
most striking difference between the ERPs to content and function words is the presence of a
much larger N400 component to the former than to the latter. This relative difference in the
N400 amplitude elicited by function and content word ERPs is maintained in the responses to
these lexical classes throughout the course of sentences. The relative reduction in the
amplitude difference between N400s to content versus function words over the course of a
sentence is mostly due to a decrease in the amplitude of N400 to content words; Van Petten
and Kutas (1991) showed that this decrease was primarily a consequence of the buildup of
semantic (and not syntactic) constraints. The data in Figure 4 show that this reduction in the
amplitude of the N400 to content words with increasing context (operationalized in terms of
ordinal word position in sentences) is present not only in the young but also in elderly subjects.
In fact, the elderly subjects generate large N400s to the first content word in the sentence, but
appear to show a faster fall-off with accumulating context.
20'S 30~S 40~S 50b 60~S 70~S
8~t
I,,,l|,,I, I,,,,,,,,. I,,.',,,I, I . . . . . . . ,. J , , I . , , I , ,.,,|...|1§
0 400800 0 400600 0 400800 0 400000 0 400600 0 400800 ms
oongruous i~u~on91mous ........
Figure 4. ERPs to visual words presented as congruous or incongruous completions of auditory contexts for
groups of subjects (each n=12) varying in age from their 20s to their 70s, with the N400 effect shaded.
These data underscore one important finding since the original reports on the N400,
namely that it is not specific to semantic anomalies. The ERPs to all words are characterized
by some N400 activity whose amplitude is largely a function of the word's expectancy within
its context. In 1984, Kutas and Hillyard demonstrated that the amplitude of the N400
tosentence final words, none of which were anomalous, was inversely correlated with doze
probability (r = -0.80-.90). This correlation has led some to propose that the N400 has
nothing to do with semantic compatibility or integration but is merely an index of predictability
(subjective conditional probability). However, in the same paper as well as others since (e.g.
Kutas, 1993), it was demonstrated that the ERPs to words with identical cloze probabilities
can have different amplitude N400s as a function of the semantic relation between the expected
word and the elieiting word. Thus, for example, if the sentence frame 'q'he pizza was too hot
t o - - " were completed not by the expected word "eat" but by a word with a subjective
conditional probability of zero such as "chew" or '~old", the ERP to both of these would
contain a hefty N400; however, the N400 to the word "chew" would be smaller, presumably
because of its semantic relation to "eat". A similar effect has been observed for outright
semantic incongruities (Kutas, Lindamood & Hillyard, 1984).
Typically, the largest and most robust N400 is elicited by an open class word that is
semantically anomalous within its context and unrelated in any way to any word in the sentence
or to the expected ending (e.g., Kutas & Hillyard, 1980ab, Kutas & Hillyard, 1982). This
finding holds whether the anomalous word occurs at the end or in the middle of a sentence
(e.g., Kutas & Hillyard, 1983). In both cases, the semantically anomalous word elicits a
significantly larger negativity which diverges from the response to a semantically appropriate
word in the same ordinal position (assuming they are matched on length and frequency of
usage) at about 200 msec, peaks between 350 and 450 msec, and has, in the average
waveform, a duration of 300 to 400 msec; these values refer to the findings in young adults. In
older adults, both the onset and the peak latency of the N400 effect during reading are delayed
(see below).
In young adults, N400-1ike responses can be recorded not only to written words but
also to semantic violations within spoken sentences (e.g., McCallum et al., 1984) and to
visually presented signs in American Sign Language (Kutas, Nevlle, & Holcomb, 1987). There
are, however, some differences in the specific characteristics of the visual and the auditory
N400s (for comparison see Holcomb & Neville, 1991). Holcomb, Coffey, and Neville (1992)
investigated the developmental timecourse of both the auditory and visual N400 to semantic
anomalies occurring at the ends of written and spoken sentences. To a large extent, the
changes in N400 across the adult lifespan have been limited to written words, although given
the current interest in semantic processing in aphasics and other brain-damaged populations,
we expect to see more data on N400s elicited by running speech in the near future.
At the moment, there is only one published study that examines aging-related changes
in the N400s elicited by semantic incongruities in spoken sentences. Woodward, Ford, and
Hammett (1993) found that the N400s to auditory stimuli, like visual N400s, are reduced in
amplitude and somewhat delayed in latency relative to that in a sample of young subjects.
However, some aspects of their design make it difficult to generalize about the effects of aging
from these results; in particular, the sentence materials were repeated across two different
conditions, and a one second delay was artificially imposed between the sentence context and
the sentence final word.
Harbin, Marsh, & Harvey (1984) were the first to examine the effects of aging on the
ERPs in a task (semantic categorization) likely to elicit and modulate the N400. They
recorded ERPs from three midline electrodes from both younger (mean age 21 years) and older
(mean age 71) subjects as they rendered decisions about the filth of a series of
visually-presented words in an Identity and a Category condition. In the Category condition,
the first four words were members of the same semantic category and subjects were asked to
indicate whether or not the fifth word was also a member; the fifth word matched the previous
four on only 15% of the trials. Both categorization times and the latency of the N400 to
mismatches were longer in the older group, although the difference N400 (mismatch minus
match ERP) peaked at approximately 540 msec for both groups. The ERPs in the elderly were
characterized by smaller N400 effects as well. Gunter, Jackson, & Mulder (1992) compared
the ERPs to congruous and incongruous endings of sentences of medium-to-high contextual
constraint from a group of young students to those from a group of highly-educated
middle-aged academics (mean age 55 years). They used both a fast and a slow rate of sentence
presentation, although this factor did not interact with age. Overall, the N400 effect in this
middle-aged group was delayed in latency by 120 msec and was somewhat reduced in
amplitude.
We have collected N400 data from young and elderly subjects both to sentences
presented visually one word at a time and to target words following short phrases. Our results
328 J.w. King and M. Kutas
are essentially the same in both cases. We have found that with advancing age N400 latencies
are prolonged and amplitudes are reduced. Since the results with short phrases were
equivalent, we used those stimuli to investigate men and women from 20 to 80 years old as
well as elderly adults suffering from senile dementia of the Alzheimer's type (Kutas, Iragui &
Salmon, submitted). Thus, rather than complete sentences, congruent and incongruent words
were flashed visually after the context (a short phrase) was spoken aloud by the experimenter.
We chose this paradigm with an auditory context and a visual target word specifically because
we have found that it is a task that can be performed by patients of various mental capabilities
and yields very robust N400s in response to semantically incongruent or unrelated words.
The stimulus set included approximately equal numbers of highly constraining antonym
contexts (e.g., '~he opposite of black") for which there is only one reasonable outcome (e.g.,
'~hite") and moderately constraining category contexts (e.g., "a piece of furniture") for which
there are several reasonable alternative members (e.g., '~able", "chair", "couch", "cabinet",
etc.). These antonymic and categorical relations map loosely onto the distinction that has been
made between semantic and associative priming (for review see de Groot, 1990) and more
directly onto the distinction between a prediction-based versus an expectancy-based strategy
for utilizing contextual information, respectively (e.g., Becker, 1980; 1982).
ERPs were recorded from a total of 72 men and women (between 20 and 80 years old)
as they performed this semantic categorization task. Subjects were asked to report the word
seen (flash duration was 265 msec) and following the report to indicate whether or not it was
appropriate given the prior context. Overall, the waveforms and their modulations with the
experimental variables were remarkably similar across the decades. These subjects seemed to
process the words in relation to the prior auditory context in a qualitatively similar manner no
matter what their ages. The ERPs to all age groups contained larger N400s to words that did
not fit with the preceding context than to those that did.
As can be seen in Figure 5, in all age groups, the N400 congruity effect in the opposite
condition was larger over the back than the front of the head and larger over the fight
than the left hemisphere, more so over the front of the head. Thus, over occipital sites, the
N400 congruity effect was nearly symmetric whereas over frontal sites it was large over the
right hemisphere and essentially absent over the left; for individuals in their 60s and 70s the
absence of an N400 congruity effect extended to the left central sites. Over the left frontal
sites, incongruity was associated with an enhanced late positivity.
While there did not appear to be any qualitative effects of advancing age on the N400
congruity effect, there were clear quantitative changes. Specifically, the amplitude of the N400
congruity effect was smaller and its onset and peak latencies were later in the older than in the
younger adults. Regression analyses revealed that there was a reliable linear decrease in the
amplitude (0.05-0.090 V/year) and a reliable linear increase (1.5-2.1 msec/year, r =.60) in the
peak latency of the N400 congruity effect with age.
We found a similar diminution in amplitude and prolongation in latency for the N400
congruity effect in a group of older adults between 63 and 83 years old (relative to college
undergraduates) when both the context and the final word were visual (see Figure 6). In that
experiment, all the subjects read 120 sentences, presented a word at a time for a duration of
200 msec once every 550 msecs; 2500 msec separated the end of one sentence from the
be~nning of the next. Half the sentences ended congruously while the remaining half ended
with a word that rendered the sentence nonsensical. In this case, the subjects were reading the
sentences not only for comprehension but also with the knowledge of an impending cued recall
experiment, all the subjects read 120

sentences, presented a word at a
YOUNG ELDERLY time for a duration of 200 msec once
every 550 msecs; 2500 msec
FRONTN. , , . ~ separated the end of one sentence
fromthe be~nning ofthenext. Half
CENTRAL . ~
the sentences ended congruously
while the remaining half ended with a
%
word that rendered the sentence
R~tI~. " w ....
nonsensical. In this case, the
%.~.r %
subjects were reading the sentences
Jk
L ~ONTAL r not only for comprehension but also
with the knowledge of an impending
=~x~lb..,~ R. FRONTAL ~ .... cued recall test wherein they would
be given each sentence context and
...,~_ ,~ L POSTERIOR asked to recall the sentence final
word. As with the short phrases, the
N400 congruity effect in the elderly
~ 9 A R. POSTERIOR _~.A ~ .
was appreciably smaller and later
than in the younger subjects. These
~-~ f i - ,_ occ.~. _~A A'" findings are consistent with less
A
.
efficient
A_ These age-related changes in
the N400 congruity effect are
;...,..., ,...,...;t" interesting in light of recent reports
4OO n~= 0 4OO 800 on the ERP word repetition effect
r -- i n c ~ (Hamberger & Friedman, 1992;
Rugg et al., 1994; Karayanidis,
Andrews, Ward, & McConaghy,
Figure 5. ERPs to congruent and incongruent final words to 1993). The ERP word repetition
visually presented sentences for Young (n=18) and Elderly effect refers to the greater positivity
(n=18) subjectswith the N400 effect shaded. elicited by items repeated (old)
relative to that occun~g for the first
time in the experiment (new).
Typically, it has been argued that the word repetition effect reflects the modulation of multiple
ERP components, of which one is the N400. The most consistent finding has been the absence
of any statistically significant differences in the size of the ERP word repetition effect between
young and elderly subjects, gugg et al. (1994) did not observe any differences among the
subject groups in the onset latency of the word repetition effect either. One methodological
difference between these experiments and ours is that we used sentences whereas these other
studies used word lists. However, if this is not the explanation, then the dissociation of the
effects of aging on the N400 congruity and the N400 word repetition effects must be taken to
mean that the two are not the same.
These age-related changes in the N400 congruity effect are interesting in light of recent
reports on the ERP word repetition effect (Hamberger & Friedman, 1992; Rugg et al., 1994;
Karayanidis, Andrews, Ward, & McConaghy, 1993). The ERP word repetition effect refers to
YOUNG ELDERLY
Frontol
A
Ak Ak @. A ,J
1 --
"4 JAI
I , I | I I , I , I I , I , I I , I , ?F"
0 200 400 0 200 400 0 200 400 0 200 400
............ First nouns Second nouns .... Third nouns
F i g u r e 6. G r a n d a v e r a g e E R P s at Frontal, Central, a n d Occipital electrode sites for Y o u n g ( n = 2 4 ) a n d E l d e r l y

( n = l 8) subjects s h o w i n g the serial position effect on the N400.
the greater positivity elicited by items repeated (old) relative to that occurring for the first time
in the experiment (new). Typically, it has been argued that the word repetition effect reflects
the modulation of multiple ERP components, of which one is the N400. The most consistent
finding has been the absence of any statistically significant differences in the size of the ERP
word repetition effect between young and elderly subjects. Rugg et al. (1994) did not observe
any differences among the subject groups in the onset latency of the word repetition effect
either. One methodological difference between these experiments and ours is that we used
sentences whereas these other studies used word lists. However, if this is not the explanation,
then the dissociation of the effects of aging on the N400 congruity and the N400 word
repetition effects must be taken to mean that the two are not the same.
2.3. Sentence-Level Effects
The ERP data we have considered so far have been time-locked to single words.
However, many language-related processes must by their nature be active at longer timescales
and analyses of their signatures in the brain's response to sentences should be quite revealing.
We have begun to examine such data with the goal of understanding the interactions between
faster, transient cognitive processes and slower, sustained processes (e.g., Kutas & King, in
Do the waves begin to waver ? 3 31
press). Note that we have akeady seen the effects of some of these interactions in the ERPs to
single words. Earlier, we hypothesized that the sustained frontal negativities characteristic of
the latter half of function word EKPs may reflect the fact that such items introduce major
syntactic constituents and generate expectations for the following content-related items.
Similarly, the contextual and serial position effects on the amplitude of the N400 during
reading are clearly dependent on the cumulative action of longer-lasting processes in
comprehension.
Changes spanning several words can also be seen in very low frequency ranges of the
ERP. Our working hypothesis in this approach is that these slow potential effects reflect
ongoing cognitive processes or changes in state caused by their continuing operation (e.g.,
fatigue). In the literature on very slow brain potentials employing non-linguistic tasks, it has
been shown that it is possible to detect systematic fluctuations in the slow potential field on
scalp regions that overlie the very neural circuits that are most heavily involved in the
processing (Roesler, 1993). In the remainder of this chapter we will discuss slow potentials
associated with three subprocesses of reading and describe how they fare during normal aging.
Specifically, we will examine slow potential effects that we hypothesize to covary with (1) the
continuous encoding of rapidly presented visual information, (2) the construction of higher
level representations (discourse) from linguistic input, and (3) the temporary storage of
currently unintegrated material within working memory. Based on all that we have argued
thus far, we expect that of these three processes, it is the latter (working memory storage) that
will show the greatest susceptibility to aging.
The negligible effects of aging on the slow potentials seen at occipital regions (left
hemisphere site shown in Figure 7) parallel those we have observed for the early visual EPs to
single words (also evident in the middle row, high pass filtered). Both young and elderly
subjects show a similar pattern during reading: a slow, negative potential shift away from the
resting baseline which appears to level out after approximately the third word in the sentence.
This slow potential ~ is relatively independent of the changes in the ERP
components that occur from word to word in the sentence, which d o vary with word class,
serial position and doze probability, among other factors. Most importantly, for the present
purpose, is that this slow potential shift over occipital sites does not seem to be si~ificantly
altered by aging. Thus, we seem to find little evidence for aging effects on the continuous
encoding of rapidly presented visual stimuli.*
In previous reports of across-sentence ERP data, a consistent but little discussed
finding has been a slow, progressive, slightly left-lateralized positive drift at anterior scalp sites
(e.g,. Kutas, Van Petten, & Besson, 1988). We have found that the amplitude of this positive
shift varies both with the structural complexity of the eliciting sentences and comprehension
skill on young adults (Kutas and King, in press). While the largest positive shifts have been
seen at electrode sites more anterior than those from which we have recorded in the elderly,
our preliminary data indicate that these frontal positivities also are relatively spared by aging.
Thus, as shown for the let~ frontal site in Figure 8a and 8b, while the elderly differ from the
young in the morphology of the higher frequency components (especially in their lack of P2
components), they are remarkably similar in their slow componentry. In these data, the slow
positive shift represents a steady positive ramp of about 0.3 microvolts per word. As far as
comprehension is concerned, performance by young and elderly subjects was virtually identical
*Further tests of this hypothesis are, of course, necessary.

YOUNG ELDERLY
Recorded ERP ~ , V~/~, v~, v~ 'VV . . . . .

(.01-20 Hz.)
High Frequency _~ t~,._~ ,~. tt .~. ~ -~ * ,, ~I ,11 ~.ll .,'l 5

(> 0.7 Hz.) IJ / V /
II I
4 !!
Low Frequency ...............

;" I , ,
"., ......... ..~ .~'" ..... I I t
(< 0.7 Hz.)
_
wl w2 w3 w4 w5 s~o ~1'oo ldSO 22'oo m,.c
Figure 7. Grand average ERPs spanning the first five words of a sentence at the left occipital electrode site for
Young (n=18) and Elderly (n=18) subjects, showing the recorded ERP and its High and Low Frequency
components.
in the Parsing Study. In the Repetition Study, the behavioral measure was tied to explicit
memory for sentence final words, and the elderly did perform more poorly on this task, as was
expected.
Precisely what cognitive processes are indexed by this positivity is unclear at the
moment, although some clues might be taken from its distribution and temporal dynamics. For
instance, its distinctly frontal distribution is consistent with processes located in more anterior
cortex, (e.g. frontal or anterior temporal regions), which, given the visual stimuli used, would
implicate higher level processing more than those indexed by the occipital negativity. Further,
the occipital negativity asymptotes relatively early, while this frontal positivity accretes in a
manner suggestive of a cumulative rather than transient underlying process.
The reasoning for this argument is as follows. If a given slow potential reflects an
increased level of electrical activity in a cortical region whose processing reaches some stable,
steady state, then the slow potential should reach some asymptotic voltage.** This case would
seem to cover the slow drift over occipital regions quite well if we assume that it represents a
change in activity level required by the low level processing of an incoming stream of words.
In the second case, if a slow potential reflects increased activity in a cortical region whose
This is because the capacitance of brain tissue is negligible, so that the instantaneous current flowing from
the brain through the scalp is a direct measure of the net current being emitted from the neural generators
involved.
Panel A:
Pccr~g St~d~l Sentences
YOUNG ELDERLY
,A .^ .A .A
(.01-20 Hz.)
iI I'~
High Frequency -,,.~ /",Yklr,,"l /'-~ " I't _ -,,' .=, .
(> 0.7 HZ.) Ii ~,"' I/ / ~'~~, I : l l~"
Low Frequency ..... ........ iT ......... . . . . . . .I. . . . . . . . . . . . . . . . . . . .I. . . . . . . . . . I ....... - ..... " ......... I I I
. .........................................
(< 0.7 Hz.) ............ ,......
3~v
I
,.,'1 ,,'2 ,'3 ,;4 .Q5 ' 6 560 10'00 15'00 20'00 msec
Panel B:
ore St~d~l Sentences

Repet'i,t'i,
YOUNG ELDERLY
Recorded ERI ~V,.~. V/,'~V~. j

(.01-20 Hz.)
High Frequency .,, ,-~9 _,.~ ..--i .q/i -, ,,Yl. ,/I ,j'~. _ , _~
(> 0.7 Hz.) 'l/"/ ''~1'' t /'' ~/J I//"
v
Low Frequency ..............)..............I............."F.............I..............

(< 0.7 Hz.)
" 1 -
3~v
i i I
,i ,./2 ,/3 ,,'4 .s 6 s~o 11'00 ldSO 22'00 n~,~c
Figure 8. Panel A and Panel B display multi-word ERP data at frontal sites for Young and Elderly subjects,
showing the recorded ERP and its High and Low frequency components
processing load increases over time, then the slow potential should track the resource demands
of the process, and may not necessarily reach an asymptotic voltage. In a sentence
comprehension task, the integration of the linguistic input into a discourse level representation
requires continual evaluation and constant linking between the current content and previous
knowledge about the topic. This ongoing, accretion of processing might therefore be expected
to result in a cumulative effect on slow potentials across the sentence. Conversely, one could
argue that, as a sentence progresses, its representation becomes more consolidated and
subsequent content becomes more predictable. On this view, the processing load at an
integrative level would progressively decrease as more input arrives and it is this that is
reflected in the ramp-like slow potentials observed.
Whichever interpretation of the slow frontal positivity is correct, the pattern we
observed in the elderly was very similar to that seen in the younger subjects. While this is
hardly conclusive proof~ it does suggest that integrative processing in structurally simple
sentences, as verified by comprehension, is little affected by the aging process. However, we
would expect to see a clear effect of aging on integration for sentences whose structure
imposes a heavy burden on working memory. This was, in fact, a primary motivation for
comparing the performance of young and elderly subjects during the processing of sentences
known to tax the limits of working memory and thereby lead to increased comprehension
difficulty in all readers, but especially in the elderly.
From the work of Kemper and her colleagues (e.g. Kemper, 1988), we know that older
adults change both their use and comprehension of various syntactic structures as they grow
older. Further, the structures most likely to cause difficulties in either production or
comprehension are precisely those that are generally argued to make the greatest demands on
working memory capacity. Investigating these ideas requires sentence types that differ in their
WMC demands but are otherwise similar enough to allow comparisons between individual
critical words and between the sentences themselves. For these reasons, psycholinguistic
investigations have frequently concentrated on two sentence types that contain relative clauses
but which differ subtly in their structure:
(la) The reporter who harshly attacked the senator admitted the error.
(lb) The reporter who the senator harshly attacked admitted the error.
Both sentences (la) and (lb) contain a relative clause modifying the subject of the sentence,
but differ in the role that the main subject noun phrase ('~he reporter") plays in the relative
clause; in (la), the main-clause subject is also the subject (and agent) o f th e verb in the relative
clause, while in (lb), it is the object (and patient). Accordingly, sentences like (la) are known
as subject-subject relative (SS) sentences, while those like (lb) are known as subject-object
relative (SO) sentences.
As any reader can readily attest to, SO sentences (lb) are generally more difficult to
process than SS sentences (la), although even SS sentences are more difficult than sentences
without relative clauses. A long history of linguistic argument starting with work by Chomsky
and Miller (1963) suggests that SO sentences tax working memory to a greater extent, and that
this load becomes especially acute at and just following the relative clause verb of SO
sentences; it is here where, in more modem theories, two separate thematic role assi,~nments
D o the waves begin to waver ? 335
YOUNG ELDERLY
Frontal 4Jl
Anterior ~,j~"~~ ~,,^..,.~ ~ ~ .__n~
Temporal ~7
Central
m
I i I ! I i I i I | I i I , I I I i I i I i I 7 r
0 200 400 0 2 0 0 400 0 200 400 0 200 400muc
2"n,d V e r b F~ller - - - 3 3 Mas Verb ............ 3 0 Mai,r~ Verb
Figure 9. Grand Average ERPs from six anterior electrode sites to SO, SS, and non-relative clause control
verbs for Young (n=24) and Elderly (n=18) subjects. The difference between the control verbs and the two
relative clause types is shaded dark grey, while the differencebetween SO and SS verbs is shaded light grey.
must be carried out. That is, it is here that readers encounter the first verb of the sentence and
must determine which noun phrase is indeed the subject. Note that, with these materials,
neither semantic nor pramnatic information can be used to make this choice. King and Just
(1991) verified that the greatest reading time differences are found at this point, and that these
differences were larger for readers with relatively small working memory capacities. While
some effect of carrying two (rather than one) noun phrases in working memory might be
expected before the end of the relative clause, such effects are generally not obtained in
reading times (e.g. King & Just, 1991; Ford 1983; Holmes and O'Regan, 1981). Perhaps under
these circumstances, RT measures are not sensitive enough to maintaining a load in WM, or,
alternatively, are sensitive to a number of different counteracting effects which therefore yields
a null effect.
We thus chose to examine the processing of SS and SO sentences in young and elderly
subjects by recording ERPs during their extent. In so doing, we uncovered ERP effects that
covary with differences in working memory use during parsing, and that also seem to
distinguish youngreaders from elderly readers as well as better comprehenders from poorer
comprehenders, presumably in part due to WMC limitations. The sentence location
immediately following the end of the relative clause ("admiRed" in (la) and lb)) where the
greatest RT effects between SO and SS sentences are generally found is also a site of large
ERP effects (see Figure 9).
In both young and elderly subjects, not only do the ERPs to main clause verbs from the
SS and SO sentences differ from each other, but, as expected, both of these differ from
comparable verbs in filler sentences that do not contain relative clauses at all. This is
consistent with the suggestion that even SS sentences tax WM relative to sentences without
relative clauses, albeit in different ways than SO sentences tax WM. In the case of SS
sentences, a WM load may arise because of the greater temporal separation between the
subject noun phrase and the (main) verb relative to sentences with simpler structures, rather
than because of any difficulties in determining which NP is the true subject. For both the
young and elderly subjects, the difference between SO and SS verbs is larger over anterior
relative to posterior sites and larger at left (than right) hemisphere sites. The difference
between SS and filler sentence verbs is also left lateralized in the young subjects, but not in the
elderly subjects; older subjects exhibit a more bilateral and distinctly more frontal difference.
We still need to see whether this particular aging difference is a replicable finding. This
difference notwithstanding, the overall pattern of ERP to the verbs from the various sentence
types is quite similar in the young and elderly subjects. In both age groups, a greater load on
working memory at the verb seems to be associated with a larger frontal, slightly
lefi-lateralized negativity.
By contrast, much greater age-related differences are revealed by the across-sentence
ERP data seen in Figure 10. In the younger subjects, the ERPs to both relative clause sentence
types are characterized by a positive frontal drLft that is larger for Good compared with Poor
comprehenders; likewise, the difference between the two relative clause types is larger for the
better comprehenders. This pattern is consistent with the notions that the good comprehenders
integrate the content of both sentence types more easily, and that they find the working
memory demands made by the two types (relative to their capacity) to be dissimilar. Poor
subjects, on the other hand, seem to experience difficulty with SS sentences so that they must
stretch their processing capacity even with these "simpler" loads. In briet, at the sentence
level, good and poor comprehenders differ in their treatment of SS sentences; their ERPs to
SO sentences are roughly similar. Turning to the ERPs of the elderly subjects, we note that the
Good comprehenders do exhibit slightly more frontal positivity (i.e. below the baseline) for
both sentences types than the Poor comprehenders. However, neither group of elderly
subjects shows as much difference between SO and SS sentence types as was present among
even the poorer young comprehenders.
Two other features of these data deserve brief mention. First, the ERP data of both the
young and elderly subjects show a very clear difference between the SO and SS sentences
much earlier in the sentence than is typically observed in RT studies; specifically, this difference
occurs at the sentence location where the second noun phrase of the SO sentences must be
loaded into working memory. Such memory-loading negativities have been seen in
non-linguistic tasks as well (e.g., Ruchkin et al., 1990). Another feature of the data from the
elderly is that the end of the SS relative clause is marked by a noticeable negative peak (around
3000 msec or word 7). Closer inspection reveals that there is a similar relative negativity for
the younger subjects, albeit smaller. We have also observed this clause-ending negativity
(CEN), with its fronto-central and left-lateralized distribution in simple declarative sentences
(Kutas and King, in press). Thus, the CEN may be an ERP feature of wider interest given the
YOUNG ELDERLY
V _ "~ "~ ~:-~"
w'l ;2 w3 w4 w5 w6 w7 w8 vl9wi0 0 ' 10'00' 20'00' 30'00'40'00 msec
. . . . . . . . . . . . . . . . . 30: ~uv relporter tulm tim senator harshl~l orltazked ~ e d the error...
83: The r ~ o r t l n " who harshly r |Ira senator ~ e 4 tim error...
Figure 10. Grand average multi-word ERPs from leR frontal sites for Young (n=24) and Elderly (n=18)
subjects in response to SO and SS sentences. Good comprehenders in each group (n=12 and n=9 respectively)
are shown in the top row and the Poor comprehenders shown in the bottom row. Word labels indicated on the
left scale correspond to the onset of words 1 through 10 in the the words in the example sentences given below
the waveforms.
known importance that clause endings have both in theoretical models of parsing (e.g., Frazier
& Fodor, 1978) and in RT and eye movement data (e.g., Just & Carpenter, 1980). Of greatest
relevance here, however, is that these processes, too, are intact in the elderly.
3. CONCLUSIONS
Like too many other topics within the field of cognitive aging research, not enough is
known about how language processes change as people age, let alone about the
electrophysiology related to these processes. What we do know is restricted to circumscribed
situations, and concerns mostly reading rather than listening or language production.
Fortunately, we can leverage this relatively scant information with the greater body of
information we have about ERPs and language processing in young adults to reach some
tentative conclusions and generate testable hypotheses for future research.
From the single word data we report, it appears that the ERPs prominent over the back
of the head such as P l and N1, which presumably reflect primarily early visual processing, are
quite similar across the lifespan. Indeed, N1-P2 amplitudes varied with comprehension status
in both young and elderly subjects alike. In contrast, both the temporal-parietal N1 and the
centro-ffontal P2 component were notably (and reproducibly) different in the older subjects, at
least under conditions where words were presented at relatively fast rates (i.e. with stimulus
onset asynchronies of either 500 or 550 msec. in these studies). While neither of these
YOUNG ELDERLY
"J"-~ ',~ x.f
./% A
]-
5~zv
1;o 2;o 3;o 4~o 6 1Ao 2;o 3Ao ~Ao I msec
Figure 11. ERPs to all open class words atthe left anterior temporal site for Young and Elderly subjects. The
top row shows traces for all individual subjects overlapped, while the other rows showtraces for approximately
matched pairs of Young and Elderly subjects.
components has been studied systematically in language tasks, the localization of an important
P2 generator to the basal temporal lobe area suggests that the marked reduction in its size may
be related to known reductions in grey matter in that region of the brain. Subcomponents of
the P2, likewise, have been implicated in studies of visual working memory, a process also
known to be affected adversely by aging. Later components such as the N400 have been better
in both young and elderly subjects and show the typical trend of becoming smaller and later
with advancing age. These changes in semantic analyses (contextual integration) are dearly
quantitative rather than qualitative in nature. The elderly are slower and more variable in their
registration of meaning. Exactly what mechanism is at the core of N400 generation remains
unclear, although both attentional and inhibitory processes have been suggested. Data from
our recordings of longer epochs suggest that much of the normal, sustained processing during
reading is essentially unchanged in the elderly, except when their reduced working memory
capacity impacts their efficiency at parsing linguistic input and at integrating the results into
their ongoing discourse representations.
Our observations on the general consequences of aging on reading notwithstanding, we
think it important to emphasize that these effects of aging are neither categorical nor absolute.
While we have taken care to exclude subjects whose physical or mental health was in question,
what we portray here as the result of '~he aging process" is, naturally, the net sum of many
influences that differ from individual to individual. As the waveforms in Figure 11 suggest,
individual variability is great even at those frontal sites where many aging-related changes are
evident on the average; taken one by one, some young and some old subjects look more alike
than one would have predicted from examining the averages alone. The grand mean is never
the grand meaning.
In the future, we expect to see much more work in the field of geriatric
psycholinguistics, not only to understand normal developmental changes in language
processing, but also to understand changes caused by diseases such as dementia of the
Alzheimer's type, Parkinson's dementia, and strokes that effect both the traditional and
nontraditional language areas (Ojemann, 1991). ERP-based research promises to be on the
forefront of such research efforts, especially if the ecological validity of ERP paradigms can be
increased by technological advances in the presentation of auditory stimuli, and in the use of
saccade-related potential research in reading paradimns (e.g. Marton & Szirtes, 1988). The
increasing availability of high quality anatomical MRI scans should also be crucial, not only to
allow the measurement of age-related changes in the brain (e.g., Jernigan et al., 1991) but also
as a way to facilitate the identification of the neural generators of ERPs (e.g., Dale & Sereno,
1993). In the end, however, it will take the efforts of more than just neuroscientists to answer
the mysteries of what it means to become older. When that story has been told, we should
expect to know more about the brain, but also more about story-telling.
REFERENCES
Allison, T., Ginter, H., McCarthy, G., Nobre, A. C., Puce, A., Luby, M., & Spencer, D. D.
(1994). Face recognition in human extrastriate cortex. Journal of Neurophysiology, 71,
821-825.
Anderson, J. K (1983). The architecture of cognition. Cambridge, MA: Harvard University
Press.
Baddeley, A. D. (1986). Working Memory. Oxford: Clarendon Press.
Becket, C. A. (1980). Semantic context effects in visual word recognition: An analysis of
semantic strategies. Memory & Cognition, 8, 493-512.
Becker, C. A. (1982). The development of semantic context effects: Two processes or two
strategies?. Reading Research Quarterly, 17, 482-502.
Berger, B. (1992). Dopaminergic innervation of the frontal cerebral cortex: evolutionary trends
and functional implications. Advances in Neurology, 57, 525-544.
Botwinick, J. (1984). Aging and behavior: a comprehensive integration of research findings
(3rd edition). New York: Springer.
Brown, A. S. (1991). A review of the tip-of-the-tongue experience. Psychological Bulletin,
109, 204-223.
Callaway, E., Tueting, P., & Koslow, S. (Eds.) (1978). Event-Related Brain Potentials in
Man. New York: Academic Press.
Chapman, R. M., McCrary, J. W., & Chapman, J. A. (1978). Short-term memory: The storage
component of human brain responses predicts recall. Science, 202, 1211-1214.
Chapman, R. M., McCrary, J. W., & Chapman, J. A. (1981). Memory processes and evoked
potentials. Canadian Journal of Psychology, 35, 201-212.
Chase, W. G., & Simon, H. A. (1973). Perception in chess. Cognitive Psychology, 4, 55-81.
Cohen, G. (1988). Age differences in memory for texts: Production deficiency or processing
limitations?. In L. L. Light & D. M. Burke (Eds.), Language, memory, and aging. New
York: Cambridge University Press.
Cohen, G. (1990). Recognition and retrieval of proper names: Age differences in the fan effect.
European Journal of Cognitive Psychology, 2, 193-204.
Dale, A. M., & Sereno, M. I. (1993). Improved localization of cortical activity by combining
EEG and MEG with MRI cortical surface reconstruction: a linear approach. Journal of
Cognitive Neuroscience, 5, 162-176.
de Groot, A. M. B. (1990). The locus of the associative-priming effect in the mental lexicon. In
D. A. Balota, G. B. Flores d'Arcais & K. Rayner (Eds.), Comprehension processes in
reading (pp. 101-123). Hillsdale, NJ: Lawrence Erlbaum Associates.
Donchin, E., & Coles, M. G. H. (1988). Is the P300 component a manifestation of context
updating?. Behavioral Brain Sciences, 11, 357-374.
Eslinger, P. J., & Grattan, L. M. (1993). Frontal lobe and frontal-striatal substrates for
different forms of human cognitive flexibility. Neuropsychologia, 31, 17-28.
Ford, M. (1983). A method of obtaining measures of local parsing complexity throughout
sentences. Journal of Verbal Learning and Verbal Behavior, 22, 203-218.
Frazier, L., & Fodor, J. D. (1978). The sausage machine: A new two-stage parsing model.
Fuster, J. M. (1989). The prefrontal cortex: anatomy, physiology, and neuropsychology of the
frontal lobe. New York: Raven Press.
Gerard, L., Zacks, R. T., Hasher, L., & Radvansky, G. A. (1991). Age deficits in retrieval: the
fan effect. Journals of Gerontology, 46, P131-P136.
Gemsbacher, M. A., & Faust, M. F. (1991). The mechanism of suppression: A component of
general comprehension skill. Journal of Experimental Psychology: Learning, Memory, &
Cognition, 17, 245-262.
Goldman-Rakic, P. S. (1987). Circuitry of primate prefrontal cortex and regulation of behavior
by representational memory. In F. Plum & V. Mountcastle (Eds.), Handbook of
Physiology: The Nervous System. Bethesda, MD: American Physiological Society.
Graybiel, A. M., Aosaki, T., Flaherty, A. W., & Kimura, M. (1994). The basal ganglia and
adaptive motor control. Science, 265, 1826-1831.
Gunter, T. C., Jackson, J. L., & Mulder, G. (1992). An electrophysiological study of semantic
processing in young and middle-aged academics. Psychophysiology, 29, 38-54.
Hamberger, M., & Friedman, D. (1992). Event-related potential correlates of repetition
priming and stimulus classification in young, middle-aged, and older adults. Journals of
Gerontology, 47, P395-405.
Harbin, T. J., Marsh, G. R., & Harvey, M. T. (1984). Differences in the late components ofthe
event-related potential due to age and to semantic and non-semantic tasks.
Electroencephalography & Clinical Neurophysiology: Evoked Potentials, 59(6), 489-496.
Hasher, L., Stokzfus, E. R., Zacks, 1L T., & Rypma, B. (1991). Age and inhibition. Journal
of Experimental Psychology: Learning, Memory, & Cognition, 17, 163-169.
Henderson, G., Tomlinson, B. E., & Gibson, P. H. (1980). Cell counts in human cerebral
cortex in normal adults throughout life using an image analysing computer. Journal of the
Neurological Sciences, 46(1), 113-136.
Hillyard, S. A. (1985). Electrophysiology of human selective attention. Trends in
Neurosciences, 8, 400-405.
HiUyard, S. A., & Picton, T. W. (1987). Eleotrophysiology of cognitive processing. Annual
Review of Psychology, 34, 33-61.
Holcomb, P. J., Coffcy, S. A., & Neville, H. J. (1992). Visual and auditory sentence
processing: A developmental analysis using event-related brain potentials. Developmental
Neuropsychology, 8(2-3), 203-241.
Holcomb, P. J., & Neville, H. J. (1991). Natural speech processing: An analysis using event-
related brain potentials. Psychobiology, 19, 286-300.
Holmes, V. M., &, Rcgan, J. K~ (1981). Eye fixation patterns during the reading of relative-
clause sentences. Journal of Verbal Learning & Verbal Behavior, 20, 417-430.
Hunt, E. B., Lunneborg, C., & Lewis, J. (1975). What does it mean to be high verbal?.
Cognitive Psychology, 2, 194-227.
Jeifreys, D. A., & Tukmachi, E. S. (1992). The vertex-positive scalp potential evoked by faces
and by objects. Experimental Brain Research, 91(2), 340-50.
Jemigan, T. L., Press, G. A., & Hesselink, J. IL (1990). Methods for measuring brain
morphologic features on magnetic resonance images. Validation and normal aging.
Archives of Neurology, 47, 27-32.
Just, M. A., & Carpenter, P. A. (1980). A theory of reading: From eye fixations to
comprehension. Psychological Review, 87, 329-354.
Just, M. A., & Carpenter, P. A. (1992). A Capacity theory of comprehension: Individual
differences in working memory. Psychological Review, 99, 122-149.
Karayanidis, F., Andrews, S., Ward, P. B., & McConaghy, N. (1993). Event-related potentials
and repetition priming in young, middle-aged and elderly normal subjects. Cognitive Brain
Research, 1, 123-134.
Kemper, S. (1986). Imitation of complex syntactic constructions by elderly adults. Applied
ERP studies of language in aging. Psycholinguistics, 7, 277-287.
Kemper, S. (1987). Constraints on psychological processes in discourse production. In H. W.
Dechert & M. Raupach (Eds.), Psycholinguistic models of production (pp. 185-188).
Norwood, NJ: Ablex Publishing Corp.
Kemper, S. (1987). Syntactic complexity and elderly adults' prose recall. Experimental Aging
Research, 13, 47-52.
Kemper, S. (1988). Geriatric psycholinguistics: Syntactic limitations of oral and written
language. In L. L. Light & D. M. Burke (Eds.), Language, memory, and aging (pp. 58-
76). New York, NY: Cambridge University Press.
Kemper, S., Rash, S., Kynette, D., & Norman, S. (1990). Telling stories: The structure of
adults' narratives. European Journal of Cognitive Psychology, 2, 205-228.
King, J., & Just, M. A. (1991). Individual differences in syntactic processing: The role of
working memory. Journal of Memory and Language, 30, 580-602.
King, J. W., & Kutas, M. (in press). Who did what and when? Using word- and clause-related
ERPs to monitor working memory usage in reading. Journal of Cognitive Neuroscience.
Kutas, M. (1993). In the company of other words: Electrophysiological evidence for single-
word and sentence context effects. Language & Cognitive Processes, 8, 533-572.
Kutas, M., & Hillyard, S. A. (1980). Event-related brain potentials to semantically
inappropriate and surprisingly large words. Biological Psychology, 11, 99-116.
Kutas, M., & Hillyard, S. A. (1980). Reading senseless sentences: brain potentials reflect
Kutas, M. Marta, & Hillyard, S. A. (1982). The lateral distribution of event-related potentials
during sentence processing. Neuropsychologia, 20, 579-590.
Kutas, M., & Hillyard, S. A. (1983). Event-related brain potentials to grammatical errors and
semantic anomalies. Memory & Cognition, 11, 539-550.
Kutas, M., & Hillyard, S. A. (1984). Brain potentials during reading reflect word expectancy
and semantic association. Nature, 307, 161-163.
Kutas, M., Lindamood, T., & Hillyard, S. A. (1984). Word expectancy and event-related brain
potentials during sentence processing. In S. Kombhm & J. Requin (Eds.), Preparatory
states and processes (pp. 217-238). Hillsdale, NJ: ErlbauI~
Kutas, M., Neville, H. J., & Holcomb, P. J. (1987). A preliminary comparison of the N400
response to semantic anomalies during reading, listening and signing.
Electroencephalography and Clinical Neurophysiology, Supplement, 39, 325-330.
Kutas, M., & Van Petten, C. (1994). Psycholinguistics Electrified. In M. A. Gemsbacher (Ed.),
Handbook ofpsycholinguistics (pp. 83-143). San Diego: Academic Press.
Kutas, M., Van Petten, C., & Besson, M. (1988). Event-related potential asymmetries during
the reading of sentences. Electroencephalography and Clinical Neurophysiology, 69, 218-
233.
Kynette, D., & Kemper, S. (1986). Aging and the loss of grammatical forms: A cross-sectional
study of language performance. Language & Communication, 6, 65-72.
Light, L. L., & Capps, J. L. (1986). Comprehension of pronouns in young and older adults.
Lima, S. D., Hale, S., & Myerson, J. (1991). How general is general slowing? Evidence from
Mangun, G. R., & Hillyard, S. A. (1991). Modulation of sensory-evoked brain potentials
indicate changes in perceptual processing during spatial priming. Journal of Experimental
Mangtm, G. R., Hillyard, S. A., & Luck, S. J. (1993). Electrocortical substrates of visual
selective attention. In D. E. Meyer & S. Kornbhm (Eds.), Attention and performance 14:
Synergies in experimental psychology, artificial intelligence, and cognitive neuroscience
(pp. 219-243). Cambridge, MA: MIT Press.
Marton, M., & Szirtes, J. (1988). Context effects on saccade-related brain potentials to words
during reading. Neuropsychologia, 26, 453-463.
McCallum, W. C., Farmer, S. F., & Pocock, P. V. (1984). The effects of physical and semantic
incongruities of auditory event-related potentials. Electroencephalography & Clinical
McGeer, P. L., McGeer, E. G., & Suzuki, J. S. (1977). Aging and extrapyramidal function.
Archives of Neurology, 34(1), 33-35.
Miller, E. I~, & Desimone, 1~ (1994). Parallel neuronal mechanisms for short-term memory.
Science, 263, 520-522.
Miller, G. A., & Chomsky, N. (1963). Finitary models of language users. In D. Luce, 1L Bush,
& E. Galanter (Eds.), Handbook of mathematical psychology. New York: John Wiley.
Mtmte, T. F., Heinze, H-J., & Mangun, G. It (1993). Dissociation of brain activity related to
syntactic and semantic aspects of language. Journal of Cognitive Neuroscience, 5, 335-
344.
Naatanen, 1L, & Michie, P. T. (1979). Early selective-attention effects on the evoked potential:
A critical review and reinterpretation. Biological Psychology, 8, 81-136.
Neville, H. J., Mills, D. L., & Lawson, D. S. (1992). Fractionating language: different neural
subsystems with different sensitive periods. Cerebral Cortex, 2, 244-58.
Nunez, P. L. (1981). Electric Fields of the Brain. New York: Oxford University Press.
Ojemann, G. A. (1991). Cortical organization of language. Journal of Neuroscience, 11, 2281-
2287.
Perfetti, C. A., & Lesgold, A. M. (1977). Discourse comprehension and sources of individual
differences. In M. A. Just & P. A. Carpenter (Eds.), Cognitive processes in
comprehension. Hillsdale, NJ: Erlbaunl
Pfefferbaum, A., Mathalon, D. H., Sullivan, E. V., Rawles, J. M., Zipursky, 1L B., & Lim, K.
O. (1994). A quantitative magnetic resonance imaging study of changes in brain
morphology from infancy to late adulthood. Archives of Neurology, 51, 874-87.
Puce, A., Nobre, A. C., Allison, T., Luby, M. L., McCarthy, K. M., & McCarthy, G. (1994).
Event-related potential correlates of visual recognition modules in extrastriate cortex.
Cognitive Neuroscience Society Meeting, 1, 93-93.
Pujol, J., Junque, C., Vendrell, P., Grau, J. M., & Capdevila, A. (1992). Reduction of the
substantia nigra width and motor decline in aging and Parkinson's disease. Archives of
Neurology, 49, 1119-1122.
Renault, B., Kutas, M., Coles, M. G. H., & Gaillard, A. W. K., (Eds.) (1989). Event-related
potential studies of cognition. Am~erdam, Holland: Elsevier.
Rosier, F. (1993). Beyond reaction time and error rate: Monitoring mental processes by means
of slow event-related brain potentials. In W. C. McCallum & S. H. Curry (Eds.), Slow
potential changes m the human brain (NATO ASI series. Series A: Life sciences, Volume
254, pp. 105-119). New York, NY: Plenum Press.
Ruchkin, D. S., Johnson, 1L, Canoune, H., & Ritter, W. (1990). Short-term memory storage
and retention: an event related brain potential study. Electroencephalography and Clinical
Rugg, M. D., Pearl, S., Walker, P., Roberts, 1L C., & Holdstock, J. S. (1994). Word repetition
effects on event-related potentials in healthy young and old subjects, and in patients with
Alzheimer-type dementia. Neuropsychologia, 32, 381-398.
Salthouse, T. A. (1985). Speed of behavior and its implications for cognition. In J. E. Birren
& I~ W. Schaie (Eds.), Handbook of the psychology of aging. New York: Van Nostrand
Reinhold Co, Inc..
Sawaguchi, T., & Goldman-Rakic, P. S. (1991). D 1 dopamine receptors in prefrontal cortex:
involvement in working memory. Science, 251, 947-950.
Tipper, S. P. (1985). The negative priming effect: Inhibitory priming by ignored objects.
Quarterly Journal of Experimental Psychology: Human Experimental Psychology, 37A,
571-590.
Tipper, S. P. (1991). Less attentional selectivity as a result of declining inhibition in older
adults. Bulletin of the Psychonomic Society, 29, 45-47.
Van Petten, C., & Kutas, M. (1991). Influences of semantic and syntactic context on open and
closed class words. Memory and Cognition, 19, 95-112.
Verleger, K (1988). Event-related potentials and cognition: A critique of the context updating
hypothesis and an alternative interpretation of P3. Behavioral & Brain Sciences, 11(3),
343-356.
Wechsler, D. (1987). Wechsler Memory ScalewRevised. New York: Psychological
Corporation.
Woodward, S. H., Ford, J. M., & Hammett, S. C. (1993). N4 to spoken sentences in young
and older subjects. Electroencephalography & Clinical Neurophysiology, 87, 306-320.
Zelinski, E. M. (1988). Integrating information from discourse: Do older adults show deficits?
In L. L. Light & D. M. Burke (Eds.), Language, memory, and aging. New York:
Cambridge University Press.
Zola-Morgan, S., & Squire, L. 1L (1993). The memory system damaged in medial temporal
lobe amnesia: Findings from humans and nonhuman primates. In O. Taketostfi, L. 1L
Squire, M. E. Raichle, D. I. Perrett & M. Fukuda (Eds.), Brain mechanisms of perception
and memory: From neuron to behavior (pp. 241-257). New York, NY: Oxford University
Press.
Memory and Aging: An Event-Related Brain Potential Perspective*
David Friedman and Monica Fabiani
Cognitive Electrophysiology Laboratory, New York State Psychiatric Institute
In this chapter we will review, from an event-related brain potential (ERP) perspective,
the results of studies on the cognitive aging of memory-related phenomena. This review will
not cover studies of "short-term memory," i.e., "primary" memory (see Waugh & Norman,
1966), as typically assessed via digit span or the Steinberg memory scanning paradi,~mn. Older
adults appear to have similar-capacity primary memory stores compared to young adults (see
Poon, 1980 for a review) and, aside from longer reaction times and P3 latencies in the memory
scanning procedure, show similar scan times per item (as assessed by P3 latency) to those of
their young adult counterparts (see Ford et al., 1979). Comprehensive reviews of this
literature have been published elsewhere (Bashore, 1990; Friedman, in press; Ford &
Pfefferbaum, 1985; Polich, 1991), and the interested reader is referred to these sources for
detailed treatments. Working memory (Baddeley, 1986) will also not be reviewed, due to a
paucity of studies dealing with ERPs and working memory (for a recent review of behavioral
studies, see Salthouse, 1994).
This chapter will concern itself primarily with the explicit/implicit memory distinction
that has recently become one of the most intensively researched areas within cognitive
neuroscience in general and the cognitive aging field in particular. In keeping with the scope of
the current volume, virtually all of the ERP studies of explicit and implict memory-related
phenomena have employed words as stimuli. Thus, the conclusions we reach at the end of this
chapter apply almost exclusively to verbal memory. We will first introduce the explicit/implicit
memory distinction and briefly review the major behaviorally-based findings with respect to
cognitive aging. We will then briefly discuss the neuropatholo~cal data, considering the extent
to which these findings can explain the age-related dissociation between performance on direct
Address correspondence and requests for reprints to: Dr. David Friedman, Cognitive Electrophysiology
Laboratory, Unit 58, New York Psychiatric Institute, 722 West 168th Street, New York City, New York 10032.
Phone: (212):960-2476. Fax: (212):781-2661. E-mail: FRIEDMAN@IRIS.RFMH.ORG
ACKNOWLEDGEMENTS: The authors express their deep appreciation to the collaborators on the various
projects described in this review, Drs. Steven Berman, Maria Hamberger, Victoria Kazmerski, Walter Ritter,
Gregory Simpson, and Joan G. Snodgrass. In addition, we thank Mr. Charles L. Brown for preliminary data
reduction and computer programming, and Ms. Charlotte Trott, Ms. Blanca Rincon, Mr. Sean Hewitt, and Mr.
Jeff Cheng for their aid in the collection and analysis of data resulting from several of the studies reported here.
We thank Ms. Rachel Yarmolinsky and Ms. Eve Vaag for the construction and photo reproduction of figures.
Many thanks to Dr. Ted Bashore for providing us with critical commentary and editorial assistance. Thanks
also to Dr. Ray Johnson, Jr. for his criticial commentary. Preparation of this chapter was supported in part by
grants AG05213 and AG09988 from the NIA, and by the New York State Department of Mental Hygieffe: The
Computer Center at New York State Psychiatric Institute is supported in part by a grant (MH-30906) from the
National Institute of Mental Health. David Friedman is supported by Research Scientist Development Award
#K02 MH00510 from NIMH.
346 D. Friedman and M. Fabiani
and indirect tests of memory. A second intensive area of investigation within cognitive aging,
in which traditional psychometric and experimental neuropsychological, as well as
psychophysiological techniques have been brought to bear, is the extent to which memory
deficits can be explained by changes in frontal lobe function in older individuals. The
performance data will be considered first, followed in a later section by ERP findings from this
laboratory. After these introductory sections concerned primarily with behaviorally-based data
are presented, we will review memory-related ERP phenomena in young adults, following this
with a review of the extant memory-related ERP aging literature.
1. DIRECT (EXPLICIT) AND INDIRECT (IMPLICIT) MEMORY AND AGING.
It is, by now, amply clear, that older individuals demonstrate deficits, relative to young
adults, on traditional episodic or explicit tests of memory, such as free recall, cued recall, and
recognition, tests which require the conscious accessing of previously experienced events (for
reviews see Light, 1991, and Moscovitch & Winocur, 1992). The evidence for this "age-
associated memory impairment" (or AAMI) for explicit recollection in otherwise normally
aging individuals is so overwhelming that a diagnostic category with this label has been
proposed (Crook et al., 1986). On the other hand, a reasonably large and burgeoning literature
has documented the relative insensitivity of aging on performance during indirect or implicit
tests of memory (for reviews, see Davis & Bemstein, 1992 and Howard & Wiggs, 1993).
Memory is demonstrated indirectly when the subject's task does not make reference to
previously experienced events, but performance shows the facilitating effect of having
experienced those events. For example, subjects might be asked to study a fist of words during
an initial study phase. This is followed after a delay by a test phase in which subjects are
exposed to a series of three-letter stems, some of which formed parts of the words they saw
during the study phase, while some were not previously presented (these are baseline or foil
stems). They are asked to complete the stem with the "first word that comes to mind." The
benefit of previous experience is demonstrated by the typical finding that more "old" stems are
completed correctly than foil stems. This performance benefit or priming on this test of word
stem completion appears to be preserved with aging.
A similarly constructed direct test, stem cued recall, can also be administered. In this
case, subjects study a highly similar list of items and are then given the test of stem cued recall.
During the test phase the only difference between word stem completion and stem cued recall
is the instructional set. For word stem completion the memorial nature of the test is
"disguised" by telling subjects they are participating in a "word puzzle" or that they are "aiding
in a normative study of word completions" and, as stated above, are to supply the first word
that pops into mind. That is, subjects do not necessarily need to consciously access a
representation of the previous event in order to show the benefit of previous experience. By
contrast, in the direct test, subjects are told that they are to complete the stem only if they
specifically remember having seen it during the study phase. In this case, subjects must
consciously access their memory for the previous episode, determine whether or not the stem
was seen during the study phase, and then supply the appropriate response.
Impetus for the investigation of performance on these two types of memory tests was
provided by studies of densely amnestic individuals who were severely impaired on direct tests
of memory but nonetheless performed as well as normals on indirect tests. Implicit
performance was preserved for the same stimuli and episodes for which the explicit deficit had
Memory and aging 347
been demonstrated (e.g., Warrington & Weisenkrantz, 1968; see Cohen & Eichenbaum, 1993,
Squire, 1987, and Squire, 1992 for reviews). Moreover, several impressive and highly
replicable performance dissociations between these two types of test were published. For
example, although the type of orienting activity (semantic versus non-semantic) at study had a
dramatic effect on explicit memory performance, it had a smaller or null effect on implicit
memory performance. Coupled with the fact that these amnestic patients had well-localized
injuries to specific parts of the brain (e.g., medial temporal lobe; diencephalic structures),
several theorists proposed that performance on each type of test is subserved by a
neuroanatomically and computationally distinct memory system. This pattern of impaired
performance on direct and spared performance on indirect memory testing is one of the most
powerful pieces of evidence offered in support of the explicit/implidt multiple memory systems
approach to human memory.
Explicit memory performance is thought to be mediated primarily via the medial
temporal lobe system (with the hippocampus playing a pivotal role--see Cohen & Eichenbaum,
1993 for a review and synthesis), whereas performance on indirect tests is thought to be
mediated via neural circuits lying outside the medial temporal lobe memory system
Presumably, in the case of perceptually-based implicit tests, these computations would be
carried out in sensory cortical areas (e.g., Tulving & Schacter, 1990), while in the case of
conceptually-based implicit tests, in neocortical association areas (see, for example, Gabfieli et
al., in press; see Roediger & McDermott, 1993 for a review).
An alternate approach, labeled transfer-appropriate processing, purports that there is
no fundamental difference between implicit and explicit memory, and explains dissociations and
associations among types of memory tests on the basis of shared processing mechanisms
(Roediger, 1990; Roediger & Blaxton, 1987; for a review see Roediger & McDermott, 1993).
Data-driven or perceptual processing results when the characteristics of the stimuli (e.g.,
morality, typography, letter case, surface form) are critical in performing the task. In contrast,
conceptually-driven processing is called on when subjects must elaborate, organize or in some
fashion meaningfully process the stimuli in order to perform adequately in the task. On this
view, the pattern of association and dissociation between direct and indirect tasks (or, for that
matter, within direct or indirect tests) will depend critically on the type of processing that
subjects must engage in. However, it is difficult to evaluate the tranfer appropriate processing
approach with respect to aging, since the majority of studies have not been directed at testing
this hypothesis.
Another distinction, item versus associative priming, has been used to account for age-
related differences on indirect memory performance (see, for example, Howard et al., 1991).
Although some studies of indirect memory for new associations do show the older adult to be
impaired relative to the young adult (i.e., associative implicit memory; e.g., Howard et al.,
1991), more recent studies have shown that the older adult is also at a disadvantage when
implicit memory for items stored in long-term semantic memory is tested (i.e., implicit item
memory; e.g., Davis et al., 1990). These latter data suggest that the item versus associative
implicit memory distinction may not account for much of the variance in explaining age-related
deficits on indirect tests of memory (see also, Friedman et al., in press).
Thus, age differences on direct and indirect tests could either be due to a breakdown
with aging in strategic processing or in the putative brain system(s) that subserve each form of
memory, or in some combination of the two. In addition, several independent and overlapping
hypotheses have been advanced to account for memory deficits in the elderly. For example,
Light (1991) considered four hypotheses for explaining memory decline in old age: 1) Failures
of metamemory: Older subjects have deficient knowledge about memory, do not employ the
most efficient strategies, and are poor at monitoring their own memories. These deficiencies
might be a consequence of frontal lobe deficits, which have been hypothesized to account for
some aspects of cognitive aging--see Moscovitch and Winocur, 1992 for a review; see also
frontal lobe section below); 2) Semantic deficit hypothesis: Older subjects do not
spontaneously employ elaborative activity to adequately encode to-be-remembered material,
leading to poorer quality memory traces that are deficient in richness, extensiveness and depth
of encoding; 3) Impairment in deliberate recollection: Older adults have an intact activation
mechanism, but show deficiencies in the processing of the spatio-temporal context in which the
information was learned, consistent with older individuals showing preserved indirect, but
impaired direct memory performance (possibly due to neuropathology in the structures that
mediate direct memory); 4) Reduced processing resources: Older subjects display reduced
attentional capacity, reduced working memory capacity, and cognitive slowing (see Salthouse,
1991). However, Light (1991) concluded that no one of these hypotheses effectively provides
a good explanatory mechanism for the empirical evidence of reduced memory performance
with increasing age.
2. NEUROANATOMICAL BASIS OF AGE-RELATED MEMORY DIFFERENCES.
Most hypotheses concerning the sources of the cognitive aging of memory have been
framed within the brain systems approach, since older subjects appear at first glance to look
like less impaired amnestics, in the sense that they are relatively impaired on direct tests of
memory, but perform similarly to young adults on indirect tests of memory. Moreover, the
extant neuropathological and imaging data appear to support this distinction, pointing to
relatively greater neuropathological involvement of structures intimately involved in explicit
memory, such as the hippocampus (e.g., Golomb et al., 1994), and relatively less involvement
of sensory and neocortical association areas (e.g., Bouras et al., 1994), which are thought to
underlie implicit memory performance. However, recent behavioral data obtained during
implicit paradigms (e.g., Davis et al., 1990; Hultsch et al., 1991) have begtm to blur this
distinction, as these studies have shown poorer implicit memory performance for older relative
to younger adults. One possibility that could account for these results might be the extent to
which frontal lobe control functions are disturbed with aging (e.g., Stuss et al., 1994; see
below). If frontal lobe deficits (which are reported to increase with age), in addition to medial
temporal lobe deficits, impact performance on both implicit and explicit memory tasks (which
has also been reported in studies of cognitive aging of memory-related phenomena--see frontal
lobe section below), this would tend to decrease the age-related implicit/explicit memory
dissociative pattern.
To what extent do the age-related neuropathological data so far reported map onto
those brain areas that are implicated in performance during direct and indirect memory?
Although changes with age in indices of what could be termed brain insult (e.g., cell loss;
neurofibrillary tangles) appear to occur diffusely throughout the brain, there are some data
suggesting that cell loss is more pronounced in the hippocampus than elsewhere (e.g., Dam,
1979). Moreover, Tomlinson et al. (1968) reported that the presence ofneurofibrillary tangles
was more prominent in the hippocampus and parahippocampal gyrus than in the neocortex.
Similarly, Scheibel et al. (1976), although based on a very small sample of brains, reported
deterioration of hippocampal dendritic systems with increasing age. In a very recent study,
Golomb et al. (1994) found a fairly strong relationship between delayed recall and the size of
the hippocampal formation (measured via magnetic resonance imaging) in a sample of normally
aging older adults, whereas smaller or null relationships were found between the delayed recall
measure and non-hippocampal measurements. All of these factors suggest that some of the
difficulty the elderly exhibit during direct memory testing could be due to hippocampal
malfunction.
With respect to those neocortical areas implicated in performance on indirect testing,
the data of Terry et al. (1987) suggest that aging has more of an effect on the frontal and
temporal lobes than on the parietal lobes, with age-related changes taking the form of a
shrinkage of large neurons, and a concomitant increase in the ratio of small to large neurons.
Just what cognitive sequelae neuronal shrinkage would have is largely unknown, but the more
anterior distribution of the shrinkage is consistent with preserved indirect memory in the
elderly being mediated via posterior cortical areas (see Heindel et al., 1989; 1990). Additional
support implicating neocortical association areas as putative sites underlying lexical priming
performance has been reported by Heindel et al. (1989). The Heindel et al. data make it clear
that different indirect tasks (with different task demands and underlying processes--Roediger &
McDermott, 1993) are most likely mediated via different brain areas. Since some age-related
changes (e.g., cell loss) are not uniform throughout the brain (Haug et al., 1983), just which
implicit task performance is preserved with age will depend upon the brain area(s) implicated in
task performance and the amount of"damage" they have suffered.
3. FRONTAL LOBE FUNCTION.
As previously alluded to, another major hypothesis used to account for cognitive aging
phenomena is that the elderly show deficits in frontal lobe function. This evidence comes from
a variety of sources, including traditional neuropsychological test performance (see, for
example, Albert & Kaplan, 1987), imaging of cerebral blood flow (e.g., Shaw et al., 1984),
experimental neuropsychological investigations (e.g., Craik et al., 1990), neuropathological
data (Kemper, 1984; Scheibel & Scheibel, 1975), and ERP investigations (reviewed below;
e.g., Friedman et al., 1993c; Fabiani & Friedman, in press). For example, Haaland et al. (1987)
reported age-related performance deficits in normally aging samples with the Wisconsin Card
Sort Test (WCST), a test typically used to assess the extent of frontal lobe dysfunction.
Haaland et al's data indicated that deficits in two response categories, number of categories and
number of errors were confined primarily to the oldest group of subjects (aged 80 to 87).
Albert et al. (1990) reported age-related decrements in performance on three tests of
abstraction ability, a function also linked with the frontal lobes. However, care must be
exercised when attempting to infer a deficit in frontal lobe function from a psychometric
instrument such as the WCST, as there is evidence that questions the sensitivity and specificity
of this putative frontal lobe test. Moreover, a so-called "frontal dysfunction" may occur
without evidence of focal frontal disturbance, and could result from diffuse cortical damage
(Stuss, 1993). Therefore, it is a good idea to employ more than one test of frontal lobe
function.
Another memory function, memory for source, or the context within which an item is
learned, also appears to depend upon the frontal lobes (Schacter et al., 1984). Source amnesia
refers to the inability to remember where information was originally acquired, while the ability
to recall the information itself is retained. In an experimental neuropsychological study of

source amnesia, Craik et al. (1990) reported that, for their sample of older subjects (aged
60-84), the degree of source amnesia was correlated with age and with deficits on measures of
frontal lobe fimction. Several other investigators have also replicated this phenomenon in
elderly samples (e.g., McIntyre & Craik, 1987; Spencer & Raz, 1994). Thus, contextual
information may not be as readily available to the elderly as to their young adult counterparts,
and this may account for some of the reported deficits in explicit remembering. Moreover, in
support of a frontal lobe role in memory function, frontal lobe contributions to both memory
encoding and retrieval have recently been given weight by studies of explicit memory using the
PET technique (Kapur et al., 1994; Shallice et al., 1994; Tulving et al., 1994a; 1994b).
Indices of frontal lobe dysfunction may also have an impact on implicit memory
performance in the elderly. For example, in the results of Davis et al. (1990), stem completion
priming performance was negatively correlated with two types of errors (number of trials
required to identify the first Wisconsin category; failure to maintain a consistent pattern of
response--failure to maintain set) that increased with increasing age on the WCST. Moreover,
Parkin & Walter (1992) reported an increase in the proportion of familiarity- compared to
contextually-based judgements with age that was correlated with measures of frontal lobe
dysfunction. This experiment was motivated by two-process theories of recognition memory
(e.g., Mandler, 1980), which postulate that familiarity or perceptual fluency as well as a
contextual episodic component contribute to recognition judgements. The fluency or
familiarity component is also thought to play a role during implicit memory tests (cf., Gardiner
& Java, 1990). Moreover, two-process theories receive some support from the finding that
recognition judgements can be partitioned into "know" (i.e., seen before in the experiment and
recoanized on the basis of familiarity) and "remember" (i.e., context-related) responses (cf.,
Gardiner & Java, 1990). Thus, these data begin to suggest the possibility that age-related
changes in performance on both implicit and explicit tests could be explained, at least partially,
by changes in frontal lobe function.
4. THE ERP ELICITED D U R I N G M E M O R Y TASKS.
In recent years, the ERP has been increasingly employed as a convergent and
complementary source of information in studies of memory. Several reviews of the basic
findings have been published (Johnson, in press; Kutas, 1988; Rugg, in press; Rugg & Doyle,
1994), and only a brief overview will be provided here. ERPs are time-locked changes (on a
millisecond time base) in the brain's ongoing electrical activity that occur in response to
sensory, motor, or cognitive events, and can provide information concerning the hierarchy,
sequencing and timing of information processing that is typically much more difficult to obtain
from behavioral indices alone. In contrast to inferring intervening stages from reaction time,
which is the final common pathway for a number of information processing stages, the
components (or voltage swings) of the ERP (each presumed to reflect a different aspect of
information processing) can be measured relatively directly (see Hillyard & Picton, 1987 for a
review). The ERP also provides information on the spatial distribution of electrical activity
recorded at the scalp. That is, the distribution of ERP component amplitudes on the scalp
surface can be assessed to determine if different conditions, for example, direct versus indirect
memory (Kazmerski & Friedman, submitted; Paller, 1990), produce different distributions (or
topographies) across the scalp. If two (or more) conditions give rise to different scalp
distributions, one inference that can be made is that the electrical activity elicited by the various
conditions is either generated by a different configuration of intracranial sources and/or an
amplitude change in a subset of those neural sources (see Johnson, 1993 for a detailed
treatment). This kind of information could figure importantly, for example, in inferring
whether explicit and implicit memory can be considered to be subserved by functionally and
anatomically unique systems, and whether young and older adults differ in the extent to which
their ERP distributions can be considered to provide evidence of frontal lobe sources.
ERPs have been recorded during direct (e.g., Bentin et al., 1992; Donchin & Fabiani,
1991; Fabiani & Donchin, in press; Friedman, 1990a; Johnson et al., 1985; Neville et al., 1986;
Paller, 1990; Rugg & Nagy, 1989; Smith & Halgren, 1989), as well as indirect (e.g., Bentin &
Moscovitch, 1990; Friedman et al., 1993a; Paller & Kutas, 1992; Rugg et al., 1988) tests of
memory. Immediately below we review the major findings resulting from many of these
studies of memory in young adults, separately for measures of encoding and retrieval. This is
followed by a review of age-related changes during similar kinds of memory tasks.
4.1. Encoding
The first investigators to compute ERP measures recorded during a study phase as a
function of subsequent memory test phase performance were Sanquist and his colleagues
(1980). Since that seminal study, several investigators have noted an association between
electrical activity recorded at study and performance on subsequent direct (e.g., Friedman,
1990b; Paller, 1990; Paller & Kutas, 1992; Paller et al., 1987a; Karis et al., 1984; Fabiani et al.,
1986, 1990) as well as indirect (e.g., Paller et al., 1987b) tests of memory (for a
comprehensive review of these studies, see Johnson, in press). In these investigations, the
ERPs elicited at study (i.e., to "new" or first presentation items) are averaged according to
whether those items were or were not subsequently correctly recalled, recognized, or primed at
test. The difference between these two ERPs is then evaluated, either by measuring separately
the two waveforms, or by subtracting the subsequently incorrect from the subsequently correct
ERPs. The resulting difference is considered to represent the ERP sign of those processes that
led to successful memory test performance (presumably related to encoding activity). Figure 1
depicts a typical subsequent memory effect recorded to new or first presentation words during
a continuous recognition memory paradigm (Friedman, 1990b). Some authors have
interpreted the subsequent memory effect in terms of the modulation of the P3b* component
(e.g., Donchin & Coles, 1988). Other authors, however, have pointed out that the distribution
of the difference between these two classes of ERP varies from that of the P3b with which it
appears to overlap (see Friedman, 1990b; Paller et al., 1987; and Johnson, in press for similar
*The label, "P3b," refers to a positive potential elicited typically by task-relevant, infrequently occurring events,
with a peak latency between 300 and 1000 ms post-stimulus (depending upon the complexity of the task). The
scalp distribution of the P3b is usually, but not always, characterized by a parietally-focused amplitude
maximum. P3b's scalp distribution can also be modulated by task requirements (see Johnson, 1993 for a
complete discussion). Infrequently occurring, task-irrelevant, unusual or novel events, elict a "novelty P3," with
a more frontaUy-oriented scalp distribution than the P3b (see discussion in the Assessments of Frontal Lobe
Function in the Elderly section of this chapter). In addition, another P3 component, the "P3a," has been
observed in the waveforms elicited by very infrequently occurring deviant events, when those events are
"unattended," as well as when they are attended (see N. Squires et al., 1975). The P3a occurs with a peak
latency of around 280 ms and is recorded with a Cz-maximum scalp distribution in a midline series of electrode
sites.
arguments). In fact, as can be seen in the figure, the memory effect (indicated by the shaded
areas) has a frontally-oriented scalp distribution, whereas the P3b displays a parietally-maximal
scalp distribution.
In their initial study, Pallet et al. (1987a) labeled this difference "Din" (for _difference in
subsequent memory) and, consistent with an interpretation that this activity reflected at least
some aspects of encoding, showed that Dm was larger for semantic compared to orthographic
orienting conditions (see also Pallet, 1990; Friedman, 1990b). The data of Friedman (1990b)
and Pallet (1990) have been interpreted as suggesting that Dm reflects elaboration, an effortful
process conducive to the later retrieval of events.
A Subsequently R e c o g n i z e d B 6" r Subsequently Recognized

..... Subsequently Unrecognized - - - O.-- Subsequently Unrecognized
0
>4
"0
Fz -~'-,: ,v,' "','/-/,,-. . . . ~3
l,t 0 -e
Q'O' 0
E
< ,'0""...0
0--2'
eO"
IZ t t
IJJ ," 900-1200

Cz " ~ O"
-4 9 9 9
Fz Cz Pz Oz F; Cz Pz Oz
6
Pz ,- a ~ " u~
m
o
>,
_c
Oz "" ~ " .8 "

o,~
"---e
+ a'O
I 10 E
r Subsequently Recognized minus
- 0- 2, Subsequently Unrecognized
l I l l l l l l i
400 800 1200
"' -4
Fz Cz Pz & Fz Cz Pz Oz
MSEC
Electrode Location
Figure 1: A. Grand mean ERPs for the young subjects elicted by new items that were or were not subsequently
correctly recognized as old when presented for the second time. Shaded areas indicate the memory effect, and
correpsond to those portions of the waveform where the subsequentlyrecognized and unrecognized voltage
measures differed significantly. The dam are depicted at 4 midline scalp sites. B. Scalp distribution of the
subsequent memory effect computed on the waveforms depicted in panel A. Adaptedfrom Friedman (1990b),
and reprintedby permission.
An aspect of the differences in the interpretation of the encoding memory effect among
investigators stems from the fact that subsequently recalled and not recalled items may differ in
more than one ERP component. In fact, Karis et al. (1984) observed that an earlier parietally-
focused activity (interpreted as the P3b) separated the subsequently recalled and not recalled
items in rote memorizers, while a slower, frontally-oriented activity differentiated the two in
subjects using elaborative mnemonic strategies. This latter activity appears more similar to
that described by Friedman (1990b) and Paller (1990). This double dissociation was replicated
by Fabiani et al. (1990). Thus, one problem that may occur in attemmpting to include ERP
encoding memory effects under the label "Dm" is that the ERP difference may be
superimposed on different components at different latencies depending on the paradigm used
and on the processing required of the subject.
Recent experiments by Paller and his associates (Paller, 1990; Paller & Kutas, 1992)
have suggested that Dm can distinguish between implicit and explicit tests of memory, since
Dm was robust for subsequent direct (i.e., free or cued recall), but not for subsequent indirect
(word stem completion priming) performance. Moreover, in Paller's (1990) data, the scalp
distributions of the two effects differed. Since differences in scalp distribution i ~ l y
differences in the underlying neural sources, those data lend some support to the theory that
the two types of memory test might be subserved by unique neural and cognitive architectures.
In addition, since Dm is recorded during study, the ERP data suggest (as noted by Paller,
1990) that an important distinction between implicit and explicit tests of memory occurs during
encoding. Similarly, Smith (1993) recorded a robust Dm effect that did not differ as a function
of whether the correctly identified study words would be subsequently associated with
contextual (i.e., "remember") or familiarity (i.e., "know") judgements (e.g., Gardiuer & Java,
1990--see frontal lobe section above) during the recognition memory test. This suggests that
the Dm effect is dependent upon processes that occur during encoding. We will return to a
consideration of this electrical activity in our review of age-related studies below.
4.2. Retrieval
One of the most robust findings to come out of the studies that have recorded ERPs
during the retrieval phases of study-test or continuous recognition memory paradigms is that
the ERPs elicited by repetitions of previously presented items (i.e., "old"), whether in direct or
indirect tests, produce greater-magnitude positive activity than the ERPs elicited by their first
presentation counterparts. This has been labeled the "ERP repetition effect," and a typical
"old/new" effect is depicted in Figure 2. The data illustrated in Figure 2 were recorded in the
same young adult subjects during both direct (continuous recognition) and indirect (semantic
decision) tests of memory. Note that in Figure 2 the ERP repetition effect appears to span two
major peaks in the waveform, a negativity in the latency range of 250-450 ms, larger to new
items, identified with the N400 of Kutas & Hillyard (1980), and a positivity, larger to old
items, in the latency range of 450-650 ms, most likely synonymous with the P3 (or P3b)
component. Because both of these components are affected similarly by repetition (i.e.,
enhanced positivity), it was not clear in the early studies of these phenomena whether
repetition modulated both components or one component that overlapped both the N400 and
the P3b. Subsequent studies have demonstrated that each component probably has a unique
fimctional significance, since they can be differentiated on the basis of temporal delay between
study and test (Rugg & Nagy, 1989; the N400 is reduced or absent at longer delays), scalp
distribution (Berman et al., 1991; P3b has a parietal-maximum scalp distribution, whereas
N400 displays a more equipotential amplitude distribution), and word frequency (Berman &
Friedman, 1993; Rugg, 1990; Van Petten & Kutas, 1990; only the P3b appears to be
modulated by frequency). Thus, the repetition effect has been fractionated into "early" and
"late" aspects, each presumably reflecting unique aspects of the processes thought to underlie
recognition memory and/or repetition priming (see discussion below).
Also noteworthy for the young adult data depicted in Figure 2 is the comparison
between the ERPs recorded during the direct (continuous recognition) and indirect (semantic
decision) tests. The repetition effects spanning the N400 and P3b portions appear similar in
the
SEMANTIC DECISION CONTINUOUS RECOGNITION
FZ FZ
r v . . . . r ~ ' "- ..... YOUNG

cz '~ <'-" P3D
~., ~-~.~j --_ . ^ , ' ~
i -r' ~ ; ' ....
PZ
:~__ PZ ~
,', O w
Wave
_ , -.., _,
Fv- -~ - iJwv -.---
N, N 4 0 0
to~v
! . .I . . . ,,_
400 coo I;~o II~oo ,,oo Boo 1 2 o o 16oo
! I O L D
CZ "" CZ
_ J ~ ~ . . . . . . A ,"'" _ ---
i v - " -T v" --
oz /V~ , ~ ' % ~ Pz ,-.
t -T" - ~ J --"
! !
400 000 S~O0 IlSO0
Msec
F~rst Presentation
..... SeconO Presentation
Figure 2" Grand mean ERPs elicited by first and second presentation correctly identified animal and non-
animal words during a semantic decision task (adapted from Hamberger & Friedman, 1992 and reprinted by
permission), and by correctly identified first and second presentation items during a continuous recognition
memory paradigm (adapted from Friedman et al., 1993b and reprinted by permission). Both young and elderly
adult data are depicted. Arrows and associated labels denote the N400, P3b, and slowwave portions of the ERP
waveform.
two sets of waveforms. However, note that for the portion following the N400-P3b interval,
there is an old/new "crossover" for the direct, but not the indirect ERPs. During this period
for the direct test waveforms, the new ERP is greater than the old ERP beginning at about
700-800 msec post-stimulus. In the continuous recognition study, subjects were asked to
explicitly identify the old/new status of each word in the series, and were informed that an item
would not be presented more than twice. Therefore, if subjects reco~ized an item as a repeat,
they should have been able to discard it from further consideration. This "crossover" effect
spans the "slow wave" region of the waveform (see Ritter & Ruchkin, 1992 for a review).
Since slow wave' activity has been linked to "further processing," of the kind that might be
involved in encoding a new item for subsequent recognition (e.g., elaboration), the greater
slow wave to new items in the ERPs of the young adults during performance of a continuous
recognition task is consistent with this kind of mental activity. Additional evidence favoring
the interpretation that encoding is reflected in positive slow wave activity can also be seen in
the young adult data shown in Figure 2. In contrast to the instructions for continuous
recognition, during the semantic classification task, subjects were instructed to respond
differentially to non-animal and animal names, and repetition was incidental to task
performance. As can be seen in Figure 2, by contrast with continuous recognition, during the
semantic condition there is no difference between new and old items for the slow wave region
of the wavefor~ From a comparison of these data, we conclude that when encoding is
necessary in order to ensure that to-be-repeated words will be subsequently recognized as old
(i.e., during continuous recognition), enhanced slow wave activity is elicited by new items. In
contrast, during the semantic condition, when deliberate encoding activity is not likely, reduced
positive slow wave activity results. One additional point worth mentioning is that this same
inference could not have been made had we only employed reaction time measures, since these
ERP differences occurred after the generation of the reaction time response (see Table 1).
From a consideration of the reaction time latencies (Table 1) and the latency of the slow wave
"crossover" depicted in Figure 2, the continuous recognition data suggest that (young) subjects
first decided whether the item was old or new and, if new, engaged in further processing of the
item (presumably encoding activity). However, whether these processes were strictly serial or
occurred in parallel is ditticult to determine from these data.
Table 1
Comparison of RT repetition effects during direct (Friedman et al., 1993b) and indirect
memory (Hamberser & Friedman, 1992) tasks.
Age Group Semantic Decision Continuous Recognition
R1 R2 R1-R2 gl R2 R1-R2
YOUNG 709 678 31 689 777 -88
OLD 705 665 40 691 794 -103
Studies of intracranially recorded ERPs, as well as studies of scalp-recorded ERPs in

patients who have undergone unilateral temporal lobectomy lend some support to the notion
that these memory-related ERP phenomena recorded during explicit tasks are generated, at
least in part, within structures located in the medial temporal lobe. Intracranial studies by
Smith and colleagues (1986) have shown that the N400 and P3b components are generated
locally within the medial temporal lobe, suggesting that these components reflect the role of
medial temporal lobe structures in memory processing. Consistent with these intracranially
recorded data, the repetition effect elicited by words in the test phase of a study-test procedure
is attenuated in patients with left temporal, but not with fight temporal, lobectomy (Smith &
Halgren, 1989). This reduction was related to the poorer verbal memory performance of the
patients with removal of the anterior portions of the medial temporal lobe in the left
hemisphere. Similarly, Johnson (in press, 1994), also using a verbal study-test paradigm~
reported that the repetition effect was absent in the scalp-recorded ERPs of patients in whom
left temporal lobe structures had been removed. In these patients, verbal memory performance
was poor relative to patients with right temporal lobectomies. Moreover, Johnson (in press)
also found that the repetition effect was eliminated in amnestic patients. Although Rugg et al.
(1991) similarly reported a reduction in the ERP repetition effect during a continuous
recognition memory task, it was reduced in both lett and right lobectomy patients and appeared
unrelated to verbal memory performance (but see Johnson, in press, and Paller & Kutas, 1992
for counter arguments). By contrast, in an indirect memory test the repetition effect evident in
the ERP was of normal magnitude in these same patients. However, Rugg et al. did not have
a behavioral index with which to assess the relationship between the ERP phenomena and
priming facilitation during this latter task. The data of Rugg et al. suggest that during their
explicit continuous recognition memory task the ERP repetition effect is dependent upon intact
medial temporal lobe structures, whereas during their implicit task, structures outside the
medial temporal lobe give rise to these electrical phenomena. Consistent with the data from
Rugg et al., recent intracranial data recorded during a direct memory task suggest that the
hippocampus may contribute to repetition effects recorded at the scalp (Cmillem et al., 1993).
Using a visual continuous recognition memory task, similar findings were reported by Puce et
al. (1991). Taken together, the intracranial and scalp-recorded data in lobectomy patients
suggest that these ERP memory-related phenomena reflect the functional role that medial
temporal lobe structures play during direct memory tasks.
Other studies also implicate memory-related ERP phenomena as reflecting behavioral
processes thought to subserve recognition memory performance. The two behavioral
components, familiarity and context-sensitivity, thought to underlie recognition memory
performance appear to have neurophysiological manifestations, although there is some
confusion engendered by recent findings. For example, Rugg (1990) reported that low
frequency repeated words were associated with larger P3b amplitudes than high frequency
repeated words, and interpreted this as reflecting the mismatch between "local" and baseline
familiarity postulated by two-process theories of recognition. However, recent studies from
Rugg's laboratory (reviewed in Rugg, in press) have provided some support for the notion that
the late aspect of the ERP repetition effect reflects the context-sensitive, recollective
component of recognition memory. Consistent with this interpretation, Smith (1993)
demonstrated that the late aspect of the ERP repetition effect was larger in association with
"remember" (context) than "know" (familiarity)judgements. Similarly, in a preliminary report,
Wilding and Rugg (1993) reported that only those items whose source (i.e., original context)
was correctly identified elicited a robust late ERP repetition effect (for more extensive data and
discussion see Rugg, in press). Considered as a whole, the Wilding and Rugg (1993) and
Smith (1993) data suggest that the late aspect of the ERP repetition effect reflects the
recollective, rather than the familiarity-based, aspect of recognition memory (see Rugg, in
press and Johnson, in press for thorough discussions).
5. ERP Studies in Normally Aging Adults.
The breadth of parametric investigation evident in the brief review above does not exist
in the ERP aging literature, which is dearly in its infancy. Nevertheless, sutficient information
from ERP studies of both direct and indirect memory has accrued to enable at least some
general conclusions concerning the cognitive aging of memory function in the normally aging
adult. The database, although limited, includes the relationship between behavioral and ERP
measures, as well as some, albeit restricted, data on age-related changes in scalp distribution as
this affects memory function. Table 2 presents all of the studies that, to our knowledge, have
so far been published in either full or preliminary form,
5.1. ERPs Recorded during Indirect Paradigms
By far, the majority of studies, as can be seen in Table 2, have attempted to assess
memory indirectly. This enterprise is built on several assumptions, some of which are
controversial. First, it has been assumed, based on several studies in the behavioral literature
Table 2
ERP, Aging, and Memory Studies and Designs.
Study Authors Memory Paradigm Stimuli Age
9 Test Type . . . differences
Karayanadis et al., Indirect Repetition Priming (short & Words Yes
1993 long lags) during lexical
decision
Rugg et al., 1994 Indirect Repetition Priming (short & Words Yes
lon~ lags)
Friedman et al., 1993a Indirect Repetition Priming (semantic Words Yes
& orthographic decision with
short, intermediate & long
lags)
Kazmerski et al., Indirect Multiple repetition priming Pictures No
submitted
Hamberger & Indirect Repetition priming (short, Words No
Friedman, 1992 intermediate & lon~ lags) .
Fabiani et al., 1994 Direct Free Recall of names in name Words Yes
oddball (male &
female
names
Friedman et al., 1993b Direct Continuous Recognition Words Yes
(short, intermediate & long
lags)
Swick & Knight, 1994 Direct & Continuous Recognition & Words Yes
Indirect Lexical Decision
Friedman et al., Direct & Stem Cued Recall and Word Words Yes
submitted Indirect Stem Completion
showing repetition priming facilitation (i.e., reaction time and/or accuracy), that, because
attention is diverted from the memorial aspects of the task, memory is being measured
"implicitly." This assumption is questionable because explicit retrieval is always a possibility
(see discussion below). For example, in a study of normally aging controls and Alzheimer's
patients, Moscovitch (1982) showed that repetition during a lexical decision task speeded
reaction time in both the patients and the controls. Because the lexical decision task of
discriminating words t~om non-words had very little in common with the requirements of
explicit memory tasks, Moscovitch interpreted this benefit of a single previous exposure as an
"implicit" memory effect. Similarly, Ober and Shenaut (1988), in a study that again compared
Alzheimer's patients and normally aging controls, reported normal reaction time priming
facilitation with event repetition in both groups of subjects (see also Hamberger & Friedman,
1992; Kazmerski et al., submitted; described below). The facilitation in reaction time in
behavioral studies has been taken as evidence that some kind of retention has occurred, since
task performance associated with the repeated item was influenced by the previous
presentation of the identical item
A second argument mustered in favor of characterizing these indirect tasks as implicit,
has been the fact that often the same manipulation (e.g., inter-item lag) has dramatically
different effects on identically constructed indirect and direct tests. For example, in the
Moscovitch (1982) study described above, the lexical decision (indirect) and continuous
recognition (direct) memory tasks were constructed similarly using short and long inter-item
lags. During lexical decision, variations in lag had no effect on the magnitude of the reaction
time facilitation. This was not so during continuous recognition; the longer the lag the more
prolonged reaction time was relative to the item's first presentation. In similar fashion,
Friedman et al. (1993b) showed that during continuous recognition, lag prolonged reaction
time to old items, especially at the longest lag (32), whereas during identically-constructed
semantic decision blocks (Hamberger & Friedman, 1992), lag had no effect on the magnitude
of reaction time facilitation. This can be seen in Table 1; whereas repetition speeds reaction
time in the semantic decision task, it prolongs it during continuous recogniton. These data
suggest that the processes underlying the reaction time repetition effect differ for the two types
of memory test.
One additional problem that may hinder unequivocal interpretation of age-related
changes in implicit memory function is the notion that one can never be absolutely certain that
explicit retrieval strategies were not employed in the putatively implicit memory test. This
factor may be particularly acute in cognitive aging studies, since it has been reported that more
young than elderly subjects are aware of the relationship between the study and test stimuli
(i.e., these are "test aware" subjects according to the criteria of Bowers & Schacter, 1990; see,
for examples, Howard et al., 1991; and Rybash, 1994). In other words, it is possible that, if
young subjects become aware of the fact that previously viewed stems ~om the study phase (in
word stem completion priming, for example) are being presented during the test phase, they
may somehow use that information to explicitly retrieve the appropriate word completion. If
fewer older subjects are aware of this relationship (and, as a consequence, cannot use the
information explicitly), it is possible that any differences between young and old are due to
differences in explicit but not implicit retrieval, or to some mixture oft he two.
However, in the one study that specifically addressed this issue for non-associative
priming (i.e., for items with pre-existing representations in long-term semantic memory--
Rybash, 1994), test awareness did not have an impact on the magnitude of age differences--
younger adults showed larger priming effects than older adults regardless of their relative level
of"awareness." Nevertheless, when associative priming was considered (i.e., priming for new
associations), both old and young test-aware subjects showed greater priming than their test
unaware counterparts. Based on several lines of evidence, Rybash argued that test awareness
may be a consequence (i.e., the "automatic intrusion of explicit recognition during the
performance of a priming task") rather than a cause of associative priming. Thus, being test
aware may not necessarily lead to contamination by explicit strategies in a presumably implicit
test.
Indeed, Rybash (1994) offerred an intriguing hypothesis to account for his finding that
more older subjects were test unaware than younger subjects. He speculated that test
unawareness may reflect a source monitoring deficit in the elderly, consistent with the evidence
discussed above (frontal lobe section) showing that older subjects display a greater amount of
source amnesia than younger subjects (e.g., Craik et al., 1990). Rybash speculated that
although older subjects might have a feeling of familiarity about items on the priming test, they
are unable to correctly attribute the source of those feelings to items from the study phase.
Bearing in mind the caveats raised above, speeding of reaction time and/or increased
accuracy to the repeated event has been theorized to be due to the automatic "activation" of
the lexical representation of the item in memory, which temporarily increases the availability of
the word (its "perceptual fluency;" Jacoby and Dallas, 1981). Mandler (1980) and Jacoby and
Dallas (1981) have attributed behavioral repetition effects to activation, assuming that
facilitation occurs because the item's attributes are still available when it is repeated,
eliminating the need to completely reprocess it (i.e., fluent reprocessing).
Motivated by this theoretical basis, Hamberger and Friedman (1992) compared ERP
and reaction time repetition priming in young (mean age = 25), middle-aged (49) and elderly
(70) adults, in a study similar to Rugg's (1985), in which young adults served as subjects.
Rugg's (1985) study had established that the enhanced positivity to the repeated item was
associated with reaction time facilitation, and thus appeared to lend support to the notion that
aspects of the ERP repetition effect could serve as indices of indirect memory. Therefore, we
(Hamberger & Friedman, 1992) sought to replicate this phenomenon in young adult controls
and determine if a similar relationship between this ERP memory-related phenomenon and
reaction time facilitation would also be present in the oldest subjects of our investigation.
Figure 2 (semantic decision waveforms) depicts the ERP data resulting from the
semantic condition in both young and old adults, and Table 1 presents the corresponding
(semantic decision) reaction time data that resulted from the Hamberger and Friedman (1992)
study. The orthographic condition (choice uppercase/lowercase response) did not produce
robust repetition effects, consistent with other research (Rugg et al., 1988), and are not
presented here. As can be clearly seen in Figure 2, the ERPs elicited by the repeated event
begin to diverge from those elicited by their first presentation counterparts between
approximately 350 and 400 ms post-stimulus. This repetition-related positive enhancement
lasts for approximately 400-500 ms, depending upon the age group. However, the shape and
scalp distribution of the effect appear to be quite similar in the two age groups. Moreover,
reaction time facilitation associated with the ERP repetition effects also appear to be highly
similar in the two groups of subjects. We concluded that, as reflected in the conjunction of the
ERP and reaction time data, that the mechanisms underlying repetition priming were intact in
the normally aging adult, were probably highly similar to those modulating the ERPs in the
young adult, and were most likely emanating from similar neural sources in both age groups
(to the extent that the repetition effect reflected the brain's indirect detection of a previously
presented event).
One difficulty with the Hamberger and Friedman (1992) study was the fact that we
required a task-relevant, choice response on each trial, which led to the presence of P3b
components in the waveforms (as can be clearly seen in Figure 2). Because P3b has a different
scalp distribution in young and old age groups (Friedman et al., 1989; Pfefferbaum et al.,
1984), age-related changes in the scalp distribution of the ERP repetition effect would have
been ditficult to identify. In addition, because repetition affects N400 and P3b in a similar
fashion (i.e., greater positivity),it would have been difficult to determine whether repetition
modulated P3b, N400 or both. Thus, a follow-up study with young (mean age = 25.5) and
elderly (68.8) adults was designed (Friedman et at, 1993a), in which reaction time responses
were not required (see also Rugg et al., 1988). During the semantic condition, subjects
responded (via reaction time) to infrequently occurring animal names, and withheld their
responses to frequently occurring nonanimal names (some of which repeated). During the
orthographic task, subjects responded (via reaction time) to infrequently occun~g words
presented in upper case lettering, or withheld their responses to frequently occurring words
presented in lower case lettering (some of which repeated). Because the Hamberger and
Friedman (1992) and Kugg (1985) studies had established that the enhanced positivity to the
repeated item was associated with reaction time facilitation, for the Friedman et al. study
(1993a), the ERP repetition effect itself was used as the index of repetition priming.
-.,,-,.
C3 x ' ",, .."," -. C ",, , ' " ' -

f v-- I ,,'v-~'- ~,~ F" I v.~w
T3 , ' " " , . . , , . , " .... . T4 "'""",'"'"

YOUNG
}-~-v--"v . . . . . " - l- - ' ~ ~ ' ~ '
',-,:.-,. OLD
P'-I v v~.u-~-,-., f"l V"vw
r--
, I
V v v'k/~,..~
P"l - -~v
i
lO)JV
1 I 1 I
400 BOO 1200 1600 ' ' "46oe6o'12'oo'ldo'o ' ' 46o'ec~o~2'ooldoo
msec
Figure 3: Grand mean difference waveforms(secondpresentation-first presentation) averaged across subjects

in the young and old age groups. The depicted data are from the semantic condition. Arrows mark stimulus
onset, with time lines every200 msee.
Consistent with our previous study (Hamberger and Friedman, 1992) and studies from
other laboratories (e.g., Rugg et at, 1988), the orthographic condition did not yield reliable
repetition effects in either age group. The data that resulted from the semantic condition are
depicted in Figure 3. These data are presented as difference waveforms, which were computed
by subtracting the ERPs elicited by the first presentation from those elicited by the second
presentation. As can be seen, both age groups show robust repetition effects that are widely
distributed across the scalp. In fact, for the 300-600 ms portion of the waveform, the
repetition effect for both groups was statistically reliable and was similarly distributed across
the scalp. We identified this aspect of the new minus old difference waveform with the N400
component (see also Rugg et al., 1988). These data suggested to us that the processes
reflected in this portion of the ERP repetition effect were similar in both young and old, and
were most likely emanating from similar brain tissue. Although lag was manipulated in this
experiment, it had no effect on the magnitude of the repetition effect in either age group, in
keeping with previous studies of ERPs recorded during indirect memory paradigms.
YOUNG OLD
READ
r .... 1- '
Pz ~-I %~
RESPOND
;~
Cz J ' ' 7-J--
Pz
--I-" " ":, ....
-41~V
LAG 0 LAG 4
! i !
Figure 4: Grand mean difference waveforms from Karayanadis et al. (1993; first presentation-second
presentation) averaged across subjects in the young and old age groups. The figure depicts the ERP repetition
effect at the two lags in both the read and respond lexical decision conditions. Reprinted by permission.
However, one aspect of the older group's repetition effect is clearly different compared
to that of the younger group's--there is a large-amplitude, long-duration positive shift of the
waveform be~nning at approximately 600 msec and lasting until the end of the recording
epoch. This aspect of the repetition effect in older adults suggests the possl"bility of a
qualitative age-related difference which, in comparison to the previous Hamberger and
Friedman (1992) study, is a new finding due, perhaps, to reduction in overlapping P3b activity.
This long-duration activity appears to be consistent with the modulation of slow wave activity
(see Ritter & Ruchkin, 1992 for a review), rather than the N400. Recall that repetition leads
to facilitated processing of the old item, which should be reflected in a physiological measure
indicative of a reduction in processing activity (see Squire et al., 1992). However, larger slow
wave activity in the older group's ERPs elicited by the item's second presentation is consistent
with additional processing (Ruchkin and Sutton, 1983) of the repeat. Although clearly
speculative, these data may lend some support to the results of behavioral studies that have
shown reduced performance on indirect memory testing and thus may be consistent with
inefficient implicit memory performance in the elderly adult (e.g., Chiarello and Hoyer, 1988;
Davis et al., 1990; Hultsch et al., 1991).
Similar findings have been reported by Karayanadis et al. (1993), who assessed the
effect of immediate and delayed (lag = 4) repetition on ERPs and behavioral indices during two
lexical decision tasks in young, middle-aged and elderly adults. In one, the respond condition,
word/nonword choice reaction time responses were required; in the read condition, a mental
count of each class (word/nonword) of items was required. The difference waveforms (new
minus old) that resulted are illustrated in Figure 4 (note that in this figure positive is down).
As can be seen in the figure, the oldest group's repetition effect shows a prolonged negative
shift, that appears to be greater at the longer lag. Karayanadis and colleagues interpreted this
to mean that when an item is immediately repeated the contents of the first presentation are still
in working memory (highly similar to an "activation" interpretation). By contrast, when an
item is repeated after 4 intervening items activation has decayed, triggering a retrieval process
to refresh working memory. These investigators concluded that the prolonged negative
and the interaction of lag and age in modulating its amplitude indicated that aging affects
processes related to episodic retrieval and the use of contextual information in "integrating a
stimulus with its context." On the other hand, they concluded that processes related to
accessing of stored representations of words (i.e., activation) remained intact with increasing
age. However, the conclusion that, at the relatively short lag of 4 items, an episodic process is
necessary to refresh the contents of working memory, is difficult to reconcile with the
Hamberger and Friedman (1992), Friedman et al. (1993a), and Rugg et al. (1994; see below)
studies, in which lags considerably longer (32 in Friedman et al., 1993b) than those used by
Karayanadis et al. (1993) did not modulate the magnitude of the repetition effect in an age-
associated fashion. Moreover, in a study by Friedman et al. (submitted), in which an average
lag of 140 intervening items was employed, there was no age-related decrement in the
magnitude of the repetition effect.
Rugg et al. (1994) assessed repetition priming with ERPs as the dependent measure in
a design similar to that used by Friedman et al. (1993a). By contrast with Karayanadis et al.
(1993), but in keeping with Friedman et ars (1993) results, Rugg et al. (1994) reported that lag
did not interact with age to affect the size of the ERP repetition effect. However, Rugg et al.
did not report finding prolonged positive shit, s in the ERPs of their oldest subjects when
elicited by the repeated items. The reasons for this discrepancy are not clear at this time,
although the task and response requirements in the three studies were quite different (choice or
silent count, lexical decision, Karayanadis et al., go/no-go, semantic classification, Friedman et
al., and Rugg et al.), and could have produced the differences among studies.
In the only study (to our knowledge) to use non-word stimuli, Kazmerski et al.
(submitted) assessed the effect of multiple picture repetitions on the ERP and reaction time
repetition effects in young and older individuals. This study was motivated by previous
findings suggesting that multiple repetitions (i.e., more than 2) modulated the magnitude of the
ERP elicited by repeated items (Bentin et al., 1992). Bentin et al. (1992) with young adult
subjects had interpreted their data as indicating that the ERP repetition effect may reflect "the
strength of a memory trace." In the Bentin et al. studies, items were repeated three times
across three blocks of trials (the middle block was always a direct, recognition test of memory,
while the first and third blocks were semantic- or lexical-decision tasks). Kazmerski et al.
(submitted) used a similar design, with the exception that all testing was performed indirectly,
repeating pictures of common objects across three blocks of trials. In each block of trials
subjects made choice (animal/non-animal) reaction time responses. Animal concepts occurred
infrequently, and only non-animal pictures were repeated across the three blocks. The major
finding was that repetition induced a positive enhancement of the ERP in association with
reaction time facilitation that did not differ for the young and older controls. However, there
was no additional ERP enhancement induced by the third presentation (i.e., second repetition),
consistent with the results of Young and Rugg (1992) with young adults, but in contrast to the
Bentin et al findings. Moreover, both young and old controls showed highly similar scalp
distributions for the repetition effect, in keeping with the data of Rugg et al. (1994) and
Friedman et al. (1993a).
With these pictorial stimuli, there was no prolongation of the repetition effect evident in
the ERPs of the oldest subjects. As previously stated, in the studies by Bentin and colleagues,
items presented in the second block were always embedded in an explicit recognition task,
while the third block was a presumed indirect assessment of memory. The use of a recognition
memory task may have induced additional processing that would have required subjects to
access directly the representations of previously presented items. Such intentional retrieval
may have enhanced the memory trace compared to incidental repetitions of items in a
semantic-decision task (as in Young and Rugg, 1992; and Kazmerski et al., submitted). This
difference in the Bentin et al. studies makes it difficult to attribute any enhancement in the
repetition effect exclusively to the number of presentations of the stimuli. Nevertheless, the
Kazmerski et al. data extend the findings with word stimuli reviewed above to repetition
priming with pictures, and suggest that similar mechanisms underlie the generation of the
pictorial repetition effect that do not appear to differ with age.
5.2. Summary of Indirect Studies
To summarize, the results of the few studies that exist are consistent with an
identification of the early aspect of the repetition effect with the N400 component and with its
generation by similar neural sources in young and elderly samples. These data suggest that
access to representations in long-term lexical memory are intact in the older adult (based on
the assumption that N400 amplitude reflects processes associated with activation). With the
exception of the Kugg et al. (1994) study, the results of two of the three studies that used
words suggest that the later aspect of the ERP repetition effect is prolonged in older relative to
younger adults. The prolonged duration of this effect in the elderly supports a qualitative
distinction between young and old. The fact that a prolonged repetition effect was not found
with pictorial stimuli (Kazmerski et al., submitted) suggests an effect of surface form (i.e.,
pictures versus words). Pictorial stimuli are known to have sensory codes that are more
differentiating and less susceptible to interference compared to word stimuli (Nelson et al.,
1977). If we assume, as described above, that the item's attributes are activated by its first
presentation and are still available when repeated, that should have led to more efficient and/or
reduced processing of the repeating words, not "further processing" (e.g., Fabiani et al., 1990;
Friedman, 1990b; Paller et al., 1987), as reflected in prolonged slow wave activity in the
elderly. Thus, the lack of this prolongation with pictorial stimuli may indicate that the more
distinctive sensory codes attributed to pictures reduced the need for fiLrther processing activity
in the older subjects. This, however, must remain a speculative interpretation until these
pictorial and verbal effects are directly compared in the same study in a within-subjects design.
One issue, raised earlier, must be considered. It is not at all clear t~om the studies
reviewed above whether the ERP repetition effect recorded during these presumably indirect
tasks reflects "implicit" retrieval mechanisms. Thus, these data do not support a strong
conclusion that implicit retrieval mechanisms are intact in the elderly. One difficulty in
concluding that this ERP effect does reflect implicit memory mechanisms is the fact that in
none of these studies was it compared with a repetition effect recorded during a direct memory
test (but see review of Kazmerski & Friedman, submitted, below). Although there are several
experimental manipulations that have been shown in behavioral studies to dissociate direct and
indirect tests (e.g., orienting activity during study; review by Richardson-Klavehn & Bjork,
1988), this same dissociative methodology has been applied in~equently in studies of the ERP
repetition effect (see also review by Rugg, in press). Such studies should enable firmer
conclusions as to whether the mechanisms underlying the generation of this effect are
differentially modulated by the same experimental manipulations that have been shown to
dissociate performance on direct and indirect tests of memory. In addition, studies of densely
amnestic individuals, who show severe explicit memory deficits, would also be enlightening.
That is, the presence of a robust repetition effect during an indirect task in these individuals,
would be good evidence that the ERP repetition effect reflects implicit mechanisms. In this
vein, studies of Alzheimer's patients, who also display severe explicit memory deficits even at
the "mild" stage, might aid this endeavor. For example, we (Friedman et al., 1992) studied
repetition priming in an indirect task and free recall for the words presented during the priming
series in Alzheimer's patients. Although the patients produced robust ERP repetition effects,
they were severely impaired on the measure of free recall for the same stimuli initially
presented during the priming series (see also Rugg et al., 1994). On this basis, we argued that,
to the extent that the ERP repetition effect reflects the brain's "indirect" identification of
previously presented information, then this putative implicit memory function was intact in
patients in the early stages of Alzheimer's disease. Because of their extremely poor free recall
performance, those data suggested to us that it would not have been likely for the patients to
have used explicit retrieval strategies in performing during the priming series, making it
unlikely that such strategies would have played a role in the generation of the ERP repetition
effect. However, as stated above, firmer conclusions await direct comparison of the repetition
effect during the two kinds of memory tests.
Scalp distribution evidence has not been used extensively in an attempt to distinguish
between implicit and explicit effects on the ERP repetition effect. Recently, however,
Kazmerski & Friedman (submitted) compared the scalp distribution of the ERP repetition
effect for within-surface form (i.e., pictures versus words) study-test pairings when elicited
during direct (yes/no recognition) and indirect (semantic decision) memory tests. For both
word-word and picture-picture study-test combinations, the indirect ERP measure was larger
than the direct measure over posterior portions of the scalp. Because fluent reprocessing (see
above) is presumed to play more of a role in indirect than direct tests, we offerred an
interpretation consistent with some data based on imaging protocols that the different scalp
topography over temporal and occipital scalp sites elicited during the indirect, compared to the
direct, test was suggestive of the involvement of cortical word form areas during the indirect
test for the word-word study-test combination (e.g., Petersen et al., 1989; Squire et al., 1992).
Studies of the neural loci of object priming, however, are lacking, although there is some
evidence that posterior cortical areas may be involved in the visual identification of common
everyday objects (reviewed by Tulving & Schacter, 1990). This would be consistent with the
scalp distribution of the repetition effect observed in the indirect picture-picture study-test
pairing. These scalp distribution data suggest that the neural loci of the scalp recorded data in
the two types of test were different and, hence, that the processes engaged were also different.
This type of data, therefore, can be used as convergent measures in an attempt to dissociate the
two types of memory test.
5.3. ERPs Recorded during Direct Paradigms
Very few studies of direct memory, ERPs and aging exist. In one of the first, Friedman
et al. (1993b) recorded ERPs and reaction times during continuous recogniton memory in
young, middle-aged and older adults. This investigation was motivated by findings in the
behavioral literature that had shown older adults to be at a disadvantage when items had to be
stored in and retrieved from secondary or long-term memory (e.g., P o o n & Fozard, 1980;
review of behavioral studies by Poon, 1980). In this study, we employed the continuous
recognition memory paradigm with verbal stimuli. In this procedure, items are presented
"continuously" and, unlike the "study-test" paradigm, "new" items (i.e., first presentation)
become "old" (i.e., second presentation) within the same stimulus series. The continuous
recognition memory paradigm (Shepherd & Teghtsoonian, 1961) thus allows the investigator
to systematically vary the lag between first and second presentations, enabling ERPs and
behavioral data to be examined continuously from the registration of a verbal item to its
recognition from both within, and outside the immediate memory span. In one of the first
behaviorally-based aging studies to employ this task, Pooh and Fozard (1980) reported that the
performance of older adults was similar to that of younger adults at the shortest lags (those
assumed to reflect storage in primary memory), whereas at longer lags (secondary memory)
the performance of older adults was deficient compared to that of younger adults. Thus, we
were considerably interested in assessing the effect of lag on the relationship between ERPs
and behavior in order to determine if manipulating lag would have more of an effect on the
reaction time and ERP indices of the older adults.
For this investigation, we compared young (mean age = 25), middle-aged (49) and
elderly (70) adults. We replicated the interaction between age and lag (2, 8, and 32 intervening
items) previously reported by Poon and Fozard (1980)--the older adults showed poorer
performance at the longer relative to the shorter lags. However, the ERP data did not mirror
this interaction, as there were no age by lag interactions for any aspect of the ERP waveform
The ERP waveforms elicited during continuous recognition (averaged across lags), along with
the repetition data from the previously described semantic decision task, are depicted in Figure
2. Two aspects of the continuous recognition waveforms are particularly noteworthy. For the
portion of the repetition effect that spans the N400 and P3b latency regions (from
approximately 300 - 700 msec), the young and the old show highly similar differences between
the first and second presentation ERPs. However, as previously mentioned, for the slow wave
portion of the response, the young display an old/new "crossover," in which the first
presentation of a stimulus elicits greater ERP amplitude than the second presntation. This
effect is not present, however, in the ERPs of the older adults. As previously described, slow
wave activity has been associated with "further processing" of the kind that might be involved
in encoding a new item for subsequent recognition. Thus, the larger slow wave to new items
in the young adult ERPs is consistent with this kind of encoding activity. The lack of slow
wave activity to new items in the older subjects' waveforms is consistent with a large body of
behavioral data that suggests that older, unlike younger adults, do not spontaneously engage in
elaboration when memorizing word lists for subsequent recall or recognition (e.g., Rankin &
Collins, 1985; Hashtroudi et al., 1989). These slow wave data implicate an age-related
difference in encoding as responsible for the poorer performance of the older adults during this
task (sensitivity values collapsed across lags, respectively for young and old of 5.1 and 4.6).
By contrast with the slow wave data, the N400 and P3b data suggest that older relative to
younger adults engage similar memory mechanisms prior to the recognition decision (as
reflected by mean RT), and it is only subsequent to that derision (i.e., within the latency
window of slow wave activity) that age-related differences emerge.
Because of the differences in slow wave activity and its putatiuve association with
encoding activity, the continuous recognition study was followed by a study (behavioral data in
Friedman et al., in press; ERP data in Friedman et al., submitted), in which encoding activity
was manipulated (either semantic or non-semantic). For this investigation, we were primarily
interested in the subsequent memory effect or Dm since, as discussed earlier in this chapter, it
appears to reflect the kind of encoding activity subjects engage in during study (Paller et al.,
1987). As also previously discussed, Dm may not be a unitary phenomenon (see section on
encoding above). This label has been used to denote an operational definition--the difference
between ERP activity elicited by items that are subsequently correctly recalled, reco~ized, or
primed and ERP activity elicited by those items that are not subsequently recalled, recognized
or primed. The difference presumably reflects activity (e.g., encoding) that led to successful
subsequent memory performance. Because older relative to younger adults are thought to be
deficient in elaborative encoding, the subsequent memory effect or Dm was assessed to
determine if it would shed any light on the age-related performance differences we expected to
obtain. The experimental design included a direct comparison between direct and indirect tests
of memory. During the study phase, subjects saw verbal items during two encoding tasks, a
structural task in which they were required to detect words whose first and last letters were in
exact alphabetic sequence, and a semantic task, in which they detected words that denoted
animals. Semantic versus non-semantic encoding was also manipulated because of the well
known dissociations between direct and indirect memory tests induced by this levels of
processing manipulation. Target items (to which subjects responded via button press)
occurred infrequently (20 percent probability), and only nontargets (80 percent probability),
which did not require a response, were tested during the subsequent memory test phase.
Eighty subjects (40 young, mean age - 26, and 40 older, mean age -- 70) completed the
study. Subjects within each age group were alternately assigned to either the direct or indirect
test. The direct test was stem cued recall and the indirect test was stem completion. In both,
previously studied "old" 3-letter stems were presented along with an equal number of foil
stems that had not been previously studied. The only difference between the tests was the
instructions. During word stem completion subjects were instructed to complete the stems
with the first word that came to mind. For stem cued recall, subjects were instructed that some
of the stems had formed parts of words that they had seen during the study phase, and that
they were to say the whole word only if they remembered seeing it during the study phase.
Thus, the only difference between the two test versions was that for stem completion the
subject's attention was directed away from conscious recollection, while for cued recall
conscious recollection was required.
.25
1 Young
jm j/jm
--ram FA Old
,,-,
t,,,)
.20
L_
0
o .15
t-
O
,1
1=
0 9
10
L0 .
a...05
Semantic Structural Semantic Structural

STEM COMPLETION CUED RECALL
Figure 5: Grand mean proportion of correctly completedstems during the indirect test (word stem completion)
and correct recalls during the direct test (cued recall) as a function of encoding task at study (semantic;
alphabetic or structural) for young and old subjects. From Friedman et al. (in press) and reprintedby
permission.,
During the study phase, old and young adults performed equivalently, thus ensuring
that subsequent age-related memory performance differences would not be due to differences
in the way in which the stimuli were processed at study. Figure 5 presents the behavioral data
recorded during the memory test phase. A few major findings are evident from the data
depicted in Figure 5. First, levels of processing (semantic vs structural) had a dramatic effect
on both indirect and direct memory performance, and was evident for both age groups.
However, the levels of processing effect was larger for~ cued recall than for stem completion
(interaction of study encoding task and memory test). Second, there appear to be small age
differences on both the direct and indirect tests (these, however, were not statistically reliable).
YOUNG OLD
Semantic Structural Semantic Structural
z
Fz
F-
_J Cz
Pz
u.l
Oz
s
Fz
_J
0 Cz
iii
cr
Pz
(D
Oz
I
i,U,l,i,i,nq',l,i, l'J'i'i'l'i'l'i"l' ~ -Trff'PTq'r~Tr~T~
~ C'0012~)1600 40080012001600 4008001200tf~]0
msec msec msec msec
+
........
Subsequently
Subsequently
correct
incorrect
I 10 l.tV
Figure 6: Grand mean E R P waveforms averaged across subjects within each age group and type of memory
test. The illustrated data were elicited at the midline electrode sites during both encoding tasks as a function of
subsequent stem completion and cued recall performance. Shaded areas indicate the subsequent memory effect.
Adapted from Friedman et al. (submitted) and reprinted by permission. Arrows m a r k stimulus onset with time
lines every 200 msec.
.. ,-.
Third, subjects in both age groups completed more stems under the indirect (stem completion)
memory instructions than under the direct (cued recall) memory instructions (which was also a
statistically significant finding). In showing a depth of processing effect on cued recall
performance, these data are consistent with a large array of experimental investigations in both
young (reviewed by Richardson-Klavehn & Bjork, 1988) and elderly (review by Light, 1991)
adults. In showing an effect of depth of processing on performance on the indirect task, they
run counter to the majority of such studies. However, recent investigations have begun to
question this pattern as either trends or significant effects of levels of processing on indirect
test performance have been obtained in several instances (reviews by Chiarello & Hoyer, 1988;
and Challis & Brodbeck, 1992).
Although, based on the large majority of studies of the cognitive aging of indirect
memory, we expected a lack of age-related performance differences, the small age-related
difference on cued recall testing was unexpected. Because the study used long lists of "old"
and foil stems (N=224 old, 224 new or foil stems presented at test) and only nontargets from
the study phase were tested during the memory phase, we concluded that an explicit retrieval
strategy would have been inefficient for good performance during the cued recall test. This led
subjects to adopt an implicit retrieval strategy. This stategy would have involved a search of
semantic memory for a likely completion and the testing of that completion via yes/no
recognition to determine if it had been previously seen during the study phase. A generate +
reco~ize model (Anderson & Bower, 1974; Kintsch, 1974; see also Jacoby & Hollingshead,
1990) was applied to the data in support of this interpretation. According to this model,
subjects first generated a likely candidate word in response to the three-letter stem, in much the
same way they would during implicit retrieval for a word stem completion implicit memory
test. This was followed by a recognition check to determine if the item was "old" or "new."
Since older subjects have been reported to perform as well as younger subjects in the majority
of studies of implicit memory (review by Light, 1991; but see Chiarello & Hoyer, 1988; and
Davis & Bernstein, 1992 for a different conclusion), and do not always show deficits when
tested in a recognition memory format (e.g., Craik & McDowd, 1987), the data and the model
were consistent with the lack of si~ificant age differences on the stem cued recall test.
Nonetheless, as previously noted, for both word stem completion and cued recall, there
were trends in the data for the young to perform better than the old. As can be seen in Figure
5, these age-related differences appear larger for items that had been studied under semantic
compared to structural encoding conditions.
Figure 6 presents the waveforms recorded at study averaged as a function of
subsequent direct and indirect memory test performance. As can be seen in the figure, the
young adult waveforms display robust subsequent memory effects (shaded areas in the figure),
whereas the older adult waveforms do not. We (Friedman et al., submitted) concluded that
these age-related differences in the subsequent memory or Dm effect may have reflected the
small age-related performance differences depicted in Figure 5. To the extent that this
subsequent memory effect reflects differences in encoding at the time of study, the data shown
in Figure 6 suggest the possibility that for the oldest subjects, such differences in electrical
activity are not related to subsequent memory performance for either the direct or indirect
memory tests. These candidate processes could include activation and/or elaboration (cf, Graf
and Mandler, 1984). Moreover, if the onset and duration of encoding processes are prolonged
in older subjects (as the results of some behavioral studies suggest, e.g., Howard et al., 1986),
then those processes would be temporally out of phase with the occurrence of Dm (Paller,
1990), and could have led to the amplitude reduction seen here in the oldest subjects.
Salthouse (1988) has suggested that slower speed of activation might result in the excitation of
fewer nodes in semantic memory, resulting in less "elaborated" memory traces for the elderly.
Thus, prolongation of activation and/or elaboration might be one mechanism accounting for
the lack of a clear Dm in the elderly for both stem completion and cued recall, and could have
led to the small-magnitude performance decrement we have observed on both types of test in
these subjects.
Figure 7A presents the preliminary results of a recently completed study conducted by

Fabiani et al. (1994). After participating in an experiment, in which visually presented male
and female names served as stimuli (50/50 probabilities; no name repeated), young and old
subjects were asked to freely recall as many of the previously seen names as possible. Sixteen
young (ago range of 22-29) and 17 old (65-88) subjects saw 50 male and 50 female names
(presented in one block of trials; randomly intermixed), and made choice male/female button
press responses to each name. ERPs were recorded from 30 scalp sites (for complete details
of EEG recording, see Fabiani & Friedman, submitted). At the end of the oddball task,
subjects were given a short break, and then were unexpectedly asked to write down as many of
the names as they could remember in any order they wished. The young subjects recalled more
names (18%) than the old subjects (15%). As can be seen in Figure 7A, when the ERPs
elicited by the names during the oddball task were averaged as a function of subsequent flee
recall memory performance, both groups showed a slightly enhanced positivity to the recalled
words. However, the scalp distribution of this memory effect appears to differ for the two age
groups. It is prominent at the midline frontal electrode site for the young, while it appears to
be smaller but equipotential across the midline sites for the older subjects.
YOUNG OLD MEMORY EFFECT
F-"VA~w-;~', FV
CZ l ...
cz /Din
.... :^ .... ,,.. ........
+
,-v -~v J-~ rw i-*

l l I
9 " "' 160 - l~O liO0- 16"00 + foe 000 l~O0 1600 "" - ,,ilOO ~ll - 12"00-'ll~i -
msec
Young
Subsequently Recalled
s not recalled ..... Old
Figure 7A: Grand mean midline ERPs in the left portion of the figure show the waveforms (averaged across
male and female names) computed as a function of subsequent recall performance. The ERPs depicted in the
fight portion are the difference waveforms (subsequently recalled- subsequently unrecalled), computed
separately for each age group. Arrows mark stimulus onset with timelines every 200 msec.
Figure 7B. Surface potential (SP) and current source density (CSD) maps shown separately for the
unsubtracted (left side of figure, subsequently recalled, subsequently unrecalled) waveforms and their difference
(fight side of figure). The maps were computed at the point in time when the positive difference between the
subsequently recalled and not recalled ERPs was maximal. For the maps based on the unsubtracted data, the
isopotential lines for the surface potential (SP) are separated by 1 microvolt for both young and old, and .183
mA/M 3 for the current source density (CSD) maps, again for both young and old. For the maps based on the
young difference waveforms, the isopotential lines are separated by .5 microvolts for the surface potential and
.092 mA/M 3, for the maps based on the difference waveforms, the isopotential lines are separated by .25
microvolts for the surface potential and .045 mA/M 3 for the current source density. Unshaded areas indicate
positive values (sources for CSD), shaded areas indicate negative values (sinks for CSD).
The rightmost portion of Figure 7B depicts the surface potential and current source
density maps computed for the subsequent memory effect using the difference waveforms
(subsequently recalled - subsequently unrecalled) recorded from all 30 electrode sites. The
surface potential maps are computed on the raw waveforms and reflect the distribution of ERP
amplitudes at a given point in time at all 30 electrode sites. Current source density is an
akemative method for analyzing scalp distributions. In contrast to the surface potential, which
reflects both superficial cortical as well as electrical activity generated at deeper levels within
the brain, the current source density reflects primarily activity generated in superficial cortical
tissue. However, neither technique in isolation can be used to determine the intracranial source
of the activity recorded at the scalp surface. As can be seen in Figure 7B, the current source
density maps show more focal activity compared to the maps of the surface potential (for
complete details and computations see Perrin et al., 1989; for a similar application, see
Friedman et al., 1993c; and Fabiani & Friedman, submitted). For both age groups, the maps
based on the difference waveforms (under the column headed "memory effect") appear to
show activity that is somewhat lateralized to scalp sites overlying the left hemisphere
(consistent with previous studies; see Johnson, in press). For the young, however, this activity
appears to be frontally focused, whereas for the old the activity shows loci overlying left
temporal scalp. As discussed earlier in the section on encoding, the memory effect shows a
scalp distribution quite different than the P3b with which it overlaps (Figure 7B). This is true
whether one compares the maps based on the difference waveforms with the maps based on
the unsubtracted "base" waveform (i.e., subsequently unrecalled; which is pre~mably less
overlapped by processes reflected in the memory effect) or with the maps computed on the
subsequently recalled waveforms (which reflect the superposition of processes involved in the
P3b as well as those involved in generating the memory effect; see Johnson, 1993 for a
complete discussion). These data imply that the two components (i.e., P3b and the memory
effect) reflect qualitatively different information processing mechanisms.
Although preliminary and requiring interpretive caution, the data depicted in Figure 7A
and B may be consistent with recent studies of cerebral blood flow recorded during the
encoding phases of direct memory paradigms using the PET technique (Kapur et al., 1994;
ShaUice et al., 1994; Tulving et al., 1994). These three PET studies all reported left prefrontal
activation in response to semantic encoding during the acquisition phases of either recall or
yes/no recognition paradi,~ns. Since the current source density reflects primarily activity
generated in superficial cortical layers (Perrin et al., 1989), the current source density maps
(for the young adults) depicted in Figure 7B are consistent with a contribution to the memory
effect from left frontal cortex. These mapping data could either be consistent with different
strategic processing on the part of young (semantic or elaborative activity) and old (non-
elaborative activity) adults, and/or deficient frontal lobe processing on the part of the older
adults. The latter interpretation would be in keeping with a large body of cognitive aging
evidence suggestive of frontal lobe deficits in older adults.
Swick and Knight (1994) also compared directly behavioral performance and ERP
indices elicited during indirect and direct tests of memory with young (mean age = 23) and
older (68) subjects. The direct test was continuous recognition memory and the indirect test
was lexical decision. Both tests were identically constructed, and both included words and
nonwords with lags of 0, 1-3, and 9-19 intervening items. By contrast with Friedman et al.'s
(1993b) study of continuous recognition, all items in the Swick and Knight study repeated, but
no more than once in each block. Because all words and nonwords repeated, this design may
have induced greater expectancy than those in which some items are seen only once and some
are never repeated (Friedman et al., 1993b; Rugg & Nagy, 1989).
During lexical decision, the old and young showed equivalent reaction time facilitation
at all lags when elicited by words. Although the old and young displayed equivalent ERP
repetition effects (reduced N400 and enhanced P3b to the repeat), for the old, onset was
delayed and duration was prolonged relative to the young. For words presented during
continuous recognition, young subjects produced faster responses to repeated items at the two
shortest lags, but not at the longest lag. By contrast, older subjects showed facilitation at the
shortest lag, no difference at the intermediate lag, and prolonged reaction times at the longest
lag. This pattern of reaction time priming was mirrored in the ERP data--whereas young
subjects showed reliable ERP repetition effects at all lags, older subjects showed these effects
reliably only at the two shorter lags, while they showed a reduction in amplitude at the longest
lag. In comparing their ERP repetition effects elicited during lexical decision and continuous
recognition, Swick and Knight noted that the young produced greater amplitude for
continuous recognition than lexical decision, whereas the opposite pattern held for the older
age group. Based on Smith's (1993) "know" and "remember" results (described above), they
interpreted this age-associated pattern as reflecting greater reliance by the young on
recollective processes during continuous recognition and less reliance on these processes by
their older subjects. This interpretation has intuitive appeal, as older subjects are known to
produce a greater number of recognition judgements based on familiarity than on context
(Parkin & Walter, 1992). However, as these investigators did not partition the recognition
judgements, it is by no means clear that a greater magnitude repetition effect necessarily
implies an underlying association with the recollective component of recognition memory.
Moreover, although this hypothesis is intriguing, the data are difficult to reconcile with the
notion that one of the mechanisms thought to underlie repetition priming is familiarity (e.g.,
Gardiner & Java, 1990), a process associated with "know" judgements during recognition
memory tasks (see earlier discussion above). Thus, to the extent that the older adult's
repetition effect during recognition memory depends more on familiarity than on context, one
would have expected equal-amplitude lexical decision and recognition repetition effects for the
older adults.
5.4. Summary of Direct Memory Paradigms
To summarize, the continuous recognition results (Friedman et al., 1993b), the

comparison of stem completion with cued recall (Friedman et al., submitted), and the name
oddball free recall data (Fabiani et al., 1994) suggest that encoding deficits may be one aspect
of the older adult's memory functioning that is responsible for producing age-related
performance deficits on both direct as well as indirect testing. With respect to mechanisms
underlying the recognition decision, the continuous recognition ERP data from Friedman et al.
suggest similarities between young and older adults. However, because of the fact that
experimental variables were not manipulated that could have influenced the degree to which
recognition judgements were made on the basis of familiarity or context, it is difficult to ascribe
the age similarity to one or the other behavioral component (or both). Swick and Knight
(1994) have suggested that the larger repetition effects they recorded (in young adults) during
continuous recognition compared to lexical decision reflect the greater contribution of
(conscious) recollection in the former. This conclusion is premature, however, without the
recording of the appropriate behavioral judgements.
6. ASSESSMENTS OF FRONTAL LOBE FUNCTION IN THE ELDERLY
Evidence we reviewed earlier from a growing literature spanning a number of fields

points to changes in frontal lobe function as a possible factor in explaining memory deficits in
older individuals. With the exception of neuropathological data, the vast majority of this
evidence is indirect. Moreover, at this stage of our knowledge, the impact of the reported
neuropathological changes on the cognitive functions thought to be subserved by the frontal
lobes is unclear. Nevertheless, based on a kind of "triangulation," in which the evidence from
several domains of research are integrated, it should be possible to assess, albeit indirectly, the
nature of the changes in frontal lobe functioning and their effect on memory function.
In our attempt to "probe" frontal lobe function and understand how frontal lobe
function might alter aspects of memory performance, we have studied the P3 component
elicited by novel task irrelevant stimuli. This component of the ERP waveform was originally
reported by Courchesne et al. (1975), who modified the standard oddball paradigm by
including, in addition to visual target (requiring a response) and standard stimuli, highly
complex, task irrelevant, visual stimuli that were difficult to label. The major finding was that
the uninstructed, novel stimuli elicited a frontally-oriented P3 component, whereas the targets
(which were equally infrequent) elicited a parietal-maximum, P3b component. This was one of
the first reports of "P3" components with different scalp distributions, implying their
generation by different brain tissue, and implicating distinct underlying cognitive functions for
each of these late positivities. For example, the novelty P3 has been interpreted as reflecting
"orienting" to novel, initially uncategorized stimuli, presumably a frontal lobe function (Luria,
1973). The fact that it was largest at scalp sites overlying frontal cortex lent some, albeit
severely limited, support to this notion. However, greater weight was given to this
interpretation by Knight's (1984) finding that patients with left and fight dorsolateral frontal
lobe lesions did not show the differentiation in scalp distribution between target and novelty P3
components shown by the young controls in Courchesne et al.'s (1975) study and by the
controls in Knight's (1984) investigation. In patients with dorsolateral frontal lobe lesions,
both stimuli elicited parietal-maximum P3 scalp distributions. This led Knight (1984) to
conclude that the dorsolateral prefrontal area is either required for the modulation of the
novelty P3 or is a generator of this electrical activity (see Knight 1990 for a review and
interpretation). Based on the conjunction of these ERP and lesion location findings, we
(Fabiani & Friedman, submitted; Friedman & Simpson, 1994; Friedman et al., 1993c) reasoned
that the novelty P3 component could be used as a "probe" of frontal lobe function in older
samples.
In the first of a series of three experiments, we (Friedman et al., 1993c)used 48 unique
auditory novel stimuli, none of which was repeated (12 percent probability; consisting of
animal calls, environmental sounds, bird calls; synthesized sounds). Older individuals are
reported to be less able to inhibit responses to task irrelevant stimuli (e.g., Rabbitt, 1965), a
function that also appears to depend upon the frontal lobes (Stuss et al., 1982; Woods &
Knight, 1986). Supporting evidence for this deficit comes from a variety of experimental
paradigms, including negative priming (Tipper, 1991) and indirect memory tasks (Hartmann &
Hasher, 1991). Thus, we reasoned that including unique, non-repeating task-irrelevant events
should increase the likelihood of finding an increase in responding to task-irrelevant events in
the oldest subjects. The novel stimuli were embedded in a series of background, frequent (76
percent probability) and target (12 percent probability) events. Subjects were asked to respond
only to the tonal targets with a speeded, button-press response. There were high- (1000z Hz)
and low-pitched (750 Hz) tones. The ERP data from the young (mean age -- 24) and older
(70) age groups are illustrated in Figure 8. Marked differences are evident between the age
groups in the scalp distributions of both the target and novelty P3s. For example, for the
young the P3 to a novel stimulus shows a fronto-central scalp distribution, whereas the P3 to a
target shows a distinct parietal-maximum distribution. By contrast, for the old, P3s to both
stimuli are characterized by large amplitudes at the frontal electrode sites. When these data
were subjected to normalization in order to eliminate amplitude differences so that only the
shape of the distributions would be statistically assessed (McCarthy & Wood, 1985), the scalp
distributions of the two P3s differed reliably for the young, but showed less differentiation for
the oldest subjects.
Consistent with our prediction, the older subjects showed a siL-,nificantly elevated false
alarm rate to the novel events. We suggested that the age-related differences in scalp
distribution in association with the age discrepancy in the false alarm rate were compatible with
TARGET
F3 ....pk,, ,t - _.
c3 ~ cz ,,~ "C4: V ~
~--.~,
v "~"/ ...................' " -v~,/
.,: - ........-,..,
-'
v - -
P3 PzJf-----~ P4
F?.v/%~ NOVEL yS.vA..c_ ~
/'~'"~ Novelty
-V./ -V.J ....... ~.." .......
Young
............. o,o T ._~V
t t ~-" t
I1'1'1'1' I1'1'1'I' I1'1'1'1 ~
200 400 600 800 200 400 600 800 200 400 600 800
msec msec msec
Figure 8: Grand mean superimposed target and novel ERPs averaged across subjects within each age group.
The epoch shown consists of 100 msec pre- and 900 msec post-stimulus periods. Arrows mark stimulus onset,
with time lines every 100 msec. The data are modified from Friedman et al. (1993c) and are reprinted by
permission.
a change in frontal lobe activity with increasing age. Specifically, the increased false alarm rate
in the elderly may have been consistent with those studies (briefly reviewed above) reporting
decreased ability with age to inhibit the processing of task-irrelevant stimuli. An alternative,
but compatible interpretation of these data, in terms of a working memory deficit is also
possible. This interpretation is based on the assumption that older adults have a working
memory deficit. Because o f this deficit, elderly individuals require increased frontal activity
(and increased effort) because their memory templates decay more rapidly. These kinds of
data may have implications for memory deficits in the elderly, since it is known that the frontal
lobes are involved in the maintenance of working memory templates (Goldman-Rakic, 1987b;
1992). Moreover, a working memory deficit can account for a variety of impairments
associated with frontal lesions, such as problems with control functions (e.g., implementation
of encoding and retrieval strategies--Moscovitch & Winocur, 1992).
Figure 9: Scalp potential (SP) and current source density (CSD) maps for the P3 components elicited by the
novel stimuli in the young and old age groups. The data are depicted as a function of stimulus number (1-6)
within the block. The isopotential lines for the surface potential maps are separated by I uV; for current source
density, they are separated by 0.183 mA/Ma. Unshaded areas indicate positive values (sources for CSD); shaded
areas indicate negative values (sinks for CSD). Numbers at the head of each column indicate the time point (in
ms) at which the maps were computed. Adapted from Friedman and Simpson (in press) and reprinted by
permission.
Further evidence for these assertions was provided by data from the second of these
studies (Friedman & Simpson, in press). In that study, we recruited additional subjects and
averaged the novel and target ERP data of Friedman et al. (1993c) according to their
numerical sequence within the block of trials (there were 6 novels and 6 targets per block). In
previous studies with young adults, Courchesne (1978) had shown that the P3 to novel events
elicited a scalp distribution that was initially frontally oriented, which then shifted to a more
posterior distribution as more novel events were presented. Those data suggest that, as
subjects gained more experience or familiarity with the novel events, they became capable of
categorizing them into a distinct class of events, and this was manmifested in the posterior
amplitude focus of the novelty P3 (i.e., a scalp dsitribution more typical of P3b-- Squires et al.,
1975). If the novelty P3 does, in fact, reflect some aspects of prefrontal processing (such as
orienting, e.g., Luria, 1973, and/or the maintenance of working memory templates--e.g.,
Goldman-Rakic, 1987b; 1992), then we expected that the shift in topography of the novelty P3
with experience, seen for the young adults would take longer to occur for the older adults.
The data that resulted are illustrated in Figure 9. As is apparent, the young show a
centrally-oriented scalp distribution to the first novel event, but a more posterior distribution to
succeeding novel events, a finding that was statistically robust. By contrast, the older adult
maps do not show a distributional shift with event recurrence. Rather, the topographies for the
older subjects are characterized by relatively greater contributions from frontal scalp
throughout the stimulus series. We interpreted this finding to mean that positive frontal
activity may reflect prefrontal processing of new events (i.e. the first presentation of target;
see discussion of Fabiani & Friedman, submitted, below) or "novel" stimuli. One possible
explanation of why the positive activity over frontal scalp diminishes with time in the young is
that they categorize the rare stimuli (targets and "novels") into distinct classes, such that
subsequent presentation of stimuli within these classes are no longer treated as "new" events.
This could involve inhibition of presumed prefrontal processes that would normally occur to
new events (e.g., orienting), with a consequent reduced need to keep the characteristics of
these items in working memory. By this reasoning, the older adults display altered inhibition of
these prefrontal processes.
Unfortunately, in the study by Friedman and Simpson, ERPs were not recorded during
the practice block, which was a standard oddball task without novels. Thus, we were not able
to examine the ERPs to the first target stimulus. However, ERPs have been recorded during
the practice condition of a recently completed study (Fabiani & Friedman, submitted) and
those data indicate that the first presentation of a target also elicits frontal positive activity in
all age groups. These data are illustrated in figure 10A and B. Visible in figure 10A is the fact
that, for the young, the targets during the practice block elict a scalp distribution for the P3
that appears remarkably similar to that elicited by novels (N=48 unique and non-repeated
environmental sounds) during novelty oddball blocks (see rightmost column of Figure 10A).
Note that, as time on task increases, the target P3 distribution becomes increasingly parietally
focused, with a relative reduction in P3 activity over frontal scalp. For the young, these
phenomena are also visible in the maps depicted in Figure 10B, in which it can be seen that,
relative to the P3 elicited at the posterior site, the frontally-oriented positive activity decreases
in magnitude from practice to novelty oddball blocks. For the old, by contrast, the target P3
distribution does not appear to change as a function of time on task. In fact the three target P3
distributions depicted at the bottom of Figure 10B appear remarkably similar to the distribution
elicited by the novelty P3. In other words, the frontal aspect of the distribution does not
diminish for the old as it does for the young.
In a study ofnonmotor learning using PET imaging of blood flow, Raichle et al. (1994)
reported the elicitation of prefrontal activation when subjects were naive with respect to task
requirements. This frontal area of activation diminished dramatically after 15 minutes of
practice. Although the Raichle et al. task and requirements were quite a bit different than the
target detection task we employed, the underlying principle, i.e., during the initial phases of
acquisition the prefrontal areas of the brain are utilized to monitor stimulus characteristics
important for task performance, would be common to both regardless of task. Once these
characteristics are learned, however, this cortical area would not play as critical a role. This
interpretation is consistent with the finding that for the older adults in the novelty oddball
investigations, frontally-oriented ERP activity did not diminish with stimulus number
(Friedman & Simpson, in press) or time on task (Friedman & Fabiani, submitted), suggesting
that the older subjects continued to utilize prefrontal cortical areas.
TARGETS NOVELS
Practice Standard Novelty
Block Oddball Blocks Oddball Blocks
YOUNG
- /~ ~ Fz
§ 10
OLD I
.... A A
- - ~ Cz
t t t t
' I ''1 ' I ' I ' " ' I ' I'' I ' i ' ' I ' I ' I ' I ' ' I ' I ' 1" I ~
200 400 600 800 200 400 600 800 200 400 600 800 200 400 600 800
msec msec msec msec
Figure 10A. Grand mean target waveformsat three midline scalp sites for the practice, standard oddball, and
novelty oddball blocks. For comparisonpurposes the ERPs elicted by the novel stimuli are depicted in the right
most column. Arrows mark stimulus onset, with time lines every 100 msec.
After an approximate 20 minute delay following the oddball task, the subjects of the
Fabiani and Friedman investigation participated in a yes/no recognition memory task for the
previously presented novel environmental sounds, randomly presented along with an equal
number o f f oil sounds. As expected the young performed better than the old (sensitivity, d, an
analog of d prime; a measure of memory strength; for young = 1.39, SD = 0.68; for old = 0.28,
SD = 0.30), and this was not due to an age-related difference in response bias. We (Fabiani &
Friedman, submitted) concluded from the conjunction of the scalp distribution and recognition
memory data that the processing of novel information involves an organized set of processes,
whose pivotal aspect may be the formation of memory templates for novel items. This process
Figure 10B. Surface potential (SP) and current source density (CSD) maps for the target P3 elicited during
practice, standard oddball and novelty oddball blocks for young and old subjects. For comparison purposes the
scalp distributions for the novelty P3 are depicted in the rightmost column. The isopotential lines for surface
potential maps are separated by 2 ~tV; for current source density, they are separated by .366 mA/m3. Unshaded
areas indicate positive values (sources for CSD); shaded areas indicate negative values (sinks for CSD).
Modified from Fabiani and Friedman (submitted), and reprinted by permission.
takes time to develop and, in young subjects, is complete once the stimuli are repeated a few
times. The early presence of frontal positivity and its diminution over time suggests that the
frontal lobes may be a generator of this activity and thus may be involved in this kind of
processing. However, in older individuals the formation and/or maintenance of these memory
templates is disrupted, and the process continues for a much longer time. As discussed earlier,
these data may have an important application to the cognitive aging o f memory function,
linking the aging of the frontal lobes with an inability to form or maintain adequate templates
(see also Salthouse, 1990), which may be a critical aspect of the memory impairment
associated with aging.
7. S U M M A R Y , C O N C L U S I O N S AND D I R E C T I O N S F O R F U T U R E R E S E A R C H
Clearly, the cognitive aging of memory as assessed by ERP measures, is in its infancy.
Not surprisingly, there are too few data points for a coherent pattern of age-related change in
memory encoding and retrieval functions to emerge. Much work remains to be done.
Nevertheless, some preliminary observations can be made at this very early stage of the
endeavor. The data based on subsequent memory performance suggest that encoding
difficulties may be responsible, at least in part, for the poorer direct and, in at least one instance
(Friedman et al., submitted), indirect memory performance in the older adult. By contrast, the
majority of studies of ERP parameters recorded during indirect tasks (in which items repeat
shortly after their initial presentation) suggest similar performance and cognitive processes in
old and young adults. In showing equivalent performance to that of younger adults on these
putatively indirect tests of priming, but reduced direct performance for these same items
relative to young adults, the data are consistent with the interpretation that older adults show a
qualitatively similar memory deficit to that displayed by amnesties. This pattern of findings is
support for the dissociation between explicit and implicit memory systems (and thus a multiple
memory systems approach to human memory). However, the ERP data from at least two
studies (Friedman et al., 1993c; Karayanadis et al., 1993) and behavioral data from other
laboratories (e.g., Chiarello & Hoyer, 1988; Davis et al., 1990; Hultsch et al., 1991), are
consistent with the conclusion that older adults also show deficits on test of implicit memory.
This raises the possibility that the above interpretation is premature. Recent studies that have
attempted to correlate implicit and explicit memory performance with frontal lobe test
performance in older adults (e.g., Craik et al., 1990) and our own novelty oddball data (Fabiani
& Friedman, in press; Friedman et al., 1993a; Friedman & Simpson, 1994) suggest that an
important source of variance in explaining age-related memory deficits may be age-related
differences in frontal lobe function.
Those studies attempting to correlate behavioral measures with indices of frontal lobe
fimction have all assessed the retrieval aspects of memory. However, unlike behavioral
assessments, the ERP provides a relatively direct, on-line measure that appears to reflect
encoding activity, the subsequent memory effect, or Dm. Moreover, evidence from PET
studies suggests that an important contribution during the acquisition phase of memory
paradigms is made by the left frontal lobe. Although preliminary, our data (Fabiani et al.,
1994) raise the possibility that the subsequent memory effect may be generated, at least in part,
by a neural system that involves this area of the brain. In addition, those data may be
consistent with an age-related difference in accessing that system during encoding.
Corroborative evidence for the role of the frontal lobes in the acquisition of to-be-remembered
material comes from patients with frontal excisions, who show defective encoding strategies
(review by Stuss et al., 1994). However, as assessed by ERP measures, retrieval has been less
often studied, so a skewed picture exists. This deficiency needs to be remedied. In addition,
scalp distribution has also been neglected in ERP studies of memory in general (for examples,
see Johnson, in press) and needs to be applied to the study of the aging of memory-related
phenomena.
What appears necessary at this stage in the study of ERPs, memory, and aging is a
convergence of neuropsychological, behavioral, ERP and (whenever possible) imaging data.
Through the convergence of these kinds of indices, including current source density analyses,
and source localization techniques (e.g., Scherg, 1990), attempts can be made to examine
further the possibility of a change in frontal lobe function with age as one means of
understanding age-related performance differences during both direct and indirect memory
tasks. For example, as stated earlier, the majority of studies of age-related changes in memory
performance support the explicit/implicit dissociative pattern seen for amnestics. However, if
frontal lobe dysfimction proves to be an important mediator of age-related performance
differences on both explicit and implicit tasks, older individuals may be better characterized as
less impaired "frontal amnestics" (e.g., Baddeley & Wilson, 1988; Stuss et al., 1994). That is,
memory for source or the context in which an item was originally learned appears to depend
upon the frontal lobes, and this component of explicit remembering is reported to be deficient
in elderly samples. In addition, another aspect of explicit remembering, recency memory (i.e.,
temporal information as to when the event was previously experienced), also appears to
depend upon the frontal lobes (Milner et al., 1991). Recent evidence from this laboratory
(Fabiani et al., unpublished observations) suggests that older individuals perform worse when
they are required to determine which of two events was most recently experienced compared
to whether the events were old or new (i.e., recognition memory). This componential
approach may prove useful in further understanding the nature of memory deficits in older
individuals. Thus, the distinction between mild forms of "medial temporal lobe" and "frontal"
amnesias, if experimentally verified, could eventually prove important in terms of remedial
strategies and drug treatments for older individuals with memory disorders. As one means of
exploring this possibility, we are currently assessing the relationship between memory for
source and cognitive aging (Trott et al., unpublished observations). We have also
hypothesized that at least some of the variance in the reduced memory performance of older
individuals may be due to an age-related change in frontal lobe activity. However, this
argument could be bolstered by finding relationships between memory performance,
topographical shifts in ERP activity, and indices of frontal lobe function (e.g., WCST; verbal
fluency; recency memory). These two approaches are currently being pursued in our
laboratory.
REFERENCES
Albert, M., F.H. Duffy, & M. Naeser (1987) Nonlinear changes in cognition with age and
their neuropsychologic correlates. Canadian Journal of Psychology, 41, 141-147.
Albert, M.S., J. Wolfe, & G. Lafleche (1990) Differences in abstraction ability with age.
Anderson, J. R., & Bower, G. H. (1974). A propositional theory of recognition memory.
Memory & Cognition, 2, 406-412.
Baddeley, A., & wilson, B. (1988). Frontal amnesia and the dysexecutive syndrome.
Brain and Language, 7, 212-230.
Bashore, T.I~ (1990). Age-related changes in mental processing revealed by analysis of
event-related brain potentials. In Rohrbaugh, JW, Parasuraman, R, Johnson, Jr, R
(Eds), Event-Related Brain Potentials: Basis Issues and Applications. New York:
Oxford University Press, pp. 242-275.
Bentin, S., & Moscovitch, M. (1990). Neurophysiological indices of implicit memory
performance. Bulletin of the Psychonomic Society,
Bentin, S., Moscovitch, M., & Heth, I. (1992). Memory with and without awareness:
Performance and electrophysiolo~cal evidence of savings. Journal of Experimental
Psychology: Learning, Memory, and Cognition, 18,
Berman, S., & Friedman, D. (1993) A developmental study of ERPs during recognition
memory: Effects of picture familiarity, word frequency, and readability. Journal of
Berman, S., Friedman, D., & Cramer, M. (1991) ERPs during continuous recognition
memory for words and pictures. Bulletin of the Psychonomic Society. 29, 113-116.
382 D. Friedmanand M. Fabiani
Bouras, C., Hot~ P.R., Giannakopoulos, P., Michel J.P., & Morrison, J.H. (1994).
Regional distribution of neurofibfiallry tangles and senile plaques in the cerebral cortex
of elderly patients: A quantitative evaluation of a one-year autopsy population from a
geriatric hospital. Cerebral Cortex, 4, 138-150.
Bowers, J.S., & Schacter, D.L. (1990) Implicit memory and test awareness. Journal of
Experimental Psychology: Learning, Memory and Cognition, 16, 404-416.
Challis, B. H., & Brodbeck, D. 1L (1992). Level of processing affects priming in word
fragment completion. Journal of Experimental Psychology: Learning, Memory, and
Cognition, 18, 595-607.
ChiareUo, C. & Hoyer, W.J. (1988). Adult age Differences in implicit and explicit memory:
Time course and encoding effects. Psychology and Aging, 3, 358-366.
Cohen, N.J., & Eichenbaum, H. (1993). Memory, amnesia, and the hippocampal system.
Cambridge: MIT Press.
Courchesae, E. (1978) Changes in P3 waves with event repetition: Long-term effects on
scalp distribution and amplitude. Electroenceph. clin. Neurophysiol., 45, 468-482.
Courchesae, E., Hillyard, S.A., & Galambos, R. (1975). Stimulus novelty, task relevance,
and the visual evoked potential in man. Electroencephalography and Clinical
Craik, F.I.M., & McDowd, J. M. (1987). Age differences in recall and recognition.
Craik, F.I.M., Morris, L.W., Morris, 1LG., & Loewen, E.R. (1990). Relations between
source amnesia and frontal lobe functioning in older adults. Psychology and Aging, 5,
148-151.
Crook, T., Bartus, 1L T., Ferris, S. H., Whitehouse, P., Cohen, G. D., & Gershon, S.
(1986). Age-associated memory impairment: Proposed diagnostic criteria and
measures of clinical change-Report of a National Institute of Mental Health workgroup.
Developmental Neuropsychology, 2, 261-276.
Dam, A.M. (1979). The density of neurons in the human hippocampus. Neuropath. Appl.
Neurobiol, 5, 249-264.
Davis, H.P., & Bernstein, P.A. (1992). Age-related changes in explicit and implicit
memory. In L.1L Squire, & N.A. Butters (Ed.), The Neuropsychology of Memory. (pp.
249-261) New York City: Cafilford Press.
Davis, H., Cohen, A., Gandy, M., et al. (1990). Lexical priming deficits as a function of
age. Behav. Neurosci, 104, 288-297.
Donchin, E., & Fabiani, M. (1991). The use of event-related brain potentials in the study
of memory: Is P300 a measure of event distinctiveness? In J. IL Jennings, & M. G. H.
Coles (Ed.), Handbook of cogniave psychophysiology: Central and autonomic nervous
system approaches (pp. 471-498). Chichester, UK: Wiley.
Fabiani, M., & Donchin, E. (in press). Encoding processes and memory organization: A
model of the von Restortf effect. Journal of Experimental Psychology: Learning,
Memory, and Cognition.
Fabiani, M., & Friedman, D. (in press) Changes in brain activity patterns in aging: The
novelty oddball. Psychophysiology.
Fabiani, M., Friedman, D., & Cheng, J. (1994) Scalp distribution and memory differences
between young and old subjects during a name oddball paradign~ Psychophysiology.
Fabiani, M., Karis, D, & Donchin, E. (1990). Effects of mnemonic strategy manipulation in
a Von gestorff paradigaL Electroencephalography and Clinical Neurophysiology, 75,
22-35.
Fabiani, M., Karis, D., & Donchin, E. (1986). P300 and recall in an incidental memory
paradignL Psychophysiology, 23, 298-308.
Ford, J.M., & Pfefferbaum, A. (1985). Age-related changes in ERPs. In P.IC Ackles, J.1L
Jennings, & M.G.H. Coles (Eds.), Advances in Psychophysiology (pp. 301-339)
Greenwich: JAI Press.
Ford, J. M., Rosenbloom, M. J., Sullivan, E. V., & Pfefferbaum, A. (1991). ERPs and
brain structure: Relationships across the adult age span in alcoholics and in a patient
with herpes simplex encephalitis. In C. H. M. Bnmia, G. Mulder, & M. N. Verbaten
(Ed.), Event-Realted Brain Research (EEG Suppl. 42, pp. 342-353). Amgerdam:
Elsevier Publishers.
Ford, J.M., Roth, W.T., Mohs, R., Hopkins, W., & Kopell, B.S. (1979) Event-related
potentials recorded from young and old adults during a memory retrieval task.
Electroencephalography and Clinical Neurophysiology, 47, 450-459,
Friedman, D. (1990a). ERPs during Continuous Recognition Memory for Words.
Friedman, D. (1990b) Endogenous event-related brain potentials during continuous
recognition memory for words, Biological Psychology, 30, 61-87. Friedman, D. (in
press) Cognition in the normal elderly: An event-related potential perspective, in F.
Boiler & J. Grafinan (Eds), Handbook ofNeuropsychology, Vol. 9.
Friedman, D., & Simpson, G. (1994) Amplitude and scalp distribution of target and novel
events: effects of temporal order in young, middle-aged and older adults. Cognitive
Brain Research, 2, 49-63.
Friedman D., Hamberger, M., & gitter, W. (1993a). Event-related potentials as indicators
of repetition priming in young and elderly adults: Amplitude, duration and scalp
distribution, Psychology and Aging,, 8, 120-125.
Friedman, D., Berman, S., and Hamberger, M. (1993b). Recognition memory and ERPs:
Age-related changes in young, middle-aged and elderly adults. Journal of
Friedman, D, Putnam, L, Sutton, S (1989) Event-related potentials in children, young
adults and senior citizens: Homologous components and scalp distribution changes.
Developmental Neuropsychology: 5, 33-60.
Friedman, D., Simpson, G., and Hamberger, M. (1993c). Age-related changes in scalp
topography to novel and target stimuli. Psychophysiology, 30, 383-396.
Friedman, D., Hamberger, M., Stem, Y., & Marder, I~ (1992). Event-related potentials
(ERPs) during repetition priming in Alzheimer's patients and young and older controls,
Journal of Clinical and Experimental. Neuropsychology, 14, 448-462,
Friedman, D., Ritter, W., Snodgrass, J.G., & Trott, C. (submitted). ERPs during explicit
and implicit memory in young and old adults.
Friedman, D. Snodgrass, J.G., & Ritter, W. (1995). Implicit retrieval in cued recall:
Implications for aging effects in memory. Journal of Clinical and Experimental
Gabrieli, J.D.E., Francis, W.S., Reminger, S.L., Veffaellie, M., Grosse, D.A., & Wilson,
IL S. (in press). A neuropsychological dissociation between different forms of repetition
priming: Intact picture-naming and impaired word-completion priming in patients with

Alzheimer's disease. Journal of Experimental Psychology: Learning, Memory, and
Cognition.
Gardiner, J. M., & Java, 1L I. (1990). Recollective, experience in word and nonword
recognition. Memory and Cognition, 18, 23-30.
Goldman-gakic, P. S. (1987). Circuitry of primate prefrontal cortex and regulation of
behavior by represenational memory. In V. B. Mountcastle, F. Plum, & Geiger (Eds.),
Handbook of physiology - The nervous system (vol. 5, pp. 373-417). Bethesda, MD:
American Physiological Association.
Goldman-Rakic, P. S. (1992). Working memory and the mind. Scientific American,
267(3), 110-117.
Golomb, J., Kluger, A., deLeon, M. J., Ferris, S. H., Convit, A., Mittelman, M. S., Cohen,
J., gusinek, H., DeSanti, S., & George, A. E. (1994). Hippocampal formation size in
normal human aging: A Correlate of delayed secondary memory performance.
Learning and Memory, 1, 45-54.
Graf~ P., & G. Mandler (1984) Activation makes words more accessible, but not
necessarily more retrievable. Journal of Verbal Learnning and Verbal Behavior, 23,
553-568.
Cnfifiem, F., N'Kaoua, B., Rougier, A., & Claverie, B. (1993). A preliminary study of the
intracranial topography of EgP repetition effects. Psychophysiology, 30, $31.
Hartman, M., & Hasher, L. (1991) Aging and suppression: Memory for previously relevant
information. Psychology and Aging, 6, 587-594.
Hashtroudi, S., Parker, E.S., Luis, J.D., & Reisen, C.A. (1989) Generation and elaboration
in older adults. Experimental Aging Research, 15, 73-78.
Haug, H., Barmwater, U., Eggers, 1L, et al. (1983). Anatomical changes in aging brain: A
morphometric analysis of the human prosencephalon. In J. Cervos-Navarro & H. I.
Sarkander (Eds.), Aging: Vol. 21. Brain Aging: Neuropathology and
Neuropharmacology (pp. 1-12) New York: Raven Press.
Heindel, W.C., Salmon, D.P., & Butters, N. (1990). Pictorial priming and cued recall in
Alzheimer's disease and Huntington's disease. Brain Cognition, 13, 282-295.
Heindel, W.C., Salmon, D.P., Shults, C.W., Walicke, P.A., & Butters, N. (1989).
Neuropsychological evidence for multiple implicit memory systems: A comparison of
Alzheimer's, Huntington's, and Parkinson's disease patients. Journal of Neuroscience,
9, 582-587.
Hillyard, S.A., & Picton, T.W. (1987). Electrophysioloy of cognition. In V.B.
Mountcastle et al. (Eds): Handbook of Physiology, VoL V: Higher functions of the
brain, Part2 (pp. 519-583), Baltimore: American Physiological Society.
Howard, D., Heisey, J., & Shaw, 1L (1986). Aging and the priming of newly learned
associations. Developmental Psychology, 22, 78-85.
Howard, D.V., Fry, A.F., & Brtme, C.M. (1991) Aging and memory for new associations:
Direct versus indirect measures. Journal of Experimental Psychology: Learning,
Hultsch, D.F., Masson, M.E.J., & Small, B.J. (1991) Adult age differences in direct and
indirect tests of memory. Journal of Gerontology: Psychological Sciences, 46,
P22-30.
Jacoby, L. L., & Dallas, M. (1981). On the relationship between autobiographical memory
and perceptual learning. Journal of Experimental Psychology: General, 110, 306-340.

Jacoby, L. L., & Hollingshead, A. (1990). Toward a generate/reco~nize model of
performance on direct and indirect tests of memory. Journal of Memory and
Language, 29, 433-454.
Johnson, 1L (in press). Event-related potential studies of working and long- term memory:
Normative and patient data. In F. Boller & J. Grafman (Eds.), Handbook of
Neuropsychology, Vol. 10
Johnson, 1L (1994). On the neural generators of the P300: Evidence from temporal
lobectomy patients. In G. Karmos, M. Molnar, V. Csepe, I Czigler, & J.E. Desmedt
(Ed~), Perspectives of event-relatedpotentials research (vol. EEG Suppl. 44, pp. 110-
129). Am~erdam: Elsevier.
Johnson, R. (1993). On the neural generators of the P300 component of the event-related
potential. Psychophysiology, 30, 90-97.
Johnson, 1L, Pfefferbaum, A., & Kopell, B.S. (1985). P300 and long-term memory:
Latency predicts recognition performance. Psychophysiology, 22, 497- 507.
Kapur, S., Craik, F. I. M., Tulving, E., Wilson, A. A., Houle, S., & Brown, G. M. (1994).
Neuroanatomical correlates of encoding in episodic memory: Levels of processing
effect. Proceedings of the National Academy of Sciences, 91, 2008-2111.
Karayanadis, F., Andrews, S., Ward, P.B., McConaghy, N. (1993) Event-related potentials
and repetition priming: Evidence for contextual processing differences in young,
middle-aged and elderly normal subjects. Cognitive Brain Research, 1, 123-134.
Karis, D., Fabiani~ M., & Donchin, E. (1984). "P300" and memory: Individual differences
in the von Restorff effect. Cognitive Psychology, 16, 177-216.
Kazmerski, V., & Friedman, D. (submitted). Direct and indirect tests of picture and word
priming in within- and cross-form conditions: Event-related potential and behavioral
measures.
Kazmerski, V., Friedman, D., & Hewitt, S. (submitted). The ERP repetition effect in
Alzheimer's patients: Multiple repetition priming with pictures
Kemper, T. (1984) Neuroanatomical and neuropathological changes in normal aging and in
dementia, in Albert, M.L. (Ed), Clinical Neurology of Aging. New York: Oxford
University Press, pp. 9- 52.
Kintsch, W. (1974). The representation of meaning in memory. Hillsdale, NJ: Erlbaum,
Knight, 1LT. (1984) Decreased response to novel stimuli after prefrontal lesions in man.
Electroenceph. clin. Neurophysiol., 59, 9-20.
Knight, 1LT. (1990). Neural mechanisms of event-related potentials: Evidence from human
lesion studies. In J.W. Rohrbaugh et al. (Eds): Event-Related Brain Potentials: Basic
Issues and Applications (pp. 3-18) New York: Oxford University Press.
Kutas, M. (1988). Review of event-related potential studies of memory. In (pp. 181-217).
Kutas, M. & HiUyard, S.A. (1980) Reading senseless sentences: Brain potentials reflect
Light, L.L. (1991). Memory and aging: Four hypotheses in search of data. Annual Review
of Psychology, 42, 333-376.
Lim, IC O., Zipursky, R. B., Watts, M. C., & Pfefferbaum, A. (1992). Decreased gray
matter in normal aging: An in vivo magnetic resonance study. Journal of Gerontology:
Biological Sciences, 47, B26-B30.
Luria, A.1L (1973). The Working Brain: An Introduction to Neuropsychology. Basic
Books, New York, 1973.

Mandler, G. (1980) Recognizing: The judgement of previous occurrence. Psychological
Review, 87, 252-271.
McCarthy, G., & Wood, C.C. (1985) Scalp distributions of event-related potentials: An
ambiguity associated with analysis of variance models. Electroencephalography and
Clinical Neurophysiology, 62, 203-208.
Mclntyre, J. S., & Craik, F. I. M. (1987). Age differences in memory for item and source
information. Canadian Journal of Psychology, 41, 175-192.
Merikle, P. M., & Reingold, E. M. (1991). Comparing direct (explicit) and indirect
(implicit) measures to study unconscious memory. Journal of Experimental
Milner, B., Corsi, P., & Leonard, G. (1991) Frontal lobe contribution to recency
judgements. Neuropsychologia, 29, 601-618.
Moscovitch, M. (1982) A neuropsychological approach to perception and memory in
normal and pathological aging. In F.I.M. Craik & S. Trehub (eds): Aging and
Cognitive Processes. New York: Plenum, pp. 55-78.
Moscovitch, M., & Winocur, G. (1992). The neuropsychology of memory and aging. In
F. I. M. Craik, & T. A. Salthouse (Ed.), The Handbook of Aging and Cognition (pp.
315-372). Hillsdale: Lawrence Earlbaum Associates.
Nelson, D.L., Reed, V.S. & McEvoy, C.L. (1977) Learning to order pictures and words: A
model of sensory and semantic encoding. Journal of Experimental Psychology:
Learning, Memory, and Cognition. 3, 485-497.
Neville, I-I, Kutas, M, Chesney, G, Schmidt, AL (1986) Event-related brain potentials
during initial encoding and recognition of congruous and incongruous words. Journal
of Memory and Language, 25, 75-92.
Ober, B.A., and Shenaut, G.K. (1988) Lexical Decision and Priming in Alzheimer's
Disease. Neuropsychologia, 26, 273-286.
Paller, K.A. (1990) Recall and stem-completion priming have different electrophysiological
correlates and are differentially modified by directed forgetting. Journal of
Experimental Psychology: Learning, Memory, and Cognition, 16, 1021-1032
Paller, K. A., & Kutas, M. (1992). Brain potentials during memory retrieval provide
neurophysiological support for the distinction between conscious recollection and
priming. Journal of Cognitive Neuroscience, 4, 375-391.
Paller, K.A., Kutas, M., & Mayes, A.R. (1987a) Neural correlates of encoding in an
incidental learning paradigm. Electroencephalography and Clinical Neurophysiology,
67, 360-371.
Parkin, A.J., Waker, B.M. (1992) Recollective experience, normal ageing, and frontal
dysfunction. Psychology and Aging, 7, 290-298.
Perrin, P., Pernier, J., Bertrand, O., Echallier, J.F. (1989). Spherical splines for scalp
potential and current density mapping. Electroencephalography and Clinical
Neurophysiology: 72; 184-187,
Petersen, S.E., Fox, P.T., Posner, M.I., Mintun, M., & Raichle, M.E. (1989) Positron
emission tomographic studies of the processing of single words. Journal of Cognitive
Neuroscience, 1, 153-170.
Pfefferbaum, A., Ford, J.M., Wenegrat, B.G., Roth, W.T., & Kopell, B.S. (1984). Clinical
application of the P3 component of the ERP I: Normal aging. Electroencephalography
and Clinical Neurophysiology, 59, 85-103.

Polich, J. (1991). P300 in the evaluation of aging and dementia. In C. H. M. Bnmia, G.
Mulder, & M. N. Verbaten (Ed.), Event-Related Brain Research (EEG Supplement 42,
pp. 304-322). Amsterdam: Elsevier Science Publishers.
Poon, L.W. (Ed.), Aging in the 1980s: Psychological issues. Washington, D.C.: American
Psychological Association.
Pooh, L.W., & Fozard, J.L. (1980) Age and word frequency effects in continuous
recognition memory. Journal of Gerontology, 35, 77-86.
Puce, A., Andrewes, D. G., Berkovic, S. F., & Bladin, P. F. (1991). Visual recognition
memory: Neurophysiological evidence for the role of temporal white matter in man.
Brain, 114, 1647-1666.
Rabbitt, P. (1965). Age-decrement in the ability to ignore irrelevant information. Journal
of Gerontology, 20, 233-238.
Raichle, M. E., Fiez, J. A., Videen, T. O., MacLeod, A. IC, Pardo, J. V., Fox, P. T., &
Petersen, S. E. (1994). Practice-related changes in human brain functional anatomy
during nonmotor learning. Cerebral Cortex, 4, 8-26.
Rankin, J.L., & Collins, M. (1985) Adult age differences in memory elaboration. Journal
Richardson-Klavehn, A., & Bjork, 1LA. (1988). Measures of memory. Annual Review of
Ritter, W., & Ruchkin, D.S. (1992) A review of event-related potential components
discovered in the context of studying P3. Annals of the New York Academy of
Sciences, 658, 1-32.
Roediger, H. L. (1990). Implicit memory: Retention without remembering. American
Psychologist, 45, 1043-1056.
Roediger, H. L., & Blaxton, T., A. (1987). Effects of varying modality, surface features,
and retention interval on priming in word-fragment completion. Memory and
Cognition, 15, 379-388.
Roediger, H. L. I., & McDermott, K. B. (1993). Implicit memory in normal human
subjects. In F. Boller,& J. Grafman (Ed.), Handbook of Neuropsychology (vol. 8, pp.
63-131). Am~erdam: Elsevier Science Publishers, B.V.
Ruchkin, D.S., & Sutton, S. (1983) Positive slow wave and P300: Association and
dissociation. In CJaillard, AWK, Ritter, W (Eds), Tutorials in ERP research:
Endogenous components. Am~erdam: North Holland, pp. 233-250.
Rugg, M. (1990). Event-related brain potentials dissociate repetition effects of high- and
low-frequency words. Memory and Cognition, 18, 367-379.
Rugg, M. D. (in press). ERP studies of memory. In M. D. Rugg, & M. G. H. Coles (Ed.),
Eleetrophysiology of Mind: Event-Related Brain Potentials and Cogniton (New York
City: Oxford university Press).
Rugg, M., & Doyle, M. (1994). Event-related potentials and stimulus repetition in direct
and indirect tests of memory. In H. Heinz, T. Munte, & G. R. Mangun (Eds.),
Cognitive eleetrophysiology (pp. i24-148). Cambridge, MA: Birkhauser.
Rugg, M., Furda, J., & Lorist, M. (1988). The effects of task on the modulation of
event-related potentials by word repetition. Psychophysiology, 25, 55-63.
Rugg, M., & Nagy, M. (1989). Event-related potentials and recognition memory for
words. Electroencephalography and Clinical Neurophysiology, 72, 395-406.
Rybash, J. M. (1994). Aging, associative priming, and test awareness. Aging and
Salthouse, T. A. (1988). Initiating the formalization of theories of cognitive aging.
Salthouse, T. A. (1991). Theoretical Perspectives on Cognitive Aging. HiUsdale:
Lawrence Erlbaum Associates.
Salthouse, T. A. (1994). The aging ofworking memory. Neuropsychology, 8, 535-543.
Schacter, D.L., Harbluk, J.L., & McLachlan, D.R. (1984). Retrieval without recollection:
An experimental analysis of source amnesia. Journal of Verbal Learning and Verbal
Behavior, 23, 593-611.
Scheibel, M.E., Lindsay, KD., Tomiyasu, U., & Scheibel, A.B. (1976) Progressive
dendritic changes in the aging human limbic system Experimental Neurology: 53; 420-
430.
Scherg, M.(1990). Fundamentals of dipole source analysis. In Hoke, M et al. (Eds),
Auditory Evoked Magnetic Fields and Potentials. Basel: Karger, pp. 40-69.
Shallice, T., Fletcher, P., Frith, C. D., Grasby, P., Frackowiak, K S. J., & Dolan, K J.
(1994). Brain regions associated with acquisition and retrieval of verbal episodic
memory. Nature, 368, 633-635.
Shaw, T.G., Mortel, I~F., Meyer, J.S., Rogers, KL., Hardenberg, J., & Cutaia, M.M.
(1984). Cerebral blood flow changes in benign aging and cerebrovascular disease.
Neurology, 34, 855-862.
Shepherd, K N., & Teghtsoonian, M. (1961). Retention of information under conditions
approaching steady state. Journal of Experimental Psychology, 62, 302-309.
Smith, M.E., & Halgren, E. (1989) Dissociation of recognition memory components
folowing temporal lobe lesions. Journal of Experimental Psychology: Learning,
Memory, and Cognition: 15; 50-60, 1989.
Snodgrass, J.G., and J. Corwin (1988) Pragmatics of measuring recognition memory:
Applications to dementia and amnesia. Journal of Experimental Psychology: General,
116, 34-50.
Spencer, W. D., & Raz, N. (1994). Memory for facts, source, and context: Can frontal
lobe dysfunction explain age-related differences? Psychology andAging, 9, 149-159.
Squire, L. R. (1987). Memory and Brain. New York City: Oxford University Press.
Squire, L. 1L (1992). Memory and the hippocampus: A synthesis from findings with rats,
monkeys, and humans. Psychological Review, 99, 195-231.
Squire, L.1L, Ojemann, J.G. Miezen, F. M., Petersen, S. E., Videen, T. O., & Raichle, M.
E. (1992). Activation of the hippocampus in normal humans: A functional anatomic
study of memory. Proceedings of the National Academy of Sciences, 89, 1837-1841.
Squires, N.I~, Squires, ICC., & Hillyard, S.A. (1975) Two varietes of long-latency positive
waves evoked by unpredictable auditory stimuli in man. Electroencephalography and
Clinical Neurophysiology: 38; 387-410, 1975.
Stuss, D. T. (1993). Assessment of neuropsychological dysfunction in frontal lobe
degenration. Dementia, 4, 220-225.
Stuss, D.T., Kaplan, E.F., Benson, D.F., Weir, W.S., Chiulli, S., & Sarazin, F.F. (1982).
Evidence for the involvement of orbitofrontal cortex in memory functions: An
interference effect. Journal of Comparative Physiology and Psychology, 96, 913-925.
Stuss, D. T., Alexander, M. P., Palumbo, C. L., Buckle, L., Sayer, L., & Pogue, J. (1994).
Organizational strategies of patients with unilateral or bilateral frontal lobe injury in

word list learning tasks. Neuropsychology, 8, 355-373.
Stuss, D. T., Eskes, G. A., & Foster, J. I~ (1994). Experimental neuropsychological
studies of frontal lobe functions. In F. Boiler, & J. Grafinan (Ed.), Handbook of
Neuropsychology (vol. 9, pp. 149-185). Amsterdam: Elsevier Publishers.
Swick, D., & Knight, 1L T. (1994). Effects of aging on ERPs and behavioral performance
in explicit and implicit memory tasks. Paper presented at the First Annual Meeting of
the Cognitive Neuroscience Society, San Fransisco, California, March.
Terry, 1LD., DeTeresa, 1L, & Hansen, L.A. (1987) Neocortical cell counts in normal
human adult aging. Annuals of Neurology, 21, 530-539.
Tipper, S.P. (1991). Less attentional selectivity as a result of declining inhibition in older
adults. Bulletin of the Psychonomic Society: 29; 45-47.
Tomlinson, B.E., Blessed, G., & Roth, M. (1968). Observations on the brains of non-
demented old people. Journal of Neurological Sciences: 7; 331-356.
Tulving, E., & Schacter, D.L. (1990) Priming and human memory systems. Science, 247;
301-306.
Tulving, E., Kapur, S., Craik, F. I. M., Moscovitch, M., & Houle, S. (1994). Hemispheric
encoding/retrieval asymmetry in episodic memory: Positron emission tomography
studies. Proceedings of the National Academy of Sciences, 91, 2016-2020.
Tulving, E., Kapur, S., Markowitsch, H. J., Craik, F. I. M., Habib, 1L, & Houle, S. (1994).
Neuroanatomical correlates of retrieval in episodic memory: Auditory sentence
recognition. Proceedings of the National Academy of Sciences, 91, 2012-2015.
Van Petten, CV, Kutas, M, Kluender, R, Mitchiner, M, Mclsaac, H: Fractionating the word
repetition effect with event-related potentials. Journal of Cognitive Neuroscience: 3;
131-150, 1991.
Warrington, E.IC, & Weiskrantz, L. (1968) New method of testing long-term retention
with sepcial reference to amnesic patients. Nature, 217, 972-974.
Waugh, N.C., & Norman, D.A. (1965) Primary memory. Psychological Review: 72;
89-104.
Woods, D.L., & Knight, 1LT. (1986) Electrophysiological evidence of increased
distractibility after dorsolateral prefrontal lesions. Neurology, 36, 212-216.
Young, M. P., & Rugg, M. D. (1992). Word frequency and multiple repetition as
determinants of the modulation of event-related potentials in a semantic classification
task. Psychophysiology, 29, 664-676.
390 1995 Elsevier Science B.V.
Do general slowing functions mask local slowing effects?

A chronopsychophysiological perspective*
Theodore 1L Bashore a and Fren Smuldersb
aDepartment of Psychology, University of Northern Colorado, Greeley, CO 80639
bDepartment of Psychology, University of Am~erdam, Amsterdam, The Netherlands
It is taken as fact that slowness of behavior with age

has been reliably observed in man and other mammals in
a wide variety of contexts. It is also taken as fact
that most of the slowness of behavior associated with
advancing age can be attributed to the time taken or
required by the central nervous system in mediating the
input and output relations in behavior. Generally, not
much of the slowness with age can properly be attributed
to motility of joints or muscular contraction. Also,
limitations of sensory input or the primary sensations
are not usually the major contributor. (pp. 191-192)
Birren (1965)
Perhaps the most ubiquitous and significant change

observed in the older organism is slowness of
behavior. Slowness is not limited to motor responses
or to peripheral sensory phenomena...; instead, it
appears to be even more evident the more complex the
behavior observed and the higher the mediating neural
structures in the nervous system,..it seems basic to
other phenomena and it in turn offers the greatest
possibility of being explained in terms of accompanying
neurobiological changes. (p. 293)
Birren, Woods, & M.V. Williams (1980)
ACKNOWLEDGEMENTS: This chapter was written while the first author was a Visiting Professor at the
University of Amsterdam, supported by the Experimental Psychology Graduate School (ElK)S) in the
Netherlands, by a visitor's grant from the Experimental Psychology Graduate School (EPOS), and by a stipend
from the Research Corporation at the University of Northern Colorado. In addition, preparation of this chapter
was supported in part by grants to the first author from the National Institute on Aging (AG04581, AG12263).
The authors would like to thank Elly Zeef and Albert Kok for providing data from their study, and Maurits van
der Molen and Albert Kok for carefully reading a previous draft of the chapter and making very thoughtful
suggestions for revision.
Do general slowing functions mask local slowing effects? 391
By the early 1960s, as expressed so eloquently in the quote from Birren (1965), there
was virtual unanimity in the psychological literature that the speed of information processing
slows as the adult nervous system ages. Indeed, few observations in this literature are as
indisputable as the observation that information processing speed, indexed by reaction time
(RT), is slowed among the elderly (Salthouse, 1985a). The view, expressed in the quotes from
Birren, that the slowing is restricted in large measure to the central components of processing
has also received considerable support in the literature. There is, however, little agreement
about the extent to which this slowing is a generalized phenomenon that influences all
components of central processing and spares none or is a differential phenomenon that
influences some components of processing while sparing others. This disagreement has given
rise to a spirited debate among investigators in cognitive aging in which the participants have
taken extreme positions, arguing either for generalized (i.e., global) slowing or for differential
(i.e., local) slowing (e.g., see Cerella, 1994; Fisk & Fisher, 1994; Myerson, WagstatI~ & Hale,
1994; Perfect, 1994). The origins of this debate canbe traced to the work of James Birren and
his associates in the 1950s and 1960s (e.g., Birren, 1955, 1956, 1964, 1965; Birren &
Botwinick, 1951a, b; Birren, Botwinick, Weiss, & Morrison, 1963; Birren, Riegel, &
Morrison, 1962). They argued that the magnitude of the slowing in response speed produced
by advancing age differed for the peripheral and central components of information processing.
Advancing age was presumed to have little or no effect on the rate of peripheral processing
speed, while at the same time producing an indiscriminant, or generalized, slowing in the rate
of central information processing speed that became more evident as processing complexity
increased (see review in Birren et al., 1980). This hypothesis is known today as the
Complexity or Birren Hypothesis (Cerella, Pooh, & Fozard, 1981; Salthouse, 1985a).
The generalized slowing in response speed evident among the elderly was contrasted by
Birren (1965) to a task-dependent slowing in response speed among the young: "...young
subjects appear to be task specific in their response speed, i.e., they are quick or slow
depending upon the nature of the task. With increasing age individuals tend to show a
characteristic slowness of response regardless of the nature of the task." (p. 193). Moreover,
he argued that with advancing age there is a growing interdependence between central stimulus
processing time (what Birren, 1965, called association time) and response output time.
According to the view of Birren and his associates, the slowing of response speed reflected a
primary change in the nervous system in which "...all or most processes mediated by the
central nervous systenl.." (Birren, 1965, p. 199) were influenced. Indeed, Birren (1965) took
the position that "A search for the physiological correlates of age changes in speed of behavior
would be less compelling if there were many, rather than one or a few, time constants involved
in the slowness of behavior." (p. 195) In its strong expression, the Birren Hypothesis asserts
that all elements of central information processing are slowed equivalently by advancing age as
information processing requirements are made more demanding; in its weak expression, it
asserts that all elements of central information processing are simply slowed (Cerella et al.,
1981).
In this chapter, we trace the history of the debate over the nature of age-related
cognitive slowing, a debate that has emerged from analyses of RT studies of information
processing speed; and review research from cognitive psychophysiology that suggests that the
decline in processing speed may have important local components that are not revealed in the
reaction time data.
392 Th.R. Bashore and F. Smulders
1. THE BRINLEY ANALYSIS: A REGRESSION-BASED APPROACH
Aside from the fact of slowing per se, the observation

that the amount of slowing is related to the ditficulty
of the task may be one of the earliest general conclusions
drawn from the gerontological literature. (pp. 332-333)
Cerella, Pooh, & D.M. Williams (1980)
1.1. Proportional slowing
Salthouse (1978) and Cerella et al. (1980) were the first to argue that the strong
version of the Birren Hypothesis could be tested using an analytic procedure modeled after the
seminal work of Joseph Brinley* (1965; for an historical review of the use of this procedure in
studies of cognitive slowing see Bashore, 1994). This procedure has a graphic and an analytic
component. It involves plotting the intersection of the mean or median response latencies of a
group of older subjects on those of a group of younger subjects across levels of task
complexity (defined operationally by variations in RT) and then completing a regression
analysis on the paired values. Thus, each pair of values (old vs young) is represented by a
point in XY space. Advocacy of this analytic procedure was motivated by the following logic.
(1) If all elements of mental processing speed are slowed equivalently by advancing age with
increases in processing complexity but not with changes in task content, the dispersion of
points in the coordinate space will be described by a linear function (i.e., one having a high r2).
(2) The relative contributions of age-related changes in the rates of peripheral (i.e.,
sensorimotor, perceptuomotor, noncomputational) and central (i.e., postperceptual,
computational) processing speed to the overall slowing are deducible from the slope and
intercept properties of the regression function: (a) peripheral slowing is invariant across tasks
of differing complexity, therefore, it is revealed in a function with an elevated intercept (i.e., an
additive function); (b) generalized central slowing is proportional across tasks of varying
complexity, therefore, it is manifest in a function that has an intercept of zero and a slope
exceeding 1.0 (i.e., a multiplicative function); and (c) slowing that encompasses both peripheral
and central processing elements is expressed in a regression function with an elevated intercept
and a slope greater than 1.0 (see discussions in Cerella, 1985a; and Salthouse, 1985a,b). Thus,
it was reasoned that the intercept of the regression function revealed peripheral processes,
whereas the slope revealed central processes. These hypothetical relationships are depicted in
Figure 1.
Salthouse (1978) used the analytic procedure introduced by Brinley (1965) to evaluate
the response latencies of older subjects relative to those of the young across levels of
processing complexity within a task. In his analyses, regression lines were fit to response
latency data obtained from (1) an experiment in his laboratory in which subjects performed
several versions of the Digit Symbol Substitution Test (DSST), and (2) five individual
experiments done in other laboratories that had at least ten different conditions. The
Brinley introduced the graphical presentation and regression analysis procedure that now bears his name by
applying it to different conditions within one experiment. Cerella et al. (1980) expanded the approach by
applying it to a wide array of tasks taken from a large number of experiments conducted in different
laboratories (i.e., making it meta-analytic).
1200, 1200 1200
lo00, 1000 ..1000
-, ~ ,o0 E,**
>,
so0, Ioo
effect 93 4o0,
1 effect
(central slowing)
20o J (peripheral slowing) 2o0, :tOO' / (periPheral and central slowing)
/" yul.0x§ 100 / y=l.Sx§
/ r y: 1 . S x + 100
- . . . . . . . . .
~ ~0 ~0 ~ " ~0 I ~ 1~0 o ~o 4o0 o0o 8oo lOOO 12oo
young latency (msec) young latency ( m s e c ) young latency (reset)
Figure 1. Hypothetical relationships between age and slowing as expressed by regressing old response
latencies on young response latencies. The function on the left depicts an additive effect that was argued to
reveal peripheral slowing; the function in the middle shows a multiplicative effect that Was assumed to express
central slowing; and the function on the right illustrates slowing that was thought to include both peripheral
and central components.
distribution of points from the DSST was described by a linear function with an intercept of-
90 msec, a slope of 1.60, and an r 2 of .972. Linear functions also described the spread of
points in the five other experiments. These functions had intercepts ranging from-35 to -415
msec, slopes ranging from 1.58 to 2.01, and coefficients of determination ranging from .901 to
.992 (for the precise properties of the regression functions in each experiment, see Table 2 in
Salthouse, 1985a). This list of studies was later expanded by Salthouse (1985b; Table 9.1) to
include 8 additional studies, the distribution of points from each of which was likewise
described by linear functions similar to those reported in the earlier analysis (intercepts ranging
from-10 to -420 reset, slopes ranging from 1.23 to 2.05, and r2s ranging from .822 to .992;
for an exception to this overall pattern see Madden, 1984, who reported a linear function,
r2=.93, with a slope of 1.07 and an intercept of 210 msec, that described the points for 10
conditions in two visual search tasks).
Cerella et al. introduced a meta-analytic variant of the Brinley analysis; reasoning that
"...although comparisons of conditions within individual experiments provide tests of the
hypothesis, comparisons of conditions between experiments offer even more opportunities for
evaluation"in which "...the hypothesis should be able to rank order, in terms of the magnitude
of the age difference, all of the conditions of all of the studies combined.". (p. 333) They took
mean or median KTs from a large number of studies in which age-contrasted groups were
assessed on a broad range of tasks of varying complexity and regressed the values of the older
subjects on those of the younger subjects. Young subjects were in their 20s; older subjects
spanned every decade from the 30s through the 80s. The distribution of points thus generated
was described by a linear function with an intercept that was slightly negative, -70 msec, a
slope of 1.36, and a coefficient of determination of .90. A series of stepwise multiple
regression analyses was completed by Cerella et al. to determine the contribution of different
variables to this overall pattern. To take a closer look at the influence of older age on the
regression function, Cerella et al. partitioned the older groups into those who were over 60 and
those who were between 30 and 60, and regressed the values from these two groups onto
those of the subjects in their 20s. The resultant functions explained 96.4% of the variance, had
slopes of 1.62 and 1.16, and intercepts of-130 and -40 msec for the over 60 and 30-60 age
groups, respectively. That this distribution of points for indiscriminantly aggregated tasks did
not distort the distribution of points for the different classes of tasks included in the meta-
analysis was suggested by the fact that regression functions with comparable properties were
derived by Cerella et al. in analyses on subsets of tasks that were segregated by similarity. Six
different task groups were generated, yielding slowing functions that accounted for 91.2% of
the variance, with slopes ranging from 1.08 to 1.70, and intercepts ranging from -260 to 150
msec (see Table 24-4 in Cerella et al. for the specific values). Thus, partitioning the data into
classes of tasks contributed little to explaining the variance in the regression function (an
increase of 1 percentage point).
However, the differential contribution of peripheral and central processing to the form
of the regression function was suggested in another set of analyses by these investigators. In
these analyses, tasks were partitioned by Cerella et al. into those that they considered primarily
sensorimotor and those that they considered primarily computational* , a partitioning that
revealed contributions of both peripheral and central slowing to the overall slowing in RT
among the older subjects (defined as 30 or older). The slowing was observed to be more
profound for central than for peripheral processing (slope of 1.62 and intercept of-10 msec for
the former; slope of 1.14 and intercept of 0 msec for the latter; R 2 of .917), supporting the
conclusion that both types of processing slowed with age but that the greatest slowing was
evident when computational mechanisms were engaged. When Cerella et al. crossed age (60
and older, 30-60) with peripheral versus central processing, they found that the function for
computational operations changed little for the subjects over 60 (slope of 1.62 versus 1.66;
intercept of-130 versus 0 msec; R 2 of 96.4 versus 96.1), but that its slope varied considerably
for sensorimotor tasks among these subjects (1.62 versus 1.25, intercept of-130 versus 0
msec, R 2 of 96.4 versus 96.1). This analysis also revealed that the magnitude of the slowing
fimctions for peripheral and central processing were similar among the 30-60-year-old group,
with the former resembling that of the over 60 group (slope of 1.18 versus 1.25) but the latter
being quite different (slope of 1.14 versus 1.66). Thus, the age-related slowing of central
processing speed was observed to be much more dramatic for central than for peripheral
processing from middle to older age.
This overall pattern of results led Cerella et al. to conclude that peripheral slowing is
small and essentially invariant across levels of processing complexity, whereas central slowing
is moderate and increases proportionately with increases in processing complexity. Thus, in
this series of analyses, a single parameter, described by the slope of the regression function,
appeared to characterize the age effect across tasks of differing complexity. Cerella et al. were
therefore encouraged to conclude that the "...slowing appears to affect all mental processes
equally and may account for the complexity effects observed in the data." (p. 339)
Identification of a single slowing function that summarized the pattern of results across a wide
range of disparate tasks was, as Cerella et al. expressed, a rather astonishing outcome. A
somewhat different conclusion was drawn, however, in work done subsequently by Cerella
(1985a). He refined and extended the original meta-analyses of Cerella et al. in a series of
regression analyses on both actual and simulated data. The outcomes of these analyses led him
to modify the single factor (slope) model to a two-factor model (intercept, slope) of age-
related cognitive slowing. According to the revised model, advancing age produces a slight
Peripheral (sensorimotor, perceptuomotor, non-computational) tasks were defined as those that included the
same stimulus presentation and response requirement of more complicated tasks, but not the intermediate
computations. All other tasks were consideredto be computational (i.e., cognitive).
Do general slowing functions mask local slowing effects ? 395
slowing of peripheral processing speed that is not associated with task complexity (factor 1)
and a concomitant moderate decline in central processing speed that is associated very closely
with task complexity (factor 2).
The regression analyses reported by Salthouse and Cerella et al. consistently generated
functions that, although linear, had a rather disconcerting property, negative intercepts (as did
the original function reported by Brinley, 1965)* . The presence of a negative intercept
implies, according to the logic underlying the regression approach, not only that there is
peripheral involvement in the aging effect, but, more importantly, that in tasks requiring little
computational processing older adults are faster than younger adults (i.e., peripheral
processing is faster among the elderly than among the young). An advantage among the
elderly on such tasks is never reported, however (Birren et al., 1980; Welford, 1977). In a
subset of the analyses completed by Cerella (1985a) an explanation was provided for this
departure from the predicted value in the intercept for additive (positive) or multiplicative
(origin) functions. He demonstrated that (1) regression functions resembling those reported in
the meta-analyses (i.e., across tasks) described the fit of points for tasks requiring
computational processing within individual studies (as was evident in Brinley, 1965; Salthouse,
1978; and Salthouse & Somberg, 1982); and (2) a systematic relationship existed between the
slope of the regression function for an individual study and its intercept when the origin of the
function was determined by the ratio of the older subject latency to that of the young subject
latency on a perceptuomotor task: within limits, as the slope of the function became steeper,
the intercept became more negative.
Thus, the magnitude of the negative intercept for an individual study was found to be
tied very closely to the degree to which computational processing demands were increased
across levels of the task under examination. Most salient, however, was Cerella's
demonstration that a pair of hypothetical factors (one peripheral, one central), each estimated
for an individual study from a global equation (i.e., one that described the regression function
in the meta-analysis for peripheral and central slowing, respectively) generated a negative
intercept in the individual analysis even though the model from which each point was derived
was multiplicative. The intercept was negative in an individual study only when the central
factor exceeded the peripheral factor, which is almost always the case. Cerella thus concluded
that a negative intercept in the individual analysis does not implicate an additive slowing
process and that the age effects in individual studies "...can be modeled by assuming a fixed
peripheral deficit, and a more severe but varied central deficit." (p. 78)
1.2. Exponential slowing.
Hale, Myerson, and Wagstaff (1987) were concerned, however, that the negative
intercept may be a mathematical artifact that results from fitting a straight line to a distribution
of points that has a positive curvature. Moreover, they were concerned that the distribution of
points used in the original meta-analyses may have been artificially dispersed because of the
wide range of tasks included. As a result, they restricted their analysis to RT tasks that were
nonverbal in nature and required only simple, discrete response outputs (e.g., a button press),
eliminating verbal tasks and those that required more complicated response outputs (e.g., card
Salthouse and Somberg (1982) evaluated the slowing phenomenon in manipulations of several factors in a
Sternberg memory scanning task. As in the other analyses, a linear function, with a slope of 2.16 and an
intercept of-.3 l, described the points well (r2=.982).
sorting). Additionally, they thought that the age range may have been sufficiently broad to
obscure important patterns in the meta-analyses of Cerella et al. (who used subject groups with
mean or median ages under 40 and over 60). Thus, they restricted the age range among the
young to 20-25 and among the old to 65-75.
The analysis by Hale et al. revealed that a positively accelerated power function
described the spread of points (O=ayl29). Transformation of the latencies to logarithmic
values produced a regression function with a slope of 1.29, indicating a positively-accelerated
power function, that accounted for 98.9% of the variance. These results suggest that age-
related slowing is nonlinear rather than linear in the nonverbal domain, implying that the
relative magnitude of the slowing increases systematically as more processing elements are
engaged (i.e., as task complexity increases). Like the linear functions revealed in the original
meta-analyses, a power function i ~ l i e s slowing is generated across tasks of differing
complexity, but it d o e s n o t imply equivalent slowing across elements of processing, as does the
multiplicative function.
Recall that Cerella et al. (1980) and Cerella (1985a) derived two different linear
fimctions for sensorimotor and computational tasks. Hale et al. argued that the coupling of
slopes and intercepts demonstrated by Cerella (1985a) for individual studies, with steeper
slopes being associated with more negative intercepts, is "...exactly what one would expect if;
as indicated by our analyses, the relationship between the performances of older and younger
groups was positively accelerated." (p. 134) Thus, a single power function explained the
variance more parsimoniously than did two linear functions. In explanation of their finding,
Hale at al. characterized the relationship between old and young response latencies across
levels of task difficulty in terms of what in biology is called an allometde relationship. This is a
relationship in which two variables (in this instance, young RT and old RT) are both
exponential functions of a third variable (in this instance, task difficulty). Hale et al. assumed
that a positively-accelerated increase in RT occurs in both young and older adults as
processing complexity is increased, but that the rate of increase is greater in older subjects.
Thus, the model they proposed, as does Cerella's (1985) two-factor model, contains two
parameters, one for the slowing produced in young subjects by increases in task complexity
and another for the slowing produced in older subjects. Unlike Cerella's model, however, the
slowing factors were exponential, not multiplicative.
Subsequent work by Hale, Myerson and colleagues has extended this first set of
findings, revealing that positively accelerated functions describe (1) the relation of the response
latencies of middle-aged adults (36-44 years old) to the response latencies of young adults
across tasks of varying complexity and domain (lexical, nonlexical; Myerson, Hale, I-Iirschman,
Hansen, & Christensen, 1989); (2) both individual and group data (Myerson et al., 1989); and
(3) the best fit for RT data sampled at different points in the latency distribution on a variety of
lexical tasks (Smith, Poon, Hale, & Myerson, 1988) or an aggregate of both lexical and
nonlexieal tasks (Myerson et al., 1989), with these functions being coninear and having
marginally better fits by power than by linear functions (Smith et al., 1988).
Myerson, Hale, Wagstafl~ Poon, and Smith (1990) gave mathematical expression to
their work in the form of the Information-Loss Model. This model was developed on the basis
of the following assumptions: (1) information is processed over a series of discrete
nonoverlapping steps (multiple occurrences of which may constitute a stage), each of which is
influenced equally by a~e; (2) the duration of each step increases as the amount of information
available to it decreases; (3) information is lost during each step in the processing sequence in
both the young and the old, producing increasingly longer steps with each new step; (4) new
steps are added to the processing sequence by increments in task complexity and the number of
steps added is comparable in the young and the old; and (5) the efficiency of the information
processing system diminishes with age, which is manifest in an accelerated loss of information
among older individuals as more steps are required in the processing sequence (i.e., as task
complexity is increased). An important element of the model is that increases in task
complexity are assumed to add comparable numbers of steps in the processing sequences of
the young and the old. That is, aging does not add new steps or engage additional steps as
processing demands are increased. Hence, the slowing evident among the aged as task
demands are increased is the product of a progressively greater loss of information in the old
than in the young with the engagement of each new processing step. This differential loss of
information is revealed as a positively accelerated function in the regression analysis. Because
this model correctly predicted the positively accelerated function relating response latencies of
older subjects to those of young subjects "...without regard for the specific nature of the
task..." (p. 484), Myerson et al. concluded that "...the view that the age difference in latencies
represents a global change" (p. 484) is dearly supported.
In more recent theoretical work, Cerella (1990) has formulated a quantitative
alternative to the Information-Loss Model that shares a kinship with it. Like the Information-
Loss Model, this alternative, the Overhead Model, derives from the meta-analytic approach.
The Overhead Model assumes, as does the Information-Loss Model, that information
processing occurs via a series of discrete steps. Unlike the Information-Loss Model, age-
related cognitive slowing is assumed to be produced by what Cerella calls an "overhead"
penalty that is extracted from older individuals at each information processing step and serves
to prolong the duration of each step. This processing burden is assumed to grow in magnitude
as the number of processing steps required by a task increases. Thus, the greater the number
of steps required to complete a task, the more time given steps consume as they are activated
(i.e., the larger the penalty). This relationship is expressed as a nonlinear, positively
accelerating parabolic function described by a quadratic equation, as opposed to the linear or
power functions of the alternative models. The Overhead Model also differs from the
Information-Loss Model in that it assumes that the basic processing mechanisms are
comparable among the young and the old. Cerella (1990) refers to this assumption as the
Correspondence Axiom. The Information-Loss Model assumes, in contrast, that the basic
mechanisms differ: Information is lost during processing by both young and older subjects, but
it is lost at a faster rate in the elderly. According to the Overhead Model, slowing among the
elderly is a manifestation of a growing processing burden imposed on them, but not on the
young, at each additional step in the sequence. Thus, a single parameter is postulated to
describe a generalized age effect on cognitive processing. Cerella (1990) demonstrated that
this single-parameter model can account for approximately the same amount of variance (for
both single study and general meta-analyses) as does the more complicated two-parameter
models and the original single-parameter model. It is important to note, however, that Hale,
Lima, and Myerson (1991) have shown that si~ificantly more variance can be explained by
adding a second parameter to the Overhead Model, prolonged duration in the basic processing
step among older adults.
Although both models postulate global change in age-related cognitive slowing, they
can be re-interpreted to implicate particular components of cognitive processing more than
others in the slowing process. Re-interpretation of the Information-Loss Model suggests the
following: (i) with each step in the information processing sequence there is progressively less
information being passed to the next step (i.e., there is growing information loss) in both the
young and the old; (ii) the magnitude of the growth of this reduction is greater in the old than
in the young; (iii) this growing decrement in the amount of information passed through the
system persists to the execution of the response, which implies (iv) that information loss is
greatest immediately prior to initiation of the response. If so, the components of cognitive
processing that mediate the translation of a stimulus input to a response output (i.e., response
selection) or aspects of motor processing subsequent to this translation (e.g., motor
programming, motor adjustment; Sanders, 1990) may be differentially susceptible to the effects
of aging. The Overhead Model, like the Information-Loss Model, can be re-interpreted:
Growing overhead as information processing proceeds, like growing information-loss, implies
the emergence of the largest burden in the elderly somewhere near initiation of the response
output.
1.3. Chinks in the armor: Domain-specific slowing
Support for the strong version of the Bitten Hypothesis began to erode with the
publication of the article by Hale et al. (1987; see also Madden, 1989) that suggested that
different slowing functions may describe lexical (linear) and nonlexical (power) tasks. The
importance of distinguishing domain-specific slowing, certainly in the form of lexical versus
nonlexical, was demonstrated most clearly in the work of Lima, Hale, and Myerson (1991).
They evaluated this distinction in a series of three meta-analyses that included a wide range of
lexical decision (i.e., make a word-nonword judgement) and lexical (e.g., naming, category
membership, upper or lower case judgement) tasks, and compared the results of these analyses
with those of a meta-analysis on nonlexical data. In the first analysis, the function derived from
a large number of experimental conditions in a variety of lexical decision tasks conducted in
several laboratories was compared with that obtained by Madden (1989) from a small number
of lexical decision tasks completed in his laboratory. This analysis, like Madden's, revealed a
linear function with a slope of about 1.5, a slightly negative intercept, and a coefficient of
determination exceeding .90. The second meta-analysis was restricted to lexical tasks that did
not require a lexical decision. Once again, the points were fit by a linear function that was very
similar to that derived for the lexical decision tasks and accounted for approximately the same
amount of variance. These results support the conclusion that "... essentially one age-related
cognitive slowing factor characterizes all lexical processing, regardless of whether the task
involves words only or both words and nonwords, and regardless of the specific variables
manipulated in the experiment." (p. 419)
However, that this pattern of slowing may not be all-encompassing was demonstrated
very convincingly in the third meta-analysis in which Lima et al. compared data directly from
lexical and nonlexical tasks, the first time that such a comparison had been made. All of the
lexical data used in the first two analyses and portions of the nonlexical data set used by Hale
et al. (1987) were included in this analysis. The latter data set was restricted to experimental
conditions with RTs that approximated the range of RTs in the lexical data set. The points for
the nonlexical tasks were described by a power function that bore an extremely close
resemblance to the function derived by Hale et al. in their analysis of the larger data set
(O=1.63Y T M vs O-1.62y129). Approximately equal portions of the variance for the nonlexical
tasks were explained by linear and power functions (.908 vs .913; slope of 2.05; intercept of-
385 msec for the linear function). The slope of the linear function differed significantly from
the slopes derived for the lexical decision and lexical tasks, which did not differ, thereby
indicating that the general slowing associated with lexical processing is significantly less than
the general slowing associated with nonlexical processing. Lima et al. concluded that the
"...precise mathematical relationship between the latencies of older and younger adults within
each domain indicates that the amount of slowing is general across the different experimental
tasks and conditions within that domain.." and "...the degree of slowing is domain-specific,
with the amount of slowing in the lexical domain being less than that in the nonlexical domain."
(p.422)
On the basis of these findings, Lima et al. suggested that Cerella's (1985a) two-factor
model (different slowing functions for sensorimotor and central processing) be elaborated to
include different levels of central slowing associated with at least two different types of
cognitive processing domains, lexical and nonlexical. Further support for this elaboration is
found in a study by Myerson, Ferraro, Hale, and Lima (1992) in which it was demonstrated
that slowing functions closely resembling the slowing functions reported by Lima et al. for the
lexical tasks characterize aggregated studies on semantic priming effects in lexical decision and
word naming tasks. These two meta-analytic studies provide relatively strong support for the
existence of a general lexical slowing mechanism that differs from a general nonlexical slowing
mechanism.
The importance of this distinction has been underscored in another study by this group.
Hale et al. (1991) compared the relative precision in estimating the slowing function in the
nonlexical domain from a series of four RT tasks done by the same groups of older and
younger subjects with that of aggregating data from different subjects across different tasks.
The fundamental intent of this comparison was to test the general and local slowing models.
The former predicts that the slowing estimate will be more precise because between-subject
variability is eliminated when every task is done by every subject; the latter predicts that the
slowing estimate will not be more precise because the variability in subject performance is
associated with inherent variability in the different components engaged by the various tasks.
Another goal of this study was to assess the variety of mathematical models, both linear and
nonlinear, proposed to explain the slowing phenomenon (the single-parameter linear model, the
two-factor linear model, the Overhead Model, and two variants of the Information-Loss
Model). Support was found by Hale et al. for the general slowing model; the within-subjects
design added considerably more precision to the slowing estimate (about 3 times more). Hale
et al. also found that the nonlinear models (overhead, information-loss) explained the general
slowing evident in the nonlexical domain better than the linear models. This pattern of results
reinforces the distinction between lexical and nonlexical slowing, suggesting that the slowing is
not only quantitatively different, but also qualitatively different, in the two domains (for a
review of their work see Myerson and Hale, 1993).
1.4. Holes in the a r m o r
Identification of two distinct cognitive domains with different slowing functions

challenges the strong version of the Bitten Hypothesis. It suggests that, at the very least, there
are two substantive aspects of cognition that age differently, one proportionally and the other
exponentially. There is a body of literature that has the seeds of a challenge to the weak
version of the Birren hypothesis as well. Madden (1992) has demonstrated, for example, that
the slowing evident among the elderly when making lexical decisions may have both
generalized and local components, and that some elements of this decision-making process may
not slow at all with age. Evidence for generalized slowing was found in the observation that
linear functions, with systematically increasing slopes of 1.06, 1.34, 1.41, 1.47, and 1.56,
characterized the Brinley plots for the decades of the 30s, 40s, 50s, 60s, and 70s, respectively
(regressed against subjects in their 20s) across the tasks used in the study. Different slowing
functions were revealed, however, for degraded as opposed to nondegraded target stimuli (i.e.,
the words). Intact targets and targets with a single blank separating each letter in the target
had a 4 msec per year rate of slowing, whereas degraded targets (with an * separating each
letter) had a 10 msec per year rate of slowing. This differential rate of slowing suggests that
the early feature extraction components of stimulus processing may be more vulnerable to the
effects of aging than other, later components of this processing. Madden also failed to find an
absolute increase in the magnitude of benefit to RT by a prime or in the RT difference between
words and nonwords across the adult lifespan, as would be predicted by a generalized slowing
model. This failure stands in contrast to the finding in the meta-analysis by Myerson et al.
(1992) that priming effects approximated the proportional decline seen in other lexical decision
and lexical tasks (around 1.5). Myerson et al. (1992) argued, however, that this difference
may reflect important sources of variability in priming studies investigating aging effects:
variations in the range of primes used, the number of trials run, and the number of subjects
tested.
That different aspects of lexical processing may be differentially sensitive to the effects
of aging is suggested further in a series of experiments by Madden, Pierce, and Allen (1993).
In a series of four experiments, they assessed the degree to which semantic priming effects in
visual word recognition, revealed in lexical decision and word pronunciation tasks, are
influenced in older subjects by the timing of the prime relative to the target. In these
experiments, the onset asynchronies of different types of primes and the target (stimulus onset
asynchrony, SOA) and the response output (manual vs vocal) were varied. In the aggregate,
these experiments revealed no age differences in the time course of semantic priming,
suggesting to Madden et al. that "...semantic activation processes are exempt from age-related
slowing." (p. 503) Similar results have been reported by Madden (1989) and Burke, White,
and Diaz (1987), but not by Howard, Shaw, and Heisey (1986) who found age influences on
semantic activation.
Madden et al. (1993) tempered their conclusions, however, by pointing out that the
reliability of the priming effects was "disappointingly low", the highest being .35, thereby
precluding strong inferences. Myerson et al. (1992) argued that this low reliability results t~om
the fact that priming effects are based on difference scores (e.g., RT to the target word
following an unrelated prime minus RT to the target word following a related prime), which
are much less reliable than the mean latencies used to derive them However, Madden et al.
(1993) derived a large number (11) of other difference scores from a variety of conditions in
their experimental series that yielded reliability coefficients ranging l~om .60 to .95. These
values suggest that the use of difference scores is not entirely responsible for the unreliable
findings. If not, Madden et al. argued that the source of the low reliability may lie in the
component processes themselves that mediate semantic priming. These processes may be
activated less reliably than the component processes engaged by other aspects of lexical tasks
and this low reliability may produce variations in the patterns of results across studies.
In a series of Brinley analyses, Madden et al. (1993) found only partial support for
generalized slowing. When all of the tasks were included in the analysis, a linear function that
closely resembled the functions reported by Lima et al. (1991) and Myerson et al. (1992) in
their meta-analyses described the spread of points (r2=.91, slope=l.58, intercept =- 183 msec).
As reported by Myerson r al. (1992), this function did not differ between the lexical decision
and word pronunciation tasks or as a function of the type of prime (related, unrelated, neutral).
However, the slowing function did differ with variations in SOA and target type (word,
nonword). A generalized slowing function characterized the data for word targets but not for
nonword targets, and the properties of the slowing function varied with SOAs above or below
200 msec. The functions were linear in both instances (<200 msec, r2=.94; >200 msec,
r2=.91). However, the function derived for SOAs less than 200 msec had a slope of 1.81 and
an intercept of-334 msec, whereas the function for SOAs greater than 200 msec had a slope of
1.39 and intercept of-64 msec. Madden at al. reasoned that these differences may reflect
differences in the "psychological refractory period" between the young and the old; that is,
older individuals may require a longer period of time to respond optimally to successive
stimuli. Thus, the task is more difficult when the SOAs are short, and this is revealed in the
steeper slope of the regression function. Madden et al. (1993) also found evidence against
generalized slowing in Brinley analyses of the priming effects (cost, benefit): regression
functions with coefficients of determination around. 15 that could not be explained by outliers
in the data set (see Myerson et al., 1992, for a discussion of the influence of outliers).
In earlier work, Madden, Pierce, and Allen (1992) introduced a technique for
transforming young response latencies from the output of the Brinley regression analysis. This
method attempts to identify age-related changes in processing speed that exceed those
associated with generalized slowing. It involves converting the raw scores of the young
subject by the slowing fimction (using both the slope and intercept values) obtained in the
regression analysis and then submitting these values to an ANOVA along with the
untransformed scores of the older subjects. Madden et al. (1992) reasoned that any Age x
Task Condition interactions obtained with this data set reveal effects whose magnitude is
greater than the generalized slowing effect. To digress a bit, it is interesting to note that a
Brinley analysis on untransformed data they collected in an attention allocation task revealed a
linear function with a slope of 1.58, an intercept of-39 msec, and an r 2 of .88. This is a
general slowing function. However, when the data were transformed and re-analyzed, the
significant interaction (age x cue type) found in the analysis of the untransformed data
remained. Persistence of this interaction in the transformed data suggests task-specific
differences between the old and the young. In this instance, the older subjects appeared to
benefit more from the cue than did the young subjects. Returning to Madden et al. (1993), 6
of 11 si,~nificant interactions remained after their data were transformed, the most meaningful
of which suggested that age produced a disproportionate effect on response selection
processes. Of further interest are their observations that (1) older subjects had longer RTs to
nonword targets than did younger subjects, and (2) older subjects had faster vocal than manual
RTs (87 msec), but the opposite was true for younger subjects (manual RTs were 14 msec
faster than vocal RTs). The nonword effect was eliminated when the data were transformed,
but the vocal-manual RT difference was not. The latter suggests that the response systems
mediating the two outputs differ in their sensitivity to aging.
The general thrust of the results originating from work in Madden's laboratory is that
the magnitude of age-induced slowing within the lexical domain may vary with differences in
task demands; that is, variations in the difficulty of encoding the target word (via degrading)
and in the time permitted for the prime to have an influence (via manipulations of SOA) may
alter the slowing effect. In addition, their introduction of transformed young latencies into the
Brinley analysis suggests that task-specific slowing may be superimposed on general slowing.
Their work to date indicates that these local slowing effects may be most evident at early
stimulus and late response ends of processing, at least for visual word identification (see Simon
& Pouraghabaghar, 1978, for evidence that stimulus encoding, but not response selection,
slows with age in a choice reaction). Indeed, Madden (1992) asserted that "...in the case of
visual word identification, relatively peripheral encoding and response processes are more
vulnerable to age-related decline than the more central processes involved in the activation and
retrieval of semantic information." (p. 505; see also Allen, Madden, Weber, & Groth, 1993; for
the differential influence of skill levels on the sensorimotor and computational elements of
cognitive slowing see Allen, Ashcrafi, & Weber, 1992; Chamess, 1987; Salthouse, 1984;
criticisms of general slowing can also be found in Baron & Matilla, 1989; Hartley, 1992;
Hertzog, 1992; Hertzog, Raskind, & Cannon, 1986). The work from Madden's group also
suggests that some processes, in this case semantic activation, may not decline with age; and
that the benefits of cuing in some forms of attention allocation may actually be larger for older
than for young adults.
The work by Madden and his colleagues has been devoted to assessing age-associated
factor effects across multiple conditions within one or a small number of experimental tasks. A
recent recta-analytic study by Laver and Burke (1993) likewise raises questions about both
general and domain-specific flowing. They evaluated the influence of age on semantic priming
in lexical decision and word pronunciation tasks in a meta-analysis that included 15 studies (10
from the Lima et al., 1991, study) and 49 experimental conditions. The priming effect was
found to be significantly greater for the old than for the young. A Brinley analysis of this effect
yielded a linear function with a slope of 1.0, a positive intercept, and an r 2 of only .69.
Similarly, a Brinley analysis of RT (for combined related and unrelated primes) revealed a
linear function with a slope of 1.01, a positive intercept of 178 msec, and an r 2 of .78. This
function differs substantially from that reported by Lima et al. (r2=.94, slope=l.52, intercept--
93 msec).
In an attempt to explain these discrepant functions, Laver and Burke completed a series
of Brinley analyses on subsets of their data set that were among the studies included in the
Lima et al. analysis. Linear regression functions were derived in two of the analyses that
differed considerably from those reported by Lima et al.: word targets in 7 lexical decision
studies, with 46 conditions (r 2 of .83, slope of 1.16 that did not differ from 1.0, and an
intercept of 152 msec); word responses in 10 lexical decision studies, with 57 conditions (r 2 of
.88, slope of 1.20, and intercept of 98 msec). However, the last analysis exposed the source of
the difference in the two analyses: Lima et al. used surrogate data for a group of young
subjects in one of the studies in their meta-analysis. When Laver and Burke did likewise, the
meta-analysis yielded the same regression function reported by Lima et al. On the basis of this
observed instability in the Brinley functions, Laver and Burke concluded that slowing estimates
based on meta-analyses are of questionable reliability and vary so widely as to suggest that
even within a particular domain slowing is process-specific.
Support for this conclusion is found in a study by Smulders (1993). Young (mean
age=24) and older (mean age=68) subjects made speeded choice reactions to stimuli that were
either intact or degraded. Subjects responded to two different types of stimuli, digits and
mmmmmmmmm mmmmmamnm
9 9 9 9
., , , . . , : 9 : .... :
Digits: Intact i ."! i :'". :
9 9 9 9 9 9
9 mmmmm 9
, ....... : : .... :
mmemmmnm 9
mum in 9 mm 9 m iBm
9, ,,,,; :; mmm 9
. . . . . ."'in.,
Digits: Degraded 9 mm
: ,,
man9 9
.:
mann-mum
:., :.:
m an
man, m e n
mm 9 mm 9 mmmmm 9
mare 9 9 9 am,. 9 9 mm
Words: Intact
-:: i "i i i":1 " , 1 1 "i "i1 1 , . i9
11 . . . . . . . . . .
9 9 en Hi
"":'
1 . 9 1 4 9 1 4 9 1'"'""
49 :,:i
mum
nnmnnmnnmmmmmmmmnummmmmmm 9149149 mannnalmm 9149149149149149
9 ann ana i m amaim | ale In a n aNanaid iN mma 9 inn ami t 9 modimamamaim ilia IBII | memml m amalm amamalm i i ida 9 am
[I,,I,l,[[I,,ll,l,.',,I,, " . . . . ." l.'''l'''.':.'O''.'_'l'l''l'l''',l

i 9149 9 9 . l9 9 , . . ,9 i9 l ..... 9 9 l:
Words: Degraded ...
""" ! ...'.r.i.....
;9 "'9 : "" "i .".."
. . . . . .".'.'q:.
. . . . . i.'"':i
9 ,.. 9149149 9 ,,;l , 9149 ,,,;, 9 , 9 1 4 9 91 14 49 9 1 4 9 1 499 1;4' 9
::::::::::::::::::::::::::::: :. ,... .......
9 milalalmim a i m a i m m i n 9 aimmmm 9
9. . . . ,:
m e n me
Figure 2. The stimulus sets used by Smulders (1993).
words, in separate blocks of trials. The digits 2 and 5 signaled responses with the left and fight
index finger, respectively; the words LINK and RECH (derivations of Dutch for "left" and
"right") signaled responses with the left and right index finger, respectively. The subjects were
shown the stimuli on a monitor at a viewing distance that was either short (80 cm) or long
(160 cm). Figure 2 shows the stimulus sets. It is apparent that the degree of degradation was
comparable for the two types of stimuli. Figure 3A depicts the factor (Age, Stimulus Quality,
Stimulus Type, and Viewing Distance) effects on mean RT. For young subjects, degradation
of the digits and words produced roughly equivalent slowing effects, particularly at the long
viewing distance. For the older subjects, the delay in response latency induced by degrading
the digits was similar in magnitude to that evident for the young subjects. However, as is very
evident, the effect of degrading the words was dramatically larger in magnitude among the
older subjects. These data were subjected to a Brinley analysis, the results of which are shown
in Figure 3B. The distribution of points was described by a linear function (r2=.90) with a
slope of 2.0 and an intercept of-342 msec. Because the ANOVAs revealed a particular
sensitivity among the older subjects to degradation of the words, separate Brinley analyses
were done on RTs for the two types of stimuli. These plots are shown in Figure 3C and D. In
both cases, the points were fit by linear functions (digits, r2=.92; words, r2=.92). However, the
slopes and the intercepts of the two functions differed considerably. The regression function
for digits had a slope of 1.40 and an intercept of-97 msec (Figure 3C); whereas the function
for words had a slope of 2.24 and an intercept of-465 msec (Figure 3D). Here, then, we have
an example of an experimental manipulation, stimulus degradation, that is thought to influence
the early elements of stimulus processing, producing dramatically different effects in older, but
not in young, subjects when the complexity of the stimulus was increased. The Brinley analysis
of the aggregated data sets concealed this difference, however. Indeed, visual examination of
the Brinley plot suggests no such differential factor effect.
404 Do general slowing functions mask local slowing effects ?
~600
L-
700
650 A 650"
600"
B7
t.
=_
550
500 ~ 550"
500'
450 ~All data:
450" y = 2.0 x - 341.6; r2 = .90

400
Intact Degraded Intact Degraded
80 cm 160 cm ,L
400
401 4.~o .~oo 55o 6o0 65o
RT Young (msec)
Q
650 650.
6O0 600.
~. 550 ~550.
"" 500 ~500.

I9 I. nl ~ Words only:
450 y = 1.4 x - 96.89; r2 = .92 450. / y = 2.24 x - 464.87; r2 = .92
400 400
400 450 500 550 600 650 400 450 500 550 6oo 65o
RT Young (reset) RT Young (msec)
Figure 3. A. The factor effects on RT from Smulders (1993). The unfilled circles and triangles indicate the
response latencies for young subjects to digits and words, respectively. The filled circles and triangles indicate
the response latencies for older subjects to digits and words, respectively. B. Brinley analysis that includes all
of the data points. C. Brinleyanalysis for the digit stimulus set. D. Brinleyanalysisfor the word stimulus set.
In the body of work reviewed in this section, evidence has been adduced that raises
questions about the general slowing model and the analytic procedure on which it rests.
Despite the obvious importance to psychological theory of articulating the effects of growing
older on information processing speed, it has only been within the past few years that a serious
dialogue has developed among investigators in the field with opposing viewpoints. We now
turn to the extant debate between proponents and opponents of the Brinley analysis and its
theoretical yield, generalized slowing.
2. THE DEBATE
Little interest was expressed in Brinley~s (1965) approach to studying cognitive

slowing, despite the exemplary work of Salthouse, Cerella, Hale and Myerson, until the last
few years. This interest has been fueled in large measure by the work of Cerella and of Hale
and Myerson. Their work has been sufficiently compelling to encourage other investigators to
give this analytic approach serious consideration. A critical look at the Brinley method vis-a-
vis the traditional ANOVA approach began most earnestly after Cerella (1991) commented on
a study by Fisk and Rogers (1991). This study examined age-related differences in the effects
of practice on visual and memory search tasks in a series of three experiments. Fisk and
Rogers found that after extensive practice under consistent mapping conditions performance
differences persisted among the two age groups on visual search but not on memory search,
whereas practice under varied mapping conditions produced no differential effects of aging.
Fisk and Rogers concluded that global slowing models cannot accommodate these task-
dependent effects. Cerella (1991) re-analyzed the data from Fisk and Rogers using the Brinley
method and followed this with a Monte Carlo simulation of these data that was likewise
submitted to a Brinley analysis. Both analyses revealed sets of simple linear regression
functions with slopes exceeding 1.0 that accounted for more than 90% of the variance. Cerella
argued that these results support the conclusion that there is global, task-independent slowing
among the elderly. Indeed, Cerella's re-analysis of the findings from the first experiment in the
series suggested that the initial performance of the older subjects on the visual search task was
actually worse than their final performance would have predicted. From this, he argued that
the elderly may have a "...particular facility for consistent visual search..." (p.222) because their
performance improved more with training than would have been expected. This conclusion
stands in dramatic contrast to the conclusion of Fisk and Rogers.
Perhaps, though, the most incendiary conclusion by Cerella was directed at the
ANOVA approach used by Fisk and Rogers, the preferred approach by most investigators in
cognitive aging research. Cerella argued that this approach assumes that age produces a main
effect and that this main effect is of less theoretical interest than are age x task interactions.
The latter are thought to reveal exceptional, rather than common, effects of age and therefore
are thought to provide greater insight into the aging process. The main effect is, then,
according to Cerella, treated as a null effect against which interactive effects are interpreted.
Cerella argued, however, that this disregard of the main effect of age is "...implicit
acknowledgment of a global deficit, a baseline against which exceptional losses are
distinguished." (p. 214) He argued fimher that this reasoning assumes an additive effect of age
for any individual task, thereby permitting only an additive global deficit in processing speed
among the elderly. This is not the case, according to Cerella, when the Bfinley method is
applied to a distribution of points in XY space. This curve-fiRing procedure permits a wide
variety of global deficits to be revealed and evaluated besides the additive. The futility of the
ANOVA approach is expressed, Cerella argued, quite clearly when age-related slowing is
proportional. When it is, age x task interactions are guaranteed and post hoc testing leads
"...to nowhere" (p. 222) because the absolute magnitude of the change in performance from
one task to another in a pair by the young subjects determines whether or not an interaction
will be significant (as in the pure search conditions for visual and memory search in the first
two experiments of Fisk and Rogers).
Cerella (1991) offered a relatively damning criticism of the ANOVA approach in
closing his commentary, arguing that this presumably atheoretical statistical procedure actually
embraces a theoretical position: normal aging adds a uniform slowing constant to response
latency. He closed his comment by asserting:
"Because the model does not comply with the data, it flags countless conditions
as exceptions in need of further explanation. Clearly the identification of
exceptions is a worthwhile pursuit that may enhance the understanding of aging;
equally clear is the value of first refining the baseline to extract a maximum
amount of variance in Y that is shared across conditions. That pursuit is better
conducted by curve-fitting in XY space than by siring ANOVAs for statistically
si~ificant effects. Given the success of the curve-fitting exercises, it remains to

be seen whether any conditions will be left over requiring process-specific
explanations. Generalized slowing may be the whole gory." (p. 223)
Fisk, Fisher, and Rogers (1991) answered these criticisms by arguing that a global view
of age-rdated slowing neglects theoretically important task-specific effects of aging on mental
processing speed. They argued for a model that incorporates the relative influences of
additive, multiplicative, and task-specific factors on cognitive slowing (an Interactive Model).
In their reply, Fisk et al. demonstrated that the amount of variance explained in the Brinley
analysis may vary importantly with the overlap in response latencies between the young and the
old (less overlap, more variance explained). They then fit the interactive and what they called
the independent (i.e., global) models to data from the memory search task reported by Fisk and
Rogers (1991). The fit was excellent for the Interactive Model but quite poor for the
Independent Model (which accounted for less than 10% of the variance). Fisk et al. concluded
their reply by arguing that research on cognitive slowing should be theory-driven and the
strong theoretical base in cognitive psychology exploited to develop task-specific slowing
slowing hypotheses to guide these research efforts.
This debate has now grown to include other participants. In a recent edition of the
Journal of Gerontology: Psychological Sciences, the contrasting analytic and theoretical
perspectives were argued by Perfect (1994)and Fisk and Fisher (1994) on the one hand, and
by Cerdla (1994) and Myerson et al. (1994) 3on the other hand. In the lead article of this
series, Perfect (1994) demonstrated through simulation that the amount of variance explained
by the linear regression varied inversely with the degree of slowing and with the amount of
overlap in the response latencies. Thus, the likelihood of explaining the least amount of the
variance in a Brinley analysis is highest, he claimed, under those conditions that are least likely
to occur in the literature: a large slowing effect and total overlap of the response latencies. Of
particular concern to Perfect were his observations (on both simulated and actual data sets)
that aggregated data sets may obscure fundamental differences among individual data sets.
Thus, like Fisk et al. (1991), the core element of Perfect's (1994) criticism of the
Brinley analysis is that the amount of variance it explained was tied very closely to the amount
of overlap in the response latencies among the old and the young on two comparison tasks (the
higher the overlap, the lower the amount of explained variance). Myerson et al. (1994)
countered this argument by demonstrating that the data sets used in the simulations by these
investigators had a si~ificant confound: as the overlap decreased the range of latencies
increased. Elimination of this confound by Myerson et al. from a hypothetical data set,
generated in accord with the specifications of Fisk et al., produced nonoverlapping response
latencies for young subjects on two tasks that yielded the lowest r 2 values. This result suggests
that a high r E in the Brinley analysis is not the inevitable outcome of a lack of overlap in the
RTs of the young subjects. Rather, the crucial variable in determining the size of r 2 is
maintenance of the relationship between the magnitude of the response latencies of the young
adults and the size of the age difference in the two tasks being compared. If it is maintained,
then r: will be high; if it is not, r E will be low. Myerson et al. then demonstrated that violation
of the fundamental assumption of the slowing hypothesis, the size of the absolute difference in
RT increases with increases in task complexity, produces a distribution of points in XY space
that reduces the explained variance in the Brinley analysis. Thus, violation of this assumption
yields precisely what general slowing anticipates--an absence of ordering in the dispersion of
data points. . . . . .
Myerson et al. argued further that both Fisk et al. and Perfect (1) failed to represent
their hypothetical data sets in graphical form, and (2) performed the regression analysis in
isolation (i.e., without taking advantage of other regression techniques). They represented the
data from both simulations in graphical form and, in so doing, made it obvious that both sets of
data could not be explained by a single slowing function. Moreover, they demonstrated that
the mix of regression functions in both the Fisk et al. and Perfect data sets could be identified
using another regression technique, indicator variables. They then demonstrated that separate
regression functions taken from Lima et al. (1991) fit the data from Experiments 1 (letter
search) and 2 (word search) by Fisk and Rogers (1991) (collapsed across test sessions 1-3
versus test sessions 4-6 practice). These two functions were consistent with the conclusion
drawn by Lima et al. that different, but general, slowing functions characterize the data in the
nonlexical (Experiment 1, letter) and lexical (Experiment 2, word) domains. Myerson et al.
also addressed the concern of Fisk et al. that since standard regression analysis assumes no
measurement error in the independent variable it is not appropriate for RT data, which does
have such error, as the independent variable. Myerson et al. countered that this measurement
error reduces r 2, making the consistently reported values of .90 or greater even more
impressive. The fundamental message conveyed by Myerson et al. is that rather than avoiding
regression techniques in cognitive aging research they should be used in conjunction with other
analytic procedures on data collected from well-designed experiments.
In his response to Perfect (1994), Cerella (1994) completed a series of simulations to
demonstrate that only a small, circumscribed subset of Brinley plots (those that are monotonic)
can be fit by a single, slowing factor. These simulations showed that when different rate
parameters apply to the processing speeds of the young and the old, the resulting Brinley plots
are primarily non-monotonic. Importantly, he demonstrated that as task complexity increases
the likelihood of producing a monotonic Brinley function diminishes. That is, as the likelihood
increases that the cognitive processing of the old is governed by different mechanisms than that
of the young, the probability of a monotonic Brinley function characterizing the data reduces.
However, as Cerella (1990) had argued earlier, if the basic mechanisms mediating information
processing are the same in the young and the old (i.e., his Correspondence Axiom is met), then
the likelihood of the Brinley analysis yielding a monotonic function is increased substantially.
From Cerella's perspective, then, the probability of a set of aggregated data points being
described by a monotonic function is low; therefore, findings like those in which quantitative
outcomes conform closely to theoretical specifications offer strong support for the theory in
question, generalized slowing (e.g., see Hale et al., 1991).
Despite the obvious disagreements among the disputants in this debate, Fisk and Fisher
(1994) identified several fundamental points of agreement among them: (1) there is consensus
that r 2 by itself cannot always discriminate general from specific theories of aging; and (2)
there is the shared view that at least two distinct domains of slowing exist, lexical and
nonlexical, which ipso facto eliminates an all-encompassing model of slowing. In their view,
many other aspects of the debate have reflected misunderstandings rather than fundamental
disagreements and, therefore, have been obfuscating rather than clarifying. Examples of these
misunderstandings are found in the dispute over what constitutes a Brinley plot, the role of RT
overlap on r E, the insensitivity of Brinley plots to task-specific effects, and the value of plotting
data during the analytic process. The issue of fundamental concern to Fisk and Fisher is the
extent to which research in cognitive aging is driven by theory. They argued that the
application of Brinley analyses to aging data is deficient in this regard, being in large measure
atheoretical and descriptive in nature. Thus, in their view, the selection of data sets for analysis
has not been done often enough on the basis of models of cognitive processing.
This criticism was leveled at Myerson et al. (1994)for collapsing across the first and
the last three sessions to derive mean RTs for the Brinley analysis described earlier. Fisk and
Fisher argued that there are strong theoretical reasons for not collapsing the data in this way.
Thus, according to these investigators, "...one can be led falsely to accept a model of general
slowing if one aggregates data across conditions (or sessions) without due consideration given
to models of performance in each of the various conditions."(p. P86) It should be noted,
however, that Myerson, Wagstafl~ and Hale (1994) have demonstrated that general slowing
functions taken from Lima et al. for the lexical and nonlexical domains also describe data from
single sessions in Fisk and Rogers (1991) for visual and memory search, respectively. Fisk and
Fisher argued further that if in a Brinley analysis identity is assumed in perfo~ance across
sessions within age, a process-specific slowing can be confused for general slowing, and if
performance is assumed to be identical across age within tasks, then a general slowing model
could be confused for a task-specific model. They reasoned that this problem is solved by
developing tests of the aging process that are derived from models of changes in performance
across test sessions and across age groups.
This debate and the research that generated it have served to frame the issues regarding
the two analytic approaches. Moreover, they have contributed to refining the conceptual space
in which hypotheses about cognitive aging are formulated, tested, and empirical results
interpreted. Investigators in cognitive aging now differentiate not only the strong from the
weak versions of the Complexity Hypothesis, global (or task-independent) from local (or task-
dependent) slowing, and general (i.e., global) from domain-specific (e.g., lexical vs nonlexical)
slowing, but they are also be~nning to distinguish task-specific (e.g., category membership vs
word naming) slowing from process-specific (e.g., stimulus encoding vs response selection)
slowing (see discussion in Fisk & Fisher, 1994). However, the extent to which these
distinctions can be elaborated may be limited if the analytic procedures depend exclusively on
RT measures. In the next section, we describe research in chronopsychophysiological aging
that may contribute to improving the precision with which these distinctions can be articulated.
3. A ROLE FOR CHRONOPSYCHOPHYSIOLOGY IN THE STUDY OF

NEUROCOGNITIVE AGING
Chronopsychophysiological analyses of age-related changes in mental processing speed

combine measures of the timing of components of the event-related brain potential (ERP),
measured at the scalp in humans, with RT. Reaction time represents the final output of what is
often a very complicated decision-making process. Thus, variations in RT express the
outcome of an aggregate of processes that occur between the presentation of a stimulus and
the execution of a response. Changes in the timing of components of the ERP, in contrast,
allow the transmission of information to be viewed as it occurred. However, the value of these
measures is questionable if experimental factor effects on component latency and RT always
covary. If so, nothing is gained by adding these measures to the assessment procedures. That
these electrophysiological measures do augment behavioral measures is revealed in
observations that factor effects on component latency and RT can be dissociated (i.e., they are
not always correlated). Moreover, some of these components can be elicited in the absence of
any overt behavior (see the chapter by Ridderinkhof & Bashore in this volUme for a brief
tutorial on ERPs).
A fundamentally important property of some components of the ERP is that factor
effects on their latency and on RT can be dissociated. For example,
chronopsychophysiological studies have revealed a dissociation of factor effects on P300
latency and RT in a variety of experimental tasks; namely, variations in stimulus processing
demands but not in response output demands influence P300 latency, whereas variations in
either of these processing demands influences RT (e.g., Brookhuis, Mulder, Mulder, Gloerich,
van Dellen, van der Meere, & Ellerman, 1981; Callaway, 1983; de Jong, Kok, & van Rooy,
1988; Duncan-Johnson & Donchin, 1982; Duncan-Johnson & Kopell, 1981; Kutas, McCarthy,
& Donchin, 1977; Magliero, Bashore, Coles, & Donchin, 1984; McCarthy & Donchin, 1981;
Mulder, Gloerich, Brookhuis, van Dellen, & Mulder 1984; Pfefferbaum, Christensen, Ford, &
Kopell, 1986; Ritter, Vaughan, & Simson, 1983; Smid, Mulder, & Mulder,, 1990). This
dissociation is especially important in studies of mental chronometry because it suggests that
the set of processes manifested by P300 latency are a subset of those manifested by RT. Thus,
the selective influence of experimental factors on the processes manifested in the P300 and
those manifested in RT can be used to articulate mental chronometric processes with more
precision than reliance on RT alone (see review in van der Molen, Bashore, Halliday, &
Callaway, 1991). If one of these factors is age, then more precise conclusions can be drawn
about its influence on the various components of information processing. Another important
property of P300 latency is that it, unlike RT, does not vary si~ificantly with changes in
speed/accuracy trade-offs (Coles, Gratton, Bashore, Eriksen, & Donchin, 1985; Kutas et al.,
1977; Pfefferbaum, Ford, Johnson, Wenegrat, & Kopell, 1983). Thus, differences in subject
strategy as it pertains to the relative importance of speed and accuracy have little effect on
P300 latency, while these differences can have large effects on RT. This is a particular concern
in studies of mental processing speed that compare the young and the old because the latter
tend to be more concerned about accuracy than the former.
3.1. Component latencies as indices of neurocognitive slowing
During the last 20 years a growing literature has demonstrated that the latencies of
some components of the ERP are prolonged significantly in the elderly (see reviews in
Bashore, 1990; Ford & Pfefferbaum, 1980, 1985; Miller, Bashore, Farwell, & Donchin, 1987;
Polich & Starr, 1984). The latencies of the P200, N200, and P300 components have been
reported to be slowed by about 35 msec, 40 msec, and 90 msec, respectively, in tasks that
require simple discriminations between target and non-target stimuli in the auditory, visual, and
somatosensory modalities (e.g., Brown, Marsh, and LaRue, 1983; Goodin, Squires &
Starr,1978; Goodin, Squires, Henderson, & Starr,1978; Pfefferbamn, Ford, Wenegrat, Roth &
Kopell, 1984; Picton, Stuss, Champagne, & Nelson, 1984; Polich, Howard, & Starr, 1985; see
review in Bashore, 1990). Of particular interest for investigators who study the mental
chronometry of aging are the observations that not only do some component latencies increase
with age, but the patterns of increase may not always covary with behavioral slowing. For
example, in Sternberg memory scanning tasks, behavioral studies have found that the
regression function for the response latencies (RT regressed on memory set size) of elderly
subjects has a higher intercept and a steeper slope than the function for young subjects (Anders
& Fozard, 1973; Anders, Fozard, & Lillyquist, 1972; Eriksen, Hamlin, & Daye, 1973; Madden,
1982; Madden & Nebes, 1980; Maniscalco & DeRosa, 1983; Salthouse & Somberg, 1982; but
see Houx, Vreeling, & Jolles, 1991). However, although in chronopsychophysiological studies
a similar function for RT has been reported, a different function has been observed for P300
latency: only the intercept is elevated (Ford, Pfefferbaum, Tinklenberg, & Kopell, 1982; Ford
et al., 1979; Marsh, 1975; Pfefferbaum, Ford, Roth, & Kopell, 1980; Pratt, Michalewski,
Patterson, & Starr, 1989a,b; Strayer, Wickens, & Braune, 1987). Thus, the slope of the
function for P300 latency is comparable for the young and the older subjects. This suggests, of
course, that serial comparison time (the time to compare the test set item to the memory set
items) is preserved in older individuals. The RT data suggest, in contrast, that all elements of
memory scanning are slowed in the aged.
This body of chronopsycholophysiological research on mental chronometry and aging
encouraged Bashore, Osm~n, & Heffley (19.89) to include measures of both P300 latency and
RT in a meta-analysis of changes in mental processing speed among the elderly, the results of
which are shown in Figure 4. Like Hale et al. (1987), the analysis only included tasks that
required simple motor responses. Unlike those investigators, but like Cerella, a mix of
experimental tasks was included. The spread of points for RT was described best by a
multiplicative function with a slope of 1.27, an intercept that did not differ from zero, and an r e
of .87 (Figure 4A). This function closely resembles the functions reported by Cerella,
Salthouse and their colleagues for response latency. The function derived from the RT data
differs from that of Hale et al. This may reflect, in part, differences in task domain (nonverbal
vs mixed), but it may also reflect differences in the range of RTs in the analysis. The response
latencies used by Hale et al. were as long as thirty seconds for older subjects and nine seconds
for young subjects, whereas those used by Bashore et al. did not exceed 1400 msec. These
values are located in the lower left quadrant of the larger space in which the latencies used by
Hale et al. were distributed. CereHa (1990) has noted that the latencies used in his earlier
analyses (Cerella, 1985a; Cerella et al., 1980) likewise fell into the lower left quadrant of this
larger space, and that this quadrant of the space in the data set for Hale et al. could be
described by a simple multiplicative function where the points suggestive of disproportionate
slowing were distributed in the upper right quadrant (and, as pointed out by Cerella, 1990,
were the consequence of one study that had decidedly longer latencies). It is important to note
that Lima et al. (1991), who we discussed earlier, restricted their analyses to a subset of the
latencies taken from Hale et al. that did not exceed 2.0 and 3.0 seconds for the young and
older subjects, respectively. Nonetheless, the points were described by a power function that
closely approximated the function reported by Hale et al. However, it must also be pointed
out that a linear, multiplicative function accounted for essentially the same amount of variance
for this smaller data set (.908 vs .913 for the power function).
Most informative, however, are the functions derived for the P300 latency data. The
first analysis included all of the experimental tasks, those that required an overt response as
well as those that did not (a P300 can be elicited in tasks that require only a covert response,
like mental counting). The function it yielded, like the liT function, was linear (r2-.91), but
unlike the RT function it had a slope that approximated 1.0 (0.95) and an intercept that was
elevated above zero (80 msec; Figure 4B). The second function was derived for P300 latency
from a data set comprised exclusively of tasks that required a manual response (i.e., were RT
tasks). Like the first P300 latency function, this function was linear (r2=0.91) with a slope that
did not differ from 1.0 (0.93) and an intercept that was additive (90 msec; Figure 4C). The last
REACTION TIME P300 LATENCY
9,,,~ ,~
I000- I0(0)'
o e
"y. - 9
m
0 0
500"
0 . . . . ! . . . . ! . . . .
0 500 I000 0 50O I000

Young Young
P300 LATENCY (RT TASKS) RT- P300
lOCK)' I000
"10
0 O
9 goo~
500" 500,
~ - - - 9 i . . . . i . . . .
0 500 I000 " s56" "t656" " "

Young Young
Figure 4. Brinley plots from Bashore et al. (1989). The upper left panel shows the Brinley plot for RT; the
upper right panel shows the function for P300 latency across all tasks; the lower left panel shows the regression
function for P300 latency in RT tasks; and the lower right panel shows the Brinley plot for RT-P300 latency.
From "Mental slowing in elderly persons: A cognitive psychophysiologicalanalysis" by Theodore R. Bashore,
Allen Osman, and Earle F. Heffiey, 1989, Psychology and Aging, 4, p. 241. Copyright 1989 by the American
Psychological Association. Reprinted by permission.
fimction was derived from difference scores (KT-P300 latency) that are thought by some to
represent the activation of response-related processes (e.g., Ford et al., 1979). This function
was linear (r2=.82) with a slope of 1.32 and an intercept of 50 msec (Figure 4D).
The results of the analysis on P300 latency challenge the logic underlying the original
KT meta-analysis. According to that reasoning, only peripheral effects produce a function with
a positive intercept and a slope of 1.0. Thus, this reasoning would support the erroneous
conclusion that the P300 originates from the peripheral nervous system Although its origins
have not been identified, there is consensus that it originates from structures in the cerebral
hemispheres (Knight, 1990; Wood, McCarthy, Squires, Vaughan, & McCallum, 1984). Thus,
the meta-analysis of P300 latency has produced a regression function that defies the logic on
which the original meta-analyses were based. Here it is important to note the distinction made
by cognitive aging theorists between peripheral (perceptuomotor in their terms) and central
(computational in their terms) components of information processing is best characterized as a
functional, not a classic structural, neuroanatomical distinction (nervous system structures
encased in bone, covered by meninges, and surrounded by cerebrospinal fluid are central;
nervous system structures outside these media are peripheral; Carpenter, 1991). Activation of
the neuroanatomically-distinguished peripheral nervous system probably only comprises about
30 to 40 msec of the total RT (Barker, 1991; Barker, Jalinous, & Freeston, 1985; Robinson &
Slimp, 1990). Consequently, most of the processing cognitive aging theorists have labeled as
perceptuomotor is probably taking place in the central, not in the peripheral nervous system
If so, the variation in the regression functions for P300 latency and RT suggest that although
the rate of mental processing speed' is generally slower among older persons, the speed of
some elements of central processing may be differentially sensitive to increases in processing
demands (i.e., those manifested by RT), whereas the speed of other processes may not be (i.e.,
those manifested by P300 latency).
This inference receives support from the chronopsychophysiological studies of memory
scanning described briefly earlier. These studies permitted more precise articulation of the
properties of the regression function than is possible using one dependent measure and, in so
doing, offered different insights into the effects of older age on memory scanning. The work
was based on the reports in the literature that revealed the relatively selective influences of
variations in stimulus processing on P300 latency, supporting the hypothesis that the latency of
this component manifests the engagement of that processing. From this research, the following
assumptions were made about the properties of the regression function for P300 latency in
memory scanning tasks: (1) The slope and intercept of the P300 latency-memory set size
function provide indices of serial comparison time and stimulus encoding time, respectively;
and (2) the slope and intercept of the (RT-P300 latency)-memory set size function reveal the
timing of binary decision and response translation and organization processes, respectively. In
these combined studies, older subjects were observed to have longer gTs across memory set
sizes than young subjects, and the regression function that described the relationship of
response latency to memory set size had higher intercepts and steeper slopes. This pattern
replicated the patterns found in RT studies. However, the P300 latency-memory set size
regression function departed from this pattern. Although the intercept of this function was
found to be elevated in the old relative to the young, the slopes of the function for the two
groups were comparable. As was indicated earlier, Ford et al. (1979) reasoned that the
interval between the peak of the P300 and the response provides an estimate of the relative
timing of response-related processes. To infer the effects of age on these processes, they
regressed this value against memory set size as well. The slope of this function was steeper
and the intercept was larger in the old than in the young.
This pattern of results argues against the inference that there is a generalized decline in
the rate of memory scanning among the elderly. Rather, following the reasoning of Ford et al.,
the results from these combined studies suggest that (1) serial comparison time (as revealed in
the P300 latency slope) does~not increase with age; (2) stimulus encoding processes (as
revealed in the P300 latency intercept) are somewhat flower in the old than in the young; (3)
response translation and organization processes (as revealed in the RT-P300 latency intercept)
are much flower in the old than in the young; and (4) the old are less confident than the young
in their response selection (binary) decisions as task difficulty increases (RT-P300 latency
slope). Similar conclusions were drawn by Strayer et aL (1987) who augmented the regression
analyses on RT and P300 latency with speed-accuracy trade-off analyses on RT.
A recent study by Smulders, Kenemans, and Kok (in submission) complements the
work on age-related effects on memory scanning. These investigators used the same word
stimuli (LINK, RECH) as did Smulders (1993), who we discussed earlier. Recall that he
required young and older subjects to make two-choice reactions to digits or words that were
either intact or degraded (see Figure 2); and found that degradation of digits produced
comparable slowing among the two age groups, but that degradation of words slowed older
subjects substantially more than young subjects. Smulders et al. required their subjects to
respond to intact or degraded presentations of LINK or RECH with either a compatible (e.g.,
left index finger to LINK) or an incompat~le (e.g., fight index finger to LINK) movement.
They found that the effects of stimulus degradation and S-R compatibility were additive, and
that the effects of S-R compatibility were larger in the older than in the young subjects. The
latter suggests, like much of the literature we have reviewed, that the performance of older
adults is more sensitive than that of young adults to variations in the ease with which a
response can be selected. Of most interest for our purposes, however, are the effects of
stimulus degradation on RT and P300 latency. Reaction times were longer for the elderly than
for the young, and for degraded as opposed to intact stimuli (see Figure 5). Moreover, the
slowing in RT produced by degradation was larger in the elderly than in the young. Like RT,
P300 latency was longer in the elderly and presentation of a degraded stimulus prolonged the
latency of this component more in older than in young subjects. The latter observation, that
the effect of degradation on P300 latency was greater among the elderly than the young,
departs l~om the results of the memory scanning studies described earlier. In these studies, the
effects of variations in memory set size were dissociated for P300 latency and RT (RT was
prolonged more in the old than in the young, but P300 latency was lengthened by the same
magnitude in both age groups). Thus, the effects on both RT and P300 latency found by
Smulders et al. suggest that there may be a processing-dependent deficit in the stimulus
encoding mechanisms of older adults that is tied, perhaps, to stimulus complexity. This pattern
Reaction time P300 latency
~ 5..
o
O Young (N=I2)
9 Old (N=il)
t,~ Dog. i.;. D4.
Figure 5. Factor effects on RT and P300 latencyfound by Smulders et al.
of results is consistent with the effects of degradation on visual word recognition reported by
Madden (1992), and it reinforces the conclusions of the chronopsychophysiological studies of
memory scanning using a direct, as opposed to an indirect, measure of stimulus encoding

processes (stimulus degradation vs an elevated intercept in the P300 latency-memory set size
regression fimction).
3.2. Unmasking local slowing effects hidden in a Brinley plot: A case study using ERPs
The research we have reviewed to this point, both t~om the cognitive psychological and
chronopsychophysiological literatures, suggests that the slowing of neurocognitive processing
induced by aging may be more complex than is revealed in the Brinley analyses of RT data.
Although the set of results produced using this approach with behavioral data is impressive, it
is important to keep in mind that RT is an aggregate measure that represents the final output of
a complicated series of information transactions. Hence, RT may be a reasonably gross
measure of mental chronometfic processes. Thus, our probe into the infrastructure of human
information processing may be sharpened if we augment RT measures with measures of the
latencies of components of the ERE This conclusion is certainly suggested in the
chronopsychophysiological studies we have discussed.
An illustration of how this combined methodology may deepen our insights into
neurocognitive aging is provided in a study by Zeef and Kok (1993). In this study, the focused
and divided attention abilities of subjects in their early 70s were compared with those of young
subjects in their early 20s. The focused attention task was a variant of a choice RT task
developed by Eriksen and Schultz (1979) in which a target stimulus is presented in an array
flanked on each side by 1 or more stimuli. In the typical variant of the task there are two
possible targets (e.g., H or N) that signal different directional responses (e.g., H signals a
button press with the left index finger; N signals a button press with the right index finger). On
any given trial, the target may be flanked either by itself (e.g., HHHHH) or by the other
stimulus (e.g., NNHNN). That is, there are arrays in which the flankers provide information
about the response that is either congruent (HHHHH) or incongruent (NNI-INN) with the
response signaled by the target. Subjects are instructed to ignore the flankers and to make
their response decisions exclusively on the bails of the identity of the target. It has been
demonstrated in numerous experiments that RT is slowed significantly when the target is
flanked by incongruent stimuli (e.g., Cerella, 1985b; Coles et al., 1985; Gratton, Coles,
Sirevaag, Eriksen, & Donchin, 1988; Eriksen & Schultz, 1979; Eriksen, Coles, Morris, &
O'Hara, 1985; Smid, Mulder, & Mulder, 1990; Wright & Elias, 1979). The examples
presented above comprise the stimulus set and responses used by Zeef and Kok.
In the divided-attention task used by Zeef and Kok, the target letter, H or N, was
presented randomly in the second or fourth position of a five-dement array. In addition to an
H or an N, the array always contained the letters FXKZ that were presented in different
combinations and never served as targets. In this task, spatial compatibility was varied in
association with the location of the target letter relative to the response hand. For example, if
H signaled a button press by the left index finger, the spatial S-R mapping was compatible
when it appeared in the second position (FHXKZ) and incompatible when it appeared in the
fourth position (FXKHZ). In addition, subjects either did or did not receive a cue indicating
the location of the target element. In the uncued condition, an asterisk appeared above the
center element in the array; in the cued condition, the asterisk appeared above the target
element. In both cases, the asterisk was presented simultaneously with the stimulus array. The
presence or absence of a cue was varied between blocks of trials.
Because of their interest in fractionating the processing engaged by these two tasks,
Zeef and Kok measured a variety of electrophysiological signals in conjunction with response
speed and accuracy. To infer the timing of response system activation they measured the onset
latency of the lateralized readiness potential (LRP) and of the electromyogram (EMG), and to
infer the engagement of stimulus processing they measured P300 latency. Two different LRPs
were derived: one time-locked to the stimulus input and the other to the response output. The
former permits inferences to be drawn about the influences of stimulus processing on response
system activation, and the latter permits influences that are unique to the response output to be
inferred. Thus, there were five dependent variables measuring information transmission: onset
latency of the EMG; onset latency of the stimulus-locked LRP; onset latency of the response-
locked LRP; peak latency of the P300; and response latency (RT).
7ooi- . . . . . . . . . . i
65O
_. 6oo I
55o1
~: 500| o S.Iocked LRP R-locked LRP
450| 9 R-locked LRP
450| , I t " - [ 9 EMG
~ 4 o o I- ~ &P300
g 4OO ~
__~ 3501
=ool
250L --
congruent
,.
~ EMG
incongruent
RT :::t "young old
t =
7501. . . . . . . . . . . 1got . . . . . . . . ) "

7001 ~ ~........-.-~ la~ D
E" 65ol C ~" 17o1
sool ~ Z 15Ol
i 550| o R.Iocked LRP/young "" 150t / .,~ ~, EMG- ,%.locked LRP
500,i 9 R-locked LRP I old 1401 f ~ 9 EMG- R-locked LRP
4501 ~ A RT/young .~ 130|
.~ 4ool 9 RT / old 12Ol
"0 t ~ 1'~ . /
300" 1O0~-- "- old
congruent incongruent young
Figure 6. Summary of data from the focused attention task by Zeef & Kok (1993). A. Main effects of
variations in Stimulus Congruency on the dependent measures. B. Main effects of Age on the dependent
measures. C. The Age x Stimulus Congruencyinteraction. D. Main effects of Age on the difference measures,
EMG-LRP onset for the stimulus- and the response-locked LRPs (filled triangle--response-locked; unfilled
triangle--stimulus-locked).
Figure 6A shows the (statistically significant) effects of varying the flankers on

transmission latencies in the focused attention task. As can be seen, there was an increase in
latency for each measure when the fl~nkers were incongruent with the target. Comparison of
the two age groups, depicted in Figure 6B, reveals that each of the transmission latencies was
longer in the older than in the young adults. However, age only had differential slowing effects
on the onset latency of the response-locked LRP and RT. This pattern of interactions is shown
in Figure 6C. The time from the onset of the LRP to the onset of EMG activation is thought
to represent the time taken from central response selection and preparation (onset of the LRP)
to peripheral activation of the motor response (onset ofthe EMG). In Figure 6D it can be seen
that this time was longer in the older than in the young subjects. The pattern of results from
this task suggests that although information is transmitted more slowly in older than in the
younger nervous system, the most dramatic slowing is at the response end of processing.
Stimulus processing, as revealed in P300 latency and the onset latency of the stimulus-locked
LRP, is slowed among the elderly, but the relative magnitude of this slowing is less than that
evidenced by the response processing components.
-ol,
S-locked LRP
t /i t i
9 It-locked LRP
P'JO0
9 RT
compatible Incompatible cued uncued
ss~ . . . . ,
Fl4ockecl LRP ~ $40 A " / P300/young/cued

9 EMG 9 9 113OOI young I uncued
3 s~ RT j S~200 P300, o l d , cued
P'J001 old I uncued
350 -
young old compabble Incompatible
220[ . . . .
~"~1 /
~
IlK) O EMG - S-locked LRP
140
lOO 1 o r
80 i _ _
young old
Figure 7. Summaryof data from the divided attention task by Zeef & Kok (1993). A. Main effect of S-R
Compatibility on the dependent measures. B. Main effect of Cuing on the dependent measures. C. Main effect
of Age on the dependent measures. D. Three-wayinteraction. E. The effect of Age on the difference measure,
EMG-LRP for the stimulus-lockedLRP.
F u l l e r support for this conclusion is found in the divided attention task. In this task,
we see that variations in spatial S-K compatibility influence KT, the onset latencies of the
stimulus- and response-locked LRPs, and the onset latency of EMG activation. These
relations are shown in Figure 7A. In contrast, P300 latency was not affected by these
variations. The presence or absence of a locational cue in the stimulus array influenced the
onset latencies of both the stimulus- and the response-locked LRPs, P300 latency, and RT, as
shown in Figure 7B. However, as is apparent in Figure 7C, older age was associated only with
slowing of the response-locked LKP, onset latency of the EMG, and P,T. There was an
interesting relationship expressed between age, type of cue, and compatibility: among younger
subjects, the absence of a locational cue shortened P300 latency when an incompatible
response was made; among older subjects, P300 latency was prolonged when an incompatible
response was made, irrespective of the presence or absence of an informative cue. This
interaction is shown in Figure 7D. The time between the onset latency of the stimulus-locked
LRP and the onset of EMG activation was longer in older than in young subjects, as revealed
in Figure 7E. Again, the overall pattern of results is consistent with the conclusion that the
most dramatic slowing among the elderly is experienced near the response end of processing.
That these differential effects may be obscured in the Brinley analysis is demonstrated
by our observation that regression of the old on the young latencies across all of the measures
produced a linear function (r2=.94), an intercept that did not differ from 0 (31 msec; t=l.61,
p=0.12), and a slope that was greater than 1.0 (1.07; t=25.82, p=0.00, see figure 8). This
function, according to the logic of the Brinley approach, supports the conclusion that slowing
is generalized, although quite modest, and equivalent across components of processing on this
set of attention tasks. This conclusion belies the pattern of results, of course, found by Zeef
and Kok. Here, then, we have a graphic example of how a Brildey analysis may obscure local
effects when the analysis of variance has revealed them to be consistently expressed across
multiple dependent measures.
00o . . . . . " " e " - "
20~y ffi 1.07X + 31.0; r2 ffi . 9 4

100~ . . . . . .
100 200 300 400 S00 f,00 700 800
young latency (reset)
Figure 8. The Brinley plot of the data from Zeef & Kok (1993).
4. CONCLUDING COMMENTS
The chronopsychophysiological data we have presented augment the cognitive data

that challenge the exclusivity of general slowing in explanation of age-related declines in
mental processing speed. Of interest is that both sets of data reveal some slowing in early
stimulus processing, but identify the response end of processing as the locus of the most
dramatic slowing. Of f u ~ e r interest is that a pattern of greater slowing near the response end
of processing may not be inconsistent with either the Information-Loss or Overhead models of
cognitive slowing. It is very important, however, not to lose fight of the fact that all of the
models that have been formulated to account for cognitive slowing were derived from analyses
of RT data collected within and across a wide range of tasks. Yet, all of the inferences that
derive from these analyses attempt to characterize the transmission of information from
stimulus input to response output within a reaction time. As we have argued at other points in
this chapter, KT is an aggregate measure that typically represents the final outcome of a
complicated decision-making process. Thus, the constituents of that transmission process may
be better characterized by measures that correspond to the activation of the components of this
transmission process. One way to achieve this end is, as we have also argued, to study the
effects of older age on component latencies in the ERP.
In a recent test of the Information-Loss Model ofMyerson et al. (1990), Molenaar and
van der Molen (1994) completed a simulation study in which both a general (i.e., all stages of
processing are slowed by age) and a specific (one stage of processing is slowed by age) version
of the model were compared in their ability to discriminate global and local age effects on KT.
The two models explained approximately equal amounts of variance in several different
simulations, suggesting that the power function on which the Information-Loss model is built
may not discriminate between global and local information loss. They demonstrated that a
specific loss may give rise to a large amount of explained variance, which suggests that local
effects may be easily obscured in the regression function. They then argued that
characterization of the slowing process may be refined by implementating Hohle's (1967)
multimethod procedure for assessing the selective influences of age on mental processing
speed. He argued that three analytic procedures should be combined. First, he advocated a
stage analysis of the reaction process using the subtraction method of Donders (1868). When
he made this recommendation, Steinberg (1969) had not yet described the additive factors
method for analyzing stage structure. This method was developed as an alternative to the
subtraction method and has proved to be a very powerful procedure for characterizing human
information processing. However, in our view, the power of this methodology has not been
exploited as it could be in studies of cognitive slowing (see our earlier discussion of memory
scanning; however, see Cerella & Hale, 1994, for a different perspective). Second, Hohle
(1967) advocated the use of psychophysiological measures to fractionate the reaction process.
Our discussion of the psychophysiological literature certainly reinforces that recommendation.
Third, he presented a distribution analysis of RT data to contribute to the decomposition of
the reaction process. He argued, and presented results consistent with his argument, that the
reaction process comprises two random variables that are characterized by different
distributions: (1) central decision (exponential), and (2) all other processing components
(linear; for an elaboration of Hohle's (1967) recommendation, see the Molenaar and van der
Molen article). If the reaction process includes only these two components, then both may
slow with advancing age. However, the degree to which the slowing is expressed may vary
importantly on the basis of the processing demands imposed on the system, and the degree to
which the slowing is revealed may vary with how it is measured. If differences do exist among
the various age groups, they may be revealed with greater precision using both behavioral and
ERP measures. Thus, the distributional analysis on RT recommended by Hohle could be
extended to include different ERP component latencies to identify age effects. Surely, the
strength of any inferences about cognitive slowing, be they global or local, would be enhanced
by converging evidence from multiple measures and multiple methods of analysis.
Recall that Birren (1965) argued that the search for the mechanisms underlying
cognitive slowing would be less compelling if there were many, rather than a few, time
constants that were affected. Recall as well that Cerella et al. (1980) thought it remarkable
that their analyses had revealed one, rather than many, slowing factors. We know of no studies
in which the response latencies in choice reactions of older adults are faster than those of
younger adults. Indeed, to our knowledge, older adults are always slower than young adults in
choice reactions. This persistent latency difference, in and of itselt~ argues that global slowing
is induced by advancing age, and the impressive orderliness of data points in the Brinley
analyses may be a manifestation of that slowing (for a recent incisive dissection of the Brinley
analysis, see Cerella & Hale, 1994). However, as we have seen, suggestive local effects are
not only observed in RT, but are also evident in the latencies of components of the ERP. Like
RT, the latencies of components of the ERP that are associated with central decision-making
are typically slower in older than young adults across factor levels; this, again, is suggestive of
global slowing. Thus, the presence or absence of local effects occurs amidst an overall slowing
in the rate of central decision-making speed. From this perspective, global slowing is not
incompatible with local slowing that is expressed differentially as a function of processing
demands. To paraphrase Cerella (1991), global slowing may provide the context in which the
entire cognitive slowing story is told.
REFERENCES
Allen, P.A., Ashcraft, M.H., & Weber, T.A. (1992). On mental multiplication and age.
Do general slowingfunctions mask local slowing effects? 419
Allen, P.A., Madden, D.J., Weber, T.A., & Groth, K.E. (1993). Influence of age and
processing stage on visual word recognition. Psychology and Aging, 8, 274-282.
Allen, P.A., Patterson, M.B., & Propper, ILE. (1994). Influence of letter size on age
differences in letter matching. Journal of Gerontology: Psychological Sciences, 49, P24-
28.
Anders, T.1L, & Fozard, J.L. (1973). Effects of age upon retrieval i~om primary and
secondary memory. Developmental Psychology, 9, 411-416.
Anders, T.1L, Fozard, J.L., & Lillyquist, T.D. (1972). Effects of age upon retrieval from
short-term memory. Developmental Psychology, 6, 214-217.
Barker, A.T. (1991). An introduction to the basic principles of magnetic neurostimulation.
Journal of Clinical Neurophysiology, 8, 26-37.
Barker, A.T., Jalinous, R., & Freeston, I.L. (1985). Noninvasive stimulation of the human
motor cortex. Lancet, 1, 1106-1107.
Baron, A., & Matilla, W.1L (1989). Response slowing of older adults: Effects of time-limited
contingencies of single- and dual-task performance. Psychology and Aging, 4, 66-72.
Bashore, T.1L (1990). Age-related changes in mental processing revealed by analyses of
event-related brain potentials. In J. Rohrbaugh, R. Parasuraman, & R. Johnson (Eds.),
Event-related brain potentials: Basic issues and applications (pp. 242-275). New York:
Oxford University Press.
Bashore, T.1L (1993). Differential effects of aging on the neurocognitive functions subserving
speeded mental processing. In J. Cerella, J. Rybash, W. Hoyer, & M.L. Commons (Eds.),
Adult information processing: Limits on loss (pp. 37-76). New York: Academic Press.
Bashore, T.R. (1994). Some thoughts on neurocognitive slowing. Acta Psychologiea, 86,
295-325.
Bashore, T.R., Osman, A., & Heffley, E.F. (1989). Mental slowing in elderly persons: A
cognitive psyehophysiological analysis. Psychology and Aging, 4, 235-244.
Bashore, T.R., & van der Molen, M.W. (1991). Discovery ofthe P300: A tribute. Biological
Birren, J.E. (1955). Age changes in speed of simple responses and perception and their
significance for complex behavior. In Old Age in the Modern World, London: E&S
Livingstone.
Birren, J.E. (1956). The si,maificance of age changes in speed of perception and psychomotor
skills. In J.E. Anderson (Ed.), Psychological Aspects of Aging (pp. 97-104). Washington,
D.C.: American Psychological Association.
Birren, J.E. (1964). The psychology of aging. Englewood Cliffs, NJ: Prentice-Hall.
Birren, J.E. (1965). Age changes in speed of behavior: Its central nature and physiological
correlates. In A.T. Welford, & J.E. Birren (Eds.), Behavior, aging and the nervous system
(pp. 191-216). Sprinfield, IL: Charles C. Thomas.
Birren, J.E., & Botwinick, J. (1951a). Rate of addition as a function of difficulty and age.
Psychometrica, 2, 219-232.
Birren, J.E., & Botwinick, J. (1951b). The relation of writing speed to age and to the senile
psychoses. Journal of Consulting Psychology, 15, 243-249.
Birren, J.E., Botwinick, J., Weiss, A., & Morrison, D.F. (1963). Interrelations of mental and
perceptual tests given to healthy older men. In J.E. Birren, 1LN. Butler, S.W. Greenhouse,
L. Sockolof~ & M. Yarrow (Eds.), Human aging: A biological and behavioral study (pp.
143-156). Washington, DC: Government Printing Office.
420 Th.R. Bashoreand F. Smulders
Bitten, J.E., Riegel, I~F., & Morrison, D.F. (1962). Age differences in response speed as a
function of controlled variations in stimulus conditions: Evidence of a general speed factor.
Gerontologia, 6, 1-18.
Birren, J.F., Woods, A.M., & Williams, M.V. (1980). Behavioral slowing with age: Causes,
organization and consequences. In L. Poon (Ed.), Aging in the 1980s: Psychological
issues (pp. 293-308). Washington, DC: American Psychological Association.
Brinley, J.F. (1965). Cognitive sets, speed and accuracy of performance in the elderly. In A.T.
Welford & J.E. Birren (Eds.), Behavior, aging and the nervous system (pp. 114-149).
Springfield, 111:C.C. Thomas.
Brookhuis, K.A., Mulder, G., Mulder, L.J.M., Gloerich, A.B.M., van Dellen, H.J., van der
Meere, J.J., & Ellerman, H.H. (1981). Late positive components and stimulus evaluation
time. Biological Psychology, 30, 107-123.
Brown, W.S., Marsh, J.T., & LaRue, A. (1983). Exponential electrophysiological aging: P3
latency. Electroencephalography and Clinical Neurophysiology, 55, 277-285.
Burke, D.M., White, H., & Diaz, D.L. (1987). Semantic priming in young and older adults:
Callaway, E. (1983). The pharmacology of human information processing. Psychophysiology,
20, 359-370.
Carpenter, M.B. (1991). Core text of neuroanatomy (4th Edition). Baltimore: Williams &
Wilkins.
Cerella, J. (1985a). Information processing rates in the elderly. Psyxhological Bulletin, 98,
67-83.
Cerella, J. (1985b). Age-related decline in extrafoveal letter-perception. Journal of
CereUa, J. (1990). Aging and information processing rate. In J.E. Birren & K.W. Schaie
(Eds.), Handbook of the psychology of aging (pp. 201-221). New York: Academic Press.
Cerella, J. (1991). Age effects may be global, not local: Comment on Fisk and Rogers.
Cerella, J. (1994). Generalized slowing in Brinley plots. Journal of Gerontology:
Psychological Sciences, 49, P65-P71.
Cerella, J., & Hale, S. (1994). The rise and fall in information-processing rates over the life
span. Acta Psychologica, 86, 109-197.
Cerella, J., Poon, L.W., & Fozard, J.L. (1981). Mental rotation and age reconsidered.
Cerella, J., Pooh, L.W., & Williams, D.M. (1980). Age and the complexity hypothesis. In L.
Pooh (Ed.), Aging in the 1980s: Psychological issues (pp. 332-340). Washington, DC:
American Psychological Association.
Charness, N. (1987). Component processes in bridge bidding and novel problem solving tasks.
Canadian dournal of Psychology, 41, 223-243.
Coles, M.G.H., Gratton, G., Bashore, T.1L, Eriksen, C.W., & Donchin, E. (1985). A
psychophysiological investigation of the continuous flow model of human information
11, 529-553.
Donchin, E., Karis, D., Bashore, T.1L, Coles, M.G.H., & Gratton, G. (1986). Cognitive
psychophysiology and human information processing. In M.G.H. Coles, E. Donchin, &
S.W. Porges (Eds.), Psychophysiology: Systems, processes, and applications (pp.

244-267). New York: The Cmilford Press.
Eriksen, C.W., Hamlin, 1LM., & Daye, C. (1973). Aging adults and rate of memory scan.
Bulletin of the Psychonomic Society, 1, 259-260.
Eriksen, C.W., & Schultz, D.W. (1979). Information processing in visual search: A
continuous flow conception and experimental results. Perception and Psychophysics, 5,
249-263.
Fisk, A.D., & Fisher, D.L. (1994). Bdnley plots and theories of aging: The explicit, muddled,
and implicit debates. Journal of Gerontology: Psychological Sciences, 49, P81-P89.
Fisk, A.D., Fisher, D.L., & Rogers, W.A. (1992). General slowing alone cannot explain age-
related search effects: Reply to Cerella (1991). Journal of Experimental Psychology:
General, 121, 73-78.
Fisk, A.D., & Rogers, W.A. (1991). Toward an understanding of age-related memory and
visual search effects. Journal of Experimental Psychology: General, 120, 13 l- 149.
Ford, J.M., & Pfefferbaum, A. (1980). The utility of brain potentials in determining age-related
changes in nervous system and cognitive functioning. In L. Pooh (Ed.), Aging in the
1980s: Psychological issues (pp. 115-124). Washington, DC: American Psychological
Association.
Ford, J.M., & Pfefferbaum, A. (1985). Age-related changes in event-related potentials. In P.
Ackles, J.R. Jennings, & M.G.H. Coles (Eds.), Advances in psychophysiology, volume 3
(pp.301-339). Greenwich, Connecticut: JAI Press.
Ford, J.M., Pfefferbaum, A., Tinklenberg, J.R., & Kopell, B.S. (1982). Effects of perceptual
and cognitive difficulty on P3 and RT in young and old adults. Electroencephalography
Ford, J.M., Roth, W.T., Mohs, R.C., Hopkins, W.F., & Kopell, B.S. (1979). Event-related
potentials recorded from young and old adults during a memory retrieval task.
Goodin, D.S., Squires, K.C., Henderson, B.H., & Starr, A. (1978). Age-related variations in
evoked potentials to auditory stimuli in normal human subjects. Electroencephalography
Goodin, D.S., Squires, K.C., & Starr, A. (1978). Long latency event-related components of
the auditory evoked potential in dementia. Brain, 101, 635-648.
Hale, S., Lima, S.D., & Myerson, J. (1991). General cognitive slowing in the nonlexical
domain: An experimental validation. Psychology and Aging, 6, 512-521.
Hale, S., Myerson, J., and Wagstait~ D. (1987). General slowing of nonverbal information
processing: Evidence for a power law. Journal of Gerontology, 42, 131-136.
Hartley, A.A. (1992). Attention. In F. Craik & T. Salthouse (Eds.), The handbook of aging
and cognition (pp. 3-49). Hillsdale, NJ: Lawrence Erlbaum Associates.
Hertzog, C. (1992). Aging, information processing, and intelligence. In K.W. Schaie (Ed.),
Annual review of gerontology and geriatrics (Volume 11, pp. 55-79). New York:
Springer.
Hertzog, C., Raskind, C.L.,& Cannon, C.J. (1986). Age-related slowing in semantic
information processing speed: An individual difference analysis. Journal of Gerontology,
41, 500-502.
HoMe, R.H. (1967). Component latencies in reaction times of children and adults. In L.P.
Lipsitt & C.C. Spiker (Eds.), Advances in child development and behavior, Volume 3 (pp.
225-261). New York: Academic Press.
Houx, P.J., Vreeling, F.W., & Jones, J. (1991). Rigorous health screening reduces age effect
on memory scanning task. Brain and Cognition, 15, 246-260.
Howard, D.V., Shaw, R.J., & Heisey, J.G. (1986). Aging and the time course of semantic
Knight, KT. (1990). Neural mechanisms of event-related potentials: Evidence from human
lesion studies. In J. Rohrbaugh, R. Parasuraman, & R. Johnson (Eds.), Event-related
potentials: Basic issues and applications (pp. 3-18). New York: Oxford University Press.
Kutas, M., McCarthy, G., & Donchin, E. (1977). Augmenting mental chronometry: The P300
as a measure of stimulus evaluation time. Science, 197, 792-795.
Laver, G.D., & Burke, D.M. (1993). Why do semantic priming effects increase in old age? A
Madden, D.J. (1982). Age differences and similarities in the improvement of controlled search.
Madden, D.J. (1984). Data-driven and memory-driven selective attention in visual search.
Madden, D.J. (1989). Visual word identification and age-related slowing. Cognitive
Madden, D.J. (1992). Four to ten milliseconds per year: Age-related slowing of visual word
identification. Journal of Gerontology: Psychological Sciences, 47, P59-P68.
Madden, D.J., & Nebes, R.D. (1980). Aging and the development of automaticity in visual
search. Developmental Psychology, 16, 377-384.
Madden, D.J., Pierce, T.W., & Allen, P.A. (1992). Adult age differences in attentional
allocation during memory search. Psychology and Aging, 7, 594-601.
Madden, D.J., Pierce, T.W., & Allen, P.A. (1993). Age-related slowing and the time course of
semantic priming in visual word identification. Psychology and Aging, 8, 490-507.
Magliero, A., Bashore, T.R., Coles, M.G.H., & Donchin, E. (1984). On the dependence of
P300 latency on stimulus evaluation processes. Psychophysiology, 21, 171-186.
Maniscalco, C.I., & DeRosa, D.V. (1983). Memory scanning of young and old adults: The
influence of rate of presentation and delay interval on recognition memory performance.
Bulletin of the Psychonomic Society, 21, 7-10.
Marsh, G.K (1975). Age differences in evoked potential correlates of a memory scanning
process. Experimental Aging Research, 1, 3-16.
McCarthy, G., & Donchin, E. (1981). A metric for thought: A comparison of P300 latency and
reaction time. Science, 211, 77-80.
Miller, G.A., Bashore, T.IL, Farwell, L.F., & Donchin, E. (1987). Review of geriatric
psychophysiology. In The annual review of gerontology and geriatrics volume 7 (pp. 1-
27), W. Schaie (Ed.), New York: Springer-Verlag.
Mulder, G., Gloerich, A.B.M., Brookhuis, K.A., Van Dellen, H.J., & Mulder, L.J.M. (1984).
Stage analysis of the reaction process using brain-evoked potentials and reaction.
Psychological Research, 46, 15-32.
Th.R. Bashoreand F. Smulders 423
Myerson, J., Ferraro, F.1L, Hale, S., & Lima, S.D. (1992). General slowing in semantic
Myerson, J., & Hale, S. (1993). General slowing and age invariance in cognitive processing:
The other side of the coin. In J. Cerella, J. gybash, W. Hoyer, & M.L. Commons (Eds.),
Adult information processing: Limits on loss (pp. 115-141). Academic Press: New York.
Myerson, J., Hale, S., Hirschman, 1L, Hansen, C., & Christensen, B. (1989). Global increase
in response latencies by early middle age: Complexity effects in individual performances.
Journal of Experimental Analysis of Behavior, 52, 353-362.
Myerson, J., Hale, S., Wagstaff~ D., Poon, L.W., & Smith, G.A. (1990). The information-loss
475-487.
Myerson, J., Wagstafl~ D., & Hale, S. (1994). Brinley plots, explained variance, and the
Myerson, J., Wagstaff, D., & Hale, S. (1994). [Brinley plots, explained variance, and the
analysis of age differences in response latencies from Fisk & Rogers (1992): The role of
averaging across sessions.] Unpublished manuscript.
Perfect, T.J. (1994). What can Brinley plots tell us about cognitive aging? Journal of
Pfefferbamn, A., Christensen, C., Ford, J.M., & Kopell, B.S. (1986). Apparent response
incompatibility effects on P3 latency depend on the task. Eleetroencephalography and
Pfefferbaum, A., Ford, J., Johnson, R., Wenegrat, B., & Kopell, B.S. (1983). Manipulation of
P3 latency: Speed vs. accuracy instructions. Electroencephalography and Clinical
Pfefferbaum, A., Ford, J.M., Roth, W.T., & Kopell, B.S. (1980). Age differences in P3-
reaction time associations. Electroencephalography and Clinical Neurophysiology, 49,
257-265.
Pfefferbaum, A., Ford, J.M., Wenegrat, B.M., Roth, W.T., & Kopell, B.S. (1984). Clinical
application of the P3 component of event-related potentials. Eleetroeneephalography and
Picton, T.W., Stuss, D.T., Champagne, S.C., & Nelson, R.F. (1984). The effects of age on
human event-related potentials. Psyehophysiology, 21, 312-325,
Polich, J., & Start, A. (1984). Evoked potentials in aging. In M.L. Alpert (Ed.), Clinical
neurology of aging (pp. 149-177). New York: Oxford University Press.
Polich, J., Howard, L., & Starr, A. (1985). Effects of age on the P300 component of the
event-related potential from auditory stimuli: Peak definition, variation, and measurement.
Pratt, H.J., Michalewski, H.J., Patterson, J.V., & Starr, A. (1989a). Brain potentials in a
memory-scanning task. II. Effects of aging on potentials to the probes.
Electroencephalography and Clinical Neurophysiology, 72, 507-517
Pratt, H.J., Michalewski, H.J., Patterson, J.V., & Starr, A. (1989b). Brain potentials in a
memory-scanning task. III. Potentials to the items being memorized.
Ritter, W., Vaughan, H.G., & Simson, 1L (1983). On relating event-related potential
components to stages of information processing. In A.W.I~ Gaillard & W. Ritter (Eds.),
Tutorials in event-related potential research (pp. 143-158). Amsterdam: North-Holland.
Robinson, L.l~, & Slimp, J.C. (1990). Fundamental concepts of somatosensory and motor
evoked potentials. Physical Atedicme and Rehabilitation Clinics of North America, 1,
133-148.
Rogers, W.A. & Fisk, A.D. (1990). A reconsideration of age-related reaction-time slowing
from a learning perspective: Age-related slowing is not just complexity based. Learning
and Individual Differences, 2, 161-179.
Salthouse, T.A. (1978). [Age and speed: The nature of the relationship.] Unpublished
manuscript.
Salthouse, T.A. (1984). Effects of age and skill in typing. Journal of Experimental
Salthouse, T.A. (1985a). Speed of behavior and its implications for cognition. In J.E. Birren
& I~W. Schaie (Eds.), Handbook of the psychology of aging, (pp. 400-426). New York:
Van Nostrand Reinhold.
Salthouse, T.A. (1985b). A theory of cognitive aging. Amsterdam: North-Holland.
Salthouse, T.A., & Somberg, B.L. (1982). Isolating the age deficit in speeded performance.
Journal of Gerontology, 3 7, 349- 357.
Simon, J.1L, & Pouraghabagher, A.1L (1978). The effect of aging on stages of processing in a
choice reaction time task. Journal of Gerontology, 33, 553-561.
Staid, H.G.O.M., Mulder, G., & Mulder, L.J.M. (1990). On the independence of stimulus
recognition and response activation. Acta Psychologica, 74, 169-202.
Smith, G.A., Poon, L.W., Hale, S., & Myerson, J. (1988). A regular relationship between old
and young adults' latencies on their best, average and worst trials. Australian Journal of
Smulders, F. (1993). [The selectivity of age effects on information processing: Response times
and electrophysiology.] Unpublished Doctoral Thesis, Department of Psychology,
University of Am~erdam
Smulders, F.T.Y., Kenemans, J.L., & Kok, A. (in submission). Stimulus degradation and
stimulus-response compatibility effects in young and old adults: An electrophysiological
approach.
Steinberg, S. (1969). The discovery of processing stages: Extensions of Donders' method. In
W.G. Koster (Ed.), Attention and Performance (Volume II, pp. 276-315). Amsterdam:
North-Holland.
Strayer, D.L., Wickens, C.D., & Braune, R. (1987). Adult age changes in the speed and
capacity of information processing. II: An electrophysiological approach. Psychology and
Aging, 2, 99-110.
Syndulko, K., Hansch, E.C., Cohen, S.N., Pearce, J.W., Goldberg, Z., Montan, B.,
Tourtellotte, C., & Potvin, A.R. (1982). Long-latency event-related potentials in normal
aging and dementia. In J. Courjon, F. Mauguiere, & M. Revol (Eds.), Clinical
applications of evoked potentials in neurology (pp. 279-293). New York: Raven Press.
van der Molen, M.W., Bashore, T.R., I-lalliday, R.F., & Callaway, E. (1991).
Chronopsychophysiology of human information processing: Mental chronometry
augmented by physiological time markers. In J.R. Jennings & M.G.H. Coles (Eds.),
Handbook ofcogmtive psychophysiology (pp. 9-178). New York: J. Wiley & Sons.
Welford, A.T. (1977). Motor perfo~ancr In J.E. Birren & K.W. Schaie (Eds.), Handbook
of the psychology of aging (pp. 450-496). New York: Van Nostrand Reinhold.
Wood, C.C., McCarthy, G., Squires, N.K., Vaughan, H.G., Jr., & McCallum, W.C. (1984).
Anatomical and physiological substrates of event-related potentials. In 1L Karrer, J.
Cohen, & P. Tueting (Eds.), Brain and information: Event-related potentials (pp.
681-721). New York: The New York Academy of Sciences.
Wright, L.L., & Elias, J.W. (1979). Age differences in the effects of perceptual noise. Journal
Zeet~ E.J., & Kok, A. (1993). Age-related differences in the timing of stimulus and response
processes during visual selective attention: Performance and psychophysiological analyses.
427
A u t h o r Index
Anders, T ..................................... 409, 419

A
Andersen, E. S ..................................... 245
Aaronson, D ................................. 180, 183 Anderson, J ............ 23, 26, 27, 35, 70, 315,
Abbenhuis, M ............... 131, 136, 274, 289 316, 339, 369, 381,419
Abbott, 1L ............................................ 271 Anderson-Garlach, M ................. 23, 26, 27
Aberdeen, J. S ...................................... 246 Andrassy, J ........................................... 185
Abrams, 1L ....................................... 24, 28 Andrewes, D. G .................................... 387
Achariyapaopan, T ............................... 309 Andrews, S ......... 72, 73, 74, 78, 82, 83,84,
Adams, M ................................. 31, 64, 291 138, 329, 330, 341,385
Akutsu, H ..................................... 204, 215 Annon, T ...................................... 179, 185
Albert, M ..... 113, 136, 188, 190, 195, 197, Aosaki, T ............................................. 340
198, 210, 216, 225, 233, 238, 239, 240, Appel, J ........................................ 210, 215
250, 268, 349, 381, 385, 390 Appell, J ............................... 224, 236, 238
Albertson, S .................................. 127, 138 Arenberg, D .......... 139, 145, 166, 169, 170
Albrecht, J .................................... 178, 184 Ashby, F.G. .................................. 6, 25, 29
Aldridge, V. J ...................................... 313 Ashcraft, M ...................... 63, 64, 402, 418
Alexander, M. P ................................... 388 Atkins, P .......................................... 31, 66
Allard, T .............................................. 242 Atkinson, 1L ................................. 173, 184
Allen, P.A .......... 30, 31, 33, 34, 35, 36, 37, Ausubel, R ........................................... 291
38, 39, 40, 41, 43, 44, 46, 47, 48, 49, 50,
51, 52, 53, 54, 55, 56, 57, 58, 59, 60, 61, B
63, 64, 65, 67, 68, 69, 70, 74, 75, 76, 77, Babcock, R ............ 171, 174, 175, 184, 186
78, 81, 83, 85, 113, 135, 138, 144, 153, Baddeley, A ...... 46, 65, 111, 136, 173, 174,
157, 162, 166, 171, 196, 197, 224, 226, 184, 287, 289, 315, 339, 345, 380, 381
227, 228, 232, 238, 242, 400, 401, 402, Bajo, M.T ............................................. 166
411, 418, 419, 422 Ball, M. J .............................................. 243
Allison, T ...................... 318, 323, 339, 343 Balota, D ...... 12, 16, 17, 19, 21, 26, 30, 42,
Altarriba, J ...................... i............. 263, 270 43, 44, 47, 50, 54, 55, 56, 58, 59, 60, 62,
Altieri, P .............................................. 185 65, 74, 84, 89, 91, 92, 93, 94, 107, 109,
Alzheimer, A ........................................ 238 113, 136, 154, 160, 163, 164, 165, 166,
Amaducci, L. A .................................... 245 167, 220, 224, 225, 226, 227, 233,234,
Amrhein, P ........... 30, 44, 64, 65, 114, 143, 238, 239, 240, 250, 268, 274, 278, 284,
144, 146, 147, 149, 151, 154, 155, 156, 289, 290, 340
157, 158, 159, 160, 162, 164, 165, 166, Baltes, P ....................... 102, 107, 210, 217
170, 214, 218, 219 Bang, S ................................................ 137
428 Author Index
Banks, K ............................... 216, 225, 240 85, 166, 184, 239, 254, 268, 343, 390,
Barbas, N ...................................... 201, 217 391,392, 395, 398, 399, 418, 419, 420,
Barclay, J ....................................... 89, 107 424, 425
Bargh, J ........................................ 262, 268 Bisiach, E ..................................... 210, 215
Barker, A ...................................... 412, 419 Bjork, R. ....... 272, 276, 292, 364, 368, 387
Barmwater, U ...................................... 384 Black, S ..................... 12, 26, 113, 136, 160
Baron, J ........................ 226, 239, 402, 419 Bladin, P. F .......................................... 387
Barr, R ............................................. 74, 85 Blake, R ....................... 200, 204, 213, 218
Bartlett, J ...................................... 124, 136 Blanks, J ................ 201,202, 215, 216, 217
Bartus, R. T ......................................... 382 Blanks, R. H. I ............................. 215, 216
Bashore, T.R ........... 21, 26, 30, 44, 47, 65, Blaxton, T ........................... 282, 289, 293,
67, 68, 69, 70, 296, 297, 298, 299, 302, 347, 387
303, 305, 307, 308, 309, 310, 311,313, Blessed, G ............................................ 389
320, 345,381,392, 409, 410, 411,419, Bloom, P. A ......................................... 309
420, 422, 424 Boner, F ................. 92, 108, 140, 160, 168,
Bassett, E ............................................ 167 198, 244, 293, 383, 385, 387, 389
Bassi, C ................ 200, 202, 203, 215, 218 Bondi, M .............. 275, 277, 278, 280, 282,
Baty, J ................................................. 239 284, 290
Bauer, J ............................................... 218 Bons, T ................................................ 169
Bayles, K ...................... 210, 215, 254, 268 Booker, J .............................................. 140
Becker, C ..... 39, 65, 89, 90, 107, 109, 209, Bosco, C .............................................. 309
215,221,223, 225, 239, 242, 262, 268, Botwinick, J ......... 145, 162, 163, 164, 166,
328, 339 316, 339, 391,419
Bell, T .................................. 235, 236, 239 Bouras, C ..................................... 348, 382
Belmore, S ..................................... 94, 107 Bourguet, M ......................................... 308
Benson, D ............ 188, 190, 197, 201,209, Bower, G.H ...... 65, 84, 108, 140, 184, 241,
214, 215, 216, 236, 240, 388 270, 369, 381
Bentin, S ....................... 351,362, 363, 381 Bowers, J ............................. 140, 290, 382
Berg, L ................ 220, 222, 223, 239, 241, Bowles, N ............ 6, 12, 15, 23, 26, 53, 60,
243, 245,269 62, 65, 74, 84, 91, 92, 94, 107, 112, 136,
Berger, H ...................... 294, 308, 319, 339 143, 144, 148, 149, 153, 154, 158, 159,
Bergman, H ......................................... 268 160, 162, 164, 165, 166, 210, 216, 225,
Berkovic, S. F ...................................... 387 226, 228, 239
Berkowsky, K ...................................... 169 Bowling, A ................................... 204, 217
Berman, S ..................... 345,353, 381,383 Brady ............. 92, 109, 160, 168, 226, 233,
Bernstein, P.A ...................................... 382 244, 256, 270
Bertrand, O .......................................... 386 Bran&, J ....................................... 276, 290
Besner, D ..... 31, 32, 38, 41, 42, 44, 65, 66, Bransford, J .................... 89, 107, 283, 292
68, 69, 72, 84, 226, 240, 244, 270 Braren, M ............................................. 312
Besson, M .................... 300, 301,302, 308, Braun, A. R .......................................... 292
331,342 Braune, R ............................................. 424
Best, M ......................... 118, 136, 160, 258 Breitmeyer, B ............................... 199, 216
Biederman, I ............. ....................... 34, 65 Bressi, S ............................................... 289
Bienkowski, M ............................... 89, 109 Briand, K ............... 115, 116, 137, 262, 268
Birren, J ............. 26, 46, 65, 66, 70, 75, 84, Brinley, J ........... 8, 9, 15, 26, 27, 28, 45, 46
Author Index 429
48, 51, 52, 53, 54, 55, 56, 59, 60, 61, 62, Carr, T ............... 31, 36, 37, 38, 39, 40, 43,
65, 66, 67, 69, 70, 143, 157, 162, 166, 51, 66, 73, 84, 168
167, 179, 180, 183, 186, 257, 258, 260, Carson, K ............................. 224, 238, 241
261,266, 268, 392, 393, 395,400, 401, CaruUo, J ...................................... 177, 184
402, 403,404, 405,406, 407, 408, 411, Cattell, J ........................... 30, 66, 147, 166
414, 416, 417, 418, 420, 421,423 Cerella, J ...... 1, 4, 5, 11, 15, 18, 23, 26, 27,
Broadbent, D .................... 13, 26, 200, 216 28, 30, 31, 36, 44, 45, 46, 47, 48, 49, 50,
Brodbeck, D. R ............................. 290, 382 56, 59, 60, 63, 64, 66, 67, 74, 75, 84, 92,
Brookhuis, K ........................ 409, 420, 422 94, 107, 113, 136, 144, 153, 156, 160,
Brooks, R ................................. 35, 65, 137 162, 166, 167, 168, 175, 184, 226, 239,
Brouwers, P ......................................... 292 244, 254, 268, 269, 391,392, 393, 394,
Brown, A ......... 68, 85, 112, 114, 117, 118, 395, 396, 397, 399, 404, 405, 406, 407,
124, 131, 133, 136, 138, 139, 141,215, 410, 414, 418, 419, 420, 421,423
316, 339, 345, 385, 409, 420 Cermak, L ............................ 139, 169, 170
Bruemmer, A ....................................... 268 Challis, B ....... 117, 140, 283, 290, 369, 382
Brtme, C.M .......................................... 384 Chambers, S ..................................... 74, 84
Bub, D .......... 187, 197, 248, 268, 286, 290 Champagne, S.C ................................... 423
Buckle, L ............................................. 388 Chancelliere, A ............................. 187, 197
Bullemer, P ........................... 278, 287, 292 Chapman, R .......................... 245, 323, 340
Buonanno, F ........................................ 218 Chapman, J. A ...................................... 340
Burke, D ........ 6, 12, 15, 22, 26, 28, 50, 63, Charness, N .................................. 402, 420
66, 68, 92, 94, 107, 108, 109, 112, 113, Chase, W ...................... 286, 292, 317, 340
133, 136, 138, 157, 158, 165, 168, 179, Chase, T. N .......................................... 292
185,223,228, 239, 255,257, 258, 269, Chenery, H ................................... 274, 290
271,340, 341,344, 400, 402, 420, 422 Cheney, M ................. 12, 26, 113, 136, 160
Buschke, H ........................... 137, 254, 269 Cheng, J ............................................... 382
Butter, C. M ........................................ 245 Cherry, K ............................................. 169
Butters, N ............ 137, 253,269, 275, 277, Chertkow, H ........ 248, 252, 253,254, 260,
278, 286, 290, 291,292, 293, 382, 384 261,264, 267, 268, 286, 290
Buzney, S ............................. 200, 204, 218 Chesney, G ........................................... 311
C Chiappa, K. H ...................................... 218
Chiarello, C ................. 129, 130, 362, 369,
Cacioppo, J .......................................... 308 380, 382
Calkins, M .................................... 215, 216 Chiesi, H ...................................... 146, 169
Callaway, E .......... 299, 308, 309, 313, 318, Childers, D. G ...................................... 309
340, 409, 420, 424 Chiu, P ................................................. 293
Campbell, M ................. 199, 201, 216, 217 Chiulli, S .............................................. 388
Canestrari, R ................................. 145, 166 Chomsky, N ................................. 334, 342
Cannon, C.J ......................................... 421 Christensen, C ...................................... 423
Canoune, H .......................................... 343 Chrosniak, L ......................................... 137
Cantor, J ....................... 177, 184, 221,239 Chumbley, J ........ 42, 43, 56, 65, 74, 84, 89,
Capdevila, A ........................................ 343 107, 154, 166, 224, 227, 238
Capps, J. L .................................... 138, 342 Clapman, R .................................. 305, 307
Caramazza, A ............................... 196, 197 Clark, M .......... 89, 109, 154, 169, 233, 240
Carpenter, P ................. 173, 177, 182, 184, Claverie, B ........................................... 384
185, 315, 317, 337, 341, 343, 412, 420 Clayton, G. M ...................................... 244
430 Author Index
Coats, M .............................................. 243 Cronin-Golomb, A ............... 202, 216, 218,

Coben, L. A .................................. 241,245 225, 240
Coffey, S. A ......................................... 341 Crook, T ...................................... 346, 382
Cohen, G. ...... 94, 107, 110, 111, 136, 216, Crossley, M .................................. 174, 184
225, 229, 240, 315, 316, 340, 346, 347, Crossman, E ..................................... 74, 84
382, 384, 424, 425 Crozier, L ................ 44, 49, 51, 55, 58, 60,
Coles, M .............. 298, 299, 302, 303, 305, 64, 74, 83, 153, 162, 166, 227, 238
306, 308, 309, 310, 311, 313, 318, 323, Cullum, C. M ....................................... 186
340, 343, 351, 382, 383, 387, 409, 414, Cummings, J .......... 204, 216, 223,236, 240
420, 421, 422, 424 Curcio, C ...................................... 201, 216
Collins, A ................. 43, 66, 114, 136, 262, Curtis, B ........................................... 31, 66
268, 366, 387 Cutaia, M.M ......................................... 388
Colombo, P ........................... 110, 111, 137
/9
Coltheart, M. 31, 37, 38, 40, 41, 42, 44,
65, 66, 72, 73, 75, 84, 85, 108, 184, 189, D'Antono, B ......................................... 268
192, 193, 197, 198, 226, 240 Dale, A ......................................... 339, 340
Commons, M ..... 26, 27, 28, 166, 167, 168, Dallas, M ...................................... 138, 384
244, 419, 423 Daly, F. L ............................................. 242
Conboy, G ....................................... 31, 67 Dam, A ........................................ 348, 382
Connelly, S ................................... 179, 184 Damasio, A .................. 277, 279, 284, 290
Connine, C ....................................... 43, 66 Daneman, M .......... 173, 174, 175, 182, 184
Connor, L ..... 220, 234, 239, 240, 278, 290 Dannenbring, G ..... 115, 116, 137, 262, 268
Conomy, J .................................... 195, 197 Danziger, W ......................................... 241
Convit, A ............................................. 384 Danziger, W. L ..................................... 269
Coon, V ........................................... 61, 70 Darofl~ R ...................................... 195, 197
Cooper, B .................... 23, 27, 67, 84, 118, Davelaar, E ........... 32, 66, 72, 84, 226, 240
140, 298, 313 Davis, H ............... 110, 111, 124, 125, 130,
Corkin, E ..... 215, 216, 218, 224, 225, 236, 137, 215, 222, 245, 247, 294, 308, 346,
240, 241,242, 272, 274, 275,277, 278, 347, 348, 350, 362, 369, 380, 382
281, 286, 290, 291 Dawson, A ................................... 117, 137
Corsi, P ................................................ 386 Daye, C ............................................... 421
Cortese, C ............................................ 140 de Groot, A... 240, 262, 263, 268, 328, 340
Corwin, J .............................. 288, 293, 388 De Mornay Davies, P ........................... 292
Costa, P ........................................ 221,240 Deecke, L ..................... 298, 305, 308, 310
Courchesne, E ....................... 374, 376, 382 deJong, R ............................. 305, 306, 308
Cox, C ................................................. 292 Dell, G ......................... 144, 148, 167, 168
Coyne, A .......................... 46, 64, 202, 216 Della Sala, S ......................................... 289
Craig, C ........................ 235, 236, 240, 244 den Heyer, K. ........ 209, 214, 216, 262, 269
Craik, F.I.M ..... 84, 85, 108, 112, 117, 118, DeRosa, D.V ........................................ 422
136, 139, 140, 171, 173, 178, 184, 185, Derrington, A ................................... 35, 66
218, 235, 244, 270, 349, 350, 359, 369, DeSanti, S ............................................ 384
380, 382, 385, 386, 389, 421 Desimone, 1L ................................ 319, 342
Crambert, 1L .................................... 49, 68 Desmedt, J ................... 297, 308, 310, 385
Cramer, M ........................................... 381 DeTeresa, IL ........................................ 389
Crawford, J ................................... 204, 216 DeValois, 1L ............................... 35, 49, 66
Cremer, J ......................................... 74, 84 Diaz, D ........... 12, 26, 50, 66, 92, 107, 165,
Author Index 431
228, 239, 400, 420 349, 351, 353, 357, 363, 370, 371, 373,
Digdon, N ..................................... 154, 169 374, 377, 378, 379, 380, 381,382, 383,
Dixon, 1~ A ......................................... 185 385
Dixon, R.A .......................................... 138 Faherty, A ........................................ 74, 85
Dobbs, A .............. 39, 40, 43, 66, 113, 137 Falmagne, J .................................. 146, 167
Dolan, R. J ........................................... 388 Farah, M ...................... 154, 167, 224, 245
Donchin, E ........... 296, 298, 299, 301,302, Farmer, S. F ................................. 311,342
303, 305, 307, 308, 309, 310, 311,323, FarweU, L ..................... 296, 309, 409, 422
340, 351,382, 383, 385,409, 414, 420, Faulconer, B ................ 143, 146, 150, 154,
422 156, 165, 169
Dowries, J. J ........................................ 292 Faulkner, D .................... 94, 107, 229, 240
Doyle, M .............. 39, 69, 72, 85, 350, 387 Faust, M ................ 184, 234, 240, 316, 340
Drachman, D ................ 223,243, 272, 292 Fedio, P ......... 224, 243, 252, 269, 286, 292
Drewnowski, A ................................ 31, 67 Feehan, M ............................................ 292
Drucker, D .................................... 201, 216 Felleman, D ...................................... 35, 70
Duchek, J ........... 12, 16, 17, 19, 21, 26, 65, Fennema, A. C ...................................... 291
91, 92, 93, 94, 107, 113, 136, 160, 163, Fera, P .................................. 41, 42, 44, 66
164, 165, 166, 167, 220, 223, 226, 234, Ferraro, F.R ....... 30, 44, 47, 50, 54, 55, 58,
238, 240, 243, 250, 268, 274, 284, 289 59, 60, 62, 65, 69, 90, 108, 113, 139, 157,
Duffy, S .................................... 44, 67, 381 168, 176, 185, 220, 225, 226, 227, 229,
Dungelhoff, F.-J ................................... 167 230, 233, 234, 235, 238, 239, 240, 242,
Dywan, J ....................................... 124, 137 244, 255, 270, 278, 279, 280, 284, 285,
290, 399, 423
E
Ferris, S ........................................ 382, 384
Echallier, J.F ........................................ 386 Fiez, J. A .............................................. 387
Eggers, R ............................................. 384 Finley, G ...................................... 113, 137
Eichenbaum, H .................................... 382 Fischler, I ..................... 263, 269, 300, 309
Elias, M ....... 149, 150, 151, 154, 158, 160, Fisher, D ........... 1, 2, 5, 6, 7, 8, 11, 12, 13,
167, 223, 241, 414, 425 14, 21, 23, 25, 26, 27, 29, 44, 45, 47, 51,
Ellerman, H.H ...................................... 420 57, 60, 62, 67, 144, 157, 160, 162, 167,
Elliott, L ............................... 235, 236, 240 258, 269, 391,406, 407, 408, 421
Ellis, A .......................... 190, 192, 193, 197 Fisk, A.D ............ 1, 3, 7, 12, 23, 24, 26, 27,
Emerson, P ........ 31, 35, 36, 37, 40, 43, 51, 29, 44, 45, 47, 57, 62, 66, 67, 144, 156,
52, 53, 58, 60, 64 160, 162, 167, 258, 269, 391,404, 405,
Engle, R ................................ 177, 182, 184 406, 407, 408, 420, 421,423, 424
Eriksen, C ............ 303, 305, 307, 308, 309, Fisman, M ............. 210, 215, 224, 236, 238
310, 409, 414, 420, 421 Fitts, P .............................................. 82, 84
Esiri, M ......................................... 237, 240 Flaherty, A. W ...................................... 340
Eskes, G. A .......................................... 389 Fletcher, P ............................................ 388
Eslinger, P .... 277, 279, 284, 290, 319, 340 Fodor, J ........ 32, 36, 67, 88, 107, 189, 190,
Evan, K. E ........................................... 240 197, 337, 340
Evett, L ........................................... 31, 68 Folstein, M...223, 243, 250, 251, 269, 272,
F 276, 290, 292
Ford, J ...... 22, 27, 327, 335, 340, 344, 345,
Fabiani, M ............ 301,302, 308, 310, 345, 383, 386, 409, 411, 412, 421,423
Forster, I~ .......... 31, 32, 39, 41, 42, 43, 44,
432 Author Index
67, 70, 73, 74, 84, 220, 225, 240 Gardiner, J ..... 117, 137, 350, 353, 373,384
Foster, J. I~ ......................................... 389 Garfield, L ........................ 74, 85,226, 245
Fox, J ............ 214, 217, 275, 291, 386, 387 Garrod, S ..................................... 177, 186
Fozard, J ............ 23, 29, 49, 66, 74, 75, 84, Gauderman, W ............................. 201, 216
92, 94, 107, 113, 133, 134, 136, 139, 141, Geary, W ................... 63, 67, 144, 157, 167
149, 151, 152, 153, 154, 158, 159, 160, Geckler, C ................................... 203, 216
161, 162, 166, 169, 170, 226, 239, 365, Gehring, W ................... 303, 307, 308, 309
387, 391,409, 419, 420 George, A. E ........................................ 384
Frackowiak, 1L S ................................. 388 Gerard, L ..................................... 316, 340
Fraisse, P ...................................... 147, 167 Gernsbacher, M ........ 97, 99, 107, 177, 184,
Francis, W ............. 43, 51, 68, 76, 85, 241, 227, 234, 238, 240, 241, 311, 316, 340,
269, 383 342
Franks, J ......................... 89, 107, 283, 292 Gershon, S ........................................... 382
Fratiglioni, L ........................................ 242 Geschwind, N ............... 188, 190, 197, 198
Frazier, L ...................................... 337, 340 Giambra, L ......................................... 1, 27
Frederiksen, J ................................... 55, 67 Giannakopoulos, P ............................... 382
Freeston, I.L ........................................ 419 Gibson, E ............. 34, 67, 72, 84, 137, 317,
Frensch, P ................ 63, 67, 144, 157, 167, 319, 341
287, 290 Gick, M. L ........................................... 244
Friedland, 1L ................................. 223, 240 Gilmore, G.C ........ 199, 202, 204, 205, 206,
Friedman, D ....... 39, 43, 66, 115, 119, 137, 207, 209, 210, 211,214, 216, 217, 219,
195, 197, 225, 226, 241, 251,269, 300, 226, 241
301, 309, 329, 330, 340, 345, 347, 349, Ginsburg, A .................................. 204, 218
350, 351, 352, 353, 354, 355, 357, 358, Ginter, H .............................................. 339
359, 360, 361, 362, 363, 364, 365, 366, Gitlin, H ............................................... 137
367, 368, 369, 370, 371, 372, 373, 374, Glaser, 1L ............... 67, 143, 144, 146, 154,
375, 376, 377, 378, 379, 380, 381, 382, 156, 167
383, 385 Glaser, M. O ......................................... 167
Friedman, A ........................................... 66 Glass, G ............................... 259, 265,269
Frith, C. D ........................................... 388 Glenberg, A .................................. 177, 184
Fry, A.F ............................................... 384 Gloerich, A.B ............................... 420, 422
Fuld, P ................................................. 269 Glosser, G. ................................... 251, 269
Fulling, K. ............................................ 243 Goddard, P.H ....................................... 309
Fulton, A ............................................. 136 Goldberg, T. E ..................................... 292
Funkenstein, H .............................. 195, 198 Goldberg, Z .......................................... 424
Furda, J ................................................ 312 Goldman-Rakic, P ....... 315, 319, 340, 343,
Fuster, J ........................ 289, 290, 319, 340 376, 377, 384
Goldstein, W..2, 6, 13, 21, 25, 27, 29, 200,
G
204, 218
Gabrieli, J ............ 274, 275, 276, 277, 278, Goodglass, H ........................................ 246
280, 282, 284, 285, 286, 289, 290, 291, Goodin, D .................................... 409, 421
292, 347, 383 Gordon, B ............. 224, 238, 241, 274, 291
Gaillard, A ........... 298, 305, 312, 318, 343, Goulet, P ...................... 149, 151, 153, 167
387, 424 Graesser, A .......................... 173, 177, 184
Galambos, 1L ....................................... 382 Grat~ P ................. 115, 137, 139, 140, 282,
Gandy, M ...................... 110, 111, 137, 382 291, 369, 384
Author Index 433
Graham, N ........................ 35, 67, 225, 244 391, 395, 396, 397, 398, 399, 404, 407,
Grainger, J ..... ....... 73, 74, 83, 84, 227, 241 408, 410, 418, 420, 421,422, 423,424
Granholm, E ........................................ 290 Halgren, E ............................................ 388
Grant, E ............................................... 243 Haller, M .......................................... 31, 66
Grant, I ................................................ 290 Halliday, 1L ......................... ................. 313
Grasby, P ............................................. 388 Halliday, 1LF ........................................ 424
Grattan, L. M ....................................... 340 Halpern, D ............ 145, 149, 150, 151, 158,
Gratton, G... 302, 303, 305, 307, 308, 309, 160, 167
310, 409, 414, 420 Hamberger, M ...... 115, 119, 137, 329, 330,
Grau, J ................................................. 343 340, 345, 354, 355, 357, 358, 359, 360,
Graves, 1L E ........................................ 186 361,362, 383
Graybiel, A ................................... 319, 340 Hamlin, R.M ......................................... 421
Greenberg, S ........................................ 291 Hamm~ V ...................................... 179, 185
Greene, H. A ....................................... 242 Hammer, M. A ..................................... 240
Greese, B ......................................... 73, 84 Hammett, S. C ...................................... 344
Gregory, R ........................ 74, 84, 247, 345 Hansch, E. C ......................................... 424
Grober, E ............. 128, 137, 254, 269, 274, Hansen, J .................. 31, 67, 299, 310, 389,
278, 284, 285, 291 396, 423
Grodzinsky, Y .............. 187, 190, 191, 193, Harbin, T ...................... 301, 310, 327, 340
194, 196, 198 Harbluk, J.L ......................................... 388
Grosjean, F ........................... 220, 235, 241 Hardenberg, J ....................................... 388
Grosse, D ............ 214, 217, 275, 277, 286, Harker, J ...................................... 145, 167
291,292, 383 Harris, P ....................... 150, 154, 165, 167
Grossman, J.L ...................................... 167 Hart, S .......... 204, 216, 223,241, 257, 269
Groth, K .... 30, 36, 44, 53, 64, 74, 83, 144, Hartley, A .................... 179, 185, 402, 421
157, 166, 197, 207, 214, 216, 226, 228, Hartley, J .......................................... 75, 84
238, 241,402, 419 Hartman, M.. 126, 127, 137, 248, 269, 272,
Growdon, J .......... 215, 216, 218, 224, 225, 286, 291,384
236, 240, 241, 274, 275, 277, 286, 290, Harvey, M. T ........................................ 310
291 Hasher, L ....... 75, 84, 92, 94, 97, 108, 109,
Guillem, F ..................................... 356, 384 126, 127, 137, 179, 184, 185, 227, 234,
Gtmter, T ...................... 300, 310, 327, 340 241,255,262, 269, 271, 315, 316, 340,
Gunther, H ....................................... 73, 84 374, 384
Guterman, Y ................................. 297, 310 Hashtroudi, S ....................... 137, 366, 384
Guynn, M ............................................ 140 Haug, H ............................................... 384
H Havinga, J ................. 40, 68, 143, 148, 168
Hayflick, L ................................... 110, 137
Haberlandt, K~ ...................... 180, 181, 185 Hcaen, H .............................................. 198
Habib, 1L ............................................. 389 Head, E ................................................ 139
Hackley, S ............................ 305, 307, 311 Healy, A ......................... 31, 35, 37, 43, 67
Haggard, M ................ 69, 72, 85, 118, 136 Hebb, D ............... 32, 43, 67, 110, 113, 137
Haider, M ..................................... 299, 310 Hedges, T ..................................... 201, 217
Hale, S ..... 1, 4, 5, 8, 11, 15, 26, 27, 28, 30, Hedreen, J .................................... 202, 219
44, 45, 66, 68, 69, 75, 85, 113, 138, 139, Heffiey, E ........... 22, 26, 65, 296, 305, 307,
144, 157, 158, 167, 168, 175, 176, 185, 309, 410, 411,419
226, 244, 254, 255, 269, 270, 317, 342, Heindel, W ........... 137, 275, 277, 278, 280,
434 Author Index
284, 285, 291, 292, 349, 384 Hoyer, W ........... 26, 27, 28, 129, 130, 166,
Heinze, H ...................... 301, 311, 318, 343 167, 168, 244, 362, 369, 380, 382, 419,
Heisey, J ...... 12, 27, 50, 67, 165, 167, 230, 423
241,384, 400, 422 H u ~ F ........... 92, 109, 160, 168, 224, 233,
Heller, H ............................... 113, 136, 137 241,244, 270, 276, 282, 285,286, 291
Henderson, L ............... 222, 226, 241, 245, Hughes, C .................... 223, 241, 250, 269
319, 341, 409, 421 Hulicka, I ..................................... 145, 167
Herman, R .................... 100, 101, 108, 109 Hultsch, D .... 118, 130, 138, 176, 185, 348,
Hertzog, C... 1, 23, 27, 130, 138, 160, 167, 362, 380, 384
185, 402, 421 Human, B. T ........................................ 290
Hess, T ................................. 137, 178, 185 Humes, L ..................................... 237, 241
Hesselink, J. R ..................................... 341 Humphreys, G ..... 31, 68, 69, 243, 244, 270
Heth, I ................................................. 381 Hunt, E ...................... 2, 27, 139, 322, 341
Hewitt, S ............................................. 385 Hutton, J ...................................... 223, 241
Hill, R .................. 137, 140, 204, 216, 218, Huy, N. T ............................................. 308
245, 247, 272
Hillyard, S ............ 298, 299, 300, 310, 311,
318, 321,323,326, 341, 342, 350, 353, Ingrain, J. C .......................................... 290
382, 384, 385, 388 Invik, R. J ............................................. 242
Hindman, J ............................ 171, 182, 186
Hinton, D .............................. 201,215, 217
Hintzman, D ................................. 177, 185 Jackson, C ............................................ 245
Hiscock, M ................................... 174, 184 Jackson, J. L ................................. 310, 340
Hitch, G ............................ 46, 65, 173, 184 Jacobs, A .......................................... 73, 84
Hockley, W ................................... 214, 219 Jacoby, L .............. 114, 117, 124, 137, 138,
Hodges, J ...................... 225, 244, 253,269 276, 281, 282, 291,293, 359, 369, 384,
Hof, P ................... 201, 202, 203, 217, 382 385
Hogaboam, T ................................. 89, 108 Jalinous, R ............................................ 419
Hogeboom, M. M ................................ 312 James, W .............. 111, 138, 171, 172, 173,
HoMe, R ....................................... 417, 422 178, 185, 227, 241, 391
Holcomb, P ........................... 327, 341,342 Jasechko, J ................................... 124, 138
Holdstock, J. S .................................... 343 Jasper, H ...................................... 295, 310
HoUen, K. M ........................................ 245 Jastrzembski, J .............................. 227, 241
Hollingshead, A ................................... 385 Java, R. I .............................................. 384
Holmes, V .................................... 335, 341 Jeffreys, D .................................... 323, 341
Hopkins, W .............. 22, 27, 94, 95, 96, 97, Jennings, J.M ........................ 136, 138, 184
98, 99, 108, 383, 421 Jernigan, T ........... 281, 285, 288, 289, 291,
Horn, L. C .................................... 218, 270 319, 323, 339, 341
Houle, S ....................................... 385, 389 John, E.R ............................................. 312
Houlihan, J .................................... 204, 216 Johnson, N.F. 31,37,43,51,52,68
Houx, P ........................................ 410, 422 Johnson, S .......... 21, 27, 88, 108, 221, 235,
Howard, D ........... 1, 12, 27, 50, 67, 92, 94, 241,242, 244, 303, 309, 310, 313,343,
108, 112, 118, 124, 125, 126, 128, 137, 345, 350, 351,355, 356, 372, 380, 381,
165, 167, 229, 230, 241, 346, 347, 358, 385, 409, 419, 422, 423
369, 384, 400, 409, 422, 423 Johnston, J.C .................. 12, 28, 31, 38, 65
Howard, J.H ........................................ 137 Jolles, J ................................................. 422
Jonasson, J ............ 32, 66, 72, 84, 226, 240
Author Index 435
Jones, T ........ 133, 136, 140, 141,202, 219 413, 424

Jonides, J ....................................... 262, 269 Keppel, G ......................................... 45, 68
Joordens, S ..................... 32, 41, 42, 44, 68 Kertesz, A ............ 210, 215, 224, 236, 238
Jorm, A ......................................... 223,242 Ketrtesz, A ................................... 187, 197
Josiassen, R. C ...................................... 310 Kimura, M ............................................. 340
Joubran, R ........................................... 292 King, J .......................................... 314, 341
Judd, C ............................................ 45, 69 Kinsbourne, M ...... 149, 150, 151, 154, 158,
Junque, C ............................................. 343 160, 167, 237, 242
Just, M ............. 2, 177, 182, 184, 185, 315, Kintsch, W ................... 177, 185, 369, 385
317, 335, 337, 341, 343, 349 Kirshner, H ................................... 210, 217
Kjelgaard, M. M ................................... 291
K
Kleigl, R ............................................. 2, 28
Kahn, H.J ............................................. 167 Kliegl, 1L ........................ 15, 25, 28, 45, 68
Kaplan, E.F .......................................... 388 Klingberg, F ......................................... 244
Kaput, S ....................... 350, 372, 385, 389 Klitz, T ................................. 206, 209, 216
Karayanadis, F ...... 357, 361, 362, 380, 385 Kluender, R .......................................... 389
Karayanidis, F ............... 138, 329, 330, 341 Kluger, A ............................................. 384
Karis, D ............... 301, 302, 309, 310, 351, Knight, R .............. 275, 292, 357, 372, 373,
352, 383, 385,420 374, 385, 389, 411, 422
Kase, C ......................................... 195, 198 Knoll, J ................................................. 242
Kaszniak, A ......... 210, 215,236, 237, 242, Knopman, D ................ 278, 279, 280, 284,
254, 268, 272, 275, 277, 278, 280, 284, 285, 291
290, 291 Kok, A ......... 296, 297, 305, 307, 310, 313,
Katz, L ................................. 201, 217, 241 313, 390, 409, 413,414, 415,416, 417,
Katzman, R .................. 222, 223, 243, 245, 424, 425
272, 292 Kolers, P ...................... 114, 138, 139, 169
Kaufman, M ..................................... 47, 64 Kopell, B .................. 22, 27, 383, 385, 386,
Kausler, D ....... 74, 85, 103, 108, 112, 117, 409, 421,423
138, 145, 168 Kornblum, S ..................... 24, 28, 311, 342
Kawas, C ............................................. 269 Kornhuber, H ............................... 298, 310
Kazmerski, V ....... 345, 350, 357, 358, 362, Koslow, S .................................... 308, 339
363, 364, 385 Koss, E ................................ 202, 210, 216
Keane, M ...... 274, 275, 276, 282, 285, 291 Kosslyn, S ............................ 154, 167, 168
Keating, C ....................................... 73, 85 Kramer, A ............................................ 309
Keefe, D ....................................... 263, 270 Krampe, 1L ....................................... 25, 28
Keenan, K ............................................ 293 Krueger, L .......... 31, 47, 68, 72, 74, 75, 85,
Kellas, G ...... 42, 50, 68, 87, 89, 90, 92, 93, 89, 97, 109, 243
94, 95, 97, 98, 100, 101, 102, 103, 104, Kucera, H ....................... 43, 51, 68, 76, 85
105, 108, 109, 226, 227, 229, 230, 235, Kuenkel, H ........................................... 311
240, 242 Kurylo, D ..................................... 236, 242
Kelley, C ....................................... 124, 138 Kushkowski, M .................................... 244
Kelly, M ........................................ 210, 217 Kutas, M ............ 89, 90, 97, 109, 298, 299,
Kemper, S ............ 103, 108, 315, 317, 334, 300, 301, 302, 303, 305, 308, 310, 311,
341, 342, 349, 385 312, 314, 318, 320, 324, 325, 326, 327,
Kempler, D .......................................... 245 328, 331, 336, 341, 342, 343,350, 351,
Kenemans, J ................. 296, 297, 310, 312, 353, 355, 385, 386, 389, 409, 422
436 Author Index
Kvak, D ................................ 39, 41, 43, 64 255, 269, 270, 317, 342, 397, 398, 399,
Kynette, D ............ 103, 108, 317, 3 4 1 , 3 4 2 401,402, 407, 408, 410, 421,422, 423
Lindamood, T ....................................... 311
Lindem, K ............................................ 184
La Heij, W ........................................... 168 Lindenberger, U ........................... 210, 217
Lachman, M ......................................... 186 Lindsay, R.D ........................................ 388
Lafleche, G .......................................... 381 Lindsley, D.B ....................................... 310
Lamain, W ........................................... 312 Liss, L .......................................... 202, 216
LaMay, J ....................................... 214, 217 Livingstone, M ..................... 204, 217, 419
Larrabee, G ................................... 145, 170 Lloyd, J ...................................... 39, 43, 66
Lasaga, M ............... 92, 108, 165, 230, 241 Lockhart, R .................................. 173, 184
Laver, G ........ 6, 12, 15, 22, 28, 63, 68, 92, Loewen, E.R ........................................ 382
108, 113, 133, 138, 157, 158, 168, 179, Loftus, G ....... 114, 136, 214, 217, 262, 268
1 8 5 , 2 5 5 , 2 5 7 , 258, 269, 402, 422 Loftus, E.F ........................................... 136
LaVoie, D ............................................ 138 Logan, G .......................... 21, 28, 306, 308
Lawson, D. S ....................................... 343 Logie, R ............................................... 289
Lawton, T ..................................... 204, 217 Long, G ..................... 27, 49, 68, 291, 316,
Laxon, V ......................................... 73, 85 382, 4 2 1 , 4 2 4
Layton, B ......................................... 74, 85 Lopez, O ...................................... 223,242
Lecluyse, K .......................................... 219 Lorch, R ............................... 185,224, 239
Lecours, A ............................................. 70 Lorge, I ............................................ 76, 85
Leddsert, B .......................................... 140 Lorist, M ...................................... 312, 387
Lee, M .............. 23, 27, 179, 185, 276, 292 LoSchiavo, F .......................................... 64
Legge, G ....... 200, 204, 207, 214, 2 1 5 , 2 1 7 Lovegrove, W .............................. 204, 217
Lehmkuhle, S ................................ 200, 215 Loveless, M .................................. 179, 186
Leiman, J ....................................... 89, 109 Luby, M ............................................... 339
Leirer, V .............................................. 185 Lucas, M ........................................ 89, 108
Lennie, P ......................................... 35, 66 Luce, P. 73, 85
Leonard, G .......................................... 386 Luce, R.D ............................. 6, 25, 28, 342
Lesgold, A .............................. 67, 322, 343 Lucj, S. J .............................................. 311
Lessell, S ............................................. 218 Luck, S. J ............................................. 342
Levelt, W ......... 40, 68, 143, 144, 148, 149, Luebker, A ................................... 214, 217
154, 167, 168 Luis, J.D ............................................... 384
Levin, H ........................................ 145, 170 Lunneborg, C ....................................... 341
Levy, J .................................. 202, 2 1 1 , 2 1 6 Lupker, S ............................................. 168
Lewis, D ....................... 201, 217, 322, 341 Luria, A ................................ 374, 377, 385
Light, L ..... 87, 94, 99, 108, 109, 112, 115, Lyons, K ............. 92, 93, 95, 104, 105, 108
127, 129, 133, 138, 223,242, 271,273,
M
292, 315,340, 341,342, 344, 346, 347,
368, 369, 385 Macar, F ....................................... 300, 308
Lillyquist, T.D ...................................... 419 Mack, L ............................................... 291
Lim, K .......................................... 343, 385 Mack, R ............................................... 269
Lima, S .......... 2, 15, 21, 22, 24, 27, 28, 31, Mackay, D .............................. 85, 113, 136
39, 44, 52, 54, 60, 61, 62, 63, 68, 69, 113, MacKinnon, G .......................... 66, 84, 198
138, 139, 157, 158, 159, 160, 161, 162, Madden, D ..... 1, 12, 28, 30, 31, 36, 37, 40,
163, 164, 167, 168, 176, 185,226, 244,
Author Index 437
44, 45, 46, 47, 48, 49, 50, 51, 52, 53, 54, McCallum, W ....... 296, 298, 300, 309, 311,
55, 56, 58, 59, 60, 61, 63, 64, 68, 69, 74, 313, 327, 342, 343, 411,425
76, 81, 83, 85, 92, 108, 113, 135, 138, McCann, R ................................. 32, 38, 65
144, 153, 157, 162, 166, 197, 224, 226, McCarreU, N .................................. 89, 107
227, 228, 230, 231,232, 238, 242, 256, McCarthy, G ........ 301,303, 305, 307, 311,
270, 393, 398, 399, 400, 401,402, 409, 312, 339, 343,374, 386, 409, 411,422,
413,419, 422 425
Magee, L ...................... 143, 146, 156, 169 McCauley, C ........................................ 168
Magliero, A .................. 303, 311,409, 422 McClelland, J ............ 6, 15, 28, 31, 32, 33,
Mahlmann, J ........................................ 291 35, 37, 38, 39, 41, 42, 44, 45, 69, 70, 88,
Mahurin, R ........................... 224, 238, 242 89, 97, 102, 108, 109, 221,225, 243,
Malec, J ........................................ 222, 242 245
Maletta, G ........................................... 244 McCloskeyM ................................ 196, 197
Mandler, G ........... 350, 359, 369, 384, 386 McConaghy, N ..................... 138, 341,385
Mangun, G ................... 298, 311,318, 321, McCrary, J ................................... 323, 340
342, 343, 387 McCulla, M .................................. 223,243
Maniscalco, C ............................... 410, 422 McCullough, B ............ 146, 150, 154, 165,
M/intyl/i, T ........................................... 139 166, 169
Marcel, A ...................................... 224, 243 McDermott, K. B ................. 140, 293, 387
Marder, K ............................................ 383 McDonald, J .............. 37, 69, 130, 138, 185
Margolin, D .......... 223,224, 243, 244, 269 McDowd, J .......... 171, 185, 234, 243, 369,
Margolis, R ................................... 236, 243 382
Marin, O ....... 194, 197, 198, 225, 243, 312 McEvoy, C ................ 43, 69, 144, 168, 386
Markowitsch, H.J ................................. 389 McFarlane, D ............................... 199, 219
Marr, D ................................ 34, 35, 49, 69 McGaw, B ............................................ 269
Marroco, R ................................... 200, 217 McGeer, P .................................... 319, 342
Marsh, G ............. 301,308, 310, 327, 340, McHugh, P ................................... 251,269
409, 420, 422 Mclntyre, J ................................... 350, 386
Marshall, J ............ 189, 191, 192, 197, 198 Mclsaac, H ........................................... 389
Martin, A ............. 169, 204, 217, 218, 221, McKann, G ........................................... 292
223,224, 237, 241,243,244, 248, 252, McKeel, D .................... 222, 223, 239, 243
253,254, 263, 269, 270, 271,286, 292 McKhann, G ......................... 223, 243, 272
Martin, M.. 89, 92, 95, 100, 101, 105, 108, McLachlan, D .............. 139, 202, 214, 217,
109, 218, 388
Martinerie, J.M .................................... 309 McLeod, B ................................... 262, 270
Marton, M .................................... 339, 342 McNamara, T ................... 31, 67, 263,270
Masson, M ............... 31, 69, 118, 138, 139, McNeal, M ........................... 39, 41, 43, 64
140, 384 McSorley, P ......................................... 290
Mathalon, D. H .................................... 343 Mead, G ............................................... 138
Matilla, W.R ........................................ 419 Meertse, K ........................................... 186
Mattis, S ............................... 248, 250, 270 Mendez, M ........................... 204, 205, 216
Max, W ........................................ 222, 243 Mergler, N ........... 149, 150, 151, 158, 159,
Mayes, A .............................. 301,312, 386 168
Maylor, E ............................. 113, 138, 139 Merikle, P ............................................ 386
Mayr, U .................................. 2, 15, 25, 28 Merling, A ............................................ 268
McAndrews, M ....... 92, 108, 165, 230, 241 Merves, J. S ......................................... 242
438 Author Index
Merz, G. S ........................................... 246 244, 382,

Messner, S ........................................... 244 Morrow, D ............ 171, 176, 179, 181, 185
Metsala, J. L ........................................ 246 Mortel, K.F .......................................... 388
Meyer, D ......... 24, 28, 29, 40, 68, 89, 109, Mortimer, J. A ...................................... 244
143, 148, 154, 168, 184, 305, 311, 342, Morton, J ....... 39, 43, 69, 73, 85, 114, 139,
388 220, 225, 244
Michalewski, H ............................. 410, 423 Moscovitch, M ..... 139, 190, 194, 198, 202,
Michel, J.P ........................................... 382 214, 217, 218, 346, 348, 351, 357, 358,
Michie, P. T ......................................... 343 376, 381,386, 389
Mickel, S. F ......................................... 290 Mueller, J ......................................... 74, 85
Miezin, F ...................................... 281,293 Muente, T .................................... 301, 311
Milberg, W ........... 136, 233, 238, 250, 268 Mulder, G. ............ 308, 310, 312, 340, 420,
Miles, J ................................................ 312 422, 424
Milgram, N ................................... 111, 139 Mulder, L.J ................................... 308, 312
Miller, G ....................................... 409, 422 Mullen, B ..................................... 119, 140
Miller, J ............... 201, 215, 217, 222, 223, Mullennix, J ...................................... 43, 66
236, 237, 239, 243, 245, 305, 307, 311, Muller, G ...................................... 238, 244
312, 319, 334, 342, Munte, T .............................. 318, 343, 387
Millikan, J ......................... 74, 85, 226, 245 Murdoch, B.E ...................................... 290
Mills, D. L ........................................... 343 Murdock, B .......................... 214, 217, 219
Milner, B ...................................... 381,386 Murphy, D.1L ....................................... 139
Miner, C. S .......................................... 290 Murray, J ...................................... 204, 218
Minsky, M ....................................... 38, 69 Mushaney, T ................................ 179, 185
Mintun, M ........................................... 386 Myerson, J ............... 1, 5, 8, 15, 27, 28, 30,
Miszczak, L .................................. 138, 185 44, 45, 52, 54, 61, 62, 63, 68, 69, 75, 85,
Mitchell, D ........... 110, 112, 113, 114, 117, 113, 138, 139, 144, 156, 157, 158, 167,
118, 124, 129, 131, 133, 136, 139, 141, 168, 175, 176, 185,226, 244, 254, 255,
149, 151, 152, 153, 158, 160, 164, 168, 257, 269, 270, 317, 342, 391, 395, 396,
186 397, 398, 399, 400, 401,404, 406, 407,
Mitchiner, M ........................................ 389 408, 417, 421,422, 423, 424
Mittelman, M. S .................................... 384
N
Mohr, J ......................................... 195, 198
Mohs, 1L ........................... 22, 27, 383,421 N'Kaoua, B ........................................... 384
Molenaar, P .......................... 157, 168, 417 Naatanen, 1L ................. 299, 311, 318, 343
Monsell, S ........ 39, 43, 69, 72, 73, 85, 143, Nagy, M ............................................... 387
170, 175, 185 Namazi, K. ....................................... 74, 83
Montan, B ........................................... 424 Naylor, L ...................................... 202, 216
Monti, L ............... 274, 277, 284, 285,292 Nebes, 1L ............... 92, 108, 109, 149, 150,
Moore, K ........ 94, 103, 108, 109, 306, 311 151, 154, 158, 160, 168, 204, 214, 218,
Morris, P ..... 150, 154, 165, 167, 220, 222, 223, 224, 226, 233,242, 244, 247, 248,
223, 235, 239, 241, 243, 244, 245, 283, 252, 254, 256, 257, 270, 286, 292, 410,
292, 382, 414 422
Morris, J. C ....... ................................... 239 Neblett, D .................................... 133, 136
Morris, P.E .......................................... 167 Neely, J ............ 43, 69, 226, 244, 257, 262,
Morrison, D .......................... 391,419, 420 263, 270
Morrison, M ........ 201, 202, 203, 217, 221, Neisser, U .................. 34, 35, 69, 199, 218
Author Index 439
Nelson, D ........ 43, 69, 204, 216, 237, 241, 381,419, 422
250, 271, 363,386, 409, 423 Pare, E ......................................... 202, 218
Nestor, P.G .......................................... 244 Parish, D .............................. 200, 207, 218
Nettlebeck, T ................................... 74, 85 Park, D .................. 127, 128, 139, 145, 169
Neville, H .... 301, 311, 324, 325, 327, 341, Parker, E.S ........................... , ............... 384
342, 343,351, 386 Parkin, A .............. 128, 139, 140, 350, 373,
Newcombe, F ................ 189, 191, 192, 198 386
Newell, A ........................................ 26, 28 Parmalee, C.M ...................................... 168
Newsome, S ................ 31, 37, 69, 202, 218 Partridge, F .......................... 275, 286, 292
Nissen, M .... 202, 218, 278, 279, 280, 284, Pashler, H ............................. 12, 28, 49, 70
285,287, 291, 292 Pate, D ................................................. 244
Nitsch, K ....................................... 89, 107 Patterson, M ......... 46, 64, 74, 83, 133, 141,
Noble, C.A .......................................... 244 189, 192, 197, 198, 224, 225,243, 244,
Nobre, A.C .......................................... 339 410, 419, 423
Noel, R ....................................... 31, 37, 69 Paul, S .......... 42, 50, 68, 87, 89, 90, 92, 94,
Nolan, B.H .......................................... 244 95, 96, 97, 98, 99, 100, 104, 106, 108,
Norman, D ................... 172, 186, 317, 341, 109, 143, 197, 198
345, 389 Paulsen, J ..................... 277, 280, 285, 292
Norris, D .............................. 209, 214, 218 Pearce, J.W .......................................... 424
Nunez, P ....................................... 318, 343 Pearl, S ................................................. 343
Pearson, 1L ................................... 237, 240
O
Pechman, T .................... 68, 143, 148, 168
O'Brien, E ..................... 177, 178, 184, 185 Peli, E .................................. 200, 204, 218
O'Regan, J ................................ 73, 84, 335 Pellegrino, J ............ 67, 146, 150, 154, 165,
Ober, B ............ 50, 69, 113, 133, 226, 234, 169
244, 247, 248, 250, 251, 253,270, 271, Pelli, D ................................. 200, 207, 217
274, 284, 286, 292, 357, 386 Perfect, J .............. 45, 52, 62, 70, 144, 157,
Obler, L ................ 210, 216, 225,239, 240 179, 186, 275, 292, 391,406, 407, 423
Ochsner, K.N ....................................... 293 Perfetti, C ....................... 89, 108, 322, 343
Ojemalm, J ............ 281, 293, 339, 343, 388 Perlmutter, M ............................... 112, 139
Oliver, W ......................................... 31, 67 Pernier, J .............................................. 386
Olson, D ........................................ 89, 109 Perrin, P ............................... 371, 372, 386
Onifer, W ....................................... 89, 109 Perry, N.W ........................................... 309
Osman, A ............. 22, 26, 30, 65, 305, 306, Persanyi, M .......................... 206, 209, 216
307, 311,410, 411, 419 Peters, L ....................................... 112, 136
Ostergaard, A ....... 281, 285, 288, 289, 291 Petersen, S ........... 281,293, 364, 386, 387,
Owsley, C ......................... 48, 69, 225, 244 388
P Pezdek, K ..................................... 145, 169
Pfefferbaum, A ............ 319, 343, 345, 359,
Paap, K ................................. 31, 37, 39, 69 383, 385, 386, 409, 421, 423
Paivio, A .............. 143, 145, 146, 154, 156, Philbin, D ..................................... 1, 24, 29
169 Phillips, G ..................................... 199, 219
Paller, K ....... 301, 311, 312, 350, 351,352, Picton, T ....... 318, 341,350, 384, 409, 423
353,355, 363, 366, 369, 386 Pierce, T ............... 48, 68, 76, 85, 113, 138,
Palmon, I~ ........................................... 186 232, 242, 400, 401, 422
Palumbo, C.L ....................................... 388 Pillsbury, W ...................................... 30, 70
Parasuraman, R. ........... 223, 237, 244, 313,
440 Author Index
Pinker, S ..................................... 38, 43, 70 R

Pirozzolo, F .......... 224, 226, 238, 242, 244
Raaijmakers, W .................... 136, 274, 289
Pisoni, D ....................... 235,237, 241,245
Raaijmakers, J ...................... 136, 274, 289
Pitts, D ............................................ 48, 70
Rabbitt, P ..................... 103, 109, 374, 387
Plouffe, L ...................................... 115, 140
Radvansky, G ............................... 316, 340
Plude, D ............................................. 1, 28
Ragot, I t ...................................... 303, 312
Pocock, P.V .................................. 311, 342
Raichle, M ............ 281, 293, 344, 377, 386,
Pogue, J ............................................... 388
387, 388
Polich, J ........................ 345, 387, 409, 423
Rajaram, S .................................... 114, 140
Pollatsek, A ............. 31, 37, 38, 39, 40, 43,
Randolph, C ................ 275, 276, 280, 284,
51, 66, 68, 73, 84
286, 292
Poon, L ............ 1, 5, 26, 28, 30, 46, 53, 60,
Raney, G .............................................. 309
62, 65, 66, 69, 74, 75, 84, 85, 91, 92, 94,
Rankin, J ...................................... 366, 387
107, 112, 136, 139, 149, 151, 152, 153,
Rapp, P.E ............................................. 309
154, 158, 159, 160, 161, 162, 165, 166,
Raskind, C.L ........................................ 421
168, 169, 170, 175, 185,221, 226, 228,
Rawles, J.M .......................................... 343
239, 244, 254, 268, 270, 345, 365, 387,
Raymond, J .................................. 204, 218
391,392, 396, 420, 4 2 1 , 4 2 3 , 4 2 4
Rayner, K ..................................... 109, 340
Porges, S.W ......................................... 308
Raz, N ................... 272, 289, 293, 350, 388
Posner, M ................ 36, 66, 197, 262, 271,
Reed, V. S ............................................. 386
312, 386
Reese, H ....................................... 145, 170
Poston, J.N .......................................... 241
Regan, J ............ 73, 84, 204, 218, 335, 341
Potter, M ............. 143, 146, 150, 154, 156,
Reingold, E .......................... 282, 293, 386
165, 169
Reinmuth, O.M ..................................... 242
Potvin, A.R. ......................................... 424
Reisen, C.A .......................................... 384
Pouraghabagher, A.R. .......................... 424
Reminger, S.L .............................. 292, 383
Powell, T ...................... 221,237, 240, 244
Renault, B .................................... 318, 343
Pratt, H ......................................... 410, 423
Requin, J ..................... 243, 305, 308, 311,
Press, G.A ........................................... 341
312, 342
Price, D ............... 202, 219, 222, 223, 243,
Reynolds, C.F ....................................... 242
245, 272, 292
Rezek, D.L ......................................... ..242
Prince, A ..................................... 38, 43, 70
Richardson-Klavelm, A ................. 292, 387
Pritchard, W ................................. 303, 312
Richter, P,.A ......................................... 244
Proctor, R ..................................... 307, 312
Ridderinkhof~ I~ .......... 298, 303, 304, 305,
Propper, P,. ............................... 46, 64, 419
306, 307, 312, 320, 409
Puce, A ................. 323, 339, 343, 356, 387
Riege, W ...................................... 145, 167
Pugh, K ........................................ 235, 242
Riegel, I~F ........................................... 420
Puglisi, J ....................................... 145, 169
Riehle, A .............................................. 311
Pujol, J .......................................... 319, 343
Rimmer, S .................................... 201, 217
Pulido, A ...................................... 118, 138
Ritchie,I~k ............................................. 140
Putnam, L ............................. 300, 309, 383
Ritter, W ...... 115, 137, 296, 309, 312, 343,
Q 345, 354, 361, 383, 387, 409, 424
Rizlo, J ................. 203, 216, 218, 225, 240
QuiUian, M ........................ 43, 66, 262, 268
Roberts, R.C ........................................ 343
Quinlan, P ........................................ 31, 68
Robinson, L .................. 226, 241, 412, 424
Robson, J ..................................... 199, 216
Author Index 441
Rocca, W ...................................... 221, 245 Salmon, D .... 137, 253, 269, 275, 276, 277,
Rodgers, W ................................. 31, 62, 67 278, 280, 282, 286, 290, 291, 292, 293,
Rodriguez, R. ...................................... 245 328, 384
Roediger, H ......... 112, 114, 116, 117, 119, Salthouse, T ........ 1, 6, 8, 9, 11, 15, 21, 28,
128, 140, 141, 272, 282, 283, 293, 347, 30, 44, 45, 46, 61, 70, 75, 84, 85, 108,
349, 387 110, 117, 118, 126, 136, 140, 171, 172,
Rogers, W ..... 3, 7, 23, 27, 44, 57, 66, 144, 174, 175, 176, 179, 184, 186, 218, 223,
167, 258, 269, 388, 404, 405,406, 407, 244, 245, 254, 256, 270, 271, 315, 343,
408, 420, 421, 423,424 345, 348, 369, 379, 386, 388, 391, 392,
Rohrbaugh, J ........ 298, 305, 312, 313,381, 393, 395,402, 404, 410, 421,424
385, 419, 422 Sandberg, M.A ..................................... 218
Romano, J ............................. 110, 111, 137 Sanford, A .................................... 177, 186
Rose, T.L ............................................. 140 Sarazin, F.F .......................................... 388
Rosen, A ....................... 200, 204, 217, 218 Sawaguchi, T ............................... 319, 343
Rosen, T.J ............................................ 218 Sayer, L ................................................ 388
Rosenbloom, M.J ................................. 383 Scarborough, H .................... 140, 180, 183
Rosenthal, R. ....................................... 139 Schacter, D .......... 111, 114, 115, 116, 118,
Rosinski, R ................................... 146, 169 126, 137, 140, 141, 272, 276, 280, 281,
Rosier, F .............................................. 343 290, 293, 347, 349, 358, 365, 382, 388,
Ross, J .......... 204, 214, 215, 217, 263, 270 389
Roth, W ........... 22, 27, 383, 386, 389, 409, Schaie, K.W ............ 1, 29, 66, 70, 85, 166,
421,423 167, 184, 239, 343,420, 421,422, 424,
Rothfleisch, J ....................................... 268 425
Roucos, S.E ......................................... 309 Scheerer, E ....................................... 73, 85
Rougier,A ............................................ 384 Scheibel, M .......................... 348, 349, 388
Royer, F ........................................ 211, 219 Scheibel, A.B ....................................... 388
Rubenstein, H ................... 74, 85, 226, 245 Scherg, M .................................... 380, 388
Rubin, E.H .................................... 239, 245 Schiefers, H ...................................... 40, 68
Rubin, G. S ........................................... 217 Schleske, M.M ..................................... 217
Ruchkin, D ........... 336, 343, 354, 361, 387 Schlotterer, G ............................... 202, 218
Rugg, M ...... 115, 140, 301, 312, 329, 330, Schmidt, A.L ........................................ 311
343, 350, 351, 353, 355, 356, 357, 359, Schmitt, F.A ......................................... 139
360, 362, 363, 364, 372, 387, 389 Schneider, W .................... 23, 29, 262, 271
Ruh, J ........................................... 182, 186 Schoenberg, B.S ................................... 245
Rumelhart, D ....... 6, 15, 28, 31, 32, 33, 35, Scholz, M .............................................. 311
37, 38, 39, 44, 69, 70, 221, 225,243 Schreiber, T ...................................... 43, 69
Rusinek, H ........................................... 384 Schriefers, H ................ 143, 148, 154, 168
Russo, R. ...................................... 128, 141 Schuebel, K .................................. 204, 215
Rybash, J ..... 21, 26, 27, 28, 166, 167, 168, Schultz, D.W ................................ 309, 421
244, 358, 388, 419, 423 Schulz, R. ..................... 113, 140, 222, 246
Rypma, B ...................... 234, 241, 315, 340 Schvaneveldt, 1~ ................. 37, 69, 89, 109
Schwartz, M ................ 137, 188, 194, 198,
S
243,290
Sadun, A ....... 201, 202, 203, 215, 217, 218 Schweickert, R. ...... 2, 6, 12, 13, 21, 25, 29
Saffran, E .............. 189, 192, 193, 194, 198 Seergobin, K. ................................... 32, 65
Salasoo, A .................................... 235, 245 Segui, J ............................................ 73, 84
442 Author Index
Seidenberg, M .......... 33, 38, 39, 41, 42, 70, 424

89, 109, 221, 225, 245, 248, 268, 286, 290 Smith, A.D ........................................... 169
Sekuler, R ......... 48, 69, 200, 204, 213, 218 Smith, G ..... 28, 69, 85, 168, 185, 242, 270,
Selffidge, O. 34, 70 423,424
Semple, J ...................... 223, 241, 257, 269 Smith, J.E.K. .......................................... 28
Sereno, M.I .......................................... 340 Smulders, F .......... 296, 297, 305, 307, 310,
Seymour, P ................... 143, 146, 156, 169 312, 402, 403, 404, 413, 424
ShaUice, T ............. 192, 198, 350, 372, 388 Snodgrass, J ......... 143, 146, 150, 153, 154,
Sharp, T ........................................ 113, 137 165, 169, 210, 218, 288, 293, 345, 383,
Shaw, R ............... 12, 27, 50, 67, 112, 127, 388
128, 139, 165, 167, 175,230, 241, 349, Snyder, C ..................................... 262, 271
384, 388, 400, 422 Soederberg, L ............................... 179, 186
Shelton, J ...................................... 263,271 Somberg, B ........ 8, 9, 11, 21, 28, 254, 271,
Shenaut, G ............... 50, 69, 113, 133, 234, 395, 410, 424
244, 248, 250, 270, 271, 274, 284, 286, Spee, C ........................................ 201, 216
292, 357, 386 Spencer, W ............ 276, 290, 339, 350, 388
Shepherd, R .................................. 365, 388 Spencer, D.D ........................................ 339
Shernoff, E ...................................... 43, 66 Spencer, M ........................................... 290
Shifffin, R ......... 23, 29, 173, 184, 262, 271 Sperber, R ............................................ 168
Shimamura, A ...... 275, 280, 282, 284, 285, Sperling, G ................... 199, 200, 207, 218
288, 291,293, 301, 312 Spinnler, H ........................................... 289
Shults, C.W ......................................... 291 Spong, P .............................................. 3 l0
Shuttleworth, E ............................. 224, 245 Squire, L .............. 112, 115, 141,269, 272,
Siegel, A ....................................... 146, 169 275, 281,282, 289, 290, 291, 293, 301,
Siemsen, D ............................................. 69 312, 319, 344, 347, 361, 364, 382, 388
Silverman, S .................................. 202, 219 Squires, N ............ 351, 377, 388, 409, 411,
Simolke, N ............................ 110, 111, 137 421,425
Simon, H ......... 26, 28, 145, 170, 184, 317, Srinivas, I~ ................................... 140, 141
340, 402, 424 Stadlan, M .................... 223, 243, 272, 292
Simpson, G ........ 89, 90, 97, 108, 109, 227, Stadtlander, L ............................. 41, 70, 72
230, 235, 242, 245, 345, 374, 376, 377, Stanger, B.Z ......................................... 291
378, 380, 383 Stanovich, K. ............... 104, 109, 236, 245,
Singer, M ...................................... 173, 184 257, 271,285,293
Singh, A ............................................... 138 Staplin, L ............................................ 1, 29
Sinha, U ................................ 236, 237, 245 Start, A ................................ 409, 421,423
Sirevaag, E.J ........................................ 310 Stelmack, R. ................. 115, 140, 301, 312
Ska, B .................................................. 167 Steinberg, S .............. 2, 5, 29, 50, 70, 204,
Skinner, B.F .......................... 110, 113, 141 218, 317, 345, 395,409, 418, 424
Skovronek, E ........................ 174, 176, 186 Stewart, D ............................................ 240
Slane, S .................................. 30, 114, 135 Stine, E .............. 46, 63, 70, 171, 176, 178,
Slimp, J.C ............................................ 424 179, 181, 182, 185, 186, 223,235, 236,
Sliwinski, M ......................................... 291 237, 246
Sloane, M ..................................... 225,244 Stojack, C .................................... 179, 185
Small, B.J ............................. 138, 185, 384 Stollery, B .................................... 103, 109
Smelcer, J ........................................ 24, 29 Stoltzfus, E .................. 234, 241, 315, 340
Staid, H ....... 303,305, 307, 312, 409, 414, Storandt, M ........... 220, 223, 239, 245, 271
Author Index 443
Strand, T ......................................... 51, 68 348, 389

Strater, L ............................................. 136 Tomoeda, C.K. ..................................... 268
Strayer, D ............................. 410, 4 1 3 , 4 2 4 Tomsak, R ............................ 206, 209, 216
SturgiU, D ............................................ 240 Torack, R ............................................. 243
Stuss, D ....... 348, 349, 374, 380, 388, 389, TorigoeY ..................... 201,202, 215, 216
409, 423 Torres, I ............................... 272, 289, 293
Sullivan, E.V ................................. 343,383 Toth, E ................................. 282, 283,293
Sutton, E ............. 117, 137, 300, 301, 302, Touchon, J ........................................... 140
309, 312, 362, 383, 387 Tourtellotte, C ...................................... 424
Suzuki, J.S ........................................... 342 Townsend, J ......... 2, 6, 12, 21, 25, 29, 202,
Swenson, M . R ..................................... 292 210, 216
Swick, D ....................... 357, 372, 373, 389 Trabasso, T .................................. 173, 184
Swihart, A.A ........................................ 242 Trahan, D ..................................... 145, 170
Swinney, D .................................... 89, 109 Treat, N ....................................... 145, 170
Syndulko, I~ ........................................ 424 Trembley, M ........................................... 70
Szafran, J ......................................... 74, 84 Trempe, C ............................ 200, 204, 218
Szirtes, J .............................................. 342 Trick, G ....................................... 202, 219
Trobe, J ........................................ 225,245
T
Trosset, M.W ....................................... 268
Tabossi, P ...................................... 89, 109 Troyer, A ..................................... 176, 186
Taft, M ........................... 31, 39, 70, 74, 85 Tueting, P .................................... 308, 312
Tainturier, M ............ 50, 55, 58, 60, 62, 70 Tukmachi, E. S ...................................... 341
Tanenhaus, M ................................ 89, 109 Tulving, E ............ 111, 112, 114, 115, 132,
Taraban, 1L ....................... 88, 97, 102, 109 140, 141, 347, 350, 365, 372, 385, 389
Tardif, T .............................................. 184 Tun, P .......................... 171, 174, 176, 186
Tate, C.S ............................................. 185 Turing, A ............................. 32, 38, 43, 70
Taylor, J.M .......................................... 243 Turner, J ............... 204, 205, 206, 210, 214,
Teghtsoonian, M .................................. 388 216, 219
Terry, R. ....... 201, 217, 222, 245, 349, 389 Twilley, L ......................................... 32, 65
Theios, J .......... 30, 64, 143, 144, 146, 147, Tyler, L ........................................ 235, 245
149, 151, 154, 155, 156, 157, 158, 159, Tym, E ......................................... 223,245
160, 162, 164, 165, 166, 168, 170, 214,
218, 219 U
Thomas, J ...... 23, 26, 29, 65, 85, 111, 133, Umlita, C .............................. 190, 194, 198
134, 139, 141, 149, 151, 152, 153, 158,
V
160, 161, 170,
Thomas, C ........... 202, 206, 207, 209, 210, Valdiserri, M ........................................ 140
211, 214, 216, 219, 226, 241, 268, 419, Vallar, G ...................................... 190, 198
420 van DeUen, H.J ............................. 420, 422
Thorndike, E .................................... 76, 85 van der Meere, J.J ................................ 420
Thurston, I ........................................... 239 van der Molen, M ........ 157, 168, 294, 298,
Tinklenberg, J.1L .................................. 421 299, 302, 303, 304, 305, 306, 307, 312,
Tipper, S ....... 234, 245, 316, 343, 374, 389 313, 390, 409, 417, 419, 424
TippeR, L ..................................... 224, 245 Van Dusseldorp, G ............... 110, 111, 137
Tomiyasu, U ........................................ 388 Van Essen, D .................................... 35, 70
Tomlinson, B ............... 221,245, 319, 341, Van Hoesen, G.W ................................ 290
Van Petten, C ..... 89, 90, 97, 109, 300, 301,
444 Author Index
308, 311,325, 331,342, 343,353, 389 Weiskrantz, L ........ 114, 141, 281,293, 389
van Woerden, G ............ 131, 136, 274, 289 Weiss, A ............................................... 419
Vanderwart, M ............................. 169, 218 Weldon, M ................... 114, 117, 140, 141
VanFleet, N ......................................... 243 Welford, A ....... 26, 47, 65, 71, 74, 85, 268,
Vaughan, H.G. ..................................... 425 395, 419, 420, 425
Vellutino, F ................................... 204, 219 Welsh, A ...................................... 235, 243
Vendrell, P ........................................... 343 Wenegrat, B.G ..................................... 386
Verfaellie, M ........................................ 383 Wenegrat, B.M ..................................... 423
Verleger, R ................... 303, 313,323, 344 Wenk, H ....................................... 202, 216
Vickers, D ....................................... 74, 85 West, R ................................................ 293
Videen, T ...................... 281,293, 387, 388 Whaley, C ........................................ 74, 86
Vorberg, D ............... 40, 68, 143, 148, 168 Wheeldon, L ................................. 143, 170
Vreeling, F.W ...................................... 422 Wheeler, D ....................................... 31, 71
Vu, H ......... 42, 50, 68, 100, 101, 108, 109 White, H ......... 12, 26, 50, 66, 92, 107, 165,
222, 228, 239, 245,400, 420
W
Whitehead, V ....................................... 268
Waag, E ................................................. 64 Whitehouse, P ....... 202, 211, 217, 219, 382
Wade, E ........................................ 113, 136 Whitfield, J.R. ...................................... 240
Wagstaff, D ........ 1, 5, 8, 27, 28, 45, 69, 75 Wickens, C ........................... 309, 410, 424
85, 157, 167, 168, 175, 185, 254, 270, Wiegersma, S ....................................... 186
3 9 1 , 3 9 5 , 3 9 6 , 408, 421,423 Wierda, M ............................................ 308
Walicke, P.A ........................................ 291 Wiggs, C .................... 1, 27, 118, 142, 346
Walker, N ......... 1, 24, 29, 67, 84, 240, 343 Wiley, J .............. 29, 63, 67, 108, 144, 157,
Wallace, B ...................... 31, 35, 36, 37, 64 167, 186, 198, 308, 309, 313,342, 382,
Waller, T .................................. 66, 84, 198 424
Walley, R ...................... 235, 246, 262, 270 Williams, D ............ 1, 26, 46, 66, 201,204,
Walter, B ............. 139, 298, 313, 345, 350, 219, 254, 268, 390, 392, 420
373,386 Williamson, G ............................... 222, 246
Ward, P.B ..................................... 138, 385 Willson, R ........................................ 74, 85
Warrington, E ...... 114, 141, 192, 198, 281, Wilson, R ......... 65, 67, 199, 200, 214, 217,
293,346, 389 219, 235, 236, 237, 242, 243,274, 275,
Watson, J ............................................. 219 277, 291,292, 380, 381,383, 385
Watts, M.C .......................................... 385 Winblad, B ........................................... 242
Waugh, N .... 23, 29, 74, 85, 133, 134, 141, Wingfield, A ................ 186, 223, 235, 236,
149, 160, 170, 172, 186, 345, 389 237, 246
Wayland, S ................................... 235, 246 Winocur, G ................. 139, 217, 346, 348,
Weale, R .......................................... 48, 70 376, 386
Webb, W ...................................... 210, 217 Winograd, E ................................. 145, 170
Weber, T ..... 30, 31, 35, 36, 37, 44, 49, 53, Winter, A.C .......................................... 313
63, 64, 83, 144, 157, 166, 197, 226, 228, Wisniewski, H .............................. 222, 246
238, 402, 418, 419 Wolfe, G .............................................. 381
Wechsler, D ...................... 45, 55, 315, 344 Wood, C ....... 301, 312, 374, 386, 411,425
Ween, J ......................................... 195, 197 Woods .......... 254, 268, 299, 313,374, 389,
Weiner, E ............................................. 139 390, 420
Weir, W.S ............................................ 388 Woods, A.M ........................................ 268
Weisbrod, S ......................................... 244 Woodward, S ............................... 327, 344
Author Index 445
Woodworth, R ................................. 31, 71 Z

Worthley, J ................................... 113, 136
Zacks, R ......... 75, 84, 92, 94, 97, 108, 109,
Wray, S ............................................... 218
179, 184, 227, 234, 241, 255, 262, 269,
Wright, C .......................... 24, 28, 414, 425
271,315, 316, 340
Zahirney, G .......................................... 268
Zandi, T ........ 149, 150, 151, 158, 159, 168
Yee, P ....................... 92, 94, 107, 228, 239
Zatorre, R.J .......................................... 242
Yelen, J ............................................ 43, 66
Zeef, E ............. 74, 84, 305, 307, 313, 390,
Yesavage, J ................................... 137, 140
414, 415, 416, 417, 425
Yonelias, A .......................................... 219
Zeki, S ......................................... 202, 219
Young..103, 118, 190, 192, 193, 197, 200,
Zelinski, E ...................... 94, 108, 315, 344
204, 212, 213, 218, 227, 228, 229, 230,
Zipursky, R.B ............................... 343,385
231,232, 233,256, 320, 321, 322, 324,
Zola-Morgan, S ............................ 319, 344
329, 330, 332, 333,335,337, 338, 363,
Zubin, J ................................................ 312
389, 393,402
447
Subject Index
392, 393, 397, 399, 402, 404, 406, 407,

A
408, 417
acquired deficits ................................... 188 anomic ................................................. 252
acquired dyslexia... 189, 190, 191, 192, 194 antecedent cause ..................................... 46
activation ........... 16, 17, 18, 19, 20, 26, 27, Anterograde amnesia ............................ 272
28, 32, 33, 39, 40, 41, 42, 43, 65, 67, 69, aphasia .................. 188, 194, 195,246, 314
70, 74, 82, 83, 88, 89, 90, 91, 92, 93, 94, Articulatory Loop ................................. 173
96, 97, 98, 99, 101, 104, 105, 107, 108, association cortex ................................. 282
109, 112, 113, 114, 135, 136, 144, 148, associative priming ....... 328, 347, 358, 388
154, 165, 166, 167, 177, 182, 183, 199, attention... 1, 25, 27, 28, 29, 37, 44, 49, 64,
238, 239, 241, 243, 247, 251, 252, 253, 90, 94, 98, 103, 104, 107, 118, 137, 165,
257, 262, 268, 270, 290, 293, 296, 299, 171, 172, 185, 186, 195, 198, 215, 223,
305, 306, 307, 308, 309, 310, 312, 315, 227, 247, 254, 263, 269, 270, 271,272,
348, 359, 362, 363, 369, 372, 377, 400, 286, 287, 289, 290, 291, 293, 297, 298,
402, 411,415,416, 417, 422, 424 299, 307, 310, 311, 313, 315, 320, 321,
adaptation to prisms ............................. 277 323, 341, 342, 343,357, 367, 401, 402,
additive factors method ........................ 418 414, 415,416, 422, 425
additive model ....................................... 22 autobiographical memories ................... 112
additive slowing .................... 143, 162, 395 Automatic cognitive processes .............. 262
adult lifespan ......................... 152, 327, 400 automatic intrusion ............................... 358
age-associated memory impairment ...... 346 autonomous models .................... 88, 89, 90
agrammatic .......................................... 194 axonal degeneration .............................. 201
agraphia ............................................... 191
alexia ................................................... 191 B
Alzheimer's disease .... 50, 69, 83, 108, 109, basal and inferior temporal lobe ............ 319
138, 139, 188, 194, 196, 199, 200, 206, baseline performance ........... 274, 275,277,
213, 215, 216, 217, 218, 219, 220, 221, 278, 284, 285, 286, 288
222, 223,224, 227, 234, 236, 237, 238, basic unit of analysis ......................... 30, 31
239, 240, 241, 242, 243, 244, 245, 246, Birren Hypothesis ......... 391, 392, 398, 399
247, 250, 268, 269, 270, 271, 272, 273, bottleneck ....................................... 35, 172
278, 279, 289, 290, 291, 292, 293,364, brain autopsy ........................................ 222
384 brain insult ............................................ 348
Amodal model ...................... 146, 147, 151 brain lesion ........................................... 187
amplitude ............................................. 199 Brinley plot ........... 8, 9, 15, 45, 48, 51, 52,
anagram ........................................ 119, 140 53, 54, 55, 61, 63, 143, 158, 179, 257,
analytic .... 10, 31, 34, 36, 41, 93, 175, 197, 403,407, 411,414, 417
448 Subject Index
Broca's aphasia .................................... 194 consciousness ........ 108, 111, 140, 172, 173
contamination hypothesis ...................... 130
C
context ......... 23, 28, 30, 35, 65, 69, 70, 88,
categorization ........ 22, 144, 154, 159, 162, 89, 90, 92, 93, 94, 95, 96, 97, 98, 99, 101,
164, 165, 169, 327, 328 104, 107, 108, 109, 112, 119, 135, 168,
category exemplars .............................. 112 175, 178, 182, 186, 187, 196, 199, 225,
cell loss ................................................ 319 230, 231, 235, 236, 237, 239, 242, 243,
central components .............................. 391 246, 247, 268, 270, 293, 296, 299, 301,
Central Executive ................................ 173 302, 313, 325, 326, 327, 328, 329, 339,
central information processing speed .... 391 340, 342, 344, 348, 349, 350, 356, 362,
cerebral blood flow ....................... 349, 372 373, 381, 387, 388, 418
cerebral hemorrhage ............................. 195 contextual instantiation ......................... 183
cerebral infarction ................................ 195 contingent negative variation ................ 298
cerebral vascular accident ..................... 195 continuous measure .............................. 314
chronop sycholophysiological ................ 410 contrast sensitivity ............... 202, 203, 204,
chronop sychophysiological ........... 390, 408, 205,209, 210, 211,213, 216, 218, 219,
409, 410, 412, 413,414, 417 223
circular reasoning ................................. 171 control processes .................... 67, 173, 184
Closed Class ................................. 323, 324 controlled cognitive processes .............. 262
cloze ............................................. 326, 331 Correspondence Axiom ........................ 397
cognition ............ 27, 38, 48, 69, 70, 84, 85, cross-modality transfer latency ...... 146, 148
108, 136, 137, 139, 143, 168, 169, 171, cross-sectional cohorts ......................... 110
175, 176, 184, 186, 215, 217, 218, 222, cued recall .... 112, 137, 272, 291, 328, 329,
242, 244, 245,270, 271, 308, 313, 315, 346, 353, 366, 367, 368, 369, 373,383,
317, 318, 339, 343, 344, 381, 384, 399, 384
421,424 Cz ........................................................ 295
cognitive impairment ..... 188, 194, 216, 285
cognitive impenetrability ...................... 190 D
Cohort model ....................................... 235 decay ....................... 16, 154, 177, 315, 376
compensating filter ............................... 213 declarative memory .............................. 319
compensatory processes ....................... 104 declining birth rates .............................. 221
completion priming .............................. 125 deep dyslexia ................................ 192, 196
complexity hypothesis .................... 46, 408 degraded stimuli ........................... 210, 213
comprehension... 84, 87, 88, 89, 90, 93, 94, degraded store hypothesis ..................... 253
96, 97, 98, 99, 105, 106, 107, 108, 109, delayed pronunciation ................... 113, 163
173, 177, 178, 179, 181, 182, 183, 184, depressed elderly .................................. 261
185, 186, 187, 188, 190, 191, 194, 195, Digit Symbol Substitution Test ............. 392
204, 213, 214, 220, 225, 235, 237, 241, direct memory ............. 272, 273, 276, 280,
243, 270, 284, 309, 316, 317, 320, 328, 282, 286, 348, 349, 356, 364, 365, 367,
329, 331, 334, 337, 340, 341, 343 372
computational slowing ......................... 162 direct priming ....................................... 117
conceptual representation ..................... 144 discourse ................ 87, 90, 97, 98, 99, 100,
confrontational naming deficits ............. 225 101, 102, 105, 106, 113, 137, 177, 178,
connectionist ........ 7, 12, 15, 18, 19, 21, 32, 180, 183, 185, 331, 334, 339, 341, 344
33, 41, 42, 43, 68, 149 Discourse priming ................................. 100
conscious introspection ........................ 112 discrimination ......... 84, 103, 168, 218, 224,
Subject Index 449
298 explicit memory .... 129, 136, 137, 138, 139,

distal stimuli ......................................... 214 140, 272, 276, 289, 290, 301, 319, 332,
distributed models ......... 39, 42, 43, 44, 149 347, 348, 350, 357, 364, 380, 382
Dm ....... 299, 301, 302, 307, 352, 353,366, explicit recollection ............................... 281
369, 380 expressive language .............................. 188
domain specificity ................................ 190 extended knowledge net ....................... 183
Dual Coding 143, 144, 145, 146, 147, 148,
F
151, 156, 157, 159
Dual stimulation ................................... 149 facilitation.. 34, 43, 95, 100, 101, 105, 106,
dual-route models ....................... 39, 40, 42 107, 114, 144, 165, 166, 240, 262, 269,
272, 278, 298, 316, 356, 357, 358, 359,
E
360, 363, 372
ecological validity ................................ 339 failures of metamemory 348
EEG ............ 294, 295, 296, 299, 302, 317, false negative ......... 2, 7, 12, 15, 19, 21, 25
320, 340, 370, 383, 385, 387 false positive...2, 7, 15, 16, 19, 21, 25, 162
electroencephalogram ................... 294, 317 familiarity ......... 42, 66, 161, 165, 169, 209,
emergent characteristics ......................... 48 210, 218, 241, 350, 353, 356, 359, 373,
encoding .......... 5, 6, 10, 22, 30, 31, 33, 34, 377, 381
35, 36, 37, 38, 40, 42, 43, 46, 47, 48, 49, fan effect ...................................... 316, 340
50, 51, 54, 56, 57, 59, 61, 63, 87, 91, featural similarity .......................... 147, 165
118, 136, 144, 145, 154, 155, 156, 158, feature extraction .......................... 209, 400
165, 171, 177, 181, 182, 199, 204, 209, field potentials ...................................... 294
213, 214, 225, 236, 242, 275, 276, 282, focal lesions .......................................... 188
283,286, 311, 312, 331, 348, 350, 351, fragmented pictures .............................. 278
352, 353, 354, 366, 367, 368, 369, 372, free association paradign~ ..................... 276
373, 376, 379, 380, 382, 385, 386, 389, free recall, ............ 112, 272, 276, 346, 364,
402, 408, 412, 413, 414 370, 373
endogenous ............ 14, 296, 297, 298, 307, frequency neighborhood ............. 77, 79, 80
309, 323 frontal dysfunction ................................ 349
episodic memory.. 107, 112, 114, 124, 126, functional equivalence .............. 2, 6, 21, 22
127, 132, 135, 138, 145, 186, 240, 292, functional localization... 188, 189, 191, 197
385, 388, 389 fusiform gyrus ...................................... 323
ERPs .... 115, 294, 296, 299, 301, 309, 314, Fz ......................................................... 295
315, 317, 318, 319, 320, 321, 322, 323,
G
324, 325, 326, 327, 328, 329, 330, 331,
332, 335, 336, 337, 338, 339, 341, 345, ganglion cell degeneration .................... 201
350, 351, 352, 353, 354, 355, 357, 359, ganglion cells ........................................ 201
360, 362, 363, 365, 370, 371, 375, 377, Gating paradigm ................................... 235
378, 380, 381, 383, 389, 409, 414 general slowing........ 2, 3, 4, 5, 6, 8, 11, 12,
event-related brain potentials .............. 294 13, 14, 15, 16, 19, 20, 21, 22, 28, 48, 61,
exception words ..................................... 42 68, 75, 104, 114, 135, 138, 168, 176,
exogenous .......... 12, 13, 14, 15, 19, 20, 2!, 254, 257, 269, 342, 390, 399, 401,402,
296, 297, 298, 307 404, 406, 408, 417, 422
expectancy... 221, 262, 263, 264, 265, 267, global aphasia ....................................... 195
269, 298, 311, 313, 324, 326, 328, 342, global models ........... 3, 4, 5, 15, 22, 23, 25
372 global precedence ................................. 200
GPC rules ........................................ 32, 40, 59
450 Subject Index
grammatical structure ............................. 87 159, 166, 167, 168, 175, 183, 184, 189,
granulovacular degeneration ................. 222 190, 191, 192, 193, 196, 197, 199, 200,
graphic code ........................................ 156 209, 213, 214, 216, 221,225, 235, 244,
254, 271,297, 299, 307, 308, 309, 319,
H
350, 372, 391, 397, 398, 404, 407, 409,
hierarchical regression ................. 45, 59, 63 412, 414, 418, 419, 420, 421,423,424
higher order processing ........ 199, 204, 210, Information-Loss Model .............. 396, 397,
215 399, 417
holistic .......... 31, 34, 36, 38, 41, 57, 58, 59 inhibition .......... 6, 15, 16, 94, 97, 144, 153,
homograph ......... 93, 97, 99, 100, 101, 109, 154, 158, 177, 200, 240, 241,243,245,
250, 251 262, 306, 315, 316, 341, 343, 377, 389
homophone ................... 119, 124, 125, 134 inhibitory deficit hypothesis. 315, 316, 325
Hybrid models ........................................ 31 inhibitory mechanism ........................ 94, 97
hybrid, horse race model ........................ 40 innately specified .................................. 190
hyperpriming ................ 248, 252, 253, 254, intelligencel, 27, 167, 217, 219, 243, 311, 342, 42
256, 257, 258, 259, 260, 261, 264, 265, Interactive Model ................................. 406
266, 267 interactive-activation models .... 90, 93, 101
intercept4, 13, 20, 61, 62, 113, 157, 162, 163,
255, 259, 392, 393, 394, 395, 398, 401,
identification ............ 22, 28, 31, 34, 35, 36, 402, 403, 409, 410, 411,412, 414, 416
37, 38, 39, 40, 43, 44, 49, 50, 51, 52, 58, interconstituent integration ................... 183
59, 60, 61, 63, 64, 65, 68, 73, 90, 91, 92, internal noise .................. 72, 74, 75, 82, 83
93, 99, 114, 115, 116, 119, 128, 131, International Electrode Placement System295
134, 138, 153, 158, 168, 193, 200, 201, intraconstituent organization ................. 183
206, 207, 209, 210, 213, 214, 216, 224, irregular words ...... 187, 191, 192, 193, 194
236, 239, 242, 274, 275, 277, 278, 279,
280, 281, 282, 283, 285, 287, 291, 292,
306, 339, 363, 364, 365, 402, 405, 422 language comprehension ...... 87, 89, 90, 93,
identification-by-components ................. 35 96, 98, 99, 106, 177, 178, 181, 183, 184,
image code ........................................... 146 187, 188, 191, 220, 317, 320
image generation task ........................... 155 language processing ....... 65, 66, 67, 68, 69,
image rotation ...................................... 159 70, 90, 93, 97, 98, 99, 101, 102, 103, 105,
Impairment in deliberate recollection .. 348 106, 108, 143, 173, 177, 178, 188, 190,
implicit memory. 1, 65, 110, 111, 112, 114, 194, 195, 196, 225, 246, 286, 299, 314,
115, 116, 117, 118, 119, 126, 127, 129, 315, 337, 339
134, 135, 136, 139, 140, 214, 272, 290, latent models ..... 1, 3, 5, 6, 7, 8, 10, 11, 12,
292, 293, 301, 345, 347, 348, 350, 351, 15, 19, 21, 24, 25, 67
358, 362, 364, 369, 380, 381, 382, 383, lateralized readiness potential ....... 298, 305,
384, 389 415
incontinence ......................................... 223 left-branching structures ....................... 317
Indirect memory ................................... 272 letter identification ......... 31, 37, 44, 50, 51,
inferential connections .......................... 177 52, 58, 59, 60, 61, 63, 206, 216
inferential processes. ............................ 182 lexical access ........... 26, 30, 32, 36, 37, 38,
information encapsulation .................... 190 39, 40, 41, 42, 43, 44, 46, 47, 50, 53, 54,
information processing ........... 6, 21, 26, 27, 57, 58, 59, 60, 61, 63, 65, 68, 74, 83, 84,
28, 29, 30, 32, 35, 38, 43, 44, 45, 46, 47,
66, 84, 145, 148, 149, 151, 154, 158,
Subject Index 451
107, 113, 135, 149, 166, 168, 182, 183, multiplicative ...... 4, 5, 6, 11, 14, 22, 23, 45,
238, 239, 243,263, 316 162, 392, 393, 395, 396, 406, 410
Lexical ambiguity ................................... 89 multiplicative model ........... 4, 5, 11, 14, 22
lexical decision task ........ 12, 31, 36, 37, 42,
N
50, 51, 52, 53, 55, 56, 58, 59, 60, 72, 73,
92, 136, 224, 225, 235, 261,267, 269, N400 ............ 299, 300, 307, 323, 324, 325,
270, 357 326, 327, 328, 329, 330, 331, 338, 342,
lexical domain ...... 1, 15, 22, 28, 52, 54, 60, 353, 354, 355, 359, 360, 361, 363, 365,
61, 62, 63, 68, 138, 168, 255, 269, 342, 372
399, 401, 422 negative intercept ......................... 395, 398
lexical instance models ........................... 38 negative priming .......... 124, 243,248, 316,
Lexical Processing Negativity ............... 324 343, 374
lexical search model ............................... 39 neocortical association areas ................. 347
life expectancy ..................................... 221 neural physiology .................................. 314
limbic system ........ 279, 280, 281,282, 388 neuritic plaques ............................ 217, 222
limited capacity .............................. 13, 173 neurofibrillary tangles ........... 201, 217, 348
lobectomy ..................................... 355, 385 neurological disorders ........................... 187
local representations ............................... 42 neuropathology .................................... 203
logogen ................. 39, 40, 41, 43, 114, 139 neurophysical ....................................... 199
long term memory ......................... 111, 316 neuropsychological ...... 139, 187, 188, 190,
LRP ....... 298, 299, 305, 306, 307, 415,416 193, 194, 196, 237, 260, 293, 346, 349,
luminance ...................................... 200, 212 350, 380, 383, 386, 388, 389
NINCDS-ADRDA ....... 223, 242, 243, 292
M
noninvasive recording ........................... 317
M-cell pathway ..................... 200, 201,202 nonlinear ............... 157, 158, 396, 397, 399
M-cells ................................................. 200 normalization algorithm .......................... 35
medial temporal lobe system .......... 319, 347 noun phrase .......................................... 315
Memory disorders ................................ 200
O
mental chronometry303, 311,409, 410, 422
mental model.. 87, 177, 178, 181, 182, 184 oddball task .................. 303, 370, 377, 378
meta-analyses157, 158, 159, 164, 165, 248, omission errors ..................................... 284
254, 259, 394, 395, 396, 397, 398, 401, on-line ...... 87, 88, 89, 90, 94, 99, 101, 104,
402, 412 105, 106, 109, 179, 182, 186, 194, 237,
meta-analysis ....... 15, 22, 28, 63, 108, 113, 288, 320, 380
117, 138, 139, 157, 159, 161, 162, 163, Open Class ................................... 323, 324
164, 179, 185, 247, 248, 255, 259, 265, operational capacity .............................. 175
394, 395,398, 400, 402, 410, 411,422 operational definition ............................ 110
middle cerebral artery ........................... 195 optic neuropathy ................................... 201
mirror-tracing task ............................... 277 orthographic code ......... 40, 68, 70, 85, 156
mismatch negativity .............................. 299 orthographic neighborhood ......... 41, 42, 72
model-based identification ...................... 35 Overhead Model ................... 397, 398, 399
modules32, 36, 87, 93, 183, 189, 190, 193, Oz ........................................................ 295
194, 196, 198, 343 P
mortality .............................................. 221
motion detection .................................. 202 P300 ....... 22, 297, 298, 299, 302, 303, 304,
MRI .............................. 281, 319, 339, 340
452 Subject Index
305, 307, 310, 311, 312, 340, 382, 383, 16, 19, 20, 21, 22, 24, 25, 26, 56, 59, 60,
385, 387, 409, 410, 411,412, 413, 414, 262, 265, 267, 408
415,416, 419, 422, 423 production rules ................................... 177
P3b ....... 351, 352, 353, 354, 355, 356, 359, pronunciation onset ................................ 55
361, 365, 372, 374, 377 prose passages ...................................... 284
Pandemonium 34 proximal stimulus ................................. 214
Parkinsons's disease 261 pseudowords... 90, 224, 226, 227, 238, 291
parsing ..... 97, 315, 317, 335, 337, 339, 340 psycholinguistics .... 238, 323, 339, 341,342
partial correlation ................................... 45 psychological present ............................ 173
pattern perception ....................... 30, 35, 38 psychophysical.............................. 199, 203
P-cell pathway ..................................... 200 pursuit rotor performance ..................... 277
P-cells .................................................. 200 Pz ......................................................... 295
perceptual grouping ............................. 200
perceptual learning ........................ 114, 283
perceptual-motor speed ........................ 113 readiness potential ................................ 298
Performance scales ............................... 145 reading comprehension... 89, 173, 185, 190
peripheral processing speed ........... 391,395 reading speed ........ 179, 204, 209, 214, 216
personality .................................... 221,223 recency memory ................................... 381
PERT networks ............ 6, 7, 12, 15, 21, 27 receptive fields ..................................... 200
PET ...................... 281,350, 372, 377, 380 recognition ......... 30, 31, 32, 33, 34, 35, 36,
phase ................................................... 199 38, 39, 40, 41, 42, 43, 44, 47, 50, 51, 53,
phonological code .......................... 40, 148 54, 58, 59, 62, 63, 64, 65, 66, 68, 69, 70,
phonological dyslexia .................... 192, 193 71, 72, 73, 74, 82, 83, 84, 85, 89, 90, 91,
picture t~agment identification .............. 128 92, 103, 105, 107, 108, 109, 112, 114,
picture processing ......... 143, 144, 151, 167 117, 124, 129, 131, 132, 137, 138, 139,
picture-naming143, 148, 149, 151, 152, 153, 140, 143, 144, 166, 167, 168, 170, 185,
154, 155, 158, 159, 164, 384 187, 188, 191, 192, 193, 196, 197, 204,
planned" contrasts ................................ 154 205, 209, 210, 212, 214, 215, 216, 217,
positron emission tomography .............. 201 218, 219, 220, 221,224, 225,226, 235,
postsynaptic discharges ........................ 294 236, 237, 238, 239, 240, 241,242, 243,
power function .............. 396, 398, 410, 417 244, 245, 246, 262, 268, 270, 272, 291,
primary memory .... 171, 172, 173, 345, 365 299, 301,307, 309, 311, 312, 322, 339,
primary visual cortex ............................ 201 343,346, 350, 351,353, 354, 356, 358,
prime-target relatedness ....................... 165 363, 364, 365, 366, 369, 372, 373,378,
problem solving .............. 28, 173,223,420 381,382, 383,384, 385, 386, 387, 388,
procedural memory ................. 63, 115, 153 389, 400, 413,419, 422, 423,424
processing capacity .............................. 175 regression line ............. 159, 162, 163, 164,
processing negativity ............................ 298 255, 260
processing resources.. 36, 47, 59, 171, 172, repetition priming ............... 115, 116, 117,
180, 183, 184, 186, 348 118, 119, 136, 137, 138, 170, 269, 273,
processing speed 45, 46, 47, 157, 159, 162, 274, 275,276, 277, 278, 280, 282, 283,
168, 171,204, 214, 299, 391,392, 394, 284, 285, 286, 287, 288, 289, 291,340,
395, 401,405,406, 408, 409, 410, 412, 341,354, 357, 359, 360, 362, 363,364,
417 373,383,384, 385
process-specific slowing3, 4, 5, 6, 8, 9, 11, 15, resource allocation. 178, 180, 182, 183, 321
retention intervals ................................. 114
Subject Index 453
retrieval ....... 21, 24, 27, 41, 61, 63, 64, 70, SOA ............. 7, 12, 13, 14, 17, 18, 20, 153,
84, 87, 99, 107, 112, 113, 114, 135, 136, 154, 158, 228, 230, 232, 234, 250, 251,
138, 140, 141, 143, 144, 152, 153, 154, 255, 261,264, 267, 400, 401,402
155, 158, 162, 164, 166, 169, 174, 179, source monitoring ................................. 359
192, 242, 247, 254, 270, 272, 275, 276, spatial frequencies ....... 199, 200, 202, 204,
281,283, 286, 287, 292, 311, 316, 340, 205, 206, 215, 218
350, 351,352, 353, 357, 358, 362, 363, spatial frequency filtering ........................ 36
364, 369, 376, 379, 380, 383,386, 388, specious present ................................... 172
389, 402, 419, 421 Speech Production System ................... 156
rule-based ........... 32, 34, 37, 38, 39, 42, 43 speed of response ................................. 284
spreading activation ..... 26, 43, 65, 93, 107,
113, 135, 136, 165, 166, 238, 239, 247,
scalp distributions ........ 351, 353, 363, 371, 252, 262, 270
374, 379 S-R compatibility .......................... 413, 416
secondary memory ........ 173, 365, 384, 419 stem completion .......... 117, 118, 119, 125,
Semantic deficit hypothesis .................. 348 129, 130, 131, 134, 136, 275, 276, 280,
Semantic dyslexia ............................... 194 281,282, 283, 284, 285, 286, 287, 289,
semantic incongruities... 311,326, 327, 342 301,346, 350, 353,358, 366, 367, 368,
semantic memory. 16, 63, 66, 84, 107, 112, 369, 373
114, 118, 134, 135, 144, 152, 153, 154, storage capacity .................................... 175
156, 162, 164, 166, 169, 192, 214, 218, striatal damage ..................................... 280
226, 247, 248, 252, 253,254, 262, 266, structural capacity ................................ 175
267, 268, 270, 290, 292, 347, 358, 369 subcortical motor system ...................... 280
semantic priming ...... 22, 28, 43, 50, 63, 68, substantia nigra ..................................... 319
69, 93, 94, 104, 108, 109, 113, 115, 116, subtraction method ............................... 418
138, 139, 140, 157, 165, 168, 185, 216, superordinate category membership ...... 252
226, 227, 241, 242, 244, 247, 248, 250, suppression135, 137, 184, 234, 314, 316, 32
251,252, 253,255, 257, 260, 262, 263, 340, 384
264, 265, 266, 267, 268, 269, 270, 271, surface dyslexia ............ 187, 191, 192, 193
285,286, 292, 293,399, 400, 402, 422, Surface Linguistic Processor ................. 156
423 Surface Pictorial Processor ................... 156
semantically anomalous ......... 299, 300, 326 symbol system ...................................... 144
sensory memory ................................... 111 syntactic structure ........................ 316, 325
sensory-motor processes ...................... 162
sensory-motor slowing ......................... 162
serial reaction time task ........................ 278 task complexity 30, 46, 158, 257, 392, 395,
short-term memory ........ 64, 173,342, 345, 396, 397, 406, 407
419 task domains ........................... 1, 2, 14, 145
shrinkage of neurons ............................ 319 template matching .................................. 35
slope ............ 13, 15, 19, 20, 56, 57, 61, 62, tip of the tongue ................... 136, 138, 316
63, 113, Tip-of-the-Tongue ................................ 113
159, 162, 163, 164, 179, 180, 183,255, total overlap ......................................... 406
258, 259, 260, 267, 306, 392, 393,394, transduction .............. 30, 38, 43, 49, 53, 54
395, 396, 398, 401,402, 403,409, 410, transfer appropriate processing .... 116, 133,
411,412, 416 282, 289, 292
slow potentials .............................. 331,334 triangulation ......................................... 373
454 Subject Index
U 238, 239, 240, 241,242, 244, 268, 270,

339, 400, 413, 419
unitary system ...................................... 183
Visuo-Spatial Scratch P a d .................... 173
voluntary eye movements ..................... 325
vascular lesion ...................................... 195 W
verbal fluency ........................ 252, 286, 381
Wernicke's aphasia ................................ 195
Verbal Scales ....................................... 145
white and grey matter ........................... 319
verification model ............... 40, 66, 69, 209
word frequency advantage ........ 37, 51, 53,
verification models ................................. 39
54, 55, 56, 58
visual deficit .................. 199, 203,210, 213
word frequency disadvantage ................. 51
visual dysfunction ......................... 203, 245
working memory ........... 46, 70, 94, 97, 98,
visual dyslexia ...................................... 191
99, 100, 104, 111, 171, 172, 173, 174,
visual masking ...................................... 203
175, 176, 177, 178, 180, 181, 182, 183,
visual word recognition ........ 30, 31, 32, 33,
184, 185, 186, 235, 244, 289, 314, 315,
34, 35, 36, 38, 39, 40, 41, 42, 43, 44, 51,
316, 317, 319, 323, 331, 334, 335, 336,
53, 54, 58, 59, 62, 63, 64, 65, 68, 69, 73,
338, 339, 341, 343, 345, 348, 362, 375,
83, 84, 85, 107, 108, 109, 166, 187, 197,
377, 388
215,218, 220, 224, 225, 226, 235, 236,

age مهم جدا

Uploaded by

Document Information

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

age مهم جدا

Uploaded by

Copyright:

Available Formats

Acknowledgments

Component cognitive processes have played a critical role in the development of

ADDRESSES OF SENIOR AUTHORS:

Phil Allen, Ph.D.

Paul Amrhein, Ph.D.

Ted Bashore, Ph.D.

Richard Ferraro, Ph.D.

Donald Fisher, Ph.D.

David Friedman, Ph.D.

Grover Cfilmore, Ph.D.

Marilyn Hartman, Ph.D.

George Kellas, Ph.D.

Johnathan King, Ph.D.

David Mitchell, Ph.D.

Beth Ober, Ph.D.

Marian Patterson, Ph.D.

Richard Ridderinkhof~ Ph.D.

Leann Stadtlander, Ph.D.

Elizabeth Stine, Ph.D.

Donald L. Fishera, Arthur D. Fiskb, and Susan A. Duffyc

bGoorgia Institute of Technology

Mount Holyoke College

2. GLOBAL AND LATENT MODELS

IV. Latent Models of Process-Specific Slowing 4

E[OI = a + #lU. (4)

Encoding Comparison Comparison Decision Response

3. MODELS OF SLOWING: PERT NETWORKS

3.1. False Positive Decisions

91 .3 .S .lr J 1.1 1.3 11.S 1.7

E[yo] = E[E] + E[C,] + E[D] + E[R] + E[Eo]. (10)

E[Y] = ta~ + tac + tad + tar' yes yes no

E[Y] = ta~,+ ta~ + tad + tar yes no yes

E[Y] = ta~,+ ta~ + tad + tar' yes no no

E[Y] : tae d- tac -~- tad' d- tar no yes yes

E[Y] : tat d- tac d- tad' -~- tar' no yes no

E[Y] : tae' -[- tac -~- tad' -}- tar no no yes

E[Y] = ta~,+ ta~ + tad' + tar' no no no

3.2. False Negatives

3.3. Global Models of General and Domain-Specific Slowing

4. MODELS OF SLOWING: CONNECTIONIST NETWORKS

4.1 False Positive Decisions

O = -201.31 + 1.63Y (13)

SOA Older Adults Younger Adults

200 674 678 4 537 539 2

800 627 645 18 510 523 13

SOA Older Adults Younger Adults

200 664 669 4 538 542 7

500 650 658 11 524 531 9

800 636 646 12 509 519 11

1Taken from Balota and Duchek (1988, Figure 1).

semantic semantic verbal

A similar equation holds in the neutral condition.

4.2 False Negative Decisions

We have spoken above of the methodological, theoretical and practical advantages

5.1 Methodological Importance

5.2 Theoretical Importance

In addition to their importance to methodology, the techniques we described for

5.3 Applied Potential

Anderson-Garlach, M. M. and Fisk, A. D. (1994, April). Age-related retention of skilled

Fisher, D. L. and Goldstein, W. M. (1983). Stochastic PERT networks as models of

Schaie, K~ W. (1989). Perceptual speed in adulthood: Cross-sectional and longitudinal studies.

Visual w o r d encoding and the effect of adult age and w o r d frequency*

Philip A. Allen a, David J. Madden b, and Steve Slane a

aCleveland State University