You are on page 1of 16

Behavioural Processes 44 (1998) 147 – 162

Time as content in Pavlovian conditioning

Hernán I. Savastano, Ralph R. Miller *


Department of Psychology, SUNY-Binghamton, Binghamton, NY 13902 -6000, USA

Received 30 April 1998; received in revised form 7 August 1998; accepted 28 August 1998

Abstract

Time has played only a limited role within the traditional theories of Pavlovian conditioning. Although temporal
factors certainly contribute to whether conditioning occurs, the traditional assumption in the associative framework
has been that associations lack temporal information. Recently, the temporal coding hypothesis has challenged that
view, arguing that animals encode temporal relationships as part of associations. That is, proximal temporal
relationships not only foster associative learning, but also are part of the content of learning. The present paper
reviews for the nonspecialist the increasing empirical evidence that temporal coding is ubiquitous in Pavlovian
paradigms, including simultaneous and backward conditioning, second-order conditioning, sensory preconditioning,
cue competition, Hall–Pearce type CS-preexposure, and conditioned inhibition. The data support the temporal
coding hypothesis’ view that contiguity is sufficient for associative learning to occur, but challenge the central
assumption of the informational hypothesis that predictive relations are necessary for learning to occur (as opposed to
predictive relationships only being necessary for the expression of knowledge). © 1998 Elsevier Science B.V. All rights
reserved.

Keywords: Conditioned stimulus; Unconditioned stimulus; Time

1. Introduction depend critically on the temporal duration of


those events as well as on their temporal relation-
Time is inextricably tied to the study of Pavlo- ship (Pavlov, 1927). For example, some degree of
vian conditioning. In Pavlovian conditioning, or- proximity in time (and space, but this article is
ganisms are said to form associations between concerned exclusively with time) must clearly exist
stimulus events that occupy temporal space. Suc- between two stimulus events in order for those
cessful conditioning has long been recognized to events (or, more precisely, their representations)
to become associated. Yet, despite the consider-
able importance of temporal variables in Pavlo-
* Corresponding author. Tel: + 1 607 7772291/7987632;
fax: + 1 607 7774890/2607; e-mail: rmiller@binghamton.edu vian conditioning, time has played a surprisingly

0376-6357/98/$ - see front matter © 1998 Elsevier Science B.V. All rights reserved.
PII: S0376-6357(98)00046-1
148 H.I. Sa6astano, R.R. Miller / Beha6ioural Processes 44 (1998) 147–162

limited role within traditional theories of associa- The temporal coding hypothesis represents a
tive learning (e.g. Hull, 1943; Rescorla and Wag- radical departure from the traditional view about
ner, 1972; Mackintosh, 1975; Pearce and Hall, the role of temporal information in associative
1980; Pearce, 1987; but see Gibbon and Balsam, learning. According to the temporal coding view,
1981; Gallistel, 1990; Sutton and Barto, 1990). an association consists of more than a mental link
The prevailing view has been that animals do not between the representations of two paired events.
learn about the temporal properties of the stimu- Rather, the temporal relationship between the
lus events in Pavlovian conditioning experiments. events that prevailed during training is encoded as
Recently, however, the temporal coding hypothesis part of the association. Thus, the temporal condi-
has questioned that assumption (Matzel et al., tions (e.g. the CS–US interstimulus interval) are
1988; Barnet et al., 1991; Miller and Barnet, not merely catalysts in the formation of associa-
1993). The present paper provides an overview of tions, but are also a part of the content of learn-
the accumulating evidence that animals do indeed ing. In Pavlovian conditioning, an animal learns
acquire (and use) temporal information in a wide not only that a US will occur, but also when the
variety of Pavlovian paradigms. We discuss the US will occur. According to the temporal coding
theoretical implications of temporal coding, while hypothesis, good contiguity (i.e. temporal proxim-
emphasizing the challenges this evidence poses for ity) is sufficient for the formation of an associa-
the prevailing view of associative learning. tion, where good contiguity is defined as similar
The prevailing framework views an association or identical onsets of the two events. Whether an
simply as a mental link between the representa- association will be expressed as observable behav-
tions of the associated events. Consider a typical ior, however, depends (in part) on the encoded
Pavlovian task: an associatively neutral condi- temporal knowledge. The temporal information
tioned stimulus (CS), such as a tone, is forward- that is acquired during training is presumed to
paired in time with (i.e. occurs immediately prior play a critical role in determining the nature,
to) an unconditioned stimulus (US), such as food, magnitude, and timing of the observed condi-
footshock, or some other event of inherent biolog- tioned response. Therefore, unlike most theories
ical significance. Following a number of these of associative learning (Rescorla and Wagner,
CS – US pairings, the presentation of the CS is 1972; Mackintosh, 1975; Pearce and Hall, 1980)
said to activate in memory a representation of which have tended to treat performance as direct
that CS, which in turn activates a representation evidence of learning, the temporal coding hypoth-
of the US. The CS comes to elicit a conditioned esis maintains a clear distinction between learning
response (e.g. salivation in anticipation of food) and the behavioral expression of that learning.
through a mediating association with the US. That is, certain temporal relationships mitigate
Traditionally, the magnitude of the conditioned against the behavioral expression of an acquired
response has been interpreted as direct evidence of CS–US association. As we shall see, the distinc-
the strength of the CS – US association. Although tion between learning and performance is justified
temporal factors, such as the proximity of the CS by data obtained with recently developed tech-
and US during training, are presumed to con- niques that have been effective in revealing associ-
tribute to the strength of the association, the ations that would otherwise remain behaviorally
assumption has been that the temporal informa- silent.
tion itself does not become a part of that associa-
tion. Thus, temporal factors serve only a
facilitative role in the formation of associations. 2. The mechanism of learning: contiguity versus
The closer the two events are in time during information
training, the more robust the resulting association
is presumed to be. However, an organism acquires The assumption of the temporal coding hypoth-
no representational knowledge about the tempo- esis that temporal contiguity is sufficient for asso-
ral relationship of the paired stimulus events. ciative learning marks a return to the early views
H.I. Sa6astano, R.R. Miller / Beha6ioural Processes 44 (1998) 147–162 149

of Pavlov (1927) and the British empiricists before association between their representations. Instead,
him (e.g. Locke, 1974), who believed that the learning is presumed to occur only when one
occurrence of two events in close temporal prox- event (nonredundantly) predicts the occurrence of
imity was the critical determinant of learning. another. Thus, a CS must provide anticipatory
Since Pavlov’s time, however, the principle of information about the US for associative learning
contiguity has encountered numerous empirical to occur. A natural account of the superiority of
challenges and, until recently, was all but aban- forward conditioning over backward and simulta-
doned. Perhaps the greatest challenge was the neous conditioning falls out of information the-
relative ineffectiveness of simultaneous (the CS ory. Stronger responding is observed in forward
and US have identical onset and termination conditioning because only in that arrangement
times) and backward (the CS onsetting at the does the CS predict the occurrence of the US.
termination of the US) conditioning procedures in Conversely, simultaneous and backward paired
producing observable conditioned responding. CSs have little or no predictive value with respect
Pavlov (1927) himself discovered that conditioned to the US, and therefore do not become associ-
responding is typically stronger after forward pair- ated with the US. By (implicitly) equating ob-
ings (in which the CS precedes the US) than after served responding with associative strength, the
training in which the CS and US are paired in a informational hypothesis attributes the relative
simultaneous (i.e. CS:US) or backwards (i.e. ineffectiveness of simultaneous and backward
US “CS) manner. On the basis of contiguity preparations to the absence of learning.
alone, one would expect simultaneous condition- The informational hypothesis garnered further
ing to produce the strongest conditioned respond- support from a classic study by Rescorla (1968).
ing because simultaneous conditioning is an In a study of conditioned suppression in rats,
instance of perfect contiguity, whereas forward Rescorla showed that an equivalent number of
and backward conditioning are instances of im- tone“ shock pairings might or might not result in
perfect contiguity. Similarly, forward and back- the tone eliciting a strong conditioned response,
ward serial pairings (with equivalent durations for depending on the number of additional shocks
the CS and US) should yield equivalent learning delivered in the absence of the tone during the
because the two temporal orderings provide conditioning session. This finding appeared trou-
equivalent contiguity. Jointly, the response deficits blesome for the contiguity view. Because the de-
observed with backward and simultaneous condi- gree of CS–US contiguity remained unchanged
tioning appeared to contradict the view that bet- by the presentation of unsignaled USs, condi-
ter temporal contiguity resulted in stronger tioned responding to the CS should not have been
associative learning. Because these deficits were affected by the presentation of additional USs.
interpreted as evidence of weak learning, grave Instead, conditioned responding was directly infl-
doubt was cast upon the view that simple contigu- uenced by the contingency between the CS and
ity was the mechanism responsible for associative US. Rescorla defined contingency as the probabil-
learning (for a review, see Moore and Gorme- ity of the US in the presence of the CS (p[US CS])
zano, 1977). As a result, various alternative mech- minus the probability of the US in the absence of
anisms were proposed to explain the superiority the CS (p[US no CS]), and reasonably argued that
of forward conditioning relative to simultaneous it served as a direct measure of the predictiveness
and backward conditioning. of the CS. (Notably, this definition of contingency
The informational hypothesis (Egger and Miller, has been adopted by several theories of human
1963; Kamin, 1969; Rescorla, 1972; Rescorla and causal judgment as a normative index of the
Wagner, 1972; Cantor, 1981) emerged as perhaps statistical correlation between two events e.g. Al-
the most-widely accepted alternative to the princi- lan, 1993; Cheng, 1997). As is discussed below,
ple of contiguity. According to the informational the notion of the informational hypothesis that a
hypothesis, the mere contiguous occurrence of nonredundant predictive relationship is critical for
two events is not sufficient for the formation of an engaging the learning mechanism was also able to
150 H.I. Sa6astano, R.R. Miller / Beha6ioural Processes 44 (1998) 147–162

account for the so-called ‘cue-competition’ phe- autoshaping (Hearst, 1989). In Hearst’s study, a
nomena, such as o6ershadowing and blocking, stimulus that had only followed brief periods of
wherein stimuli that are trained together, accord- available food proved much easier to convert into
ing to the informational view, compete for asso- an effective signal for food than did a stimulus
ciative linkage to a common US. In contrast, that had been uncorrelated with food. Therefore,
simple contiguity theory cannot explain the re- although responding to a forward conditioned CS
sponse deficits that result from cue competition is generally stronger than to a backward or simul-
because the competing stimuli are contiguous with taneous conditioned CS, the foregoing demonstra-
the US and consequently should become associ- tions of successful backward and simultaneous
ated with it. Cue competition phenomena served conditioning are problematic for the view that
as the basis for many of the current models of predictive value is essential for learning to occur.
Pavlovian conditioning (Rescorla and Wagner, The informational hypothesis has appealed to
1972; Mackintosh, 1975; Pearce and Hall, 1980), researchers in large part because it conforms
all of which hold (nonredundant) prediction to be nicely to an evolutionary framework that empha-
a requirement for learning to occur. sizes the functional nature of learning. Predictive
The informational hypothesis, however, en- learning is seemingly functional to the extent that
counters problems when applied to backward and it enables an organism to prepare for (and per-
simultaneous conditioning. The theory maintains haps modify) the occurrence of biologically sig-
that only stimuli that have predictive value should nificant events. As argued by Mackintosh (1983),
acquire excitatory status. However, several results conditioning, from a functional standpoint,
in the human and animal learning literatures should provide the subject with information about
provide evidence that a forward temporal rela- the impending reinforcer. However, one problem
tionship is not necessary for the establishment of with an evolutionary approach is that ultimate
an association between stimuli. For instance, hu- functions of behavior are sometimes confused
man learning theorists are less concerned with the with proximate mechanisms. Whereas a likely ul-
temporal relationship between stimuli because timate function of learning is to allow organisms
strong associations are seemingly formed between to anticipate significant events, the proximate
stimuli presented simultaneously in preparations mechanism responsible for the ability to predict
such as the paired-associates task in which a list significant events might well be a simpler mecha-
of word pairs is presented to subjects (e.g. Gib- nism that is sensitive to temporal contiguity rather
son, 1940). Within the animal literature, Rescorla than predictive value. But, if temporal contiguity
(1980) controlled for certain problems that con- is to be considered a viable mechanism of associa-
founded earlier studies of simultaneous condition- tive learning, we need an explanation of why
ing (i.e. distraction from one stimulus by another behavioral control resulting from simultaneous
of higher salience), and found that the association and backward pairings is generally weaker than
between simultaneously presented neutral stimuli that resulting from forward pairings, despite for-
is stronger than the association formed between ward associations being weaker than simultaneous
an antecedent cue and an immediately following and equal to backward associations in terms of
stimulus. Consistent with this, Burkhardt and contiguity.
Ayres (1978), using a conditioned suppression
procedure with rats, reported that, at least early in
training (i.e. after few training trials), a CS that is 3. The temporal coding hypothesis
paired simultaneously with a shock US will sup-
port reliable excitatory responding (see also Heth, Most Pavlovian tasks are designed to measure
1976; Heth and Rescorla, 1973). Moreover, evi- responses that are anticipatory in nature (e.g.
dence of robust backward associations has been salivating in anticipation of food, blinking the
reported both in conditioned suppression prepara- eyes in anticipation of an air puff, freezing in
tions with rats (Ayres et al., 1987) and pigeon anticipation of footshock). Thus, one alternative
H.I. Sa6astano, R.R. Miller / Beha6ioural Processes 44 (1998) 147–162 151

to the informational account is the possibility that US training, is attributed to different temporal
simultaneous and backward pairings result in well information being encoded during simultaneous
learned associations, but that the CS does not or backward pairings, rather than to a failure in
generally evoke a conditioned response with con- associative acquisition per se.
ventional measures of Pavlovian association be-
cause it does not bear a forward (predictive)
relationship to the US. Conventional measures of 4. Assessing simultaneous and backward
Pavlovian associative strength may be appropriate associations
for the expression of forward associations, but
they would likely be inadequate for the expression If simultaneous and backward associations are
of backward or simultaneous associations. Thus, latent, the question arises about how their exis-
the weak conditioned response observed in those tence can be assessed. It is unclear whether a
situations might not reflect a failure to engage the direct measure is possible because it is not certain
learning system during training, but rather an what response systems are specifically appropriate
inability to engage the appropriate response sys- to the expression of simultaneous and backward
tem. associations. However, Farley and Alkon (1987)
This is essentially the position taken by the obtained some intriguing data with the mollusk
temporal coding hypothesis (Matzel et al., 1988; Hermissenda. They paired a light CS which nor-
Barnet et al., 1991; Miller and Barnet, 1993), mally elicits a positive phototactic response in this
which can be summarized as follows; (1) Contigu- species with high-speed rotation of the subject, an
ity alone is necessary and sufficient for the forma- aversive US that normally suppresses phototaxis.
tion of an association. (2) The temporal Depolarization of the Hermissenda’s photorecep-
relationship between the associated events is auto- tor to the light served as the index of associative
matically encoded as part of the association (i.e. strength. They found stronger conditioning (i.e.
subjects form temporal maps that link events in greater suppression [decreased photoreceptor de-
memory). (3) This temporal information plays a polarization] to the light) with simultaneous
critical role in the nature, magnitude, and timing CS:US pairings than with forward CS“ US pair-
of the conditioned response elicited when one of ings. These studies are interesting because, rather
the associates is subsequently presented. Finally, than monitoring a response system that depends
(4) animals can superimpose temporal maps when on the anticipation of the US for its expression,
elements common to these maps are presented Farley and Alkon indexed associative strength
together, even when the elements were trained more directly, that is, at the neural level prior to
separately. That is, temporal information from the translation of that information into higher-or-
different training situations can be integrated. der response systems. Similar results have been
According to this view, simultaneous and back- reported in other studies in which associative
ward associations, as well as forward associations, learning was also monitored at the neural level
can be learned as long as good CS – US contiguity (Kandel and Schwartz, 1982).
exists. However, the expression of that learning In more conventional Pavlovian preparations,
requires a predictive relationship between the CS the existence of simultaneous and backward asso-
and the US, at least when the typical anticipatory ciations can be assessed indirectly, through the
response systems are measured. Thus, information use of a second-order conditioning procedure. In a
about an upcoming US supports anticipatory re- second-order conditioning procedure, a first-order
sponding to the CS, which ordinarily prepares the stimulus (CS1) is forward-paired with the US in
subject for the impending US. However, a predic- Phase 1 (CS1“ US), and subsequently a second-
tive relationship is not essential for associative order stimulus is forward-paired with CS1 (in the
acquisition. Accordingly, the commonly observed absence of the US) during Phase 2 (CS2“ CS1).
response deficit following simultaneous or back- During testing, a strong conditioned response is
ward CS–US pairings, relative to forward CS – observed to CS2 (relative to control groups lack-
152 H.I. Sa6astano, R.R. Miller / Beha6ioural Processes 44 (1998) 147–162

ing either the Phase 1 or Phase 2 experience), even training. For the purpose of assessing associations
though CS2 was never itself directly paired with that are latent because of their putative temporal
the US. For the purpose of assessing the existence content, sensory preconditioning provides another
of simultaneous or backward associations, sec- means of creating a predictive association that
ond-order conditioning is useful because it allows provides a measure of a simultaneous or back-
for testing a simultaneous or backward CS1 – US ward association. For example, Matzel et al.
association through observation of responding to (1988) initially exposed rats to forward CS2“
CS2, which has a predictive relationship with CS1 pairings followed by either forward, simulta-
CS1, thereby allowing the expression of the CS1 – neous, or backward pairings of CS1 with the US.
US association. For example, in a study of condi- Once again, the expected simultaneous and back-
tioned lick-suppression in water-deprived rats, ward response deficits were observed to CS1.
Barnet et al. (1991) trained the first-order cue However, CS2, which had been established as a
(CS1, an auditory cue) simultaneously with the predictor of CS1 (and hence presumably a predic-
US (CS1:US, the US being a mild footshock), and tor of the US), evoked a strong conditioned re-
subsequently paired a second-order cue (CS2, a sponse regardless of the manner in which CS1 had
different [counterbalanced] auditory cue) in a for- been paired with the US.
ward (predictive) manner with CS1 (CS2 “ CS1). Jointly, the studies of Barnet et al. (1991) and
Even though CS1 elicited little responding (i.e. Matzel et al. (1988) provide clear evidence that
low conditioned suppression of ongoing drink- animals acquire strong simultaneous and back-
ing), which was consistent with the expected ward associations in conventional Pavlovian
simultaneous conditioning deficit, a strong condi- preparations. However, that information was ex-
tioned response was observed to CS2. That is, a pressed when and only when the acquired infor-
strong simultaneous (CS1 – US) association was mation was of functional, predictive value
revealed by creating a predictive relationship be- appropriate for the anticipatory response system
tween CS2 and CS1, and hence indirectly between that was monitored. Thus, these findings are con-
CS2 and the US. Presumably, the CS2-US tempo- sistent with the view of the temporal coding hy-
ral relationship was the consequence of integra- pothesis that simple contiguity is sufficient for
tion of the CS1–US and CS2 – CS1 temporal associative acquisition, but that the expression of
relationships because CS2 was never directly learning ordinarily requires a predictive relation-
paired with the US. (Note that this view can ship when assessed by an anticipatory measure. In
account for the previously described data of Res- these studies, higher-order stimuli were estab-
corla (1980) without recourse to the notion of lished as predictors of first-order stimuli that had
distraction. That is, Rescorla may have obtained backward or simultaneous associations with the
strong responding to his test stimulus, not because US. The implication is that the temporal relation-
there was no simultaneous US, but because his ship between a paired CS and US was encoded
test stimulus had an anticipatory relationship to and integrated with the memory for the higher-or-
the first-order simultaneous CS.) der stimulus (i.e. CS2). In this manner, the tempo-
Another procedure that is useful for detecting ral coding hypothesis not only explains the
simultaneous and backwards associations is sen- behavioral silence of first-order backward and
sory preconditioning (Brogden, 1939). This proce- simultaneous associations, but also offers a plau-
dure is similar to second-order conditioning, sible account of second-order conditioning and
except that the order of the training phases is sensory preconditioning.
reversed (i.e. CS2–CS1 pairings precede the first- Furthermore, the strong simultaneous and
order conditioning CS1 – US trials). The phe- backward associations revealed in those studies
nomenon of sensory preconditioning is are consistent with previous reports of simulta-
theoretically intriguing because it demonstrates neous and backward associations (e.g. Hearst,
that organisms can associate and time events that 1989). Collectively, these findings attest to the
are not biologically significant at the time of sufficiency of contiguity as the basic requirement
H.I. Sa6astano, R.R. Miller / Beha6ioural Processes 44 (1998) 147–162 153

of learning, and contradict the view that predic- forward (CS1“ US) first-order pairings. The
tion is necessary to support an association be- magnitude of responding to the higher-order con-
tween stimulus events. Moreover, they are ditioned stimulus was determined by a memory
incompatible with the conclusion that the rela- representation, constructed from the two phases
tively weak responding generally observed after of training, that provided information about the
simultaneous and backward conditioning reflects expected temporal location of the US.
a learning deficit. Contrary to the informational Cole et al. (1995) reported another intriguing
accounts and ‘prevailing wisdom’, the magnitude demonstration of temporal integration in trace
of the conditioned response in those cases is not conditioning. In trace conditioning, a temporal
necessarily a reliable indicator of underlying asso- gap is inserted between the offset of the CS and
ciative strength. Thus, the learning-performance the onset of the US. This typically results in
distinction proposed by the temporal coding hy- weaker conditioned responding than does a stan-
pothesis appears to be justified. dard forward control (with no gap; i.e. delay
conditioning). Cole et al. gave first-order CS1–US

5. Temporal integration

The results of the higher-order conditioning


studies just described imply that animals integrate
temporal relationships from different associations
that share a common stimulus. A rather com-
pelling demonstration of temporal integration,
congruent with the temporal coding hypothesis,
was reported by Barnet et al. (1997). With both
second-order conditioning (Experiment 1) and
sensory preconditioning (Experiment 2) proce-
dures, they exposed rats to either backward
(US “ CS1) or forward (CS1“ US) first-order
training. CS2 was always conditioned in a for-
ward manner with CS1 (CS2“CS1). Not surpris-
ingly, responding to CS1 was superior in the
forward case than the backward case. However,
the magnitude of conditioned responding to CS2
was actually greater with the backward first-order
pairings than with the forward first-order pair-
ings. This superiority of backward pairings is
entirely consistent with the idea that temporal
relationships from different phases of training are
encoded as memory representations and inte-
grated into a single representation that determines Fig. 1. Presumed event representations and temporal relation-
ships of a Forward (A) or Backward (B) first-order association
the nature and strength of the conditioned re- (Phase 1) integrated (superimposed) with those of a forward
sponse. Fig. 1 illustrates the presumed integration second-order (CS2 “CS1) association (Phase 2). Arrow length
of first-order and second-order associations based represents the temporal expectancy for the onset of the US
on superpositioning of their common element signaled by the onset of CS2. Font size under Test CS2
(CS1). As denoted by the length of the arrows, the indicates the magnitude of the predicted and observed condi-
tioned response. Barnet et al. (1997) assessed conditioning by
expected temporal location of the US following the ability of CS2 to suppress ongoing drinking in water-de-
the onset of CS2 is more proximate with back- prived rats. Mean times to reinitiate drinking are presented in
ward (US“ CS1) first-order pairings than with parentheses.
154 H.I. Sa6astano, R.R. Miller / Beha6ioural Processes 44 (1998) 147–162

element from the two phases of training (CS1).


The onset of CS2 signals a predictive relationship
to the US only in the trace condition, not in the
delay condition. Thus, the greater second-order
responding observed in the trace condition can be
accounted for by assuming that responding to
CS2 depended upon the integration of the memo-
ries (i.e. temporal maps) for the two phases of
training and the resultant expected temporal loca-
tion of the US with respect to CS2.

6. Temporal coding in cue competition

In addition to second-order conditioning and


sensory preconditioning, there is emerging evi-
dence that temporal attributes are also encoded in
other Pavlovian preparations. Particularly, a
number of recent studies have examined the role
of temporal encoding in the cue competition
paradigms. ‘Cue competition’ is said to occur
when two (or more) conditioned stimuli are
trained in compound with a common US and are
seen to compete with each other for behavioral
Fig. 2. Presumed event representations and temporal relation- control based on the occurrence of the US. Spe-
ships of a delay (A) or trace (B) first-order association (Phase cifically, an impaired conditioned response is seen
1) integrated (superimposed) with those of a backward second- to a CS that is trained in the presence of another
order (CS1 “ CS2) association (Phase 2). Arrow length repre-
stimulus that is more salient (i.e. overshadowing)
sents the temporal expectancy for the onset of the US signaled
by the onset of CS2 in the Trace condition; there is no arrow or a better predictor of the US (i.e. blocking).
for the Delay condition because CS2 and the US have simulta- (Notably, the effect of degrading the CS-US con-
neous onsets in the integrated representation. Font size under tingency by adding unsignaled US’s can be viewed
Test CS2 indicates the magnitude of the predicted and ob- as an instance of cue competition, wherein contex-
served conditioned response. Cole et al. (1995) assessed condi-
tioning by the ability of CS2 to suppress ongoing drinking in
tual cues provided by the experimental chamber
water-deprived rats. Mean times to reinitiate drinking are acquire predictive value and thus come to block
presented in parentheses. responding to the CS.) As previously stated, the
cue competition phenomena have played an im-
portant role in the development of the informa-
pairings with either a 5-s gap or no gap, followed
tion-based theories of associative learning, such as
in Phase 2 by backward second-order condition- the influential Rescorla and Wagner (1972) model
ing (CS1“ CS2). They observed, relative to delay and many of its modern competitors (Mackin-
conditioning of CS1, the expected trace deficit in tosh, 1975; Pearce and Hall, 1980). According to
conditioned responding to CS1. Moreover, they all of these theories, the stimulus that has greater
confirmed the counterintuitive prediction of the predictive value will acquire a larger portion of
temporal coding hypothesis that greater second- the available associative strength at the expense of
order conditioning should occur following trace the less predictive (or salient) stimulus. Thus,
first-order pairings than following delay first-or- these theories explain the response deficits ob-
der pairings. Fig. 2 illustrates this prediction by served following cue competition treatments as
superimposing the representations of the common failures of learning.
H.I. Sa6astano, R.R. Miller / Beha6ioural Processes 44 (1998) 147–162 155

In the typical blocking experiment, for example, overshadowing. Overshadowing refers to the
a stimulus (A) is initially paired with the US deficit in conditioned responding observed to a
(A “US). In a second phase of training, Stimulus CS that was trained in the presence of a more
A is presented together with another stimulus (X) salient CS (i.e. of higher intensity). Blaisdell et al.
and paired with the US (AX “ US). During sub- (1998a), in a study of conditioned lick-suppression
sequent testing, responding to X alone is found to with rats, used a preparation in which a bright
be impaired relative to responding to X in sub- flashing light would overshadow a soft tone. In a
jects that received the AX“ US trials without the series of experiments, they systematically varied
prior A “US trials. Information theories at- the overshadowed CS–US temporal relationship
tribute this effect to the previously accrued predic- and the overshadowing CS–US temporal rela-
tive value of Stimulus A, which is said to have tionship. They found support for the prediction of
‘blocked’ acquisition of an X – US association. In the temporal coding hypothesis that overshadow-
a series of experiments, Barnet et al. (1993) (see ing should be maximal when the overshadowing
also Schreurs and Westbrook, 1982) examined the and overshadowed CSs share the same temporal
role of temporal coding in blocking. During Phase relationship with the US. That is, overshadowing
1, the blocking Stimulus A was either simulta- of a CS with a forward, simultaneous, or back-
neously or forward paired with the US. In Phase ward relationship with the US was greatest when
2, Stimulus A was presented in simultaneous com- the overshadowing CS also had a forward, simul-
pound with the target Stimulus X that was paired taneous, or backward relationship with the US,
in a simultaneous or forward manner with the respectively. (Again, all assessment was through
US. Robust blocking of a forward X “ US associ- examination of a second-order CS which revealed
ation was evident if the blocking Stimulus A had otherwise silent simultaneous and backward asso-
been pretrained in a forward manner. Similarly ciations.) In a more recent series of experiments,
robust blocking of a simultaneous X-US associa- Blaisdell et al. (1998b) demonstrated that the
tion was evident if A had been trained simulta- overshadowing deficit could be alleviated by
neously in Phase 1. However, a forward A“ US changing the temporal relationship between the
association produced little blocking of a simulta- overshadowing stimulus and the US after over-
neous X–US association, and a simultaneous A – shadowing treatment. That is, overshadowing of
US association produced little blocking of a X was attenuated if the temporal relationship
forward X“ US association. (Second-order con- between the overshadowing CS (A) and US was
ditioning was used to assess all associations be- altered following AX“ US training, such that the
cause first-order simultaneous associations are A–US temporal relationship was no longer the
usually behaviorally silent.) These results suggest same as the X–US temporal relationship.
that information concerning the temporal location To the extent that temporal information is au-
of the US provided by the pretrained stimulus tomatically encoded as part of any association, as
determined the amount of blocking observed. is asserted by the temporal coding hypothesis,
Blocking was greatest when the blocking and conditioned responding to a stimulus with a par-
blocked CSs shared the same temporal relation- ticular temporal relationship with a US should be
ship with the US. Although blocking clearly does better blocked or overshadowed by a stimulus
occur with nonsimultaneous CSs (Kehoe et al., that has the same temporal relationship with the
1981), Barnet et al.’s data suggest that blocking US than by a stimulus that has a different tempo-
will be greater with simultaneous CSs than non- ral relationship with the US. The studies reviewed
simultaneous CSs. Such a dependence on CS – US here varied the temporal relationships between the
temporal relationships provides strong evidence competing events and the US in cue competition
that such temporal information is encoded as part paradigms and found cue competition to be sensi-
of CS–US associations. tive to the similarity of these temporal relation-
Temporal encoding has also been shown to ships. These observations cannot be explained in
play an important role in the phenomenon of terms of the CS–US intervals merely influencing
156 H.I. Sa6astano, R.R. Miller / Beha6ioural Processes 44 (1998) 147–162

the associative strengths of the competing cues vian situations). This rule for the expression of
because in some situations longer intervals be- associations with respect to cue competition is
tween CS onset and US onset produced more basic to the comparator hypothesis (Miller and
competition than did shorter intervals. Hence, Schachtman, 1985; Miller and Matzel, 1988),
these findings further document the encoding of which is a response rule that complements the
temporal relationships as part of the content of temporal coding hypothesis. The temporal coding
simple associations. hypothesis provides both a simple rule for acquisi-
Moreover, these studies (e.g. Barnet et al., 1993; tion (i.e. contiguity) and is one of the several rules
Blaisdell et al., 1998a) demonstrate that cue com- that jointly apply to associative expression (i.e. an
petition can occur between cues that are paired in anticipatory relationship to the US). Thus, the
a non-predictive fashion as effectively as stimuli temporal coding hypothesis is able to explain why
that are paired with a US in an anticipatory responding to a second-order CS is effective in
(forward) manner. Consequently, these findings assessing simultaneous and backward first-order
also challenge the central premise of the informa- associations, whereas responding to the first-order
tional hypothesis that a cue must have predictive CS is not, as well as why CSs with different
value in order for it to enter into an association temporal relationships to a common US compete
with a US. Other research has confirmed that less than do CSs with a common temporal rela-
blocked or overshadowed stimuli enter into asso- tionship to the US. However, the comparator
ciations, but such associations are ordinarily la- hypothesis (or some other rule for the expression
tent unless special steps are taken to reveal them of associations) is required to explain cue compe-
[e.g. recovery from overshadowing through re- tition [including Rescorla (1968) degraded contin-
minder treatments (Kasprow et al., 1982), post- gency effect] in the first place. Therefore, the
training extinction of the overshadowing stimulus temporal coding hypothesis and the comparator
(e.g. Kaufman and Bolles, 1982; Matzel et al., hypothesis complement each other in providing a
1985), and spontaneous recovery (Kraemer et al., complete explanation of Pavlovian conditioning.
1988), and recovery from blocking through post-
training extinction of the blocking stimulus (Blais-
dell et al., 1998c)]. Therefore, associative 7. Temporal coding in the Hall–Pearce
acquisition does not seem to require a nonredun- CS-preexposure effect
dant predictive relationship between the events to
occur. A well-known finding in the Pavlovian litera-
Instead, the data are consistent with contiguity ture is the CS-preexposure effect, also known as
as the guiding principle for the acquisition of ‘latent inhibition’ (Lubow and Moore, 1959). If a
associations, but with an anticipatory relationship CS is presented alone for a number of trials prior
being required as a rule for the expression of to the start of conditioning (i.e. pairing of the CS
those associations when anticipatory measures are with a US), the emergence of conditioned re-
used. Associations (which we assert always in- sponding is retarded (i.e. requires more CS–US
clude temporal information) will be formed as pairings) relative to a control group lacking the
long as there is sufficient contiguity between the CS-preexposure trials. Hall and Pearce (1979) re-
events to be associated, regardless of whether the ported a phenomenon similar to the CS-preexpo-
stimulus in question precedes the onset of the US. sure effect. In their study, the preexposure phase
However, whether the association will be ex- consisted of pairing the CS with a low intensity
pressed in the animal’s behavior appears to de- version of the US in Phase 1 (CS “ USweak) fol-
pend largely on the saliency, prior conditioning lowed by pairings with a higher-intensity version
history, and temporal relationship with the US of of the US in Phase 2 (CS“ USstrong). This resulted
other cues that were present during training with in weaker responding to the CS than was ob-
the target CS (as well as the test stimulus having served in subjects lacking the Phase 1 experience
a predictive relationship to the US in most Pavlo- with the target CS. In a recent study, Savastano et
H.I. Sa6astano, R.R. Miller / Beha6ioural Processes 44 (1998) 147–162 157

al. (1998) examined the role of temporal encoding Pavlovian conditioned inhibition training (A-US/
in this preexposure effect. They either varied or AX-) in which (1) the intended inhibitor (X) and
kept constant the temporal gap between CS termi- the training excitor (A) were paired either in a
nation and US onset across the two training simultaneous (AX-) or forward manner (X“ A-),
phases. They found that varying the gap, regard- and (2) the training excitor was either simulta-
less of whether it was lengthened or shortened neously (A:US) or forward-paired (A“US) with
between Phase 1 and Phase 2, resulted in less of the US. The transfer excitor (B) was always for-
an impairment of conditioned responding than ward-paired with the US (B“ US). Inhibitory
did equivalent gaps in the two training phases, conditioning was found to be maximal when the
thereby suggesting that the temporal relationship time at which the US was expected to be omitted
learned in Phase 1 had been encoded. Thus, these (the temporal information activated by the in-
observations further support the view that the hibitor based both on its relationship to the train-
temporal relationship between a CS and a US is ing excitor and the training excitor’s relationship
automatically encoded as part of the content of a to the US) was the same as the time at which the
Pavlovian association. US was expected to be presented (the information
activated by the transfer excitor based on its
temporal relationship to the US). Specifically,
8. Temporal coding in conditioned inhibition conditioned inhibition was greatest in the condi-
tions in which a simultaneous inhibitor was
Thus far, our discussion has been limited to the trained with a forward-paired excitor or a for-
role of temporal information in excitatory condi- ward-paired inhibitor was trained with a simulta-
tioning, that is, how subjects use temporal infor- neously-paired excitor. These finding are
mation in anticipating when a US is going to consistent with the idea that a conditioned in-
occur. However, recent evidence suggests that hibitor provides information about the temporal
temporal encoding also plays a role in inhibitory location of an omitted US (see also Denniston et
conditioning. A conditioned inhibitor is a stimu- al. 1998a; Denniston et al., 1998b; for additional
lus that signals the omission of a US that in the demonstrations of temporal coding in inhibitory
absence of the signal would be expected. In a conditioning).
typical Pavlovian inhibition procedure, one stimu-
lus (the training excitor) is paired with the US
when it is presented alone (A“US), but not when 9. Time as associative content
it is presented in compound with the intended
inhibitor (AX-). Typically, A“US and AX- trials Our review of the role of time in Pavlovian
are interspersed within a session. The inhibitory conditioning has revealed that, across a wide ar-
power of the putative inhibitor (X) is then com- ray of Pavlovian paradigms, the temporal rela-
monly assessed through a summation test, which tionships between associated stimulus events are
entails presenting X in the presence of an indepen- encoded by the animal as an integral part of the
dently trained excitatory stimulus (B, which is interstimulus association. This view differs from
called a transfer excitor). Conditioned inhibition the traditional assumption that associations can
is evidenced by a reduced conditioned response to vary only along the dimension of strength. The
the BX compound relative to the transfer excitor solution offered by the temporal coding hypothe-
(B) alone. The question here is whether an in- sis is that associations vary along at least two
hibitory stimulus provides information about the dimensions—associative strength and the tempo-
temporal location of the omitted US in the same ral relationship between the associated stimulus
way that an excitatory CS provides temporal in- events. In other words, traditional associative
formation about the presentation of the US. In a strength and temporal relationships can be viewed
recent study of conditioned-lick suppression with as two dimensions of the bond between event
rats, Barnet and Miller (1996) gave subjects representations. In this sense, the temporal coding
158 H.I. Sa6astano, R.R. Miller / Beha6ioural Processes 44 (1998) 147–162

hypothesis differs from simple contiguity theory however, have been applied to the range of phe-
by asserting that, although contiguity is sufficient nomena described here. Like the temporal coding
for the acquisition of an association, the behav- hypothesis, some real-time models (e.g. Sutton
ioral expression of that association depends on and Barto, 1990) correctly predict that cue com-
the temporal relationship of the associates. The petition will be maximal if the competing CSs
distinction between ‘what is learned’ and ‘the share the same temporal relationship to the US.
behavior observed’ appears to be among the sim- However, because most of these models were
plest modifications of contiguity theory that can derived directly from information theory and have
permit it to account both for the behavioral si- adopted a Rescorla–Wagner learning rule, they
lence of simultaneous and backward associations, are hindered by the constraint that a predictive
and for the observation that subjects learn tempo- relationship is necessary for associative acquisi-
ral relationships in a wide array of Pavlovian tion. Thus, these models would not expect the
situations. strong simultaneous and backward associations
Other recent theoretical developments also at- revealed in the second-order conditioning and
test to the importance of temporal coding in sensory preconditioning studies described above.
Pavlovian conditioning. A more extreme position As such, the real-time models are unable to en-
than that of the temporal coding hypothesis is compass the central distinction within the tempo-
adopted by Gallistel (1990). His model abandons ral coding hypothesis between learning (through
the traditional notion of associative strength alto- contiguity) and expression of acquired informa-
gether. Rather than viewing temporal relation- tion (through learned forward temporal relation-
ships as one of several attributes of an ships).
association, the model assumes that Pavlovian Perhaps it should not be surprising that animals
associations exclusively encode temporal relation- encode temporal relationships in Pavlovian tasks.
ships. In Gallistel’s view, organisms simply ac- A large literature indicates that both human and
quire representations of the temporal intervals nonhuman subjects learn to time their behavior
between events, and these temporal relationships when doing so maximizes reward or minimizes
serve as the basis of behavior. However, it may be effort on instrumental tasks (e.g. Church, 1978;
premature to relinquish the notion of associative Roberts, 1981). This is evident in many situations,
strength. Consider the everyday example of our for example the scalloped response patterns gener-
association between a table and a chair. The ated when a fixed-interval schedule of reinforce-
temporal relationship between them is so varied ment is in effect. Even in Pavlovian conditioning,
that there is effectively no temporal relationship, it has long been known that subjects may time
yet a strong associative linkage surely exists. their conditioned responses in accordance with
Moreover, reducing all learning to the temporal the duration of the CS (Pavlov, 1927). For exam-
dimension neglects other dimensions that are ple, in the phenomenon called inhibition of delay,
seemingly encoded as part of an association. For subjects with sufficient training learn to withhold
example, spatial information would be acquired their responses until near the end of a long CS,
from the strong spatial relationship that a chair just before the US is due to be presented. Thus, it
ordinarily has to a table. We prefer to think of an has long been recognized that when reinforcement
association as being the sum total of all of the is involved in an association, subjects can and do
relationships that link two events, with the tempo- learn the temporal relationship between a signal
ral relationship being of considerable import but and reinforcement.
not the sole basis of the association. Sensory preconditioning demonstrates that ani-
Several real-time models of conditioning in mals can learn to associate relatively neutral stim-
which the temporal relationship between events is ulus events. But the important question for the
in some way encoded have been formulated re- present framework is whether temporal relation-
cently (e.g. Wagner, 1981; Sutton and Barto, ships between neutral stimulus events are auto-
1990; Grossberg, 1991). None of these models, matically encoded. The series of studies by Matzel
H.I. Sa6astano, R.R. Miller / Beha6ioural Processes 44 (1998) 147–162 159

et al. (1988) that was discussed previously con- signal for the impending biologically significant
tained an additional experiment that addresses event (analogous to the second-order conditioning
this question directly. Rats were first exposed to procedure of Barnet et al., 1991). The organism
either simultaneous or forward pairings of two would be well served to prepare for the biologi-
neutral CSs (i.e. CS2:CS1 or CS2“ CS1, respec- cally significant event when it later encounters the
tively). In Phase 2, CS1 was made biologically second-order cue. However, the organism would
relevant by presenting it simultaneously with the be able to make a preparatory response to the
US (CS1:US). Not surprisingly, responding to second-order stimulus only if: (a) a forward asso-
CS1 was uniformly weak because of its simulta- ciation was formed between the second-order
neous relationship to the US. However, strong stimulus and the first-order stimulus, and (b) an
conditioned responding to CS2 was evident, but association was formed between the first-order
only when CS2 had a forward (i.e. anticipatory), stimulus and the biologically significant event.
rather than a simultaneous, relationship with CS1. The latter (b) would only occur if simultaneous
These findings indicate that animals had encoded associations could be acquired. Thus, organisms
the temporal relationship between CS1 and CS2 that were not able to encode nonanticipatory as
even though neither of these cues was biologically well as anticipatory temporal relationships may
relevant at the time they were paired. The sugges- have had a lower adaptive fitness relative to or-
tion is that encoding of temporal relationships is ganisms that were able to learn both types of
an automatic process within associative learning, temporal relationships. The latter organisms
and does not require that one of the associated would have learned on the basis of contiguity
events be of biological significance. alone, but responded on the basis of integrated
information from each of the associations that
jointly allow second-order CSs to convey anticipa-
10. Evolutionary adaptiveness tory information about biologically significant
events. Thus, a compelling argument can be made
Earlier we considered the possible evolutionary for the functional value of learning by contiguity,
advantage of an information-based learning rather than by prediction, as the critical variable
mechanism over a simpler contiguity mechanism. that underlies the learning mechanism. However,
Part of the appeal of information theory lies in at this point it is premature to categorically reject
the functional value of predictiveness and its abil- the possibility that animals’ learning by contiguity
ity to explain why organisms make preparatory as opposed to learning by predictive value merely
responses in anticipation of biologically significant reflects evolutionary inertia. That is, there may be
events, but not following biologically significant insufficient cost to animals in learning simulta-
events. However, an argument could also be made neous and backward associations that natural se-
that forming associations between events on the lection would not favor a learning mechanism
basis of contiguity alone is adaptive. Such a posi- sensitive to prediction over one sensitive to conti-
tion would have to propose an evolutionary ad- guity. More significantly, the fact that compelling
vantage of learning simultaneous or backward evolutionary arguments exist for contiguity-based
associations as well as predictive associations. learning as well as for prediction-based learning
Conceivably, in our evolutionary history, situa- points out the dangers of allowing the analysis of
tions arose in which organisms encountered a proximate processes to be overly influenced by
biologically significant event simultaneously with theorizing at the level of ultimate mechanisms.
an initially neutral (first-order) stimulus event. At
some later point, an organism might encounter a
second neutral stimulus followed immediately by 11. Bidirectional associations
the first-order event (i.e. an anticipatory relation-
ship). Thus, the second-order stimulus may accrue Finally, a clear implication of the temporal
the potential to act indirectly as an anticipatory coding hypothesis is that associations are bidirec-
160 H.I. Sa6astano, R.R. Miller / Beha6ioural Processes 44 (1998) 147–162

tional in nature. Consider the case in which discrimination task to teach pigeons that Stimu-
event A is followed by event B during training lus A leads to Outcome X and Stimulus B leads
(A“ B). Because the temporal coding hypothesis to Outcome Y. Subsequently, they observed fa-
assumes that associative learning depends on cilitated transfer in their subjects’ learning that
contiguity rather than prediction, and further- X leads to A and Y leads to B (i.e. the reversed
more assumes that temporal relationships are order of previous paired events) relative to
encoded, it suggests that presenting one event learning that X leads to B and Y leads to A
should activate the representation in memory of (i.e. events that were not previously paired).
the other event (as well as its temporal location This outcome suggests that backward associa-
relative to the signal). For this to occur, the tions were acquired at the same time that the
association between A and B must be function- forward associations were learned.
ally bidirectional (AlB). Thus, the presentation
of A alone should activate a prospective (predic-
tive) representation of B, and testing on B alone 12. Conclusions
should activate a retrospective representation of
A. We have provided an overview of the avail-
Anecdotally, it is clear that many of our ev- able data suggesting that temporal coding is
eryday associations appear to be bidirectional. ubiquitous in a wide array of Pavlovian condi-
When we hear the word ‘Paris’, we might come tioning paradigms. Those data support the con-
to imagine the popular sights of the city, such clusion that temporal contiguity between the
as the Eiffel tower. If we were later shown a representation of stimulus events is sufficient for
picture of the Eiffel tower, we would have no the formation of an association. Organisms ap-
problem supplying the name of the city. In a pear to use temporal relationships as a basis for
simple, but elegant, study, Gerolin and Matute determining the nature, magnitude, and timing
(1998) found empirical support for the existence of the expression of acquired information. This
of functional bidirectional associations in human distinction between learning and performance
subjects. Their subjects were trained on pairs of poses a serious challenge for the traditional the-
neutral stimulus events (colors and figures) in a ories of associative learning that are based on
forward direction (color“ figure). At test, sub- the notion of prediction being required for
jects were presented with a figure, and were able learning to occur, but it is consistent with the
to report the associated color, thereby suggest- basic assumptions of the temporal coding hy-
ing learning of a backward association. It is un- pothesis.
clear whether this result reflects the action of a
single bidirectional association or two opposing
unidirectional associations. However, the finding Acknowledgements
was replicated even when a single training trial
was used, making less plausible any explanation Preparation of this manuscript was supported
in terms of an independent figure“color associ- by NIMH Grants 11704-02 (H.I. Savastano)
ation being formed by the onset of the color and 33881 (R.R. Miller). Thanks are extended
activating a representation of the figure, which to Francisco Arcediano, Aaron Blaisdell, Dan
could then be forwardly associated with the still Burger, Jim Denniston, and Martha Escobar for
present color (i.e. retrospective revaluation, Van their comments on earlier versions of the
Hamme and Wasserman, 1994). manuscript. Requests for reprints should be ad-
Recent findings also appear to lend strong dressed to Ralph R. Miller, Department of Psy-
support to the existence of functional bidirec- chology, SUNY-Binghamton, Binghamton, NY
tional associations in other species. For example 13902-6000, USA, E-mail: rmiller@binghamton.
Zentall et al. (1992) used a delayed conditional edu.
H.I. Sa6astano, R.R. Miller / Beha6ioural Processes 44 (1998) 147–162 161

References Egger, M.D., Miller, N.E., 1963. When is a reward reinforc-


ing?: an experimental study of the information hypothesis.
Allan, L.G., 1993. Human contingency judgments: rule based J. Comp. Physiol. Psychol. 56, 132 – 137.
or associative? Psychol. Bull. 114, 435–448. Farley, J., Alkon, D.L., 1987. In vitro associative conditioning
Ayres, J.J.B., Haddad, C., Albert, M., 1987. One-trial excita- of Hermissenda: cumulative depolarization of Type B pho-
tory backward conditioning assessed by conditioned sup- toreceptors and short-term associative behavioral changes.
pression of licking in rats: concurrent observations of lick J. Neurophysiol. 57, 1639 – 1668.
suppression and defensive behaviors. Anim. Learn. Behav. Gallistel, C.R., 1990. The Organization of Learning. MIT
15, 212 – 217. Press, Cambridge, MA.
Barnet, R.C., Arnold, H.M., Miller, R.R., 1991. Simultaneous Gerolin, M., Matute, H., 1998. Evidence of bidirectional asso-
conditioning demonstrated in second-order conditioning: ciations. Anim. Learn. Behav. (in press).
evidence for similar associative structure in forward and Gibson, E.J., 1940. A systematic application of the concepts of
simultaneous conditioning. Learn. Motiv. 22, 253–268. generalization and differentiation to verbal learning. Psy-
Barnet, R.C., Cole, R.P., Miller, R.R., 1997. Temporal inte- chol. Rev. 47, 196 – 229.
gration in second-order conditioning and sensory precondi- Gibbon, J., Balsam, P., 1981. Spreading association in time.
tioning. Anim. Learn. Behav. 25, 221–233. In: Locurto, C.M., Terrace, H.S., Gibbon, J. (Eds.), Au-
Barnet, R.C., Grahame, N.J., Miller, R.R., 1993. Temporal toshaping and Conditioning Theory. Academic Press, New
encoding as a determinant of blocking. J. Exp. Psychol.: York, pp. 219 – 253.
Anim. Behav. Process. 19, 327–341. Grossberg, S., 1991. A neural network architecture for Pavlo-
Barnet, R.C., Miller, R.R., 1996. Temporal encoding as a vian conditioning: Reinforcement, attention, forgetting,
determinant of inhibitory control. Learn. Motiv. 27, 73– timing. In: Commons, M.L., Grossberg, S., Staddon,
91. J.E.R. (Eds.), Neural Networks Models of Conditioning
Blaisdell, A.P., Denniston, J.C., Miller, R.R., 1998a. Temporal and Action. Lawrence Erlbaum Associates, Hillsdale, NJ,
encoding as a determinant of overshadowing. J. Exp. Psy- pp. 69 – 122.
chol.: Anim. Behav. Process. 24, 72–83. Hall, G., Pearce, J.M., 1979. Latent inhibition of a CS during
Blaisdell, A.P., Denniston, J.C., Miller, R.R. Posttraining CS – US pairings. J. Exp. Psychol.: Anim. Behav. Process.
shifts in the overshadowing CS–US interval alleviates the 5, 31 – 42.
overshadowing deficit. J. Exp. Psychol.: Anim. Behav. Hearst, E.H., 1989. Backward associations: differential learn-
Process. (in press). ing about stimuli that follow the presence versus the ab-
Blaisdell, A.P., Gunther, L.M., Miller, R.R. Recovery from sence of food in pigeons. Anim. Learn. Behav. 17,
blocking through posttraining extinction of the blocking 280 – 290.
stimulus. Learn. Motiv. (in press). Heth, C.D., 1976. Simultaneous and backward fear condition-
Burkhardt, P.E., Ayres, J.J.B., 1978. CS and US duration in
ing as a function of the number of CS – US pairings. J.
one-trial simultaneous fear conditioning as assessed by
Exp. Psychol.: Anim. Behav. Process. 2, 117 – 129.
conditioned suppression of licking in rats. Anim. Learn.
Heth, C.D., Rescorla, R.A., 1973. Simultaneous and backward
Behav. 6, 225 – 230.
fear conditioning in the rat. J. Comp. Physiol. Psychol. 82,
Brogden, W.J., 1939. Sensory pre-conditioning. J. Exp. Psy-
434 – 443.
chol.: 25, 323 – 332.
Hull, C.L., 1943. Principles of Behaviour. Appleton – Century –
Cantor, M.B., 1981. Information theory: a solution to two big
Crofts, New York.
problems in the analysis of behavior. In: Harzem, P.,
Kamin, L.J., 1969. Predictability, surprise, attention, and con-
Zeiler, M.D. (Eds.), Predictability, Correlation, and Conti-
guity. Wiley, Chichester, UK, pp. 287–320. ditioning. In: Campbell, B.A., Church, R.M. (Eds.), Pun-
Cheng, P.W., 1997. From covariation to causation: a causal ishment and Aversive Behavior. Appleton – Century –
power theory. Psychol. Rev. 104, 367–405. Crofts, New York, pp. 279 – 296.
Church, R.M., 1978. The internal clock. In: Hulse, S.H., Kandel, E.R., Schwartz, J.H., 1982. Molecular biology of
Honig, W.K. (Eds.), Cognitive Processes in Animal Behav- learning: modulation of transmitter release. Science 218,
ior. Erblaun, Hillsdale, NJ, pp. 265–327. 433 – 443.
Cole, R.P., Barnet, R.C., Miller, R.R., 1995. Temporal encod- Kasprow, W.J., Cacheiro, H., Balaz, M.A., Miller, R.R., 1982.
ing in trace conditioning. Anim. Learn. Behav. 23, 144– Reminder-induced recovery of associations to an overshad-
153. owed stimulus. Learn. Motiv. 13, 155 – 166.
Denniston, J.C., Blaisdell, A.P., Miller, R.R., 1998. Temporal Kaufman, M.A., Bolles, R.C., 1982. A nonassociative aspect
coding affects transfer of serial and simultaneous in- of overshadowing. Bull. Psychon. Soc. 18, 318 – 320.
hibitors. Anim. Learn. Behav. 26, 336–350. Kehoe, E.J., Schreurs, B.G., Amodei, N., 1981. Blocking
Denniston, J.C., Cole, R.P., Miller, R.R., 1998b. The role of acquisition of the rabbit’s nictitating membrane response
temporal relationships in the transfer of conditioned inhi- to serial conditioned stimuli. Learn. Motiv. 12, 92 – 108.
bition. J. Exp. Psychol.: Anim. Behav. Process. 24, 200– Kraemer, P.J., Lariviere, N.A., Spear, N.E., 1988. Expression
214. of a taste aversion conditioned with an odor-taste com-
162 H.I. Sa6astano, R.R. Miller / Beha6ioural Processes 44 (1998) 147–162

pound: overshadowing is relatively weak in weanlings and Pearce, J.M., Hall, G., 1980. A model of Pavlovian learning:
decreases over a retention interval in adults. Anim. Learn. variations in the effectiveness of conditioned but not of
Behav. 16, 164 – 168. unconditioned stimuli. Psychol. Rev. 87, 532 – 552.
Locke, J. An Essay Concerning Human Understanding. A.D. Rescorla, R.A., 1968. Probability of shock in the presence
Woozley, Editor. New York: New American Library. and absence of CS in fear conditioning. J. Comp. Physiol.
(Original work published 1706). Psychol. 66, 1 – 5.
Lubow, R.E., Moore, A.U., 1959. Latent inhibition: the ef- Rescorla, R.A., 1972. Informational variables in Pavlovian
fect of nonreinforced preexposure to the conditioned stim- conditioning. In: Bower, G.H. (Ed.), The Psychology of
ulus. J. Comp. Physiol. Psychol. 52, 415–419. Learning and Motivation, vol. 6. Academic Press, New
Mackintosh, N.J., 1975. A theory of attention: variations in York, pp. 1 – 46.
the associability of stimuli with reinforcements. Psychol. Rescorla, R.A., 1980. Simultaneous and successive associa-
Rev. 82, 276 – 298. tions in sensory preconditioning. J. Exp. Psychol.: Anim.
Mackintosh, N.J., 1983. Conditioning and Associative Learn- Behav. Proc. 6, 207 – 216.
ing. Oxford University Press, London. Rescorla, R.A., Wagner, A.R. A theory of Pavlovian condi-
Matzel, L.D., Held, F.P., Miller, R.R., 1988. Information tioning: variations in the effectiveness of reinforcement
and the expression of simultaneous and backward associa- and nonreinforcement. In: A.H. Black, W.F. Prokasy
tions: Implications for contiguity theory. Learn. Motiv. (Eds.) Classical Conditioning II: Current Research and
19, 317 – 344. Theory. Appleton-Century-Crofts, New York, pp. 64 – 69.
Roberts, S., 1981. Isolation of an internal clock. J. Exp.
Matzel, L.D., Schachtman, T.R., Miller, R.R., 1985. Recov-
Psychol.: Anim. Behav. Process. 7, 242 – 268.
ery of an overshadowed association achieved by extinc-
Savastano, H.I., Yin, H., Barnet, R.C., Miller, R.R., 1998.
tion of the overshadowing stimulus. Learn. Motiv. 16,
Temporal coding in Pavlovian conditioning: Hall – Pearce
398 – 412.
negative transfer. Q. J. Exp. Psychol. 51B, 139 – 153.
Miller, R.R., Barnet, R.C., 1993. The role of time in elemen-
Schreurs, B.G., Westbrook, R.F., 1982. The effect of changes
tary associations. Curr. Dir. Psychol. Sci. 2, 106–111.
in the CS – US interval during compound conditioning
Miller, R.R., Matzel, L.D., 1988. The comparator hypothesis:
upon an otherwise blocked element. Q. J. Exp. Psychol.
a response rule for the expression of associations. In: 34B, 19 – 30.
Bower, G.H. (Ed.), The Psychology of Learning and Mo- Sutton, R.S., Barto, A.G., 1990. Time-derivative models of
tivation, vol. 22. Academic Press, San Diego, CA, pp. Pavlovian reinforcement. In: Gabriel, M., Moore, J.
51– 92. (Eds.), Learning and computational neuroscience: founda-
Miller, R.R., Schachtman, T.R., 1985. Conditioning context tions of adaptive networks. MIT Press, Cambridge, MA,
as an associative baseline: Implications for response gen- pp. 497 – 537.
eration and the nature of conditioned inhibition. In: Wagner, A.R., 1981. SOP: a model of automatic processing
Miller, R.R., Spear, N.E. (Eds.), Information processing in animal behavior. In: Spear, N.E., Miller, R.R. (Eds.),
in animals: Conditioned inhibition. Erlbaum, Hillsdale, Information Processing in Animals: Memory Mechanisms.
NJ, pp. 51 – 88. Erlbaum, Hillsdale, New Jersey, pp. 5 – 47.
Moore, J.W., Gormezano, I., 1977. Classical conditioning. In: Van Hamme, L.J., Wasserman, E.A., 1994. Cue competition
Marx, M.H., Bunch, M.E. (Eds.), Fundamentals and Ap- in causality judgments: the role of nonpresentation of
plications of Learning. MacMillan, New York, pp. 87– compound stimulus elements. Learn. Motiv. 25, 127 – 151.
120. Zentall, T.R., Sherburne, L.M., Steirn, J.N., 1992. Develop-
Pavlov, I.P., 1927. Conditioned reflexes. Oxford University ment of excitatory backward associations during the es-
Press, London. tablishment of forward associations in a delayed
Pearce, J.M., 1987. A model of stimulus generalization for conditional discrimination by pigeons. Anim. Learn. Be-
Pavlovian conditioning. Psychological Rev. 94, 61–73. hav. 20, 199 – 206.

You might also like