You are on page 1of 38

See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/12649656

A New Model of Verbal Short-Term Memory

Article  in  Journal of Experimental Child Psychology · April 2000


DOI: 10.1006/jecp.1999.2536 · Source: PubMed

CITATIONS READS

47 613

1 author:

Sergio Morra
Università degli Studi di Genova
74 PUBLICATIONS   1,023 CITATIONS   

SEE PROFILE

All content following this page was uploaded by Sergio Morra on 17 October 2017.

The user has requested enhancement of the downloaded file.


Journal of Experimental Child Psychology 75, 191–227 (2000)
doi:10.1006/jecp.1999.2536, available online at http://www.idealibrary.com on

A New Model of Verbal Short-Term Memory


Sergio Morra

Università di Padova, Padua, Italy

Two experiments tested a neo-Piagetian model of verbal short-term memory and


compared it with the articulatory loop model. Experiment 1 (n ⫽ 113, age range 9 –11)
tested word span for 2-, 3-, and 4-syllable words, with both visual and auditory presen-
tation. Experiment 2 (with the same participants) tested recall of visually presented
supraspan lists. Measures of M capacity (as conceived in Pascual-Leone’s neo-Piagetian
theory) and articulation rate were also used. The proposed model can account for the
effects of M capacity, word length, and presentation modality. The fit of this model to the
data was acceptable, and parameter estimates were consistent across experiments. Fur-
thermore, a correlation was found between M capacity and word span which resisted
partialling out of age and articulation rate. © 2000 Academic Press
Key Words: short-term memory; model; neo-Piagetian; visual presentation; auditory
presentation; word length; M capacity; goodness of fit.

Neo-Piagetian theories (e.g., Case, 1985; Demetriou, 1988; Halford, 1993)


have emphasized the role of working memory in cognitive development, but
traditionally have not been connected with the dominant approach in working
memory research (e.g., Baddeley, 1986). More recently, investigators (e.g., de
Ribaupierre & Bailleux, 1994) have suggested that it would be desirable to relate
these two approaches. In this article I provide such an integrative approach by
using a neo-Piagetian model that takes into account the effects of word length,
presentation modality, and use of supraspan lists, which are often reported in
working memory research.
Baddeley’s model includes a set of peripheral short-term stores—including an
articulatory loop, a phonological store, and a visuospatial sketch pad. The model
also includes a central executive, which is assumed to have both attentional and
storage functions, although its capacity and the way it works still seem to be
largely undetermined (see Allport, Styles, & Hsieh, 1994; Baddeley, 1996;
Lehto, 1996). Among the several short-term stores suggested, only the articula-

Thanks are due to Chiara Stoffel, Francesca Rizzotti, and Antonia Sartori for collecting and scoring
data, and to Sandro Bettella for computer programming. The author now works at the University of
Genova, Italy.
Address correspondence and reprint requests to Sergio Morra, DISA—sezione Psicologia, vico S.
Antonio 5/7, 16126 Genoa, Italy. E-mail: morra@nous.unige.it.

191
0022-0965/00 $35.00
Copyright © 2000 by Academic Press
All rights of reproduction in any form reserved.
192 SERGIO MORRA

tory loop’s capacity has been estimated. It was suggested that verbal short-term
memory (STM) is time-limited; that is, one can recall as much as can be spoken
in about 1.5 or 2 s (Baddeley, 1986, 1992; Baddeley, Lewis, & Vallar, 1984;
Baddeley, Thomson, & Buchanan, 1975; Hulme & Tordoff, 1989; Schweickert
& Boruff, 1986). Given the finding that there are differences in recall when word
length varies, it appeared that span depends on stimuli rehearsal rate. Baddeley
et al. (1984) also suggested that two components—an articulatory loop and a
phonological store—are needed to account for different verbal STM phenomena.
However, they attributed the word-length effect, and therefore the time-limited
capacity, to the articulatory loop, not to the phonological store. Nicolson (1981)
and subsequent studies also suggested that the articulatory loop can account for
the development of verbal STM from the age of 4 to adulthood.
However, this version of the loop model faces conflicting evidence. In contrast
to early studies cited above, recent researchers (e.g., Cowan et al., 1994; Henry,
1994) have shown that the ratio of recall to articulation rate is not constant and
that the intercept of the regression line of short-term memory span on articulation
rate is often above zero. The articulatory loop cannot, therefore, fully account for
verbal STM. Hulme, Maugham, and Brown (1991) found that word familiarity
affects the intercept of the span-to-speech-rate regression line. They concluded
that their results were incompatible with the widespread unitary view of STM
span in which all storage occurs within an articulatory loop. Hitch, Halliday, and
Littler (1989) and Morra (1990) also drew similar conclusions.
Furthermore, several authors have questioned whether articulation time really
does account for word-length effect in subvocal rehearsal, offering alternative
explanations that include verbal output interference (Cowan et al., 1992; Henry,
1991b), proactive interference (Nairne, Neath, & Serra, 1997), and complexity of
speech programming (Caplan, Rochon, & Waters, 1992; Lovatt, Avons, &
Masterson, 2000; Service, 1998). Brown and Hulme (1995) and Neath and
Nairne (1995) provided computational evidence that a word-length effect could
be produced by any of several mechanisms of forgetting. Moreover, verbal STM
span is affected by variables that have little or no effect on articulation rate, such
as semantic variables (e.g., Poirier & Saint-Aubin, 1995), grammar class (Tehan
& Humphreys, 1988), word frequency or familiarity (e.g., Henry & Millar, 1991;
Hulme et al., 1997), and order of the stimulus words (Brooks & Watkins, 1990).
Finally, developmental research shows that rehearsal skill only partly accounts
for age-related changes in STM span (Cowan, Cartwright, Winterowd, & Sherk,
1987; Cowan et al., 1994; Henry, 1991a; Hitch, Halliday, & Littler, 1989). Kail
and Park (1994) suggested that articulation rate increase is an outcome of the
increase of processing speed with age, but they also found an age effect on
memory span not explained by either processing speed or articulation rate.
Various modified versions of the articulatory loop model have been proposed
(Brown & Hulme, 1995; Burgess & Hitch, 1992; Cowan, 1992; Cowan et al.,
1992, 1994; Gathercole, 1997; Henry, 1991b; Hulme et al., 1991). These will be
VERBAL STM MODEL 193

considered in the general discussion. In revising Baddeley’s model, some authors


also considered the finding of an above-zero intercept in the regression line of
STM span on articulation rate. Hitch, Halliday, and Littler (1989) suggested that
a positive intercept may reflect a contribution of the central executive to memory
span, Nicolson and Fawcett (1991) proposed instead a phonological store, and
Hulme et al. (1991) argued that it represents a long-term memory contribution.
The model presented here is closest to the suggestion of Hitch et al. However,
it also tries to increase conceptual and quantitative precision by using neo-
Piagetian theoretical constructs to replace that of a homunculus-like central
executive. Rehearsal is also considered important; but no form of specifically
time-limited storage is assumed.
The present model elaborates upon the idea that activation can be a valid
alternative to storage as the basic metaphor for short-term memory (Cowan,
1988; Engle, Cantor, & Carullo, 1992). Instead of positing the existence of
time-limited capacity stores, these authors consider how activation of some
cognitive representations decreases during processing.
This model rests on a neo-Piagetian theory—the theory of constructive oper-
ators (Pascual-Leone, 1987; Pascual-Leone & Goodman, 1979)—which includes
two types of constructs: schemes and general-purpose operators. Following
Piaget, figurative schemes represent states of affairs, while operative schemes
represent procedures that transform either states of affairs or their mental repre-
sentations. Schemes can be activated either by perceptual input or by various
internal sources of activation. Rehearsal mechanisms are conceived not as limited
stores (such as the articulatory loop) but rather as specific operative schemes
(e.g., a verbal rehearsal operative). As such, they have no “capacity” of their own,
but their activation follows the same rules as that of any other scheme.
The theory of constructive operators posits a number of general-purpose
mechanisms that serve to increase or decrease activation of schemes or to
generate new ones. Four are relevant here. The M operator (Mental energy) is
considered as a limited amount of attentional resources that increases with age
and can be used by executive schemes to activate other schemes. It is conceived
as a general-purpose attentional mechanism, but not as “the single” resource. The
I operator (Interrupt) has a complementary control function; that is, it inhibits (or
deactivates) schemes. The F operator (Field) consists in increased activation of
those schemes that are facilitated by stimulus–response compatibility or Gestalt
principles. The L operator (Learning) allows activation, at reduced attentional
cost, of schemes belonging to a learned superscheme or structure (as in autom-
atization).
In line with the view that every scheme functions as an integrated cognitive
unit, Pascual-Leone (1970) assumed that activation by the M operator of any
scheme requires the same amount of attentional energy. Further, he suggested
that 3-year-olds have an M operator capacity sufficient to activate executive
schemes for the current task plus one (operative or figurative) scheme (i.e., e ⫹
194 SERGIO MORRA

1) and that capacity increases by one additional scheme approximately every 2


years through adolescence.
Building on Pascual-Leone’s work, Burtis (1982) modeled encoding and
retrieval processes in a short-term memory task. His model successfully ac-
counted for the effects of M capacity and chunking in recall of visually presented
supraspan sets of consonants. It did not, however, consider the word-length effect
or other effects often attributed to phonological encoding. The purpose of the
model described here was thus to preserve Burtis’s theoretical structure, revising
and extending it to consider the roles of rehearsal, presentation modality, and
differences between materials (such as varying word lengths). The main goals of
this paper are to describe the model and to test its goodness of fit.
In a pilot experiment, children aged 6 to 9 years heard, for immediate recall,
lists of two-, three-, and four-syllable words, following a span procedure.
Experiment 1 compared memory span for lists of different-length words pre-
sented visually or aurally to children aged 9 to 11. In Experiment 2 supraspan
lists were used instead of a conventional span procedure. In all experiments, the
model’s predictions depend on the participants’ M capacity, and on a single free
parameter (explained in the Model section). The model is evaluated by its
goodness of fit to the empirical data, and by the consistency of estimates across
experiments.
A further aim of these experiments was to compare articulation rate and M
capacity as predictors of individual differences in verbal STM. The correlation
between articulation rate and STM performance is often regarded as evidence for
the articulatory loop model (e.g., Baddeley, 1986; Hulme & Tordoff, 1989; but
see Cowan et al., 1994). Although articulation rate has already been compared
with time for item identification (Hitch, Halliday, & Littler, 1989, 1993) and with
general processing speed (Kail & Park, 1994), it has not been compared with M
capacity.

MODEL

While the time-limited articulatory loop model accounts easily for the word-
length effect, it can hardly accommodate those findings, reviewed in the intro-
duction, which suggest that verbal short-term memory is not time-limited. By
contrast, Burtis (1982) accounts well for relationships between M capacity,
chunking, and STM but did not explain other phenomena. What is therefore
needed is a new developmental model that integrates both M capacity and
experimental conditions as sources of variance. This section presents a general
overview of such a model, and the Appendix supplies more formal detail. For the
sake of readability, Table 1 presents a list of symbols used in this paper, most of
which are commonly used in the operator-logic notation of the theory of con-
structive operators.
The experiments reported below used unrelated, well-learned words for con-
crete referents, so that children may be presumed to have figurative schemes that
VERBAL STM MODEL 195

TABLE 1
List of Symbols Used in the Theory of Constructive Operators and in this Model

Symbol Meaning

M (operator) A central attentional resource that can activate a limited number of schemes.
F (operator) Processes that enhance activation of schemes according to Gestalt field
principles or stimulus–response compatibility.
L (operator) Processes that enhance activation of schemes as a consequence of
overlearning (i.e., a spread of activation among constituents of a well-
learned structure).
I (operator) A central attentional control process (“interrupt”) that reduces activation of
currently irrelevant schemes.
␧ Executive schemes (e.g., current goals or monitoring processes). Specifically,
␧ Lisn, ␧ Read, and ␧ Rec stand for the goals of listening to, reading, and
recalling the word lists.
␾ Figurative schemes (representations of objects or states of affairs).
␺ Operative schemes (mental blueprints for transformations or actions).
Specifically, ␺ Code, ␺ Reh, and ␺ Retr stand for the operations of encoding,
rehearsing, and retrieving words.
e A small, constant amount of the M operator capacity, necessary to activate
the executive schemes.
k The number of figurative and operative schemes that can be simultaneously
activated by a subject’s M operator.
i A free parameter, specific to this model, that represents activation decrease.
W1, W2, etc. The first, second, etc., word in a list.
␾ W1, ␾ W2, etc. The representations of the first, second, etc., word in a list.
U W1, U W2, etc. Utterances of the first, second, etc., word.

represent them. When the items are presented, the child’s encoding operations
enable activation of these figurative schemes. Some schemes are kept fully
activated by the M operator during the retention interval, but once available
resources are exceeded, activation decreases for the remaining items. Probability
of recalling the latter items depends on the relative amount of activation that
schemes still have.
At recall, there is no longer direct activation from input, because the stimuli
are absent. In a short-term memory task in which items are unrelated, neither F
nor L activation occurs. (Such activation would occur if the items could be
chunked together.) I therefore assume that the schemes representing stimulus
words can either be kept fully activated by the M operator or be partly active
because their activation has not yet dropped to zero.
Some operative schemes are activated for encoding, rehearsal, and retrieval in
STM tasks. Operative schemes do not represent items to be reported, but instead
represent processing operations. They, too, consume part of the M capacity in
order to be activated (except when they have other sources of activation).
In serial recall, specific operative schemes could keep track of presentation
order. Subvocal rehearsal is a convenient option, though not the only possible
196 SERGIO MORRA

one. I suggest that rehearsal serves not only to refresh stimuli traces, but also
mainly to encode serial order with less effort than would be required by other
strategies. For instance, activating a set of ordering tags could be more effortful.
(For a discussion of order representation, see also Brown, 1997.)

Modeling Performance with Rehearsal

Suppose that a person subvocally rehearses articulatory codes. As the first


word is presented, an encoding operative scheme and a figurative scheme for that
word are activated. As the second word is presented, the encoding operative is
still necessary, as well as a second figurative scheme. At the same time, a
rehearsal operative scheme can be activated to rehearse the scheme that repre-
sents the first word, and so on. If that person’s M operator has the capacity to
activate k operative or figurative schemes, as soon as it becomes fully engaged
with the two encode and rehearse operatives plus k ⫺ 2 figuratives (i.e., the
mental representations of k ⫺ 2 words), other previously activated schemes start
to lose activation.
In the case of auditory presentation, however, I assume that articulatory and/or
acoustic encoding of the stimuli is highly compatible and automatized. This
assumption can be justified, for example, with reference to Penney’s claim that
“a subject cannot voluntarily prevent entry of an auditory item into short-term
memory,” while “in the visual modality, . . . generation of the phonological code
is not automatic” (Penney, 1989, pp. 399 – 400; see also Baddeley et al., 1984).
Assuming that item encoding is automatic with auditory presentation, only the
rehearse (not the encode) operative requires M capacity. Thus, k ⫺ 1 (rather than
k ⫺ 2) figurative schemes can be kept active by the M operator during stimuli
presentation. Another difference between visual and auditory presentation con-
cerns the end-of-list signal. In visual presentation the signal is obviously arbitrary
and conventional. Thus, participants must encode it (as a figurative scheme) in
order to decide to report the items. In contrast, in auditory presentation, the
end-of-list signal may be either a conventional signal (e.g., a tone) or a habitual
component of human verbal communication (e.g., the experimenter turning to
look at the participant or slightly stressing the last item in the list). In auditory
presentation, if the end-of-list signal involves such nonarbitrary components,
then one can assume that it is automatically encoded. In other words, it would not
require M capacity.
As soon as some schemes start losing activation, probability of recalling them
decreases. I assume that the higher the number of decaying schemes at any given
point, the more they will interfere with each other. Thus, probability of recalling
a partly activated scheme depends both on how long it has been decaying and on
how many partially activated schemes have interfered with it. In the present
model, I assume that activation decays at the same rate during input and recall,
although this may well be an oversimplification.
Decreased activation of figurative schemes is expressed as parameter i (the
VERBAL STM MODEL 197

model’s only free parameter), defined as the decrease in probability of recall of


an item per processing step and per decaying scheme. I assume that parameter i
is specific to materials, such as articulatory codes for long or short words,
confusable or nonconfusable consonants, digits, semantic codes, and so on. (It is
conceivable that i may also vary according to some characteristic of the subject
population, e.g., resistance to interference [Dempster, 1992], although this pos-
sibility is not explored here.) The assumption that i is materials-specific implies
that the word-length effect can be accounted for by a greater drop in activation
of articulatory codes for long words than for short words during each processing
step.
The notion that a parameter can express decreasing recall probability is
common to many models. What is specific to this model, however, are the
following assumptions: (a) activation will decrease only for those representations
that are not focused on by a central attentional mechanism, that is, the M
operator; (b) activation decrease is proportional not only to time but also to the
number of currently decaying schemes (e.g., if there are three decaying schemes
at a given step, each of them will lose an amount of activation of 3i during that
step); (c) being specific to materials, parameter i does not vary with other
experimental conditions (such as presentation modality) or the individuals’ M
capacity (which is independently represented in the model).
It might be asked why the model includes only one material-specific free
parameter but is silent about whether differences among materials are due to
rehearsal time, output interference, planning complexity, or a combination of
these and other factors. The reason is twofold. First, it is desirable to keep the
model as simple as possible (a model with too many free parameters would
probably fit any set of data). Second, this model is based on two main ideas—(a)
the role of a central attentional mechanism whose limited capacity is divided
between operative and figurative schemes and (b) decreasing recall probability of
those schemes that are not kept activated by the M operator. If the model fits the
data well, then future research can make detailed exploration of the processing
factors affecting the value of i. The advantage of this model is that it highlights
and quantifies the role of a specific, central attentional resource, the M operator,
which most current models seem to neglect. Although testing the extent to which
free parameter i is related to rehearsal time, output interference, and other factors
would be feasible, it goes beyond the scope of this series of experiments.
Modeling Performance without Rehearsal
Let us now consider an individual who does not use a subvocal rehearsal
strategy. In this case, I assume that if individuals can recall the first and the last
items at all, they will recall them in the appropriate positions. Such positions are
assumed to be salient, and as such, are activated by the F operator. Burgess and
Hitch (1992) make a similar assumption within a theoretically different account,
while Cowan et al. (1992, Fig. 4) and Hulme et al. (1997, Fig. 5) also reported
evidence that the first and last position have a special advantage.
198 SERGIO MORRA

Serial positions other than the first and last could be encoded by separate
operative schemes for each position in the sequence, or possibly guessed. In this
case, in a three-item list, the first and third positions are salient and the second
is completely determined; in a four-item list, the first and fourth are salient, and
there is a .50 probability of guessing which item was the second and which the
third. In a five-item list, this probability becomes 1/6, and so on, with combina-
torial increase.
If people tend to choose a strategy that maximizes performance while mini-
mizing effort, then most adults will use subvocal rehearsal. There may be
exceptions, but at present it is hard to make clear predictions about how far adult
performance is affected by alternative strategies. We can easily predict, however,
that children do not use articulatory rehearsal if this consumes too much of their
limited attentional resources (M capacity). It has been established that young
children seldom rehearse and that, although it is possible to train them to do so,
such training is not completely successful (e.g., Cowan, Saults, Winterowd, &
Sherk, 1991; Henry, 1991a; Hitch, Halliday, Dodd, & Littler, 1989). Guttentag
(1984) demonstrated that rehearsal requires mental effort. Children may therefore
not follow this strategy because it demands excessive expenditure of attentional
capacity.
The present model implies that individuals with an M capacity of less than e
⫹ 3 do not rehearse in the case of visual presentation—not because the rehearsal
operation is difficult in itself but because it requires too much of their attentional
capacity: They do not have enough capacity to monitor rehearsal and at the same
time encode the current item. The encode and rehearse operative schemes and
the representation of the current word are three distinct psychological units, each
of which demands its share of M capacity. In the case of auditory presentation,
since the encoding operative scheme would not require a unit of M capacity the
minimum M capacity needed for rehearsal would be e ⫹ 2. The conditions for
rehearsal in the present model are also consistent with the minimal ages for
rehearsal reported in the relevant literature (e.g., Hitch et al., 1993; Hitch,
Halliday, Schaafstal, & Heffernan, 1991), since e ⫹ 2 is the modal M capacity
at 5– 6 years, and e ⫹ 3 is the modal capacity at 7– 8 years of age (e.g.,
Pascual-Leone, 1970).
Throughout this article, the model’s goodness of fit is tested on the basis of the
strict assumption that when item ordering creates some difficulty (i.e., with lists
of four or more words), individuals who have enough M capacity to rehearse do
so consistently, while those who do not have enough M capacity never use
rehearsal. No probabilistic correction is allowed for this assumption.

A PILOT EXPERIMENT

In a pilot experiment, from a total pool of 124 participants (57 girls and 67
boys, mean age 8;0, range 5;10 to 9;9, living in Northern Italy, in a small town
or its surrounding Alpine villages) 96 children were selected and tested for
VERBAL STM MODEL 199

memory span, articulation rate, M capacity, and verbal ability. In order to


maximize accuracy of M-capacity measurement, we selected 24 children in each
grade (1– 4), those who had the smallest variation coefficient (i.e., standard
deviation/mean ratio) in the M-capacity measures. M capacity was operationally
defined as the average of the scores in three tests—Figural Intersections Test, Mr.
Cucumber Test, and Counting Span Test (see method of Experiment 1 for
details). For the memory span and the rapid articulation tasks, three sets of 10
Italian nouns (of two, three, and four syllables, respectively) were used in this
and the subsequent experiments. The Vocabulary subtest from the Italian version
of the Wechsler Intelligence Scale for Children (Wechsler, 1974) and an adap-
tation of the Verbal Fluency subtest from the Primary Mental Abilities (Thur-
stone, 1941) were also used. The M-capacity tests were administered by one
experimenter, and all other tasks by another.
The memory-span scores were analyzed with a 4 (age groups) ⫻ 3 (word
lengths) mixed-design ANOVA, which yielded significant effects of age, F(3,
92) ⫽ 25.84, p ⬍ .001, and word length, F(2, 184) ⫽ 78.50, p ⬍ .001, and a
nonsignificant interaction. Although there was a word-length effect, the results
were only partly consistent with the articulatory loop model, since performance
was only worse in the case of four-syllable words. The difference between two-
and three-syllable words was not significant. The ratio between span and artic-
ulation rate (AR) increased from 1.67 s for the shortest words to 2.56 s for the
longest. The best fitting regression line, across 12 data points (four age groups ⫻
three word lengths) was Span ⫽ 2.50 ⫹ 0.685AR. This equation accounted for
51.2% variance across word lengths and age groups; both its intercept (p ⬍ .001)
and its slope (p ⬍ .02) were significantly different from zero. The finding of a
relationship between memory span and articulation rate was thus replicated. The
fact that the intercept was much higher than zero contradicts a prediction of
Baddeley’s original time-limited articulatory loop model but confirms more
recent findings (e.g., Hulme et al., 1991).
The respective contribution of articulation rate and M capacity to word span
development was assessed by correlational analyses. Children’s memory span for
words was positively correlated with both M capacity, r(94) ⫽ .65, p ⬍ .001, and
articulation rate, r(94) ⫽ .55, p ⬍ .001. I then tested whether each of these two
variables still correlated with word span, with the other one and age partialled
out. This analysis is necessary because memory span, articulation rate, and M
capacity all increase with age. Since M capacity is conceived as a developmental
variable, it may be noted that partialling out of age eliminates true variance.
However, there are also individual differences in M capacity among subjects of
the same age. Partialling out age eliminates one source of variance while leaving
the other intact, so partial correlations can be regarded as a particularly strict test
of the relationship between M capacity and word memory span. With age
partialled out, memory span was correlated with both M capacity, r(93) ⫽ .50,
p ⬍ .001, and articulation rate, r(93) ⫽ .32, p ⬍ .01. The correlation between
200 SERGIO MORRA

span and M capacity remained significant even with age and articulation rate
partialled out, r(92) ⫽ .41, p ⬍ .001. However, the correlation between span and
articulation rate, with age and M capacity partialled out, was nonsignificant,
r(92) ⫽ .09. In order to control for verbal ability, vocabulary and verbal fluency
scores were also considered. With age, verbal fluency, and vocabulary partialled
out, the correlation between memory span and M capacity was still significant,
r(91) ⫽ .41, p ⬍ .001, whereas between span and articulation rate it was not,
r(91) ⫽ .15.
The possibility can also be ruled out that the correlation between M capacity
and word span is an artifact due to a high correlation between word span and one
particular test among those included in the M-capacity battery. With both age and
articulation rate partialled out, word span correlated with the Figural Intersec-
tions Test, r(92) ⫽ .23, p ⬍ .02, with the Counting Span Test, r(92) ⫽ .47, p ⬍
.001, and with the Mr. Cucumber Test, r(92) ⫽ .22, p ⬍ .02.
These results suggest that the positive correlation between M capacity and
memory span is a robust phenomenon, at least in childhood, while the correlation
between span and articulation rate might be due to other intervening variables. In
any case, it seems that articulation rate makes only a small contribution to
individual differences in memory span during childhood.
The model presented above was tested for goodness of fit to these data. Since
it was assumed that i has different values for different materials, three separate
estimations of this parameter were made, one for each word length. (See
Experiment 1 below for the procedure for estimating parameters and testing for
goodness of fit.) The resulting values were i ⫽ .0124 for two-syllable, i ⫽ .0169
for three-syllable, and i ⫽ .0298 for four-syllable words. The model’s goodness
of fit was tested in relation to three groups of subjects with M capacity of e ⫹ 2,
e ⫹ 3, and e ⫹ 4, respectively. There were nine means (i.e., three M-capacity
groups by three word lengths), seven of which were correctly predicted by the
model. Only two of nine differences were significant (p ⬍ .05)—the model
predicted a higher mean than observed with four-syllable words for subjects with
M ⫽ e ⫹ 3, and a lower mean than observed with three-syllable words for
subjects with M ⫽ e ⫹ 4. The variance of each distribution was also considered.
The fit was worse for variances than for means, since observed variances tended
to be smaller than predicted by the model, five out of nine comparisons being
significant. Apart from this tendency to overestimate variances, this experiment
provided preliminary support for two conclusions: (a) M capacity seems to
predict individual differences in word span better than does articulation rate, and
(b) the proposed model shows a reasonable approximation to the data, at least as
far as the group means are concerned. In short, it accurately predicts the amount
of memory-span increase as a function of M capacity.
However, this pilot experiment was limited in that it included only one
presentation modality, and only a span measure of short-term memory (but no
measure with supraspan lists). In addition, subject selection ensured reliable
VERBAL STM MODEL 201

measurement of M capacity, but could have affected the results in other, un-
known ways. The following experiments were designed to test the model within
a broader range of experimental conditions and with unselected samples.

EXPERIMENT 1
This experiment replicated the pilot experiment with an unselected sample,
and also included visual presentation. It was designed to replicate the results
found in the pilot study, and to test whether the model can account for differences
between visual and auditory presentation. It was suggested earlier that the
encoding operative scheme and the end-of-list signal do not require attentional
resources (M capacity) in auditory presentation. Fourth and fifth graders were
recruited as subjects, because they can easily read the visually presented stimulus
words.

Method
Participants. The participants were 58 fourth graders and 55 fifth graders
(mean age 10;3, age range 8;11 to 11;5) from two middle-size towns in Northern
Italy. They comprised 53 girls and 60 boys. All children had normal or corrected-
to-normal vision and no other relevant impairments. The sample included all
fourth and fifth graders enrolled in the schools, with the exception of 7 children
who had reading difficulties, were mentally retarded, or were nonnative Italian
speakers.
Materials. Three sets of 10 concrete high-frequency Italian nouns (of two,
three, and four syllables, respectively) were used. The word sets, drawn from 10
semantic categories, do not differ in mean frequency according to De Mauro,
Mancini, Vedovelli, and Voghera (1993), F(2, 27) ⫽ 1.01, p ⫽ .38. The
two-syllable words were Zio (uncle), Merlo (blackbird), Pesca (peach), Sole
(sun), Treno (train), Vaso (vase), Scarpe (shoes), Bagno (bathroom), Ape (bee),
Rosa (rose). The three-syllable words were Sorella (sister), Rondine (swallow),
Banana (banana), Nuvole (clouds), Corriera (long-distance bus), Pettine (comb),
Maglione (sweater), Cucina (kitchen), Lumaca (snail), Geranio (geranium). The
four-syllable words were Genitori (parents), Pappagallo (parrot), Mandarino
(tangerine), Temporale (storm), Motoscafo (motorboat), Sigaretta (cigarette),
Pantaloni (trousers), Corridoio (corridor), Tartatuga (turtle), Garofano (carna-
tion). These were the same word sets previously used in the pilot experiment.
Another set of 10 concrete nouns of different lengths was used for practice.
For the span task, 18 two-word sequences, 18 three-word sequences, etc., up
to a maximum of seven words, were created by random selection—that is, six
lists of each length from each set. Increasingly long lists were used to assess word
span; half of them were acoustically and half visually presented. The materials
for each span trial consisted of one two-word list, one three-word list, etc., drawn
from the same set. For the articulation task, each set was divided into five word
pairs.
202 SERGIO MORRA

Three M-capacity tests were used: the Figural Intersections Test, Mr. Cucum-
ber Test, and Counting Span Test. M capacity was operationally defined as the
average of the three test scores (see Morra, 1994; Morra & Scopesi, 1997, for
justification of this procedure). Each of the M-capacity tests is based on the
principle that all of its items have the same content but differ in processing load
(Case, 1985; Morra, 1994; Pascual-Leone & Baillargeon, 1994). Since M capac-
ity is considered a general resource, it is convenient to average across tests with
different content and response demands, although this means that the correlations
between different tests cannot be very high (Case, 1985; Morra, 1994).
The Figural Intersections Test comprises 36 items, each of which requires the
intersection of a set of shapes to be found. The level of an item is determined by
the number of presented shapes, which ranges from two to nine. An individual’s
score is the number of levels at which at least 75% of responses are correct, plus
one (for “level one,” not included in the test because no intersection is possible
with one shape). For details, see Morra (1994) and Pascual-Leone and Baillar-
geon (1994).
The Mr. Cucumber Test presents outline drawings of an extraterrestrial char-
acter, with a number (from 1 to 8) of colored stickers attached to it. There are
three items per level, in ascending order. The child is shown a colorless shape and
asked to indicate the positions of the stickers in the previously presented figure.
One point is given for each consecutive level on which a person gets at least two
items correct, plus one third of a point for each correct item beyond that level.
Scores are rounded up/down to the nearest unit. For details, see Case (1985) and
Morra (1994).
The Counting Span Test requires the subject to rapidly count aloud sets of
colored dots and then recall the number of each set. The level of an item is
determined by the number of sets it comprises, ranging from one to eight. The
score is the highest level on which a child gets at least two items correct out of
three. For details, see Case (1985) and Morra (1994).
Note that, if the articulatory loop model is correct, then these tests should have
low correlations with word span. Out of these tests, only the Counting Span
requires verbal responses, and none of them is scored for serial recall of
information. Both the Figural Intersections Test and the Mr. Cucumber Test have
visuospatial content; the Figural Intersections Test does not demand any recall,
and the Mr. Cucumber Test involves recall of simultaneously presented posi-
tions. Only the Counting Span Test demands recall of verbal information;
however, order errors are disregarded, and, more important, it involves counting
aloud during item presentation, which prevents participants from rehearsing.
Hence, according to Baddeley’s model of working memory, a correlation be-
tween M capacity and word span is hardly to be expected.
Procedure. There were three sessions. In the first, the Figural Intersections
Test was group-administered. In the second, Counting Span, Mr. Cucumber Test,
VERBAL STM MODEL 203

TABLE 2
Means (and Standard Deviations) of Word Span by Grade, Presentation Modality,
and Word Length in Experiment 1

Number of syllables

Grade Modality 2 3 4

4 Visual 4.14 (0.65) 3.80 (0.75) 3.44 (0.55)


Auditory 4.53 (0.68) 4.26 (0.53) 4.06 (0.53)
5 Visual 4.41 (0.76) 3.98 (0.67) 3.55 (0.64)
Auditory 4.76 (0.68) 4.50 (0.59) 4.07 (0.62)

and Articulation Rate were individually administered in that order. The first two
tests were discontinued when a child failed all items at one level.
To measure articulation rate, participants were required to repeat each word
pair five times as fast as possible. Two practice trials were carried out first, using
pairs of words from the practice set. Next, the five word pairs of each length were
presented in one of the six possible word-length orders. A voice key triggered the
computer clock when the child started articulating, and the experimenter stopped
it by pressing a key at the end of the last repetition. This has been shown to be
a very reliable procedure (e.g., Nicolson & Fawcett, 1991).
The third session included word memory span in six experimental conditions
(visual vs auditory presentation of three word lengths), preceded by two practice
trials (one visual and one auditory) with another word set. Half of the participants
started with auditory and half with visual presentation. The six possible orderings
of word lengths were also balanced over participants.
For each modality and word length, three memory-span trials were performed.
Each trial started with a list of two words, then a list of three, and so on (up to
seven words), until the child failed on a list. Each trial was scored according to
the longest list correctly recalled, and a child’s average score across three trials
was his/her span score for words of a given length in a given modality.
The words appeared one at a time in the center of a screen in 8-mm capital
letters for 1500 ms each, followed by a blank screen interval of 500 ms—that is,
at a presentation rate of one word every 2 s, as in the main experiment of
Baddeley et al. (1975). During visual presentation, subjects silently read the
words on the screen. For auditory presentation, the screen was turned around
toward the experimenter, who read out the words as they appeared.

Results and Discussion


Preliminary analyses. A 2 (age group) ⫻ 3 (word length) ⫻ 2 (presentation
modality) mixed-design ANOVA of word span scores (see Table 2) yielded
significant main effects of word length, F(2, 222) ⫽ 130.73, p ⬍ .001, and
modality, F(1, 111) ⫽ 126.48, p ⬍ .001. The main effect of age was only
204 SERGIO MORRA

TABLE 3
Means (and Standard Deviations) of Articulation Rate in Words/s by Grade
and Word Length in Experiment 1

Number of syllables

Grade 2 3 4 Mean

4 2.84 (0.38) 2.03 (0.24) 1.64 (0.19) 2.17 (0.25)


5 2.98 (0.31) 2.20 (0.23) 1.76 (0.18) 2.31 (0.21)
All 2.91 (0.36) 2.11 (0.24) 1.70 (0.19) 2.24 (0.24)

marginally significant, F(1, 111) ⫽ 3.42, p ⬍ .07. The Modality ⫻ Length


interaction was significant, F(2, 222) ⫽ 3.30, p ⬍ .04, and the word-length effect
was greater in visual than in auditory presentation (consistent with Watkins &
Watkins, 1973). However, for each word length, performance was higher with
auditory than with visual presentation (all ps ⬍ .001), and conversely, within
each modality, span decreased as word length increased. Thus, the word-length
effect and the modality effect were replicated.
Descriptive statistics for the articulation rate are shown in Table 3. In the
visual condition, the ratio between mean span and articulation rate increased with
word length, from 1.47 s for the shortest words to 2.06 s for the longest. In the
auditory condition, it increased from 1.60 s for the shortest words to 2.39 s for
the longest.
In visual presentation, the regression line across six data points (two age
groups ⫻ three word lengths) was Span ⫽ 2.47 ⫹ 0.632AR, while in auditory
presentation, it was Span ⫽ 3.29 ⫹ 0.478AR. These equations accounted for
92.3% and 85.7% variance, respectively. Their intercepts (p ⬍ .001 in both
equations) and slopes (p ⬍ .002 in visual and p ⬍ .01 in auditory conditions)
were significantly different from zero. The linear relationship of memory span to
articulation rate replicates the findings of Baddeley et al. (1975). However, as in
the pilot experiment, the intercepts were well above zero, and the span to
articulation rate ratio was not constant. These aspects are inconsistent with the
findings and model of Baddeley et al. (1975), although they agree with other
published data (e.g., Hulme et al., 1991).
With age partialled out, the Figural Intersections Test correlated r(110) ⫽ .30,
p ⬍ .001, with the Counting Span Test and r(110) ⫽ .34, p ⬍ .001, with the Mr.
Cucumber Test. The partial correlation between Counting Span and Mr. Cucum-
ber was r(110) ⫽ .27, p ⬍ .002. These partial correlations among M-capacity
measures replicate those reported by Morra (1994), which were of the order of
magnitude of .25. Note that these three tests have very different content and
response requirements, and that partialling out of age eliminates developmental
(i.e., true) variance. Hence, the average score of these three tests was used as a
measure of M capacity.
VERBAL STM MODEL 205

For the purpose of correlational analysis, each child’s articulation rate scores
with two-, three-, and four-syllable words were averaged; so were their span
scores for visually presented words of three lengths, as well as the three span
scores for acoustically presented words.
Memory span for words was positively correlated with M capacity, r(111) ⫽
.30, p ⬍ .001, in the visual condition and r(111) ⫽ .32, p ⬍ .001, in the auditory
condition. Memory span also correlated with articulation rate, r(111) ⫽ .19, p ⬍
.03, in the visual condition and r(111) ⫽ .18, p ⬍ .04, in the auditory condition.
The correlation between span and M capacity resisted the partialling out of
both age and articulation rate, r(109) ⫽ .28, p ⬍ .002, in the visual condition and
r(109) ⫽ .31, p ⬍ .001, in the auditory condition. A smaller but significant
contribution of articulation rate to memory span was also found, since the
correlation between span and articulation rate with age and M capacity partialled
out was r(109) ⫽ .17, p ⬍ .04, in both visual and auditory conditions.
As in the previous experiment, correlation between M capacity and word span
was not due to a high correlation between word span and one particular M-
capacity test. With both age and articulation rate partialled out, the Mr. Cucum-
ber Test correlated r(109) ⫽ .27, p ⬍ .01, with visual and r(109) ⫽ .29, p ⬍ .002,
with auditory word span. The Counting Span Test correlated r(109) ⫽ .29, p ⬍
.002, with visual and r(109) ⫽ .27, p ⬍ .01, with auditory word span. Only the
Figural Intersections Test showed lower partial correlations, r(109) ⫽ .10, ns,
with visual and r(109) ⫽ .15, p ⬍ .06, with auditory word span.
It can be seen that the overall magnitude of correlations was somewhat lower
than in the pilot experiment. Nevertheless, their basic pattern was replicated and
generalized to visual presentation. The correlation between M capacity and word
span is confirmed as a robust phenomenon, while articulation rate gives a
significant but smaller contribution to children’s individual differences in span.
Goodness of fit of the model. The main aim of this experiment was to test the
model’s goodness of fit in different presentation modalities. This model assumes
that only in visual presentation must the encoding operative scheme be activated
by the M operator throughout stimuli presentation, and that the end-of-list signal
also requires M capacity.
In order to use independent observations in testing for goodness of fit, only the
second of three memory-span trials administered to each child for each word
length and modality was considered.
For the purpose of grouping participants by M capacity, their M-capacity
scores (i.e., the k values in the formula e ⫹ k) were rounded up/down to the
nearest unit. These scores ranged from e ⫹ 2 to e ⫹ 6. The probability
distribution (according to the model) of memory-span scores was computed for
each value of k from 2 to 6. Of course, these probabilities were expressed as
functions of free parameter i (see Appendix, Table 12). At this stage, the
probability distributions were weighted, according to the proportions of subjects
with an M-capacity score from e ⫹ 2 to e ⫹ 6, and then totaled in order to obtain
206 SERGIO MORRA

a probability distribution of memory-span scores in the whole sample. Conse-


quently, an expected value of the memory-span score in the whole sample
(expressed as a function of i) was obtained.
Since it was assumed that i values differ for different materials, regardless of
M capacity or presentation modality, three separate i parameter estimates were
made, one for each word length. These estimates were made by equating the
expected memory-span score value derived from the model (weighting for M
capacity and averaging visual and auditory conditions) to the observed mean
score (averaging visual and auditory presentations) in the second trial for each
word length in the whole sample. A single estimate of i was made across visual
and auditory presentation for each set of materials, since the model posits that
presentation modality affects M-capacity demand but not the value of i. The
estimates thus obtained were i ⫽ .0127 for two-syllable, i ⫽ .0170 for three-
syllable, and i ⫽ .0247 for four-syllable words. It may be noted that the values
obtained with both two- and three-syllable words are almost identical to those
found in the pilot experiment, while only the estimate for four-syllable words is
slightly lower (by about 17%). Such findings give further credibility to the
model.
These i estimates were entered in the probability distributions to thus obtain
predicted memory-score distributions for each word length, presentation modal-
ity, and M-capacity score. There were 26 subjects with a measured M capacity
of e ⫹ 3, 56 with a capacity of e ⫹ 4, and 25 with a capacity of e ⫹ 5. The six
subjects whose capacity fell out of this range (one with M ⫽ e ⫹ 2 and five with
e ⫹ 6) were excluded from the following analyses.
Table 4 shows expected and observed means, along with the relevant t tests.
Only 6 of 18 differences were significant. Moreover, 7 of 18 t values were lower
than 1 in absolute value. This suggests that the model’s goodness of fit for the
means is acceptable. On closer inspection, it can be seen that almost all signif-
icant differences from expected means concern higher-than-expected means in
the children with an M capacity of e ⫹ 3. This may be due to some inaccuracy
in M measurement (the participants in this experiment were unselected for M
capacity), or approximation in rounding up/down to the nearest unit. In support
of the latter interpretation, it can be noted that the actual mean of M-capacity tests
in this group is 3.11 (i.e., somewhat higher than 3), while it is 3.99 and 4.98 in
the other two groups (i.e., very close to 4 and 5, respectively). However, for the
two groups of children classified as having the appropriate M capacity for their
age (i.e., e ⫹ 4 or e ⫹ 5), the fit of expected to observed means is very good.
To test a model’s goodness of fit, means as well as other parameters of the
measure should be considered. Table 5 shows expected and observed variances,
along with the chi-square tests for the ratios between them. The fit of expected
to observed variances in this experiment was very good, in that only 1 of 18 ratios
was significant (i.e., precisely what is expected by chance). Ten observed
variances were smaller and eight were greater than expected. Out of four
VERBAL STM MODEL 207

TABLE 4
Observed and Expected Word-Memory-Span Means by Presentation Modality, Word Length,
and M-Capacity Score in Experiment 1

Mean span

No. of syllables M capacity Modality Expected Observed t df p

2 e ⫹ 3 Visual 3.46 3.92 2.78 24 ⬍.02


Auditory 4.14 4.27 0.58 24 ns
e ⫹ 4 Visual 3.96 4.11 1.04 54 ns
Auditory 4.66 4.50 ⫺1.32 54 ns
e ⫹ 5 Visual 4.51 4.64 0.59 23 ns
Auditory 5.32 5.16 ⫺0.84 23 ns
3 e ⫹ 3 Visual 3.20 3.65 3.07 24 ⬍.01
Auditory 3.89 4.27 3.21 24 ⬍.004
e ⫹ 4 Visual 3.70 3.71 0.09 54 ns
Auditory 4.39 4.46 0.87 54 ns
e ⫹ 5 Visual 4.24 4.24 ⫺0.02 23 ns
Auditory 5.05 4.68 ⫺1.81 23 ⬍.09
4 e ⫹ 3 Visual 2.91 3.50 3.34 24 ⬍.01
Auditory 3.60 4.11 3.03 24 ⬍.01
e ⫹ 4 Visual 3.41 3.30 ⫺0.99 54 ns
Auditory 4.08 4.20 1.29 54 ns
e ⫹ 5 Visual 3.93 3.72 ⫺1.26 23 ns
Auditory 4.73 4.36 ⫺2.13 23 ⬍.05

Note. The significance values of the t tests are two-tailed.

variances whose observed/expected ratio was significant or marginally signifi-


cant (i.e., p ⬍ .10), two were smaller and two greater than expected.
It might be questioned whether the range of M-capacity values considered in
testing for goodness of fit (i.e., from e ⫹ 3 to e ⫹ 5) was too limited. One
possible way of circumventing this problem is to pool together the auditory
condition of this experiment with the pilot experiment data (despite minor
differences in procedure and sampling). This involves using four groups of
children with M capacity of e ⫹ 2 (n ⫽ 41), e ⫹ 3 (n ⫽ 65), e ⫹ 4 (n ⫽ 69),
and e ⫹ 5 (n ⫽ 26), excluding only three children with M ⫽ e ⫹ 1 from the pilot
experiment, and five with M ⫽ e ⫹ 6 from this experiment. These results are
obviously limited to auditory presentation. The obtained i values were i ⫽ .0136
for two-syllable, i ⫽ .0169 for three-syllable, and i ⫽ .0261 for four-syllable
words. The results are shown in Tables 6 and 7. The model’s fit also seems very
good in this larger, pooled sample. Actually, only 1 of 12 t values for observed
and expected means and 4 of 12 observed/expected variance ratio chi-squares
were significant. It is also remarkable that 6 of 12 t values were smaller than 1.
This suggests that the model predicts well not only a correlation between M
capacity and memory span, but also, even more remarkably, the actual size of
memory-span increase as a function of M capacity.
208 SERGIO MORRA

TABLE 5
Observed and Expected Variances in Word Memory Span by Word Length,
Presentation Modality, and M Capacity in Experiment 1

Span variance

No. of syllables M capacity Modality Expected Observed ␹2 df p

2 e ⫹ 3 Visual 0.99 0.69 18.01 24 ns


Auditory 0.89 1.20 34.92 24 ns
e ⫹ 4 Visual 0.92 1.10 66.99 54 ns
Auditory 1.03 0.82 44.74 54 ns
e ⫹ 5 Visual 1.00 1.11 27.76 23 ns
Auditory 1.05 0.85 20.40 23 ns
3 e ⫹ 3 Visual 0.82 0.53 16.82 24 ns
Auditory 0.71 0.35 12.85 24 ⬍.07
e ⫹ 4 Visual 0.73 0.67 51.58 54 ns
Auditory 0.84 0.39 25.95 54 ⬍.001
e ⫹ 5 Visual 0.81 0.66 20.51 23 ns
Auditory 0.92 1.02 27.76 23 ns
4 e ⫹ 3 Visual 0.66 0.79 30.87 24 ns
Auditory 0.51 0.71 36.45 24 ⬍.10
e ⫹ 4 Visual 0.53 0.71 75.44 54 ⬍.06
Auditory 0.64 0.48 41.83 54 ns
e ⫹ 5 Visual 0.60 0.68 28.59 23 ns
Auditory 0.74 0.71 24.06 23 ns

Note. The significance values of the chi-squares are two-tailed.

EXPERIMENT 2
A second experiment tested the model with supraspan five-word lists, the
method used in most studies supporting Baddeley’s model. Because there is
evidence that a conventional span procedure yields different results (e.g., Morra,
1990; Morra, Mazzoni, & Sava, 1993; Nicolson & Fawcett, 1991), it seemed
necessary to test the current model also in a supraspan condition.
First, the model predicts that performance is positively correlated with M
capacity. Second, the model allows calculation of the expected probability of
recalling a whole supraspan list of a given number of words. Analyses tested
these two predictions, as well as for replication of known findings.
Participants were the same children tested in Experiment 1. The primary
reason for testing the same children was not one of convenience, but rather was
driven by design advantage. First, these children had already practiced short-term
memory tasks and refined the rehearsal strategy, so they were likely to employ
it with maximum efficiency. Second, and more important, the i parameter had
already been estimated in Experiment 1 with the same word sets. It was assumed
that the i parameter depends on the materials but not on presentation modality or
span versus supraspan lists. This meant the model could be tested in a most
rigorous way, because no free parameters were estimated in this experiment.
VERBAL STM MODEL 209

TABLE 6
Observed and Expected Means in Word Memory Span by Word Length and M Capacity:
Pooled Data of Pilot Experiment and Auditory Condition of Experiment 1

Mean span

No. of syllables M capacity Expected Observed t df p

2 e ⫹ 2 3.60 3.54 ⫺0.61 39 ns


e ⫹ 3 4.08 4.32 2.26 63 ⬍.03
e ⫹ 4 4.60 4.55 ⫺0.42 67 ns
e ⫹ 5 5.26 5.12 ⫺0.75 24 ns
3 e ⫹ 2 3.41 3.42 0.05 39 ns
e ⫹ 3 3.89 4.06 1.76 63 ⬍.09
e ⫹ 4 4.40 4.51 1.52 67 ns
e ⫹ 5 5.06 4.65 ⫺2.01 24 ⬍.06
4 e ⫹ 2 3.07 3.10 0.33 39 ns
e ⫹ 3 3.56 3.62 0.53 63 ns
e ⫹ 4 4.03 4.16 1.50 67 ns
e ⫹ 5 4.68 4.39 ⫺1.78 24 ⬍.09

Note. The significance values of the t tests are two-tailed.

Method
Participants and materials. The 113 children who took part in Experiment 1
were tested no more than 1 week after the prior session.
The same three sets of two-, three-, and four-syllable nouns as in the previous
experiments were used. Ten lists of five words were randomly created from each
set, with the constraints that each word appeared once in each position and the
last word in a list could not be the first in the following one. In addition, four
practice lists of five words were created from the practice word set.
Procedure. Performance with five-word lists was measured in a single session.
Values of M capacity and articulation rate were taken from Experiment 1.
Following practice, children were given 10 lists, each presented for immediate
recall, in each of three experimental conditions (i.e., two-, three-, and four-
syllable words). The six orderings of word lengths were balanced over subjects,
as in Experiment 1. The word lists were presented only visually, both for
practical reasons of time and because five words might not be supraspan in
auditory presentation (see Tables 2 and 4). As in the visual condition of
Experiment 1, the words appeared one at a time in the center of a screen at a rate
of one word every 2 s.
Children were required to recall words orally in the appropriate positions,
saying “niente” for those positions in which they could not recall the word.
“Niente” means “nothing” and its sound is not too dissimilar from “blank,” which
is often used for this purpose in experiments with English words.
Recall of a list was scored as correct or incorrect for each single position, and
also as correct or incorrect for the whole list. Thus, each subject received 3 scores
210 SERGIO MORRA

TABLE 7
Observed and Expected Variances in Word Memory Span for Each Word Length and M-Capacity
Score: Pooled Data of Pilot Experiment and Auditory Condition of Experiment 1

Span variance

No. of syllables M capacity Expected Observed ␹2 df p

2 e ⫹ 2 0.92 0.40 17.64 39 ⬍.01


e ⫹ 3 0.85 0.74 56.95 63 ns
e ⫹ 4 0.98 0.80 55.97 67 ns
e ⫹ 5 1.02 0.87 22.23 24 ns
3 e ⫹ 2 0.80 0.44 22.52 39 ⬍.04
e ⫹ 3 0.71 0.58 52.95 63 ns
e ⫹ 4 0.85 0.37 29.75 67 ⬍.001
e ⫹ 5 0.92 1.00 28.16 24 ns
4 e ⫹ 2 0.61 0.23 15.81 39 ⬍.001
e ⫹ 3 0.49 0.64 85.15 63 ns
e ⫹ 4 0.62 0.48 54.08 67 ns
e ⫹ 5 0.71 0.70 25.43 24 ns

Note. The significance values of the chi-squares are two-tailed.

for the number (out of 10) of whole lists correctly recalled for each word length,
and 15 scores for the number (out of 10) of words correctly recalled in each
position for each word length.

Results and Discussion


Preliminary analyses. Table 8 shows descriptive statistics for recall scores. A
2 (age group) ⫻ 3 (word length) ⫻ 5 (position) mixed-design ANOVA of recall
scores yielded significant main effects of age, F(1, 111) ⫽ 6.53, p ⬍ .02, word
length, F(2, 222) ⫽ 117.93, p ⬍ .001, and position, F(4, 444) ⫽ 296.15, p ⬍
.001. Two interactions were significant. One was Age ⫻ Word Length, F(2,
222) ⫽ 3.16, p ⬍ .05, due to a larger performance decrease with four-syllable
words for the younger group. The other was Word Length ⫻ Position, F(8,
888) ⫽ 7.19, p ⬍ .001, showing a stronger position effect with longer words. The
mean numbers of recalled words per list were 3.65, 3.28, and 2.80 for two- to
four-syllable words. All differences between word lengths were significant at p
⬍ .001. This pattern replicates the word-length effect often reported with
supraspan lists. The means are lower than those found in the visual condition of
Experiment 1, which also replicates the cliff beyond span phenomenon (Drach-
man & Zaks, 1967).
The effect of position was significant for each word length, F(4, 444) ⫽
121.11, p ⬍ .001, F(4, 444) ⫽ 157.57, p ⬍ .001, and F(4, 444) ⫽ 187.01, p ⬍
.001, for two-, three-, and four-syllable words, respectively. Bonferroni t tests
with p ⬍ .01 showed a decrease in recall from the first to the second, from the
VERBAL STM MODEL 211

TABLE 8
Means (and Standard Deviations) of Recall Scores by Grade, Word Length,
and Serial Position in Experiment 2

Serial position

No. of syllables Grade 1 2 3 4 5

2 4 9.00 (1.23) 8.02 (1.72) 6.55 (2.24) 5.62 (2.49) 5.83 (2.56)
5 9.36 (1.01) 8.71 (1.47) 7.65 (2.21) 6.25 (2.40) 6.18 (2.62)
3 4 8.71 (1.61) 7.05 (2.07) 6.24 (2.11) 5.00 (2.51) 4.33 (2.64)
5 9.24 (1.32) 7.89 (1.78) 6.87 (2.05) 5.36 (2.66) 4.96 (3.01)
4 4 7.78 (1.89) 6.31 (2.17) 4.88 (2.20) 3.22 (2.34) 3.14 (2.20)
5 8.73 (1.50) 7.51 (2.13) 5.98 (2.42) 4.60 (2.58) 4.00 (2.89)

Note. Maximum possible score ⫽ 10.

second to the third, and from the third to the fourth position in the list, for each
word length. Only with three-syllable words was the decrease from the fourth to
the fifth position also significant. These results broadly replicate those obtained
with adults (Cowan et al., 1992).
The ratio between mean span and articulation rate increased from 1.25 s for the
shortest words to 1.64 s for the longest. The regression line across six data points
(two age groups ⫻ three word lengths) was Recall ⫽ 1.65 ⫹ 0.712AR. It
accounted for 75.3% variance across word lengths and age groups, and both
intercept and slope were significantly different from zero (p ⬍ .02 in each case).
Once more, the linear relationship of memory span to articulation rate was
replicated, while the intercept was clearly above zero. (As already noted, only the
first of these two results is consistent with Baddeley’s classic model.)
Each child’s scores with two-, three-, and four-syllable words were averaged
to compute correlations between span and other variables. As mentioned above,
two different scoring criteria were used (i.e., single words and whole lists
correctly recalled). The number of words correctly recalled in their positions was
positively correlated with M capacity, r(111) ⫽ .38, p ⬍ .001, and with
articulation rate, r(111) ⫽ .26, p ⫽ .002. The number of whole lists recalled also
correlated with M capacity, r(111) ⫽ .32, p ⬍ .001, and with articulation rate,
r(111) ⫽ .23, p ⬍ .01.
With both age and articulation rate partialled out, M capacity correlated
r(109) ⫽ .37, p ⬍ .001, with recalled words and r(109) ⫽ .34, p ⬍ .001, with
recalled lists. Conversely, with both age and M capacity partialled out, articula-
tion rate correlated r(109) ⫽ .25, p ⬍ .01, with recalled words and r(109) ⫽ .24,
p ⬍ .01, with recalled lists. In short, correlations with M capacity were also
higher in this experiment, but a significant contribution of articulation rate to
recall of supraspan lists was found, too.
Once again, correlation between span and M capacity was not spurious due to
some specific test. With age and articulation rate partialled out, recalled lists
212 SERGIO MORRA

TABLE 9
Expected and Observed Means (and Standard Deviations) in Recall of Whole Five-Word Lists
for Each Word Length and M-Capacity Score in Experiment 2

Mean recall

No. of syllables Group Expected Observed t df p

2 All 4.497 4.203 (2.771) ⫺1.13 112 ns


e ⫹ 3 3.005 3.346 (2.591) 0.67 25 ns
e ⫹ 4 4.292 3.714 (2.528) ⫺1.71 55 ⬍.10
e ⫹ 5 5.998 6.280 (2.542) 0.55 24 ns
3 All 3.363 2.929 (2.744) ⫺1.68 112 ⬍.10
e ⫹ 3 1.851 2.077 (2.331) 0.49 25 ns
e ⫹ 4 3.077 2.500 (2.435) ⫺1.77 55 ⬍.09
e ⫹ 5 4.936 4.840 (2.779) ⫺0.17 24 ns
4 All 1.943 1.770 (2.248) ⫺0.82 112 ns
e ⫹ 3 0.674 1.154 (1.736) 1.41 25 ns
e ⫹ 4 1.565 1.304 (1.877) ⫺1.04 55 ns
e ⫹ 5 3.361 3.520 (2.663) 0.30 24 ns

Note. Maximum possible score ⫽ 10. The significance values of the t tests are two-tailed.

correlated r(109) ⫽ .26, p ⬍ .01, with Mr. Cucumber, r(109) ⫽ .30, p ⬍ .001,
with the Counting Span, and r(109) ⫽ .21, p ⬍ .02, with the Figural Intersections
Test, whereas recalled words correlated r(109) ⫽ .29, p ⬍ .002, with Mr.
Cucumber, r(109) ⫽ .34, p ⬍ .001, with the Counting Span, and r(109) ⫽ .22,
p ⬍ .01, with the Figural Intersections Test.
Goodness of fit of the model. Expected recall probabilities for a visually
presented five-word list are shown in the Appendix (see Table 12). The expected
recall scores are 10 times such probabilities. As mentioned, this prediction was
put to a very strict test—parameter i was not freely estimated, but the values of
.0127, .0170, and .0247, found in Experiment 1 for two-, three-, and four-syllable
words, respectively, directly substituted for i in the equations. The expected and
observed recall means are shown in Table 9, along with the goodness of fit of the
model.
The fit of the model is very good: Neither in the whole sample nor in any group
with a specific M capacity is there a significant difference between observed and
expected means. Only 3 of 12 t values were marginally significant (p ⬍ .10).
Indeed, in 6 cases (i.e., half of the comparisons) there is a result of t ⬍ 1. It may
be that the model fits the data even better than in the previous experiments
because each child’s score was not obtained from a single trial but from 10, thus
reducing random variation. In addition, participants’ previous experience with
span tasks may have improved experimental control over strategies. In any case,
it can be concluded that predictive power of this model generalizes to supraspan
lists presented visually—a conclusion that was reached without estimating any
VERBAL STM MODEL 213

free parameter (because M capacity was independently measured and i had


previously been estimated in the same sample from different span tasks).

GENERAL DISCUSSION

The experiments reported in this article were designed to compare and inte-
grate different lines of research and to test a new model of verbal STM.
Consistent with Baddeley’s articulatory loop model were the findings that there
was a word-length effect and that there was a linear relation between group
means of articulation rate and short-term memory. However, inconsistent with
Baddeley’s model were the findings that the intercepts of such linear functions
were well above zero and that the ratio of span to articulation rate was not
constant. Although positive correlations between individuals’ STM span and
articulation rate are consistent with the loop model, these correlations were not
large and were not always significant after age, M capacity, or verbal ability was
statistically removed. The estimates of the articulatory loop capacity were in-
consistent. The linear function slope ranged from .48 s in the auditory condition
of Experiment 1 to .71 s in Experiment 2, while the ratio of recall to articulation
rate varied from 1.25 s (Experiment 2, two-syllable words) to 2.56 s (pilot
experiment, four-syllable words).
On the one hand, these results suggest that rehearsal skill may make an
independent contribution to span. In both Experiment 1 and Experiment 2, the
partial correlations between articulation rate and short-term memory were low
but significant.
On the other hand, the pattern of group means was inconsistent with the
articulatory loop model, at least in its standard form. Hulme et al. (1991)
concluded that their results (similar to these) were incompatible with the wide-
spread unitary view of short-term memory span, in which all storage is said to
occur within an articulatory loop. The inconsistent estimates of the loop capacity,
reported above, seem to push this conclusion still further, raising doubts about
the existence of a time-limited component dedicated to articulatory storage. The
concept of a time-limited specialized store may have been an excessive gener-
alization from results obtained under specific conditions. Apart from one study
on digit span in English and Chinese (Stigler, Lee, & Stevenson, 1986), all
experiments that found a near-zero intercept shared two features: They involved
the English language and the use of supraspan lists. Thus, the findings, which are
often interpreted as evidence for a time-limited articulatory loop, might actually
be produced by both language and the span-assessing procedures (see also
Cheung & Kemper, 1993; Morra et al., 1993).
Although the present results corroborate the view that the articulatory loop
model needs to be revised, they are not totally incompatible with some models
derived from it (see Gathercole, 1997). Other authors report above-zero inter-
cepts and suggest that they could reflect the contribution to memory span of
storage in the central executive (Hitch, Halliday, & Littler, 1989) or in the
214 SERGIO MORRA

phonological store (Nicolson & Fawcett, 1991), or of retrieval from long-term


memory (Hulme et al., 1991). Each of these accounts is consistent in itself and
with the data from which it originated. However, there are notable differences
among them, and as Kuhn (1962/1969) noted, when too many versions of a
paradigm are proposed, it often signals that the paradigm itself is in crisis.
Moreover, some accounts may be criticized for assuming that verbal materials
are held in two separate containers, one of which is time-limited and articulatory,
and the other of which holds a very small number of words (or similar chunks).
Some models (e.g., Burgess & Hitch, 1992; Cowan et al., 1994; Hulme et al.,
1991, 1997), however, do not make this unlikely assumption. Although they still
assume a loop that is time-limited, this does not seem to be a central feature in
them. The results presented here might therefore be compatible with these
models, if they were in turn modified by stating that the articulatory rehearsal
process need not be time-limited.
Burtis’s (1982) model can account well for the correlation between STM span
and M capacity. This correlation ranged from .30 in the visual condition of
Experiment 1 to .65 in the pilot experiment. It withstood partialling out of a
number of variables, and it was significant not only with a compound M-capacity
score, but also when separate scores in different tests were considered. Never-
theless, Burtis (1982) does not account for word-length and modality effects,
which are often reported in the experiments and also replicated here.
The new model proposed here—a revision of Burtis’s—may offer a solution
to the shortcomings of previous models. There are four major differences
between the model proposed by Burtis (1982) and the present one. First, Burtis’s
main experiment used very long supraspan lists and did not require strictly
ordered recall. Hence, he did not need to posit a specific operative scheme for
subvocal rehearsal, and he did not need to be concerned with whether subjects
used articulatory coding. Second, in this paper I posited two different strategies,
one based on subvocal rehearsal and the other one based on end-anchoring and
partial guessing of order, probably used by those subjects whose M capacity is
insufficient for using a resource-consuming rehearsal. Third, because Burtis used
only consonants he did not posit that the i parameter would change as a function
of materials (e.g., as a function of word length). Fourth, Burtis assumed that the
probability of recalling a partially activated scheme is a function of how long it
has been decaying, whereas I assume that it is also a function of how many
partially activated schemes interfered with it.
The present model offers some hope of overcoming the limitations of other
models because it represents (a) M capacity in the same way as Burtis’s; (b)
modality effects in terms of automaticity, which yields differential M-capacity
demands; (c) rehearsal as an operative scheme with the main function of keeping
order information; and (d) word-length effects as the outcome of differential
activation decrease per processing step. With few exceptions, this model yielded
accurate predictions of memory-span increase as a function of M capacity
VERBAL STM MODEL 215

throughout a series of experiments with different age groups, presentation mo-


dalities, and tasks. In addition to showing goodness of fit to the data, also
favoring the new model was consistency across experiments of i parameter
estimates. Estimated i value for two-syllable words was .0124 in the pilot
experiment and .0127 in Experiment 1; for three-syllable words, it was .0169 and
.0170; and for four-syllable words, it was .0298 and .0247. In Experiment 2 it
was not even necessary to estimate its values, because those obtained in Exper-
iment 1 from the same materials and children (but from different tasks and
presentation modalities) were used. Such consistent estimates across experiments
are perhaps more indicative of the model’s power than are any tests of goodness
of fit.
Despite success, of course, this research has its limitations. These experiments
studied the effects of M capacity, presentation modality, word length, and span
vs supraspan tasks. Other variables, such as phonological confusion or articula-
tory suppression, were not manipulated. It may be asked how far the results can
be generalized beyond the specific conditions studied here. Further research
should tackle this issue. This would involve modifying the model so as to
incorporate the constraints of other experimental conditions. To give but one
example, articulatory suppression should force a person to choose the nonre-
hearsal strategy, and also require one unit of M capacity to monitor the flow of
irrelevant utterances. In a similar way, one can derive new predictions regarding
other variables, and test them.
I suggested in this article that automatic encoding in auditory presentation
allows a subject to economize M capacity. This is a novel explanation for the
modality effect. Although there is agreement on automatic encoding of heard
speech (e.g., Baddeley, 1986; Penney, 1989), the suggested implication about
economizing M capacity is new. This idea could also help to explain modality
differences in the suffix effect. This effect is weak in visual presentation (e.g.,
Penney, 1989) possibly because encoding a visually presented suffix does not
demand more M capacity than encoding a visually presented end-of-list
signal. Instead, encoding an auditory suffix demands M capacity, which
would not be required to encode the end-of-list signal. Furthermore, the
long-established but often-neglected finding that the auditory suffix effect
extends over a number of words, regardless of word length or stimuli duration
(Watkins & Watkins, 1973), is neatly suited to the role of a series of encoding
operations (one for each word) posited here. Thus, also further research on
the suffix effect is advisable.
The modality effect had been explained in various ways. The classic account,
which posits a short-lived precategorical store, has met with abundant conflicting
evidence (e.g., Longoni, Richardson, & Aiello, 1993; Penney, 1989; Watkins &
Watkins, 1973). Others (e.g., Nairne, 1990; Penney, 1989) have posited specific
traces derived from perceptual processing. The present model does not represent
lower level perceptual processes. However, I do not exclude the possibility that
216 SERGIO MORRA

they have an additional influence, nor do I suggest that economized M capacity


is the only source of the modality effect. What I argue is only that this contributes
importantly to an explanation of it.
Another remaining problem is the nature of parameter i, which was defined as
decrease in probability of recalling a partially activated scheme and was assumed
to be materials-specific. However, this definition tells us little about psycholog-
ical mechanisms of forgetting. Decreased activation of schemes may result from
various causes, such as decay, output interference, input interference from new
items, and automatic interruption. The last term refers to intervention of the I
operator, which is assumed to occur after an M operation and to reduce the
activation of schemes not boosted by the M operator (e.g., Pascual-Leone, 1987).
All these causes could affect short-term forgetting. In this article I compounded
all possible causes of activation decrease in a single parameter. This is parsimo-
nious, and the presence of only one free parameter makes the test of the model
stringent. One could find the model too complex and too unconstrained, if several
free parameters were included in it for different causes of deactivation. Yet, it is
possible to distinguish experimentally various causes, and represent them by
separate parameters. Future research may establish whether the model can be
improved in this way.
However, these experiments were not intended to clarify which psychological
mechanisms cause across-materials variation of the i parameter. Rather, they
showed the role of M capacity, how it interacts with modality and span vs
supraspan tasks, and how the word-length effect can be easily accommodated
within this framework. Indeed, a model with only two basic mechanisms (M
capacity and activation decrease) generated good predictions across word
lengths. The first of these basic mechanisms modeled in quantitative terms the
role of a central attentional resource. In my view, this is important because a role
of central components in verbal short-term memory has seldom been mentioned
(e.g., by Hitch, Halliday, & Littler, 1989), and then only generically. Most
current models seem to neglect it.
A final point to be discussed is a comparison between my model and other,
recent ones. Readers may note conceptual similarities to other activation-based
models (Anderson & Matessa, 1997; Conway & Engle, 1994; Cowan, 1988,
1993; Engle et al., 1992; Just & Carpenter, 1992). In different ways, these authors
suggested that short-term memory consists of long-term memory units, currently
activated either by central executive processes or by attended stimuli in the
sensory input. In particular, Moscovitch and Umiltà (1991) conceive of working
memory as “whatever processes are currently active,” and suggest that its limits
are set by the resources necessary for maintaining information and operating on
it. The present model, however, takes one step further—that is, its theoretical
framework allows specification of the schemes and general-purpose mechanisms
that contribute to performance in a given short-term memory task, and this in turn
allows more detailed quantitative predictions.
VERBAL STM MODEL 217

An important revision of the articulatory loop was advanced by Burgess and


Hitch (1992), whose model differs from the present one in several ways.
Nevertheless, the two models are not in principle incompatible, because Burgess
and Hitch do not consider time-limited capacity as a basic property, but as one
emerging from the particular connections and parameters set in their simulations.
Their way of conceiving the articulatory loop, as a set of weighted connections
among phoneme representations, is more detailed than my view of the rehearsal
process as an operative scheme that includes order information. Burgess and
Hitch (1992), however, concluded that this aspect of their model should be
modified. Consequently, how to model in detail rehearsal and order information
remains an open issue (see also Brown, 1997; Cowan, 1992; Gathercole, 1997).
What Burgess and Hitch’s (1992) model still lacks is a mechanism comparable
to the M operator. In addition, their model seems to exclude the possibility that
a person rehearses items before the whole list has been presented, so that
rehearsal is equivalent to covertly recalling the entire list. A central attentional
component such as the M operator and the assumption that people also rehearse
during stimuli presentation are two features of my model that other researchers
may choose to incorporate into their own models.
Other models, such as Brown and Hulme (1995) or Neath and Nairne (1995),
are even more different. They do not include constructs like the M operator, nor
any role of rehearsal, and may therefore find it difficult to account for the
correlations reported here. Their articles provide clear support for the claim that
the word-length effect need not be explained by a time-limited rehearsal device.
Nevertheless, the evidence for their models rests on estimating at least nine
parameters per experiment (Hulme et al., 1997), or on making numerous, perhaps
unwarranted quantitative assumptions (Neath & Nairne, 1995). Anderson and
Matessa (1997) advanced a model that considers both activation and time
constraints, but it does not consider either rehearsal or strategic allocation of
limited attentional resources. Also these authors obtained a good fit for their
model, but at the cost of estimating several free parameters, so that one may
question whether it is sufficiently constrained. In this article, as discussed above,
an opposite choice was made, that is, starting to test the goodness of fit of a model
with a single free parameter (leaving to subsequent research the task of improv-
ing it).
In conclusion, while new models are flourishing in short-term memory re-
search, none of them (including the present one) is likely to be fully satisfactory,
as the authors themselves often recognize. However, after two decades in which
the articulatory loop model had a status of paradigm (Kuhn, 1962/1969), it seems
that theoretical progress is being made, and that it will probably be consolidated
in research aimed at comparing or refining models, such as those discussed here.
Two main ideas have been put forward in this paper: the role of a central
attentional mechanism, whose limited capacity is divided between operative and
figurative schemes, and a decreasing recall probability only for those schemes
218 SERGIO MORRA

TABLE 10
Hypothetical Sequence of Processing Steps by a Subject with M Capacity ⫽ e ⫹ 4,
Auditorily Presented with a List of Five Words

Schemes activated by the M operator Activation weight of the schemes Probability


of correct
Step e k Output ␸ W1 ␸ W2 ␸ W3 ␸ W4 ␸ W5 recall

1 ␧ Lisn [␺ Code ] W1 ␸ W1 1
2 ␧ Lisn ␺ Reh ␸ W1 [␺ Code ] W2 ␸ W2 1 1
3 ␧ Lisn ␺ Reh ␸ W1 ␸ W2 [␺ Code ] W3 ␸ W3 1 1 1
4 ␧ Lisn ␺ Reh ␸ W1 ␸ W2 [␺ Code ] W4 ␸ W4 1 1 1 ⫺ i 1
5 ␧ Lisn ␺ Reh ␸ W1 ␸ W2 [␺ Code ] W5 ␸ W5 1 1 1 ⫺ 3i 1 ⫺ 2i 1
6 ␧ Lisn ␸ W5 ␺ Reh ␸ W1 ␸ W2 [␺ Code END] ␧ Rec 1 1 1 ⫺ 5i 1 ⫺ 4i 1
7 ␧ Rec ␺ Retr ␸ W1 ␺ Reh ␸ W2 U W1 1 1 ⫺ 8i 1 ⫺ 7i 1 ⫺ 3i 1
8 ␧ Rec ␺ Retr ␸ W2 U W2 1 ⫺ 11i 1 ⫺ 10i 1 ⫺ 6i 1
9 ␧ Rec ␺ Retr ␸ W3 ? U W3 ? 1 ⫺ 12i 1 ⫺ 8i 1 ⫺ 11i
10 ␧ Rec ␺ Retr ␸ W4 ? U W4 ? 1 ⫺ 9i 1 ⫺ 12i
11 ␧ Rec ␺ Retr ␸ W5 ? U W5 ? 1 ⫺ 9i

Note. Probability of correctly recalling the list ⫽ (1 ⫺ 11i) (1 ⫺ 12i) (1 ⫺ 9i). [Square brackets
point to the schemes activated by the F and L operators.]

which are out of the focus of central attention. In addition, it was suggested (and
formally expressed in a model) that differential demands on M capacity may
account for differences between visual and auditory presentation and between
span and supraspan tasks. If the reader agrees that this paper has provided some
evidence to support these ideas, but prefers classical human information pro-
cessing or parallel distributed processing frameworks to a neo-Piagetian one, the
challenge is to translate the ideas advanced here into the language of one’s
favorite framework.

APPENDIX
Formal aspects of this model of verbal short-term memory can best be
illustrated by means of two examples. The first involves subvocal rehearsal,
while the second considers a different strategy.

An Example Involving Rehearsal


Table 10 refers to the mental processing of a subject with M capacity of e ⫹
4 (i.e., an average 10-year-old capable of activating a task executive plus 4
operative or figurative schemes), who is auditorily presented a list of five words
and tries to remember it by rehearsing. Because a number of symbols are used,
it is convenient to explain their meaning first (see also Table 1).
W1, W2, . . . , etc., stand for the first, second, . . . , etc., word in a list. END
stands for an end-of-list signal (which in auditory presentation might simply be
the experimenter turning to look at the subject).
VERBAL STM MODEL 219

The Greek letter ␧ stands for a relatively simple task executive (e.g., a
representation of the current goals). As in Burtis (1982), ␧ Read and ␧ Rec represent
the task executives for the input and recall phases of an STM task with visual
presentation of stimuli, while ␧ Lisn denotes a task executive for the input phase
with auditory presentation.
The letter ␺ stands for an operative scheme. Three different operative schemes
are considered in the model. The first is ␺ Code, the word-encoding operation. The
second is ␺ Reh, the operation of ordered subvocal rehearsal of all the figurative
schemes currently activated by the M operator. Depending on the capacity of the
M operator, these figurative schemes may represent some or all of the words
presented to the subject. The third is ␺ Retr, the operation of retrieving the next
word in a list.
The letter ␾ stands for a figurative scheme. In particular ␾ W1, ␾ W2, . . . , etc.,
stand for mental representations of the first, second, . . . , etc., word in a list. The
symbol ␾ END stands for a mental representation of the end-of-list signal.
An important concept is the degree of activation of figurative schemes,
particularly the loss of activation of schemes that had previously been fully
activated. In the far right column of Table 10, full activation of a scheme is
indicated by the number 1, and decrease of activation is expressed in terms of i,
the only free parameter in the model. Thus, the table contains expressions such
as 1 ⫺ i, 1 ⫺ 3i, and 1 ⫺ 8i. Parameter i is defined as the decrease in probability
of recall of an item per processing step and per decaying scheme. In other words,
decrease of activation is proportional to both number of processing steps and
number of partially activated schemes. Thus, if at a given point only one scheme
is decaying, then it loses an amount i of activation during that step; if two
schemes are decaying, each of them loses an amount 2i of activation during that
step, and so on. Parameter i is assumed to be specific to each type of encoded
material (e.g., long or short words, confusable or nonconfusable consonants,
digits, semantic– conceptual codes).
Finally, U W1, U W2, . . . , etc., represent utterances of recalled words. A question
mark in the subscript, as in U W5?, indicates that the utterance of, for example, the
fifth word occurs with a specified probability smaller than 1 because ␾ W5 is only
partly activated.
In the first six steps, the subject has the goal of listening to the stimuli (␧ Lisn is
activated). In particular, in Step 1, the subject has to encode (␺ Code) the first word
(W1). Only one of the subject’s four units of M capacity is used for this, since
articulatory encoding of spoken words does not demand any (i.e., it is automatic).
The outcome of this processing step is that a specific articulatory code (␾ W1) is
activated at its highest degree (i.e., 1).
Articulatory recoding of words is assumed to be effortless in the case of
auditory presentation; if the stimuli are read aloud by a human experimenter, the
same assumption is also made for understanding the end-of-list signal (i.e., any
cues that cause a subject to start recalling the items). Square brackets, as in [␺ Code]
220 SERGIO MORRA

in Steps 1 to 5 and [␺ Code END] in Step 6, are used to represent the assumption
that these schemes do not consume any resources of the M operator to be
activated. The sources of their activation would be operators F and L due to
stimulus–response compatibility and overlearning.
In Step 2, three M-capacity units are used. Two of them activate the schemes
(␺ Reh, ␾ W1) involved in rehearsal of the previous item, and only one is allocated
to the two schemes (␺ Code, W2) involved in articulatory recoding of the second
word. Thus, four schemes are involved in all, but only three M-capacity units are
used because activation of ␺ Code is automatic. At this point, both ␾ W1 and ␾ W2 are
fully activated. Step 3 is similar to Step 2, except that four units of M capacity
are used because three schemes (i.e., ␺ Reh, ␾ W1, and ␾ W2) are now involved in
rehearsal. Thus, ␾ W1, ␾ W2, and ␾ W3 are fully activated.
However, at Step 4 it is not possible to continue activating one more articu-
latory code at its highest degree while rehearsing everything. To do so, an M
capacity of e ⫹ 5 would be needed, but we are now modeling the processing of
a subject with a capacity of e ⫹ 4, who can use one unit of the M operator in the
rehearsal process (␺ Reh), two to keep the articulatory representations of the first
two words activated (␾ W1, ␾ W2), and the last unit in recoding the fourth word. The
schemes ␾ W1, ␾ W2, and ␾ W4 are now fully activated, but activation of ␾ W3 begins
to drop. Since only one scheme is decaying at this step, its activation decreases
by an amount corresponding to free parameter i. This is the meaning of the
expression 1 ⫺ i shown in Table 10, Step 4, in the ␾ W3 activation column.
Step 5 is similar to Step 4, except that two articulatory codes, ␾ W3 and ␾ W4, are
now losing activation. Therefore, the activation of each of them is assumed to
decrease by an amount of 2i. Thus, activation of ␾ W3 drops from 1 ⫺ i to 1 ⫺ 3i,
and activation of ␾ W4 drops from 1 to 1 ⫺ 2i.
At Step 6 the end-of-list signal is encoded, so that the subject can set
him-/herself the goal of recalling words. The two relevant schemes (␺ Code, END)
are activated without any expenditure of M capacity. Therefore, three units of M
capacity can be used to rehearse the first two words (␺ Reh, ␾ W1, and ␾ W2), and one
unit can still be used to keep ␾ W5 activated. Two schemes (␾ W3 and ␾ W4) are now
decaying so each of them loses an amount of activation of 2i, dropping to 1 ⫺
5i and 1 ⫺ 4i, respectively.
Steps 7 to 11 represent the recall phase. In particular, at Steps 7 and 8 the
first two words are safely recalled (with probability ⫽ 1) because they had
been kept fully active throughout the process. At Step 7, the four units of M
capacity are allocated to schemes ␺ Retr , ␾ W1 , ␺ Reh , and ␾ W2 . In the meantime,
schemes ␾ W3 , ␾ W4 , and ␾ W5 continue losing activation. Activation of these
schemes decreases by an amount of 3i during Step 7 because there are three
decaying schemes, so it drops to 1 ⫺ 8i, 1 ⫺7 i, and 1 ⫺ 3i, respectively.
They continue to decay during Step 8, so that their activation drops to 1 ⫺
11i, 1 ⫺ 10i, and 1 ⫺ 6i, respectively.
At Step 9, the third word is eventually recalled with probability ⫽ 1 ⫺ 11i (i.e.,
VERBAL STM MODEL 221

TABLE 11
Hypothetical Sequence of Processing Steps by a Subject with M Capacity ⫽ e ⫹ 2,
Visually Presented with a List of Four Words

Schemes
activated by the
M operator Activation weight of the schemes Probability
of correct
Step e k Output ␾ W1 ␾ W2 ␾ W3 ␾ W4 recall

1 ␧ Read ␺ Code W1 ␸ W1 1
2 ␧ Read ␺ Code W2 ␸ W2 1 ⫺ i 1
3 ␧ Read ␺ Code W3 ␸ W3 1 ⫺ 3i 1 ⫺ 2i 1
4 ␧ Read ␺ Code W4 ␸ W4 1 ⫺ 6i 1 ⫺ 5i 1 ⫺ 3i 1
5 ␧ Read ␺ Code END ␧ Rec 1 ⫺ 10i 1 ⫺ 9i 1 ⫺ 7i 1 ⫺ 4i
6 ␧ Rec ␺ Retr ␸ W1? U W1? 1 ⫺ 12i 1 ⫺ 10i 1 ⫺ 7i 1 ⫺ 10i
7 ␧ Rec ␺ Retr ␸ W2? U W2? 1 ⫺ 12i 1 ⫺ 9i (1 ⫺ 12i)/2
8 ␧ Rec ␺ Retr ␸ W3? U W3? 1 ⫺ 10i 1 ⫺ 12i
9 ␧ Rec ␺ Retr ␸ W4? U W4? 1 ⫺ 10i

Note. Probability of correctly recalling the list ⫽ (1 ⫺ 10i) 2 (1 ⫺ 12i) 2 / 2.

equal to the current degree of activation of its scheme). Meanwhile, schemes ␾ W4


and ␾ W5 continue losing activation, so that at Steps 10 and 11 the last two words
are recalled with probabilities of 1 ⫺ 12i and 1 ⫺ 9i, respectively.
In short, the probability of correctly recalling the whole list is the product of
the probabilities of serially recalling each of its component words. These prob-
abilities are 1 for W1 and W2, 1 ⫺ 11i for W3, 1 ⫺ 12i for W4, and 1 ⫺ 9i for
W5.

An Example without Rehearsal


Table 11 exemplifies the mental processing of a subject with M capacity of e ⫹
2, visually presented with a list of four words, who tries to remember them
without rehearsing. Since it is very similar to Table 10, it is not commented on
step by step. The main differences are the following. In the input phase, the
subject’s goal is to read rather than to listen; hence, the executive ␧ Read appears.
Articulatory recoding of words and understanding the end-of-list signal are
assumed to require effort in visual presentation (i.e., to demand M capacity).
Thus, ␺ Code in Steps 1 to 5 and END in Step 5 do not appear in square brackets
but are included among the schemes activated by the M operator.
Some schemes begin to show activation decrease at Step 3, when the two units
of the M operator are fully engaged in keeping active the encoding operative
scheme (␺ Code) and processing the third word (W3). The final probability of
recalling each of the single words in this list is 1 ⫺ 10i, 1 ⫺ 12i, 1 ⫺ 12i, and
1 ⫺ 10i, respectively.
222 SERGIO MORRA

TABLE 12
Equations Expressing Probability of Recall of a List of a Given Length as a Function
of M Capacity, Presentation Modality, and the Free Parameter i

Modality of List
M capacity presentation length Probability of recall

e ⫹ 1 Auditory 2 (1 ⫺ i)(1 ⫺ 2i)


3 (1 ⫺ 3i)(1 ⫺ 5i)(1 ⫺ 6i)
4 (1 ⫺ 6i)(1 ⫺ 9i)(1 ⫺ 11i) 2 /2
5 (1 ⫺ 10i)(1 ⫺ 14i)(1 ⫺ 17i) 2 (1 ⫺ 18i)/6
6 (1 ⫺ 14i)(1 ⫺ 20i)(1 ⫺ 24i) 2 (1 ⫺ 26i) 2 /24
7 (1 ⫺ 21i)(1 ⫺ 27i)(1 ⫺ 32i) 2 (1 ⫺ 35i) 2 (1 ⫺ 36i)/120
e ⫹ 2 Auditory 2 1 ⫺ i
3 (1 ⫺ 3i)(1 ⫺ 4i)
4 (1 ⫺ 6i)(1 ⫺ 8i)(1 ⫺ 11i) 2
5 (1 ⫺ 10i)(1 ⫺ 13i)(1 ⫺ 17i) 2 (1 ⫺ 18i)
6 (1 ⫺ 15i)(1 ⫺ 20i)(1 ⫺ 24i) 2 (1 ⫺ 26i) 2
7 (1 ⫺ 21i)(1 ⫺ 26i)(1 ⫺ 32i) 2 (1 ⫺ 35i) 2 (1 ⫺ 36i)
Visual 2 1 ⫺ 3i
3 (1 ⫺ 6i) 2 (1 ⫺ 7i)
4 (1 ⫺ 10i) 2 (1 ⫺ 12i) 2 /2
5 (1 ⫺ 15i) 2 (1 ⫺ 18i) 2 (1 ⫺ 19i)/6
6 (1 ⫺ 21i) 2 (1 ⫺ 25i) 2 (1 ⫺ 27i) 2 /24
7 (1 ⫺ 28i) 2 (1 ⫺ 33i) 2 (1 ⫺ 36i) 2 (1 ⫺ 37i)/120
e ⫹ 3 Auditory 2 1
3 1 ⫺ i
4 (1 ⫺ 6i)(1 ⫺ 8i)(1 ⫺ 9i)
5 (1 ⫺ 10i)(1 ⫺ 13i)(1 ⫺ 15i) 2
6 (1 ⫺ 15i)(1 ⫺ 19i)(1 ⫺ 22i) 2 (1 ⫺ 23i)
7 (1 ⫺ 21i)(1 ⫺ 26i)(1 ⫺ 30i) 2 (1 ⫺ 32i) 2
Visual 2 1 ⫺ 2i
3 (1 ⫺ 5i) 2
4 (1 ⫺ 9i)(1 ⫺ 10i)(1 ⫺ 12i) 2
5 (1 ⫺ 14i)(1 ⫺ 15i)(1 ⫺ 18i) 2 (1 ⫺ 19i)
6 (1 ⫺ 20i)(1 ⫺ 21i)(1 ⫺ 25i) 2 (1 ⫺ 27i) 2
7 (1 ⫺ 27i)(1 ⫺ 28i)(1 ⫺ 33i) 2 (1 ⫺ 36i) 2 (1 ⫺ 37i)
e ⫹ 4 Auditory 2 1
3 1
4 (1 ⫺ 5i)(1 ⫺ 6i)
5 (1 ⫺ 9i)(1 ⫺ 11i)(1 ⫺ 12i)
6 (1 ⫺ 14i)(1 ⫺ 17i)(1 ⫺ 19i) 2
7 (1 ⫺ 20i)(1 ⫺ 24i)(1 ⫺ 27i) 2 (1 ⫺ 28i)
Visual 2 1
3 1 ⫺ 3i
4 (1 ⫺ 9i) 2 (1 ⫺ 10i)
5 (1 ⫺ 14i) 2 (1 ⫺ 16i) 2
6 (1 ⫺ 20i) 2 (1 ⫺ 23i) 2 (1 ⫺ 24i)
7 (1 ⫺ 27i) 2 (1 ⫺ 31i) 2 (1 ⫺ 33i) 2
VERBAL STM MODEL 223

TABLE 12—Continued

Modality of List
M capacity presentation length Probability of recall

e ⫹ 5 Auditory 2 1
3 1
4 1 ⫺ 3i
5 (1 ⫺ 7i)(1 ⫺ 8i)
6 (1 ⫺ 12i)(1 ⫺ 14i)(1 ⫺ 15i)
7 (1 ⫺ 18i)(1 ⫺ 21i)(1 ⫺ 23i) 2
Visual 2 1
3 1
4 (1 ⫺ 7i) 2
5 (1 ⫺ 12i) 2 (1 ⫺ 13i)
6 (1 ⫺ 18i) 2 (1 ⫺ 20i) 2
7 (1 ⫺ 25i) 2 (1 ⫺ 28i) 2 (1 ⫺ 29i)
e ⫹ 6 Auditory 2 1
3 1
4 1
5 1 ⫺ 4i
6 (1 ⫺ 9i)(1 ⫺ 10i)
7 (1 ⫺ 15i)(1 ⫺ 17i)(1 ⫺ 18i) 2
Visual 2 1
3 1
4 (1 ⫺ 4i)
5 (1 ⫺ 9i) 2
6 (1 ⫺ 15i) 2 (1 ⫺ 16i)
7 (1 ⫺ 22i) 2 (1 ⫺ 24i) 2

However (and this is the main difference between Tables 10 and 11), it
cannot be taken for granted that the ordering of items is correct. Without
rehearsal to ensure serial-order encoding, if W2 and W3 are retrieved, then
the probability of their being reported in the correct order is 1/2. For this
reason, the probability of correctly recalling a visually presented four-word
list by a subject with an M capacity of e ⫹ 2 is given as (1 ⫺ 10i) 2 (1 ⫺
12i) 2 /2.
In a conventional memory-span procedure, the probability that a subject has a
span of at least n items is, of course, equal to the product of the probabilities of
correctly recalling the lists ranging from the shortest to the one including n items.
Table 12 shows the equations expressing the probability that a list of a given
length is correctly recalled. Of course, a subject can obtain a span score of, say,
at least four words only if the lists of two words and three words have already
been recalled. Thus, the probability of span ⱖ 4 is equal to the product of the
probabilities of recalling the lists of two, three, and four words. In turn, the
probability that a subject’s span score is exactly four words is equal to the
difference p(span ⱖ 4) ⫺ p(span ⱖ 5).
224 SERGIO MORRA

REFERENCES
Allport, D. A., Styles, E. A., & Hsieh, S. (1994). Shifting intentional set: Exploring the dynamic
control of tasks. In C. Umiltà & M. Moscovitch (Eds.), Attention and performance, XV (pp.
421– 452). Cambridge, MA: MIT Press.
Anderson, J. R., & Matessa, M. (1997). A production system theory of serial memory. Psychological
Review, 104, 728 –748.
Baddeley, A. D. (1986). Working memory. London: Oxford Univ. Press.
Baddeley, A. D. (1992). Is working memory working? The fifteenth Bartlett lecture. Quarterly
Journal of Experimental Psychology, 44A, 1–31.
Baddeley, A. D. (1996). Exploring the central executive. Quarterly Journal of Experimental Psy-
chology, 49A, 5–28.
Baddeley, A. D., Lewis, V. J., & Vallar, G. (1984). Exploring the articulatory loop. Quarterly Journal
of Experimental Psychology, 36, 233–252.
Baddeley, A. D., Thomson, N., & Buchanan, M. (1975). Word length and the structure of short term
memory. Journal of Verbal Learning and Verbal Behaviour, 14, 575–589.
Brooks, J. O., & Watkins, M. J. (1990). Further evidence of the intricacy of memory span. Journal
of Experimental Psychology: Learning, Memory and Cognition, 16, 1134 –1141.
Brown, G. D. A. (1997). Formal models of memory for serial order: A review. In M. A. Conway
(Ed.), Cognitive models of memory (pp. 47–77). Bristol: Psychology Press.
Brown, G. D. A., & Hulme, C. (1995). Modeling item length effects in memory span: No rehearsal
needed? Journal of Memory and Language, 34, 594 – 621.
Burgess, N., & Hitch, G. J. (1992). Toward a network model of the articulatory loop. Journal of
Memory and Language, 31, 429 – 460.
Burtis, P. J. (1982). Capacity increase and chunking in the development of short term memory.
Journal of Experimental Child Psychology, 34, 387– 413.
Caplan, D., Rochon, E., & Waters, G. S. (1992). Articulatory and phonological determinants of word
length in span tasks. Quarterly Journal of Experimental Psychology, 45A, 177–192.
Case, R. (1985). Intellectual development: Birth to adulthood. New York: Academic Press.
Cheung, H., & Kemper, S. (1993). Recall and articulation of English and Chinese words by
Chinese-English bilinguals. Memory and Cognition, 21, 666 – 670.
Conway, A. R. A., & Engle, R. W. (1994). Working memory and retrieval: A resource-dependent
inhibition model. Journal of Experimental Psychology: General, 123, 354 –373.
Cowan, N. (1988). Evolving conceptions of memory storage, selective attention, and their mutual
constraints within the human information-processing system. Psychological Bulletin, 104,
163–191.
Cowan, N. (1992). Verbal memory span and the timing of spoken recall. Journal of Memory and
Language, 31, 668 – 684.
Cowan, N. (1993). Activation, attention, and short-term memory. Memory and Cognition, 21,
162–167.
Cowan, N., Cartwright, C., Winterowd, C., & Sherk, M. (1987). An adult model of preschool
children’s speech memory. Memory and Cognition, 15, 511–517.
Cowan, N., Day, L., Saults, J. S., Keller, T. A., Johnson, T., & Flores, L. (1992). The role of verbal
output time in the effect of word length on immediate memory. Journal of Memory and
Language, 31, 1–17.
Cowan, N., Keller, T. A., Hulme, C., Roodenrys, S., McDougall, S., & Rack, J. (1994). Verbal
memory span in children: Speech timing clues to the mechanisms underlying age and word
length effects. Journal of Memory and Language, 33, 234 –250.
Cowan, N., Saults, J. S., Winterowd, C., & Sherk, M. (1991). Enhancement of 4-year-old children’s
memory span for phonologically similar and dissimilar word lists. Journal of Experimental
Child Psychology, 51, 30 –52.
De Mauro, T., Mancini, F., Vedovelli, M., & Voghera, M. (1993). Lessico di frequenza dell’italiano
parlato [Frequency lexicon of spoken Italian]. Milano: Etaslibri.
VERBAL STM MODEL 225

Demetriou, A. (1988). The neo-Piagetian theories of cognitive development: Towards an integration.


Amsterdam: North Holland.
Dempster, F. M. (1992). The rise and fall of the inhibitory mechanism: Toward a unified theory of
cognitive development and aging. Developmental Review, 12, 45–75.
Drachman, D. A., & Zaks, M. S. (1967). The “memory cliff” beyond span in immediate recall.
Psychological Reports, 21, 105–112.
Engle, R. W., Cantor, J., & Carullo, J. J. (1992). Individual differences in working memory and
comprehension: A test of four hypotheses. Journal of Experimental Psychology: Learning,
Memory and Cognition, 18, 972–992.
Gathercole, S. E. (1997). Models of verbal short-term memory. In M. A. Conway (Ed.), Cognitive
models of memory (pp. 13– 45). Bristol: Psychology Press.
Guttentag, R. E. (1984). The mental effort requirement of cumulative rehearsal: A developmental
study. Journal of Experimental Child Psychology, 37, 92–106.
Halford, G. S. (1993). Children’s understanding: The development of mental models. Hillsdale, NJ:
Erlbaum.
Henry, L. (1991a). Development of auditory memory span: The role of rehearsal. British Journal of
Developmental Psychology, 9, 493–511.
Henry, L. (1991b). The effects of word length and phonemic similarity in young children’s short-term
memory. Quarterly Journal of Experimental Psychology, 43A, 35–52.
Henry, L. (1994). The relationship between speech rate and memory span in children. International
Journal of Behavioural Development, 17, 37–56.
Henry, L. A., & Millar, S. (1991). Memory span increase with age: A test of two hypotheses. Journal
of Experimental Child Psychology, 51, 459 – 484.
Hitch, G. J., Halliday, M. S., Dodd, A., & Littler, J. E. (1989). Development of rehearsal in short-term
memory: Differences between pictorial and spoken stimuli. British Journal of Developmental
Psychology, 7, 347–362.
Hitch, G., Halliday, M. S., & Littler, J. E. (1989). Item identification time and rehearsal rate as
predictors of memory span in children. Quarterly Journal of Experimental Psychology, 41A,
321–337.
Hitch, G., Halliday, M. S., & Littler, J. E. (1993). Development of memory span for spoken words:
The role of rehearsal and item identification processes. British Journal of Developmental
Psychology, 11, 159 –169.
Hitch, G. J., Halliday, M. S., Schaafstal, A. M., & Heffernan, T. M. (1991). Speech, “inner speech,”
and the development of short-term memory: Effects of picture-labelling on recall. Journal of
Experimental Child Psychology, 51, 220 –234.
Hulme, C., Maugham, S., & Brown, G. D. (1991). Memory for familiar and unfamiliar words:
Evidence for a long-term memory contribution to short term memory span. Journal of Memory
and Language, 30, 685–701.
Hulme, C., Roodenrys, S., Schweickert, R., Brown, G. D. A., Martin, S., & Stuart, G. (1997).
Word-frequency effects on short-term memory tasks: Evidence for a redintegration process in
immediate serial recall. Journal of Experimental Psychology: Learning, Memory and Cognition,
23, 1217–1232.
Hulme, C., & Tordoff, V. (1989). Working memory development: The effects of speech rate, word
length and acoustic similarity on serial recall. Journal of Experimental Child Psychology, 47,
72– 87.
Just, M. A., & Carpenter, P. A. (1992). A capacity theory of comprehension: Individual differences
in working memory. Psychological Review, 99, 122–149.
Kail, R., & Park, Y. S. (1994). Processing time, articulation time, and memory span. Journal of
Experimental Child Psychology, 57, 281–291.
Kuhn, T. S. (1969). La struttura delle rivoluzioni scientifiche [Italian trans. of The structure of
scientific revolutions]. Torino: Einaudi. (Original work published 1962)
226 SERGIO MORRA

Lehto, J. (1996). Are executive function tests dependent on working memory capacity? Quarterly
Journal of Experimental Psychology, 49A, 29 –50.
Longoni, A. M., Richardson, J. T. E., & Aiello, A. (1993). Articulatory rehearsal and phonological
similarity in working memory. Memory and Cognition, 21, 11–22.
Lovatt, P. J., Avons, S. E., & Masterson, J. (2000). The word length effect and disyllabic words.
Quarterly Journal of Experimental Psychology, 53A.
Morra, S. (1990, September). Why, another model of working memory? Paper presented at the IV
Conference of the European Society for Cognitive Psychology, Como, Italy.
Morra, S. (1994). Issues in working memory measurement: Testing for M capacity. International
Journal of Behavioural Development, 17, 143–159.
Morra, S., Mazzoni, G., & Sava, D. (1993). Esiste un loop articolatorio di capacità limitata? [Is there
a time-limited articulatory loop?]. Proceedings of the SIPs Research Division Conference
(Rome, September), 94 –95.
Morra, S., & Scopesi, A. (1997). Issues in testing for processing capacity: Portrait of convergent
validity, with oblique rotations in the background. International Journal of Behavioural Devel-
opment, 20, 739 –742.
Moscovitch, M., & Umiltà, C. (1991). Modularity and neuropsychology: Implications for the
organization of attention and memory in normal and brain-damaged people. In M. Schwartz
(Ed.), Modular processes in dementia (pp. 1–59). Cambridge, MA: MIT/Bradford.
Nairne, J. S. (1990). A feature model of immediate memory. Memory and Cognition, 18,
251–269.
Nairne, J. S., Neath, I., & Serra, M. (1997). Proactive interference plays a role in the word-length
effect. Psychonomic Bulletin & Review, 4, 541–545.
Neath, I., & Nairne, J. S. (1995). Word-length effects in immediate memory: Overwriting trace decay
theory. Psychonomic Bulletin & Review, 2, 429 – 441.
Nicolson, R. (1981). The relationship between memory span and processing speed. In M. P. Friedman
& N. O’Connor (Eds.), Intelligence and learning (pp. 179 –183). New York: Plenum.
Nicolson, R., & Fawcett, A. (1991, July). Decomposing working memory: New evidence
from memory span. Paper presented at the International Conference on Memory, Lancaster,
UK.
Pascual-Leone, J. (1970). A mathematical model for the transition rule in Piaget’s developmental
stages. Acta Psychologica, 63, 301–345.
Pascual-Leone, J. (1987). Organismic processes for neo-Piagetian theories: A dialectical causal
account of cognitive development. International Journal of Psychology, 22, 531–570.
Pascual-Leone, J., & Baillargeon, R. (1994). Developmental measurement of mental attention.
International Journal of Behavioral Development, 17, 161–200.
Pascual-Leone, J., & Goodman, D. (1979). Intelligence and experience: A neo-Piagetian approach.
Instructional Science, 8, 301–367.
Penney, C. G. (1989). Modality effects and the structure of short-term verbal memory. Memory and
Cognition, 17, 398 – 422.
Poirier, M., & Saint-Aubin, J. (1995). Memory for related and unrelated words: Further evidence on
the influence of semantic factors in immediate serial recall. Quarterly Journal of Experimental
Psychology, 48A, 384 – 404.
Ribaupierre, A. de, & Bailleux, C. (1994). Developmental change in a spatial task of attentional
capacity: An essay toward an integration of two working memory models. International Journal
of Behavioral Development, 17, 5–35.
Schweickert, R., & Boruff, B. (1986). Short-term memory capacity: Magic number or magic spell?
Journal of Experimental Psychology: Learning, Memory and Cognition, 12, 419 – 425.
Service, E. (1998). The effect of word-length on immediate serial recall depends on phonological
complexity, not articulatory duration. Quarterly Journal of Experimental Psychology, 51A,
283–304.
VERBAL STM MODEL 227

Stigler, J. W., Lee, S. Y., & Stevenson, H. W. (1986). Digit memory in Chinese and English:
Evidence for a temporally limited store. Cognition, 23, 1–20.
Tehan, G., & Humphreys, M. S. (1988). Articulatory loop explanations of memory span and
pronunciation rate correspondences: A cautionary note. Bulletin of the Psychonomic Society, 26,
293–296.
Thurstone, T. G. (1941). Primary mental abilities of children. Educational and Psychological
Measurement, 1, 105–116.
Watkins, M. J., & Watkins, O. C. (1973). The postcategorical status of the modality effect in serial
recall. Journal of Experimental Psychology, 99, 226 –230.
Wechsler, D. (1974). Intelligence Scale for Children–Revised. New York: Psychological Corporation.
[Italian version: Orsini, A. (Ed.). (1987). WISC-R: Scala d’intelligenza Wechsler per bambini
riveduta. Firenze: Organizzazioni Speciali].

Received October 10, 1997; revised September 14, 1999

View publication stats

You might also like