ArticlePDF Available

Abstract and Figures

People can create temporal contexts, or episodes, and stimuli that belong to the same context can later be used to retrieve the memory of other events that occurred at the same time. This can occur in the absence of direct contingency and contiguity between the events, which poses a challenge to associative theories of learning and memory. Because this is a learning and memory problem, we propose an integrated approach. Theories of temporal contexts developed in the memory tradition provide interesting predictions that we test using the methods of associative learning to assess their generality and applicability to different settings and dependent variables. In 4 experiments, the integration of these 2 areas allows us to show that (a) participants spontaneously create temporal contexts in the absence of explicit instructions; (b) cues can be used to retrieve an old temporal context and the information associated with other cues that were trained in that context; and (c) the memory of a retrieved temporal context can be updated with information from the current situation that does not fit well with the retrieved memory, thereby helping participants to best adapt their behavior to the future changes of the environment.
Content may be subject to copyright.
Temporal Contexts: Filling the Gap Between
Episodic Memory and Associative Learning
Helena Matute
Universidad de Deusto Ottmar V. Lipp
University of Queensland
Miguel A. Vadillo
Universidad de Deusto Michael S. Humphreys
University of Queensland
People can create temporal contexts, or episodes, and stimuli that belong to the same context can
later be used to retrieve the memory of other events that occurred at the same time. This can occur
in the absence of direct contingency and contiguity between the events, which poses a challenge to
associative theories of learning and memory. Because this is a learning and memory problem, we
propose an integrated approach. Theories of temporal contexts developed in the memory tradition
provide interesting predictions that we test using the methods of associative learning to assess their
generality and applicability to different settings and dependent variables. In 4 experiments, the
integration of these 2 areas allows us to show that (a) participants spontaneously create temporal
contexts in the absence of explicit instructions; (b) cues can be used to retrieve an old temporal
context and the information associated with other cues that were trained in that context; and (c) the
memory of a retrieved temporal context can be updated with information from the current situation
that does not fit well with the retrieved memory, thereby helping participants to best adapt their
behavior to the future changes of the environment.
Keywords: associative learning, episodic memory, temporal contexts, paired associate, predictive
learning
When people remember something that they experienced at a
particular time of their life, they may also remember other things
that occurred in the same epoch. For instance, over the course of
a conversation, people may retrieve memories of childhood friends
they made during a family holiday, and the revival of some
episodes may also lead them to retrieve the memories of a histor-
ical event that also took place during that wonderful holiday they
spent at the coast with their parents.
The phenomenon we just described may look at first glance like
a simple associative effect. If the presentation of a given stimulus
can activate the mental representation of another stimulus, this
must mean that the two stimuli share an associative link. The
challenge comes when we think about the way in which theories of
learning and memory predict that links between stimuli are
formed. Ideally, one stimulus must be consistently and contigu-
ously followed by the other (contiguity) and should not occur in
the absence of the other (contingency). Indeed, the most widely
cited theories of associative learning (e.g., Rescorla & Wagner,
1972) assume that as contiguity or contingency are reduced, the
strength of the association will weaken.
1
Theories of memory
make a similar prediction, with some of the most popular views of
memory assuming a buffer in which contiguous events become
associated (e.g., search of associative memory; Raaijmakers &
Shiffrin, 1980, 1981). Thus, most theories of learning and memory
predict that direct associations should not form between isolated
stimuli that are neither contiguous nor contingent on each other.
It appears quite obvious, however, that events that were expe-
rienced in the same temporal context can activate each other’s
mental representation. The question is how they do it. This is the
focus of the present study. Because this is a problem of learning
1
There are exceptions to the contiguity principle, such as taste aversion
learning, which occurs when the consumption of a given food is followed
by illness. The association between food and illness is a very special one
because poisonous foods can produce illness after several hours. Thus, to
be effective, this type of learning needs to bridge very long delays (Garcia,
Ervin, & Koelling, 1966). Nevertheless, even for taste aversion learning,
the general contiguity principle still holds that the longer the interval, the
weaker the association.
This article was published Online First July 11, 2011.
Helena Matute and Miguel A. Vadillo, Departamento de Fundamentos y
Me´todos de la Psicologı´a, Universidad de Deusto, Bilbao, Spain; Ottmar V.
Lipp and Michael S. Humphreys, School of Psychology, University of
Queensland, Brisbane, Queensland, Australia.
Support for this research was provided by Grant SEJ2007-63691/PSIC
from the Spanish Ministry of Science and Innovation; Grant IT363-10 from
the Departamento de Educacio´n, Universidades e Investigacio´n of the
Basque Government; and Grant DP0770844 from the Australian Research
Council. We thank M. Teresa Bajo, Mark E. Bouton, Pedro L. Cobos,
Samuel D. Hannah, Marc W. Howard, Francisco J. Lo´pez, Thomas Sud-
dendorf, and Jason M. Tangen for their very helpful comments and sug-
gestions on this research.
Correspondence concerning this article should be addressed to Helena
Matute, Departamento de Psicologı´a, Universidad de Deusto, Apartado 1,
48080 Bilbao, Spain. E-mail: matute@deusto.es
Journal of Experimental Psychology: General © 2011 American Psychological Association
2011, Vol. 140, No. 4, 660– 673 0096-3445/11/$12.00 DOI: 10.1037/a0023862
660
and memory, we first describe how this problem can best be
addressed from each of those perspectives. After that, we suggest
an integrated framework.
Within the associative learning tradition, Bouton’s (1993, 1997)
theory is the one most clearly devoted to understanding how
contexts, physical or temporal, affect learning and behavior. It is
well-known that the presentation of a cue can retrieve the mental
representation of a second stimulus (i.e., an outcome) that has been
associated with it. However, there are times in which the cue has
been paired with different outcomes in different contexts. Accord-
ing to Bouton (1997), in those cases, the cue becomes ambiguous
and the context can be used to disambiguate it. The association
acquired second becomes context specific, which means that it will
only be retrieved in the context in which it was acquired. If
organisms are tested in the context in which the association ac-
quired first was trained or in a new context, the association
acquired first will prevail. A revision of this theory advanced by
Rosas, Callejas-Aguilera, Ramos-A
´lvarez, and Ferna´ndez-Abad
(2006) makes similar predictions, although it states that both the
first- and the second-learned associations could, in principle, be
specific to the context in which they were trained (see also
Gawronski, Rydell, Vervliet, & De Houwer, 2010; Rosas &
Callejas-Aguilera, 2006). According to this version of the theory,
the context will play a role in the retrieval of the outcome if the
organism has paid attention to the context. Attention may be
devoted to the context not only because the cue is ambiguous, as
suggested by Bouton (1993), but also because the participant has
been instructed to attend to the context or for other reasons.
Bouton’s model, as well as Rosas et al.’s revision, has shown great
heuristic value in guiding new research and in providing successful
predictions in situations in which temporal contexts are involved,
situations that are normally outside of the scope of simpler theories
of learning (e.g., Rescorla & Wagner, 1972).
Consider, for example, an experiment by Brooks and Bouton
(1993). Rats were first exposed to several pairings of a cue, A, with
an outcome (e.g., A–O1). Then this learning was extinguished.
That is, Cue A was now paired with nothing (i.e., A–O2). This
normally produces responding appropriate for O2 if testing takes
place immediately after A–O2 training (i.e., a recency effect; in
this case, weak responding to A). So far, this can be simulated with
very simple associative models of learning such as that of Rescorla
and Wagner (1972). However, the presentation of A may also
produce recovery of the response appropriate to O1 if testing takes
place in a new physical context or after some time (i.e., in a new
temporal context). This latter phenomenon is called spontaneous
recovery (Pavlov, 1927). Because the Rescorla and Wagner model
conceptualizes extinction as catastrophic forgetting of the associ-
ation acquired first, the model cannot account for spontaneous
recovery. Bouton’s (1997) theory, however, does so by assuming
that because the extinction phase contradicts what was learned in
Phase 1, extinction becomes context specific. Thus, extinction-
related behavior will only be retrieved if the organism is tested in
the extinction context. Otherwise, what was learned during Phase
1, before the cue became ambiguous, will prevail. This explains
the recovery of responding when testing occurs in new temporal
(or physical) contexts. The main point of the study by Brooks and
Bouton was to show that a retrieval cue for extinction presented
after some time was able to induce behavior appropriate for
extinction instead of the default spontaneous recovery that should
have been observed at that time. That is, when rats were prompted
to retrieve the context of extinction, behavior appropriate for the
extinction phase was shown.
We know of no other theory of learning that could predict these
and many other experimental results in which the retrieval of
temporal contexts is involved. It must be noted, however, that in
Bouton’s model, the temporal context is not defined by explicit
stimuli but by the mere passage of time. That is, even though
external cues that are presented in a particular phase of training can
be used as retrieval cues, it is the passage of time, rather than the
changing external stimuli, that defines the temporal contexts. De-
spite their enormous heuristic value, neither Bouton’s (1993, 1997)
model nor Rosas et al.’s (2006) revision specifies how temporal
contexts are represented, how the representation of old temporal
contexts can be reactivated in memory, or how the presentation of
a cue can reactivate a temporal context and the memories of other
stimuli that were trained in that context. To our knowledge, these
questions have not yet been addressed in the associative learning
tradition.
These questions have, however, been thoroughly addressed in
the memory tradition and, thus, both areas can complement
each other. The temporal context model (TCM) of Howard and
Kahana (2002), for instance, provides interesting insights into
the mechanisms by which temporal contexts can be created and
represented. According to TCM, stimuli become associated
during training with the current state of a gradually changing
representation of the temporal context. This context, in turn,
gets associated with the stimuli. Later on, a presentation of the
stimulus will enable the participant to retrieve this temporal
context, and, by the same reasoning, the reactivation of the
temporal context will also enable the participant to retrieve the
representation of other stimuli that were linked to it. It is
important to note that the temporal context is not an abstract
concept here but a flexible and evolving representation that
depends on, among other things, the stimuli (both internal and
external to the organism) that are present at each moment. This
means that stimuli that are present at a given time do indeed
become part of the temporal context for other stimuli present at
that time. Howard and Kahana’s (2002) model focuses on
simulating the evolution of the temporal contexts and the re-
trieval processes. Moreover, it provides a comprehensive the-
oretical framework of how a given cue can retrieve an entire
episode or temporal context. The most recent version of this
model (Sederberg, Howard, & Kahana, 2008) maintains the
assumption of gradually evolving contexts while incorporating
a new retrieval rule that is based on the leaky-accumulator
decision model of Usher and McClelland (2001). In this new
version, it is the relative activation of each of the different
associations between context and stimuli that determine which
memories are most readily retrieved at each time. The manner
in which contexts are represented and retrieved is clearly spec-
ified in this model. Thus, the TCM can be used to complement
the theories developed in the associative learning tradition.
In support of the TCM, researchers have shown, for example,
that participants create temporal contexts that later guide their
retrieval of the episodes. For instance, when studying paired as-
sociates, participants not only associate the required pairs (e.g.,
AB) but also form associations between these pairs and other pairs
presented nearby in the list (e.g., Davis, Geller, Rizzuto, & Ka-
661
EPISODIC MEMORY AND ASSOCIATIVE LEARNING
hana, 2008). This also suggests that associations are formed not
only among items but also between items and contexts and vice
versa (see also Ranganath, 2010; Schwartz, Howard, Jing, &
Kahana, 2005).
Convergent evidence for this proposal can be found in a study
by Howard, Jing, Rao, Provyn, and Datey (2009), in which par-
ticipants were presented with lists of paired associates (e.g., AB
DE, BC EF) in a random order; the associations between items that
were not presented closely together in time but were presented in
similar contexts (i.e., with context in this case being the other
member of the study pair such that A and C occur in the same
context, namely, B) were subsequently examined. As should be
expected from the TCM, participants were able to associate distant
events that shared a common context. Similar effects were also
evident when the stimuli were presented in separate lists or sepa-
rated by several hundred seconds, that is, in the absence of conti-
guity (Howard, Youker, & Venkatadass, 2008). Associative-like
effects similar to these have been demonstrated even for stimuli
separated by distracters (Howard & Kahana, 1999). These findings
are consistent with the idea that the presentation of a cue can
retrieve the mental representation of the temporal context, which,
in turn, can retrieve the representation of other stimuli that were
also trained during the same temporal context.
It is important to note that if we are claiming that theories about
temporal contexts developed in the memory tradition can be used
to complement theories developed in the associative learning tra-
dition, it should be shown that predictions derived from the mem-
ory literature can be verified in paradigms used in associative
learning. Indeed, although the theories in the memory tradition
have reached a greater level of development, they have been tested
almost exclusively on paired associates and word lists. The asso-
ciative learning methodology can provide convergent evidence
using nonverbal behavior as the dependent variable (e.g., behav-
ioral adaptation in simple videogames; see Arcediano, Ortega, &
Matute, 1996; Costa & Boakes, 2011; Franssen, Clarysse, Beckers,
van Vooren, & Baeyens, 2010; Lipp & Dal Santo, 2002). This, in
turn, should speak to the generality of the effects, and new pre-
dictions should arise from the integration of both research areas.
Using associative learning methods will also allow us to extend the
findings from the memory tradition to new situations, such as those
that lack a time interval within phases (something that is difficult
to implement in the memory paradigms), or to situations that lack
explicit instructions about the temporal contexts. Thus, the asso-
ciative learning methods permit the investigation of how partici-
pants construct temporal contexts in the absence of external cues,
such as changing learning lists or instructions.
Experiment 1
A recent experiment in the memory tradition by Humphreys,
Murray, and Maguire (2009) is our starting point. Simplifying
tremendously, participants first learned two different word lists, as
in the AB, CD, interference paradigm. At test, participants were
more accurate when a C cue followed another C cue than when it
followed an A cue. This suggests that cues can retrieve temporal
contexts, which, in turn, can retrieve other stimuli that occurred in
that context. Humphreys et al. used both instructions and temporal
grouping to establish the lists, or temporal contexts. They also used
both instructions and a cue to reinstate the contexts at test. This
result, we believe, resembles the findings by Brooks and Bouton
(1993) described above, although, of course, Brooks and Bouton
did not use instructions with their nonhuman participants. Never-
theless, Humphreys et al.’s study is more comprehensive: Infor-
mation from both Phase 1 and Phase 2 was retrieved using retrieval
cues, whereas Brooks and Bouton had argued that only the second
phase learning (extinction) was subject to selective retrieval.
Moreover, the possibility exists that Brooks and Bouton’s results
may be exclusive to a conditioning paradigm with nonhuman
animals, which involves the use of biologically significant events
such as unconditioned stimuli (see Gunther, Miller, & Matute,
1997). Our intent is therefore to follow up on Humphreys et al.’s
study using the methods of human associative learning. Thus, we
used symbolic rather than biologically significant events, and the
dependent variable was predictive behavior. Predictive behavior
and predictive judgments are often used in the human associative
learning tradition as indices by which to determine that an asso-
ciation has been learned, and they are also regarded as analogues
of conditioned responding in animals (e.g., Arcediano et al., 1996;
Costa & Boakes, 2011; Franssen et al., 2010; Lipp & Dal Santo,
2002; Shanks & Dickinson, 1987; Vadillo, Ba´rcena, & Matute,
2006; Wasserman, 1990). According to this view, once an associ-
ation has been acquired between a cue and an outcome, partici-
pants should be able to predict the outcome when the cue is
presented, which means that anticipatory behavior (and judgments)
appropriate to that particular outcome should occur in response to
the cue.
In the current study, Target Cue X predicted outcomes that
required different responses in different phases. X was associated
with a negative outcome (O1) during Phase 1 and with a positive
outcome (O2) during Phase 2, so that participants were trained not
to respond to Cue X during Phase 1 and to respond to Cue X
during Phase 2. Each phase of the study, including the test phase,
was conducted immediately after termination of the previous phase
and without interruption. Therefore, strong responding to X should
be the default behavior observed at test, given that the temporal
context of the test trial was identical to that of Phase 2 and very
different from that of Phase 1. However, if we prime the temporal
context of Phase 1 just before testing, then participants should
retrieve the association trained during Phase 1 instead and should
thus show weak responding to X at test, as if they were back in
Phase 1. It should be noted that the current predictions go beyond
those tested by Humphreys et al. (2009). First, we are testing
whether a temporal context will be naturally formed with the
different associations learned in different phases of the experiment.
Second, we are testing whether a cue on its own can reinstate the
temporal context of Phase 1.
Method
Participants and apparatus. Forty-one students from the
University of Queensland, Australia, received course credit for
their participation in the experiment. A computer program ran-
domly assigned participants to one of two groups. This resulted in
20 participants in Group Retrieve 1 and 21 in Group Default. All
of the materials were presented in an HTML document that in-
cluded JavaScript functions to manage the presentation of the
stimuli on the computer screen and to collect participants’ re-
sponses.
662 MATUTE, LIPP, VADILLO, AND HUMPHREYS
Design and procedure. For this experiment, we used the
spy-radio task (Matute, Vadillo, & Ba´rcena, 2007; Pinen˜o, Ortega,
& Matute, 2000).
2
In this task, participants are asked to imagine
that they are soldiers ordered to rescue refugees that are hidden in
a ramshackle building. On each trial, participants are given the
opportunity to place a number of refugees on a truck and to take
them to safety. Participants can place people on the truck by
pressing the space bar repeatedly: The more they press the space
bar, the more people they place. However, the refugees placed on
the truck do not always arrive safely at their destination. In some
trials, the road the truck has to drive on landmines that can
explode. Participants can predict whether the road will be safe on
a given trial by paying attention to the colored lights in a spy radio
installed in the truck. In each trial, a cue (i.e., a colored light) is
presented for 3 s during which the participant can respond. Certain
colors in the spy radio predict that the road will be safe (and,
therefore, indicate that participants should place as many refugees
as possible on the truck while that cue is on), whereas other colors
predict that the road will be mined (and, therefore, indicate that
participants should avoid placing refugees on the truck during the
presentation of those cues).
Immediately after the cue turned off, the outcome was pre-
sented: A message on the screen indicated whether the refugees
had arrived safely at their destination; it also indicated the number
of points earned or lost in that trial. Participants lost 1 point for
each refugee placed on the truck on trials in which the road was
mined (negative outcome; O1) and earned 1 point for each refugee
placed on the truck on trials in which the road was safe (positive
outcome; O2). They were not told which color predicted which
outcome. Their main task was to learn this by paying attention to
what happened during the learning trials. In each trial, the number
of refugees placed on the truck during the 3-s interval for which
the light was on was our dependent variable. This variable reflects
the extent to which participants have learned that the cue (i.e., the
color of the light) presented on a particular trial predicts that the
road will be safe.
The design of the experiment is shown in Table 1. During Phase
1, two groups of participants received 10 trials with the target cue,
X, which was always followed by O1, and 10 trials with Cue A,
which was always followed by O2. These 20 trials were randomly
intermixed. Then, during Phase 2, both groups received 10 addi-
tional trials with X, which was now followed by O2, and 10 trials
with a new cue, B, which was followed by O1. The 20 trials of this
phase were also randomly ordered. The colors that served as Cues
A, B, and X were blue, yellow, and red, counterbalanced across
participants. The duration of the intertrial interval was random,
ranging from 3 to 7 s. During this time, all lights were off (i.e.,
grey) and pressing the space bar produced no effect.
There were no breaks between phases. Thus, if participants were
to separate the continuous sequence of trials into different tempo-
ral contexts, they needed to spontaneously create temporal con-
texts themselves that coincided with those designed by the exper-
imenters (see Table 1). At the end of the experiment, both groups
received one test trial with the target cue, X. Before that, however,
the critical manipulation occurred during Phase 3, which consisted
of just one trial. This trial was presented immediately after the last
trial of Phase 2 and immediately before the test trial (with a regular
intertrial interval separating these trials). In Group Default, the
Phase 3 trial consisted of a regular B–O1 trial (i.e., an additional
Phase 2 trial). In Group Retrieve 1, however, this trial was an
A–O2 trial (i.e., a Phase 1 trial). This manipulation should cue
participants in Group Retrieve 1 to recall the context of Phase 1
just before testing. This, in turn, should prime them to respond to
X at test with the response that was appropriate during Phase 1
(i.e., low responding to X) rather than with the response that was
most recently acquired. That is, even though strong responding to
X should be displayed by the end of Phase 2 (and this should be
evident in Group Default), retrieving the temporal context of Phase
1 should lead participants in Group Retrieve 1 to respond weakly
to X at test.
At this point, it is important to note that there was no clear
contiguity or contingency between A and X in this experiment.
Cues X and A were each contiguously followed by their respective
outcomes, but the different X and A trials were separated from
each other by the intertrial intervals and were randomly ordered
within each phase. Therefore, the only relationship that existed
between Cue A and the meaning that Cue X had during Phase 1 is
that they were trained in the same phase. Indeed, participants were
not even told that there were different phases in the experiment.
However, we predicted that participants would be able to create the
temporal context that corresponded to the phases planned by the
experimenters, as the different cue–outcome associations pre-
sented in each phase remained constant for the duration of each
phase and changed with each new phase.
Results and Discussion
The data selection criterion usually applied when using this and
similar tasks in human associative learning is that at the end of
training, there must be more responses to the positive cues than to
the negative cues. This is a very lenient criterion that discards the
data of participants who show no signs of learning during the
training session. Such lack of learning can occur for a variety of
reasons, such as color blindness or lack of attention. If the target
associations were not acquired, it makes little sense to test for their
retrieval. In the present experiment, this criterion requires that the
number of responses given during the last positive trial of Phase 1
and Phase 2 be higher than the number of responses during the last
2
A demo version of this program can be downloaded from http://
www.labpsico.deusto.es/en/resources/
Table 1
Design Summary of Experiment 1
Group
Training
TestPhase 1 Phase 2 Phase 3
Retrieve 1 X–O1/A–O2 X–O2/B–O1 A–O2 X
Default X–O1/A–O2 X–O2/B–O1 B–O1 X
Note. Cues A, B, and X are color lights in the spy radio that can predict
two different outcomes: O1 (the participants lose points if they perform the
response) and O2 (participants earn points if they perform the response). In
each phase, the different trial types were randomly intermixed. Phase 3
consists of just one trial. In the actual experiment, the different phases are
not separated (i.e., a regular intertrial interval separates the last trial of one
phase and the first trial of the next one).
663
EPISODIC MEMORY AND ASSOCIATIVE LEARNING
negative trials of Phases 1 and 2 (before the critical trial is
introduced in Phase 3). Following this criterion, we eliminated two
participants from Group Retrieve 1 and one participant from
Group Default. This resulted in 18 participants in Group Retrieve
1 and 20 in Group Default.
The results are shown in Figure 1. They confirmed our predic-
tions and replicated the basic findings of Humphreys et al. (2009)
using a very different methodology and dependent variable. The
training phases proceeded as expected, with the learning curves
showing a gradual increase in the number of responses to the
stimuli that were associated with gaining points and a decrease of
responses to stimuli that were associated with losing points. This
is hardly surprising, not only because the task was relatively easy
but also because, as mentioned above, the data from those partic-
ipants who did not learn to respond more to the positive than to the
negative stimuli were discarded.
The critical results are those of the test phase. As expected, they
show that participants in Group Retrieve 1 responded significantly
less to X at test than did participants in Group Default, t(36)
5.07, p.001. As previously mentioned, the test trial is effec-
tively an additional Phase 2 trial in Group Default and, thus, strong
responding is observed in that group. The interesting result is that
of Group Retrieve 1, in which participants, as expected, behaved
differently. Presenting participants just before testing with the
stimulus that had been trained along with X during Phase 1 made
them respond to X as if they were back in Phase 1. These results
suggest that memories of independent associations that were
trained in the same temporal context can be used to retrieve each
other even when there is no contiguity or contingency between
them.
Experiment 2
Experiment 1 suggests that the presentation of a stimulus can be
used to retrieve other stimuli or experiences that were trained at the
same time even when there is no contiguity or contingency be-
tween the two stimuli. In the first experiment, it was important to
use a condition in which strong responding to X was evident by the
end of Phase 2 so that any changes in behavior shown at test could
not be attributed to preasymptotic learning. In Experiment 2,
however, to make certain that the results were not due to this
particular aspect of the design, participants were exposed to the
same treatment but with the order of phases reversed. That is, by
the end of Phase 2, groups should now show little or no responding
to the target cue, X (i.e., as in extinction). Like in Experiment 1,
the test phase was conducted at the end of training and without
interruption, which, in this case, was planned to minimize the
likelihood of spontaneous recovery. Thus, testing should show
extinction in Group Default. Group Retrieve 1, by contrast, was,
just before testing, cued to retrieve the strong responding to X that
in this experiment occurred during Phase 1. If our hypothesis is
correct, Group Retrieve 1 should now show higher rather than
lower responding than Group Default.
Method
Participants and apparatus. A total of 51 anonymous Internet
users who visited our virtual laboratory (http://www.labpsico.com)
volunteered for the study. The computer program randomly assigned
participants to one of two groups. This resulted in 26 participants in
Group Retrieve 1 and 25 in Group Default. To comply with ethical
regulations in the conduct of Internet research, we did not request any
personal data, nor did we use cookies or other software to obtain
personal information without the participants’ consent. All stimuli
involved in the experiment were preloaded in the computer’s memory
before participants could start the experiment, so differences in the
connection speed did not influence the pace of the experiment.
Although we did not record the computers’ Internet addresses
(i.e., IPs) in the present series of experiments, we have done so in
the past and have verified that the rate of data sets coming from the
same computer is negligible (around 2%), which confirms that
multiple data submission is unlikely to pose a problem for the
Figure 1. Mean number of responses during the different phases of Experiment 1 as a function of group and
cue. Error bars represent the standard errors of the mean.
664 MATUTE, LIPP, VADILLO, AND HUMPHREYS
validity of these studies (see also Reips, 2002). Moreover, even
those cases can represent different people using the same computer
(e.g., family members often share the same computer, and different
students in the same university can use the same computer at
different moments). Thus, duplicate IPs do not necessarily prove
that the same person is participating several times in the same
experiment. Perhaps most important, previous experiments have
shown that very similar results are normally obtained in the lab-
oratory and through the Internet (for a review, see Kraut et al.,
2004). This convergence between online and offline results has
also been confirmed for experiments in associative learning that
use procedures very similar to the ones we are using here (e.g.,
Matute et al., 2007; Vadillo et al., 2006; Vadillo & Matute, 2011).
In any case and as a general rule of prudence, two of the experi-
ments in this series were conducted through the Internet but the
other two were conducted in the laboratory.
Design and procedure. As can be seen in Table 2, the only
difference between this experiment and Experiment 1 is that the
order of the first two phases was reversed. The critical manipula-
tion in this study, which again took place during Phase 3, consisted
of presenting an additional Phase 2 trial in Group Default and a
Phase 1 trial in Group Retrieve 1. Therefore, Group Default should
show weak responding to X at test. A comparatively stronger
response should be shown in Group Retrieve 1 if, as suggested by
Experiment 1, participants in this group behave at test as if they
were back to the temporal context of Phase 1.
Results and Discussion
Applying the same data selection criterion used in Experiment 1,
we eliminated three participants from Group Retrieve 1 and one
from Group Default from the analyses. The results are shown in
Figure 2. The training phase proceeded smoothly, with participants
learning to respond more to the positive than to the negative
stimuli. The critical results are those of the test phase. As expected,
participants in Group Retrieve 1 responded to Cue X at test
significantly more than did participants in Group Default, t(45)
3.62, p.001. The weak responding of Group Default at test is
not surprising, given that responding to X had been extinguished
by the end of Phase 2 and that the test trial was conducted in an
identical temporal context. However, in line with the results ob-
served in Experiment 1, participants in Group Retrieve 1 behaved
differently. One presentation of the stimulus that had been trained
along with X during Phase 1 just before test led these participants
to behave at test as if they were back in Phase 1.
Experiment 3
Experiments 1 and 2 used the strategy of showing in Group
Default the behavior that should normally be expected at the end
of Phase 2 and comparing this with behavior prompted by the
presentation of a cue that reactivated the temporal context of Phase
1. However, it is possible that the presentation of a retrieval cue
from Phase 1 at the end of Phase 2 had the additional (or alterna-
tive) effect of producing a disruption between Phase 2 and test.
This disruption between training and testing was not present in
Group Default, and it could, in principle, be responsible for the
differential responding observed at test. Thus, in Experiment 3, we
used two retrieve groups and omitted Group Default. The two
groups used in this experiment were equally distracted at the end
of Phase 2. Then, after both groups had been distracted, we cued
either Phase 1 or Phase 2 for Group Retrieve 1 and Group Retrieve
2, respectively. A test trial with the target cue, X, was then
presented to assess whether responding appropriate to Phase 1 or
Phase 2 took place. If the disruption between Phase 2 and test was
responsible for the observed effects, those effects should not be
reproduced in the present experiment.
Therefore, in this experiment, the test phase took place well
after Phase 1 and Phase 2 had finished. This normally produces
spontaneous recovery of the association that was trained during
Phase 1 (e.g., Bouton, 1997; Pavlov, 1927). Even though it could
be argued that according to some theories (e.g., Bouton, 1993),
excitatory associations should prevail at test, regardless of the
phase in which they were trained, spontaneous recovery should
nevertheless affect both groups to the same extent. That is, in our
experiment, participants should either recover the excitatory asso-
ciation in both groups or the association trained first in both groups
(i.e., retrieving X–O2 more strongly than X–O1 or vice versa). Our
prediction, however, holds that the groups will behave differen-
tially as a function of whether the contexts of Phase 1 or Phase 2
are primed for retrieval.
Method
Participants and apparatus. This experiment was con-
ducted through the Internet with 105 anonymous volunteers. The
computer program randomly allocated participants to Group Re-
trieve 1 (49 participants) and Group Retrieve 2 (56 participants).
Design and procedure. The design summary is shown in
Table 3. The order of phases used in Experiment 1 was replicated
because that experiment had been conducted only in the labora-
tory. Thus, an additional purpose of Experiment 3 was to make
sure those results would replicate in the noisier domain of the
Internet. The design was almost identical to that of Experiment 1
except that a new phase was added after Phase 2 to create a
disruption between Phase 2 and test in both groups. In this new
phase, 10 trials were provided in which a novel cue, C, was paired
with O2. Cue C was a white light for all participants. Given that
some time had elapsed since the end of Phase 2, there is no reason
now to expect that the participants’ behavior should continue
reflecting the contingencies trained during Phase 2. Thus, we now
tried to induce the retrieval of the temporal context of either Phase
Table 2
Design Summary of Experiment 2
Group
Training
TestPhase 1 Phase 2 Phase 3
Retrieve 1 X–O2/A–O1 X–O1/B–O2 A–O1 X
Default X–O2/A–O1 X–O1/B–O2 B–O2 X
Note. Cues A, B, and X are color lights in the spy radio that can predict
two different outcomes: O1 (the participants lose points if they perform the
response) and O2 (participants earn points if they perform the response). In
each phase, the different trial types were randomly intermixed. Phase 3
consists of just one trial. In the actual experiment, the different phases are
not separated (i.e., a regular intertrial interval separates the last trial of one
phase and the first trial of the next one).
665
EPISODIC MEMORY AND ASSOCIATIVE LEARNING
1 or Phase 2 in Group Retrieve 1 and Group Retrieve 2, respec-
tively. To do so, we presented, just before testing, one trial with the
stimulus (A or B) that had been trained along with X in Phase 1 or
in Phase 2. Participants should now respond to the target cue X
during the test trial as if they were back in Phase 1 or Phase 2,
depending on whether they are being cued with A or B. All other
procedural details were the same as in the previous experiments.
Results and Discussion
Two participants (one from each group) failed to meet the data
selection criterion used in the previous experiments and were not
included in subsequent analyses. The results are shown in Figure
3. As in the previous experiments, training proceeded smoothly,
with the participants learning to respond more to the positive than
to the negative stimuli by the end of training. Most important, the
results show that participants in each group responded to X during
test, as expected. Group Retrieve 1 responded significantly less
than did Group Retrieve 2, t(101) 2.55, p.05. Thus, the cue
that was presented before testing effectively activated the temporal
context of one or the other phase.
Experiment 4
The results of Experiments 1–3 are compatible with the idea that
people spontaneously construct temporal contexts using the stimuli
that are present at a given time. They also show that people can use
stimuli that are present in the current situation to cue older tem-
poral contexts and the stimuli that had been associated with those
contexts. Moreover, these experiments show that this cuing makes
participants behave as if they were back in the old temporal
context. But taking the idea of gradually evolving temporal con-
texts one step further, we can also make some new predictions.
Many memory researchers agree that the purpose of retrieving
an old temporal context or episode and the memories associated
with it is to guide adaptive behavior at present and to plan for
future situations that might benefit from this knowledge (Schacter,
Addis, & Buckne, 2007; Suddendorf & Corballis, 1997, 2007;
Tulving, 2005). For this reason, when old episodes are retrieved,
they are updated by integrating the current information into the old
episode. Their function is not to provide an exact reproduction of
the past but to provide a useful guide for the future. As such, we
can assume quite straightforwardly that these episodes will be-
come more useful on retrieval the more information they manage
to integrate from the training situation.
The next experiment was a test of this idea. It was similar to the
previous ones but tested the hypothesis that the mental represen-
tation of temporal contexts is updated with new information that is
available when they are retrieved. To do so, we cued half of the
participants in this experiment to recall the temporal context of
Phase 1. However, we included a change so that the episode
actually presented was different from what the participant had
retrieved and was therefore expecting to occur. This should cause
participants to update the episode and their representation of the
Figure 2. Mean number of responses during the different phases of Experiment 2 as a function of group and
cue. Error bars represent the standard errors of the mean.
Table 3
Design Summary of Experiment 3
Group
Training
TestPhase 1 Phase 2 Phase 3 Phase 4
Retrieve 1 X–O1/A–O2 X–O2/B–O1 C–O2 A–O2 X
Retrieve 2 X–O1/A–O2 X–O2/B–O1 C–O2 B–O1 X
Note. Cues A, B, C, and X are color lights in the spy radio that can
predict two different outcomes: O1 (the participants lose points if they
perform the response) and O2 (participants earn points if they perform the
response). In each phase, the different trial types were randomly inter-
mixed. Phase 4 consists of just one trial. In the actual experiment, the
different phases are not separated (i.e., a regular intertrial interval separates
the last trial of one phase and the first trial of the next one).
666 MATUTE, LIPP, VADILLO, AND HUMPHREYS
Phase 1 temporal context. If this is to be transferred into behavior,
as our previous experiments suggest, participants should no longer
show the behavior appropriate to the temporal context of Phase 1
but a behavior that includes the update induced by our manipula-
tion.
To this end, we changed our design to a miscuing paradigm
(e.g., Lipp & Dal Santo, 2002; Lipp, Siddle, & Dall, 1993), which
provides the ideal ingredients to test our prediction. In the standard
miscuing procedure, the participants first receive one phase of
training in which a given cue, X, is always followed by a given
outcome (O1) and another cue, A, is followed by a second out-
come (O2). Once this training is established, a single miscuing trial
follows. In this trial, Cue A that was originally paired with O2 is
now paired with O1, which is contrary to the participant’s expec-
tations. The effect of interest is that, as a consequence of this
miscuing trial, participants update their expectations, and their
behavior with respect to the other stimulus, X, changes.
The design is shown in Table 4. During Phase 1, participants
learned that X produced a given outcome, O1, and that A produced
O2. The X–O1 and A–O2 trials were randomly presented. During
Phase 2, participants learned that a new target cue, Y, produced
O1, whereas another cue, B, produced O2. Then, during Phase 3,
participants received one trial with either Cue A or Cue B, the cues
that had been trained as predictors of O2 during either Phase 1 or
Phase 2. Therefore, this should retrieve the corresponding tempo-
ral context of either Phase 1 or Phase 2. It is important to note that
Cues A and B were now followed by the other outcome, O1. Thus,
A and B at this point were no longer mere reminders of an episode;
they retrieved the episode and, because of their pairing with a
different outcome, caused it to be updated. Thus, the old repre-
sentation of the temporal context of Phase 1 should now be
updated with this new evidence. What should participants do now
when presented with Target Cues X and Y at test? If our reasoning
is correct, participants should adapt their behavior so that the
change in the predictive status of Cue A should also change how
participants respond to the target cue that was originally trained in
the same temporal context. In other words, the miscuing effect
should affect selectively one target cue or the other, depending on
which priming cue, A or B, was presented immediately before the
test. These predictions go beyond those of any individual model of
memory or learning that we are aware of.
An additional purpose of this experiment was to ensure that the
observed effects were not specific to the particular task that we
used in Experiments 1–3. Thus, we used an alternative task that
resembles a Martians videogame for this experiment (see footnote
2). Finally, the experiment was aimed at verifying that the present
results cannot only be observed in experiments run via the Internet,
as in Experiments 2 and 3, but also in the laboratory, as in
Experiment 1.
Figure 3. Mean number of responses during the different phases of Experiment 3 as a function of group and
cue. Because of a programming error, the data from the distracting C–O2 trials in this experiment were not
recorded. Error bars represent the standard errors of the mean.
Table 4
Design Summary of Experiment 4
Group
Training
TestPhase 1 Phase 2 Phase 3
M1T1 X–O1/A–O2 Y–O1/B–O2 A–O1 X
M1T2 X–O1/A–O2 Y–O1/B–O2 A–O1 Y
M2T1 X–O1/A–O2 Y–O1/B–O2 B–O1 X
M2T2 X–O1/A–O2 Y–O1/B–O2 B–O1 Y
Note. Cues A–Y were background colors in the Martians task and could
predict two different outcomes: O1 (a Martian invasion) and O2 (no
invasion). In each phase, the different trial types were randomly inter-
mixed. Phase 3 consists of just one trial. There was no separation between
phases. M1 and M2 refer to whether Phase 1 or Phase 2 was miscued; T1
and T2 indicate whether testing took place with the target cue from Phase
1 or from Phase 2.
667
EPISODIC MEMORY AND ASSOCIATIVE LEARNING
Method
Participants and apparatus. Sixty-four undergraduate stu-
dents from the University of Queensland volunteered to participate
for course credit. On arrival at the laboratory, participants were
randomly assigned to one of four experimental groups.
Design and procedure. Participants were run individually
using a Martians videogame (see footnote 2) that resembles con-
ditioning and is often used to study associative learning with
humans (Arcediano et al., 1996; Costa & Boakes, 2011; Franssen
et al., 2010; Lipp & Dal Santo, 2002). The aim of participants in
this game is to prevent an invasion of Martians. Martians appear on
the screen and can be destroyed by firing a laser gun (i.e., pressing
the space bar) just before they appear. A successful shot is indi-
cated by the appearance of an explosion. Martians appear at a
steady rate on the screen of approximately four to five per second.
Participants are first trained to press the space bar regularly to
destroy as many Martians as possible. Thus, the purpose of this
preliminary phase is simply to produce stable baseline behavior of
about four to five responses per second, against which the focal
behavioral changes will be assessed during the subsequent exper-
iment.
Once this behavior is well established, participants receive a
new instructional set and the experiment proper starts. They are
now informed that the Martians have developed an antilaser shield
and that, if they fire the gun while the shield (a white flashing
screen) is connected, hundreds of Martians will invade their screen
immediately. The flashing screen lasts just for 0.5 s, which means
that participants need to find a way to predict its activation to be
able to cease responding before the flashing occurs. Otherwise,
they will suffer an invasion. They are explicitly encouraged to
learn to predict when the Martians are about to connect the shield,
although they are not told how to do it. Thus, in this paradigm,
participants are responding regularly to Martians during the inter-
trial intervals and a trial begins when a cue is presented. The cues
are changes of the screen background color. Some of these cues
signal the activation of the shield and, thus, a potential invasion
(O1). Others are followed by nothing (O2).
3
The dependent mea-
sure is the degree to which ongoing behavior is suppressed when
a cue is presented, which is indicative of the participant’s expec-
tation that O1 will follow.
In this experiment, cues were presented for 1 s during training
with the exception of the last presentation of each cue, which
lasted for 2 s, and the test trial presentations, which lasted for 3 s
each. The reason for doing so is that1sisnormally not enough to
properly assess the degree of suppression. However, if participants
are always given 3 s for each cue, they learn this temporal pattern
and keep responding as much as they can during the presentation
of the cue (Arcediano et al., 1996). In this case, no assessment of
the dependent variable can take place either. The solution sug-
gested by Arcediano et al. (1996) was to use a longer presentation
of cues only in trials in which assessment of the dependent variable
was most critical and keep the other trials to a shorter duration. In
this way, the procedure is most sensitive. Given that the test trial
and the latest training trial were the most critical in this experi-
ment, we opted for using longer cues that would allow assessment
only in those trials.
During Phase 1, all participants received eight trials with Cue X,
which always signaled danger (i.e., O1), and 10 trials with Cue A,
which signaled nothing (O2). Then, during Phase 2, they received
10 trials with Cue Y, which always signaled danger (i.e., O1), and
10 trials with Cue B, which signaled nothing (i.e., O2). The order
with which the different types of trials were presented within each
phase was randomized, as was the duration of the intertrial inter-
vals, which ranged from 6 to 12 s. The colors that served as Cues
A, B, X, and Y were yellow, blue, light blue, and red, all coun-
terbalanced. The background color of the screen was black during
the intertrial intervals.
Phase 3 consisted of just one trial. As in the preceding experi-
ments and because we used no breaks between phases, this Phase
3 trial was actually the last trial of Phase 2 and the one immedi-
ately preceding test. This critical Phase 3 manipulation consisted
of a miscuing trial in which the Martian invasion was now signaled
by one of the cues (i.e., A or B) that in a previous temporal context
(Phase 1 or Phase 2) had predicted nothing. This miscuing trial had
the purpose of (a) cuing the temporal context of either Phase 1 or
Phase 2 and (b) making participants integrate the current informa-
tion with the representation of the older temporal context, as the
cue that they would expect to predict nothing was now a predictor
for danger. If our hypothesis is correct and participants do update
the representation of an old temporal context as they get new
information, this update should be reflected in a corresponding
behavioral change at test that should affect other stimuli trained in
that context (in this case, Cues X or Y). Thus, during a subsequent
test trial, participants should no longer respond to the target cues,
X and Y, as they had learned previously. Instead, they should
respond to them in just the opposite way. It is important to note
that this change should occur only if the cue presented in Phase 3
had been trained in the same temporal context as the target cue.
Thus, half of the participants in each group were tested with the
target cue from Phase 1, X, whereas the other half were tested with
the target cue from Phase 2, Y. In total, there were four groups
(M1T1, M1T2, M2T1, and M2T2) as a function of whether they
received miscuing with a cue from Phase 1 (M1) or Phase 2 (M2)
and as a function of whether they were tested with a target cue
from Phase 1 (T1) or Phase 2 (T2).
Results and Discussion
In associative learning studies in which the target behavior
consists of suppressing ongoing behavior, a suppression ratio is
normally calculated so that the number of responses (i.e., in this
case, presses of the space bar) that occur during the cue presenta-
tion can be compared with the base rate of responses immediately
preceding the presentation of the cue. The dependent variable is
thus computed as the ratio of (responses during cue)/(responses
during cue responses during an equivalent time period preced-
ing the cue).
The critical result that we expect in this experiment is not a mere
difference between the groups at test, as in the previous experi-
ments. Instead, we expect the miscuing trial to change the manner
in which participants respond to the target cue relative to the last
3
It should be noted that in this experiment, O2 is not a positive outcome,
as in the previous experiments. However, given that the participants’
motivation throughout this experiment is to maintain a stable and high rate
of responding, the fact that some cues are followed by nothing means that
participants should keep responding during those cues.
668 MATUTE, LIPP, VADILLO, AND HUMPHREYS
training trial. Thus, the result of interest is the degree to which
responding at test differs from responding during the last training
trial, which, as stated above, is the reason why these two trials
were given longer durations than the other ones, thereby allowing
for the critical suppression ratios to be assessed. Figure 4 shows
these results.
As can be seen in Figure 4, a reduction in suppression from the
last training trial to test was evident when the stimuli presented
during the miscuing trial and during test had been trained together
in the same phase but not when the stimuli were trained in different
phases. This impression was confirmed in a 2 (last training trial vs.
test trial) 2 (miscuing cue from Phase 1 vs. miscuing cue from
Phase 2) 2 (target cue from Phase 1 vs. target cue from Phase
2) factorial analysis of variance, which yielded a main effect for
trial, F(1, 60) 18.09, p.001, and Miscuing Cue Test Cue,
F(1, 60) 13.18, p.001, and Trial Miscuing Cue Test Cue
interactions, F(1, 60) 5.45, p.05. Follow-up tests found less
suppression during test than during the last training trial in partic-
ipants presented with miscuing and test cues that had been trained
together in Phase 1, t(60) 4.33, or Phase 2, t(60) 2.80. There
was no effect of miscuing on suppression during test if the mis-
cuing cue and the test cue had been trained in different phases,
both ts2.1. This suggests that the miscuing trial led to the
updating of other memories and predictions associated with cues
trained in the same temporal context but left behavior dependent
on cues trained in different temporal contexts untouched.
General Discussion
The results of the present experiments show that events that are
trained in the same phase of the experiment can activate the mental
representation of each other even when there is no direct contiguity
or contingency between them. The most widely accepted theories
of associative learning and memory do not anticipate this result
(e.g., Raaijmakers & Shiffrin, 1980; Rescorla & Wagner, 1972).
The results of our Experiments 1–3 were predicted by the TCM
model of Howard and Kahana (2002; Sederberg et al., 2008). They
were obtained using a task that is very different from those that
have traditionally been used to test this model. More specifically,
we used a methodology that comes from the associative learning
tradition, which speaks to the notion of a common framework for
episodic memory and associative learning. In essence, the TCM
explains the results of Experiments 1–3 by assuming that people
are continuously forming and updating temporal contexts. To do
so, people use all stimuli that are available at any given time. Each
stimulus becomes associatively linked to these contexts, and then,
as the contexts evolve with time and get linked to other stimuli, the
older stimuli, if presented again, can be used to cue the old context
along with all of the information and memories from the original
training episode. The results of Experiments 1–3 support this view.
Priming a given temporal context at test with one of the stimuli
that had been trained in that context made participants retrieve the
meaning of the target cue trained in that context. Thus, responding
to the target cue at test was dependent on the retrieved context
information.
Results of Experiment 4 support that view as well but go beyond
the predictions of the TCM. Here, we used a cue from a particular
temporal context to retrieve the memory of that context. Once
participants had retrieved that memory and were, therefore, ex-
pecting that a given outcome followed the cue, we presented the
other outcome. This led participants to update the representation of
the old temporal context. As a consequence, they also changed
their behavior when the second stimulus from that context was
presented at test. This behavioral change did not transfer to stimuli
that had been trained in a different temporal context. Note that the
TCM predicts only part of this. It does predict the update of the
temporal context representation, but it does not specify whether
and how this should affect the behavior to the other stimulus.
Indeed, these results suggest that the episodes were created by the
sets of stimuli that were trained in a temporal window. Moreover,
the episode is encoded in such a way that retrieving it and altering
what traditional theories of associative learning would regard as a
single independent association trained in that episode can also alter
other associations trained independently in the same episode.
We are not claiming that this has to be an automatic process of
episodic restructuring, but, functionally, it works like it. The
possibility that deliberate processes might have caused this result
instead should be subject to future research. Indeed, it could be
argued that our participants were creating different rules for each
phase and then going back to these rules on retrieval of one or the
other context so that they could then flexibly use or modify these
rules as needed. Note that this rule-based account does not depart
from the main point we are making. That is, the rules that apply
during a given time period can also create a context or episode.
What is clear, however, is that regardless of whether this process
is deliberate or automatic, the results show that there is a psycho-
logical reality to the contexts in Phase 1 and Phase 2. Otherwise,
the disconfirmation of one expectation during miscuing should
have changed all other expectations in the experiment. The fact
that it changed the expectation only for the second item that had
Figure 4. Mean suppression ratios during the last trial of training and the
test trial of Experiment 4 as a function of whether the miscuing and test cue
had been trained in the same temporal context (M1/T1, M2/T2) or in
different temporal contexts (M1/T2, M2/T1; error bars represent the stan-
dard error of the mean). Note that this figure needs to be read in the
opposite way as the previous ones. Values closer to zero indicate stronger
suppression (with zero being complete suppression of ongoing behavior in
response to the stimulus, therefore, a very strong response). In contrast,
values closer to .500 indicate that the number of responses during the cue
differs little from the number of responses during an equivalent baseline
period. Thus, .500 indicates null responding to the stimulus.
669
EPISODIC MEMORY AND ASSOCIATIVE LEARNING
occurred in that phase but not the expectation for the items that
occurred in the other phase supports this psychological reality.
This result was not predicted by any of the standard models of
learning or memory and is a good example of how an integration
of the learning and memory traditions can produce fruitful ad-
vances.
Our experiments show that people can update their representa-
tion of a retrieved temporal context as a function of new informa-
tion that is available in the current situation. Experiment 4 dem-
onstrated that retrieving an old temporal context and updating it
with the changes that occurred in the current situation helped
participants adjust their future behavior, even though this adapta-
tion required responding to the target cue in just the opposite way
as was done during the acquisition stage. This suggests the exis-
tence of a very direct link between remembering the past and
planning for the future, as has already been suggested by many
other researchers (e.g., Schacter et al., 2007; Suddendorf & Cor-
ballis, 1997, 2007; Tulving, 2005). That is, memory provides
people with knowledge and experiences that feed learning, and this
integration of memory and learning helps people plan for the
future and adapt future behavior to the most likely changes in the
environment.
Bouton’s theory of contexts in associative learning cannot ex-
plain our results. It predicts that only the associations acquired
second (Bouton, 1997; or only the inhibitory associations; see
Bouton, 1993) become context specific. This means that in our
experiments, only the associations acquired second (or the inhib-
itory associations) should have been subject to selective retrieval.
This was not the case. Rosas et al.’s (2006) revision of Bouton’s
theory might perhaps be better equipped to assume that either the
associations acquired first or the associations acquired second may
be subject to selective retrieval. This theory, however, is as silent
as Bouton’s concerning how temporal contexts may be created and
represented.
It should be noted, however, that several components of the
TCM were foreshadowed in the associative learning literature.
Bouton’s (1993, 1997) theory assumes time-dependent changes in
context (although the temporal context according to Bouton is
provided by time itself rather than explicit stimuli that occur within
a period of time) and Capaldi’s (1994) sequential theory of ex-
tinction assumes that trials can provide contexts for other trials in
a series (see also Bouton, Woods, & Pinen˜o, 2004). From this
perspective, it could be argued that what was manipulated in the
present experiments was not time but explicit stimuli and trials.
Admittedly, we cannot manipulate time per se and travel back in
time to a previous phase. For this reason, we manipulated stimuli
and trials instead, under the assumption that, as suggested by the
TCM, the stimuli that are present at a given time are what define
that temporal context. This, we believe, is a powerful approach. It
allows specific predictions concerning (backward and forward)
mental time travel. It makes a clear prediction about how contexts
should be represented, how they should change with time, and how
their representation might be accessed. The TCM also makes a
specific prediction about associations being formed between the
stimuli that occur in a given temporal context. Our results clearly
confirm this prediction of the TCM.
However, it may be possible that some other mechanism differ-
ent from those proposed by the TCM be involved. One alternative
would be to assume that Cues A and B defined two distinct
physical contexts. Physical contexts, however, are generally de-
fined by cues that are continuously present and that do not change
from one trial to the next. Indeed, physical contexts are assumed to
get compound, simultaneous training with the target cues accord-
ing to mainstream theories of learning (e.g., Bouton, 1993; Miller
& Matzel, 1988; Pearce, 1994; Rescorla & Wagner, 1972). This
was not the case in our experiments. The notion of temporal (rather
than physical) contexts seems to be, at least in principle, a more
parsimonious explanation for our results.
A second alternative is to assume that the majority of the stimuli
that are trained in the same phase are more contiguous (and
contingent) to each other than to the majority of the stimuli trained
in a different phase. This would allow the formation of direct
associations between the stimuli that were trained in the same
phase, on the basis of their relative contiguity and contingency to
each other as compared with stimuli trained in different phases.
That is, contiguity and contingency could probably be relative to
the temporal window that the participant is using at a given time.
It should be noted, however, that the design that we used
explicitly assured that the training of, say, Target Cue X with
Outcome O1 was independent of the training of the second cue that
was trained in the same phase, say, Target Cue A with Outcome
O2. The order in which those two types of trials occurred in each
phase was randomized. Sometimes the X–O1 pairings preceded
the pairings of A–O2, sometimes this order was reversed, yet at
other times several trials of X–O1 or A–O2 occurred in succession.
In essence, the design that we used can be thought of as identical
to that of a typical discrimination learning design in which a given
cue is followed by an outcome and a different cue is followed by
another outcome, with the order of these two trial types random-
ized (e.g., Arcediano et al., 1996; Chiszar & Spear, 1969; Vila,
Romero, & Rosas, 2002). It is also identical to many within-
subject designs that train several independent conditions, such as
an experimental condition and a control condition, by randomly
intermixing different types of learning trials during the same
phase. In those cases, traditional learning theories would hold that
no contiguity or contingency exists between the different trial
types (i.e., between the experimental and control conditions) and
that no associations are formed between them.
Nevertheless, some might argue that even such weak and rela-
tive contiguity and contingency might be sufficient to create as-
sociations. This view would require specification of the conditions
under which relative contiguity would be sufficient to produce
associations among stimuli and how those associations would
develop. For instance, why should the latest trials of Phase 1
become associated with other trials of Phase 1 rather than with the
more contiguous trials presented at the beginning of Phase 2? In
essence, developing such a proposal would be very similar to
suggesting that participants created temporal contexts that coin-
cided with those designed by the experimenters. The advantage
that we see in applying the TCM is that it has already formalized
those ideas and can be used to derive testable predictions.
Although we have used two different tasks that differed in the
dependent variable and several other details, it is also true that both
tasks involved the gain or loss of points in a computer game. It
would be interesting to know whether similar results can be
observed in other situations that do not require the gain or loss of
points. Indeed, there is a tradition within the associative learning
literature that assumes that when different stimuli are trained in the
670 MATUTE, LIPP, VADILLO, AND HUMPHREYS
same way or are associated with the same set of objects, they
become functionally equivalent. This is called mediated general-
ization and it has been used to explain concept formation and
different aspects of language acquisition (Keller & Schoenfeld,
1950; Wasserman, DeVolder, & Coppage, 1992). This account
could perhaps be used to explain the results of our Experiment 4.
Given that A and X were trained in the same temporal context
(here playing the role of the common set of objects), this kind of
mediated generalization might explain why the new information
about A generalizes to X but not to other cues trained in other
temporal contexts. Our view, however, is that this is a description
of the results more than an explanation. Moreover, this account
could also accommodate the opposite result: If whatever is learned
about A is generalized to X, we should expect less miscuing in
Experiment 4, which is contrary to the results of the experiment.
Our results suggest that participants were able to construct
different temporal windows, one for each of the phases designed
by the experimenters. Exploring how they do it seems to be a very
exciting venue for future research. TCM provides a possible ex-
planation for temporal context formation, and experiments from
the associative tradition can shed light on how this process takes
place. Indeed, the spontaneous separation of two experimental
phases into separate temporal contexts has often been observed in
human learning experiments (e.g., Alvarado, Jara, Vila, & Rosas,
2006; Collins & Shanks, 2002; Lipp & Purkis, 2006; Matute,
Vegas, & De Marez, 2002; Vadillo, Vegas, & Matute, 2004). It is
important to note that although those experiments were not explic-
itly designed to test this particular question, many of them showed
that participants can use subtle, nonphysical cues to construct
efficient and flexible temporal contexts (Collins & Shanks, 2002;
Lipp & Purkis, 2006; Matute et al., 2002; Vadillo et al., 2004). For
instance, Vadillo et al. (2004) used a contingency judgment task in
which the cues and outcomes were fictitious medicines and aller-
gic reactions, respectively. These tasks are very different from the
ones used in the present article (e.g., the dependent variable is the
number of presses of the space bar in the present research, whereas
it is a subjective judgment of causality or prediction in contingency
judgment studies). Participants were required to provide their
judgment either in every trial or at the end of training. The
(physical) context in which these judgments occurred was manip-
ulated orthogonally.
It is interesting that the results of those experiments showed that
changing the frequency with which the judgments were requested
during testing with respect to training (e.g., changing from no
request during training to a request at test) produced an effect that
was similar to changing the physical context between training and
testing. That is, participants created flexible temporal contexts as a
function of when and how often they were requested to provide
their judgments. When the physical contexts did not change and
participants were asked to provide only one judgment at the end of
training, participants assumed that the relevant temporal context to
consider was that of the entire experiment, that is, the one includ-
ing the two learning phases. When the physical contexts did not
change but participants were asked to provide frequent judgments
through the training phases, participants assumed that the relevant
temporal context to consider at test was that of the most recent
phase. These results suggest that participants used the differences
in the frequency with which they were asked to provide their
judgments in the different phases of the experiment as a subtle clue
that helped them to decide when exactly to segment the continuous
flow of information, so that they could create the different tem-
poral contexts or episodes (see also Kurby & Zacks, 2008, for
discussion on how segmentation of events may take place).
The present results should also be linked to the experimental
tradition of associative interference in paired associate learning.
There is substantial evidence that in the AB AC paradigm, the first
list association can coexist with the second list association. When
participants are asked to recall the items associated with A in the
first list and in the second list, the recall probabilities appear to be
independent (Martin, 1971). Thus, there is little or no evidence that
there is any destructive interference (Dyne, Humphreys, Bain, &
Pike, 1990), and it has been shown that this pattern of results can
be replicated by a model that depends heavily on context to keep
the two responses separate (Chappell & Humphreys, 1994). This
lack of evidence for destructive interference is compatible with the
account presented in this article and with associative learning
accounts of the effect of extinction training and contexts on ac-
quired associations (Bouton, 1993, 1997).
In summary, our experiments add to the convergent evidence
originating from findings in the memory tradition using paired
associate paradigms and word lists (e.g., Humphreys et al., 2009;
Humphreys, Bain, & Pike, 1989; Sederberg et al., 2008) and from
the associative learning tradition (Bouton, 1993, 1997; Capaldi,
1994) by providing evidence (a) for the retrieval of the represen-
tation of a temporal context through a cue trained in that context
(as opposed to retrieving it using instructions; Humphreys et al.,
2009) and (b) for the spontaneous creation of a context or an
episode through the stimulus contingencies found in a temporal
window. Also, using nonverbal behavior as the dependent variable
provides interesting confirming evidence that speaks of the gen-
erality of these effects across different tasks and measures. These
results indicate the promise of our approach to integrate concepts
and procedures from the associative learning and episodic memory
traditions to advance the understanding of the basic processes
involved in using past experiences for adaptive behavior in the
present and future.
References
Alvarado, A., Jara, E., Vila, J., & Rosas, J. M. (2006). Time and order
effects on causal learning. Learning and Motivation, 37, 324–345.
doi:10.1016/j.lmot.2005.11.001
Arcediano, F., Ortega, N., & Matute, H. (1996). A behavioural preparation
for the study of human Pavlovian conditioning. The Quarterly Journal of
Experimental Psychology: Comparative and Physiological Psychology,
49(B), 270–283.
Bouton, M. E. (1993). Context, time, and memory retrieval in the inter-
ference paradigms of Pavlovian conditioning. Psychological Bulletin,
114, 80–99. doi:10.1037/0033-2909.114.1.80
Bouton, M. E. (1997). Signals for whether versus when an event will occur.
In M. E. Bouton & M. S. Fanselow (Eds.), Learning, motivation, and
cognition: The functional behaviorism of Robert C. Bolles (pp. 385–
409). Washington, DC: American Psychological Association. doi:
10.1037/10223-019
Bouton, M. E., Woods, A. M., & Pinen˜o, O. (2004). Occasional reinforced
trials during extinction can slow the rate of rapid reacquisition. Learning
and Motivation, 35, 371–390. doi:10.1016/j.lmot.2004.05.001
Brooks, D. C., & Bouton, M. E. (1993). A retrieval cue for extinction
attenuates spontaneous recovery. Journal of Experimental Psychology:
Animal Behavior Processes, 19, 77–89. doi:10.1037/0097-7403.19.1.77
671
EPISODIC MEMORY AND ASSOCIATIVE LEARNING
Capaldi, E. (1994). The sequential view: From rapidly fading stimulus
traces to the organization of memory and the abstract concept of number.
Psychonomic Bulletin & Review, 1, 156–181. doi:10.3758/BF03200771
Chappell, M., & Humphreys, M. S. (1994). An auto-associative neural
network for sparse representations: Analysis and application to models
of recognition and cued recall. Psychological Review, 101, 103–128.
doi:10.1037/0033-295X.101.1.103
Chiszar, D. A., & Spear, N. E. (1969). Stimulus change reversal learning,
and retention in the rat. Journal of Comparative and Physiological
Psychology, 69, 190–195. doi:10.1037/h0027947
Collins, D. J., & Shanks, D. R. (2002). Momentary and integrative re-
sponse strategies in causal judgment. Memory & Cognition, 30, 1138
1147. doi:10.3758/BF03194331
Costa, D. S. J., & Boakes, R. A. (2011). Varying temporal contiguity and
interference in a human avoidance task. Journal of Experimental Psy-
chology: Animal Behavior Processes, 37, 71–78. doi:10.1037/a0021192
Davis, O. C., Geller, A. S., Rizzuto, D. S., & Kahana, M. J. (2008).
Temporal associative processes revealed by intrusions in paired-
associate recall. Psychonomic Bulletin & Review, 15, 6469. doi:
10.3758/PBR.15.1.64
Dyne, A. M., Humphreys, M. S., Bain, J. D., & Pike, R. (1990). Associa-
tive interference effects in recognition and recall. Journal of Experimen-
tal Psychology: Learning, Memory, and Cognition, 16, 813–824. doi:
10.1037/0278-7393.16.5.813
Franssen, M., Clarysse, J., Beckers, T., van Vooren, P., & Baeyens, F.
(2010). A free software package for a human online-conditioned sup-
pression preparation. Behavior Research Methods, 42, 311–317. doi:
10.3758/BRM.42.1.311
Garcia, J., Ervin, F. R., & Koelling, R. A. (1966). Learning with prolonged
delay of reinforcement. Psychonomic Science, 5, 121–122.
Gawronski, B., Rydell, R. J., Vervliet, B., & De Houwer, J. (2010).
Generalization versus contextualization in automatic evaluation. Journal
of Experimental Psychology: General, 139, 683–701. doi:10.1037/
a0020315
Gunther, L. M., Miller, R. R., & Matute, H. (1997). CSs and USs: What’s
the difference? Journal of Experimental Psychology: Animal Behavior
Processes, 23, 15–30. doi:10.1037/0097-7403.23.1.15
Howard, M. W., Jing, B., Rao, V. A., Provyn, J. P., & Datey, A. V. (2009).
Bridging the gap: Transitive associations between items presented in
similar temporal contexts. Journal of Experimental Psychology: Learn-
ing, Memory, and Cognition, 35, 391–407. doi:10.1037/a0015002
Howard, M. W., & Kahana, M. J. (1999). Contextual variability and serial
position effects in free recall. Journal of Experimental Psychology:
Learning, Memory, and Cognition, 25, 923–941. doi:10.1037/0278-
7393.25.4.923
Howard, M. W., & Kahana, M. J. (2002). A distributed representation of
temporal context. Journal of Mathematical Psychology, 46, 269–299.
doi:10.1006/jmps.2001.1388
Howard, M. W., Youker, T. E., & Venkatadass, V. (2008). The persistence
of memory: Contiguity effects across several minutes. Psychonomic
Bulletin & Review, 15, 5863. doi:10.3758/PBR.15.1.58
Humphreys, M. S., Bain, J. D., & Pike, R. (1989). Different ways to cue a
coherent memory system: A theory for episodic, semantic and proce-
dural tasks. Psychological Review, 96, 208–233. doi:10.1037/0033-
295X.96.2.208
Humphreys, M. S., Murray, K. L., & Maguire, A. M. (2009). Context and
control operations used in accessing list-specific, generalized, and se-
mantic memories. Cognitive Psychology, 58, 311–337. doi:10.1016/
j.cogpsych.2008.10.001
Keller, F. S., & Schoenfeld, W. N. (1950). Principles of psychology. New
York, NY: Appleton-Century-Crofts.
Kraut, R., Olson, J., Banaji, M., Bruckman, A., Cohen, J., & Couper, M.
(2004). Psychological research online: Report of Board of Scientific
Affairs’ Advisory Group on the conduct of research on the Internet.
American Psychologist, 59, 105–117. doi:10.1037/0003-066X.59.2.105
Kurby, C. A., & Zacks, J. M. (2008). Segmentation in the perception and
memory of events. Trends in Cognitive Sciences, 12, 72–79. doi:
10.1016/j.tics.2007.11.004
Lipp, O. V., & Dal Santo, L. A. (2002). Cue competition between elemen-
tary trained stimuli: US miscuing, interference, and US omission. Learn-
ing and Motivation, 33, 327–346. doi:10.1016/S0023-9690(02)00001-2
Lipp, O. V., & Purkis, H. M. (2006). The effects of assessment type on
verbal ratings of conditional stimulus valence and contingency judg-
ments: Implications for the extinction of evaluative learning. Journal of
Experimental Psychology: Animal Behavior Processes, 32, 431–440.
doi:10.1037/0097-7403.32.4.431
Lipp, O. V., Siddle, D. A. T., & Dall, P. J. (1993). Effects of miscuing
on Pavlovian conditioned responding and on probe reaction time.
Australian Journal of Psychology, 45, 161–167. doi:10.1080/
00049539308259134
Martin, E. (1971). Verbal learning theory and independent retrieval phe-
nomena. Psychological Review, 78, 314–332. doi:10.1037/h0031030
Matute, H., Vadillo, M. A., & Ba´rcena, R. (2007). Web-based experiment
control software for research and teaching on human learning. Behavior
Research Methods, 39, 689693. doi:10.3758/BF03193041
Matute, H., Vegas, S., & De Marez, P. J. (2002). Flexible use of recent
information in causal and predictive judgments. Journal of Experimental
Psychology: Learning, Memory, and Cognition, 28, 714–725. doi:
10.1037/0278-7393.28.4.714
Miller, R. R., & Matzel, L. D. (1988). The comparator hypothesis: A
response rule for the expression of associations. In G. H. Bower (Ed.),
The psychology of learning and motivation (Vol. 22, pp. 51–92). San
Diego, CA: Academic Press.
Pavlov, I. P. (1927). Conditioned reflexes. London, England: Clarendon
Press.
Pearce, J. M. (1994). Similarity and discrimination: A selective review and
a connectionist model. Psychological Review, 101, 587–607. doi:
10.1037/0033-295X.101.4.587
Pinen˜o, O., Ortega, N., & Matute, H. (2000). The relative activation of the
associations modulates interference between elementally-trained cues.
Learning and Motivation, 31, 128–152. doi:10.1006/lmot.1999.1047
Raaijmakers, J. G. W., & Shiffrin, R. M. (1980). SAM: A theory of
probabilistic search of associative memory. In G. H. Bower (Ed.), The
psychology of learning and motivation: Advances in research and theory
(Vol.14, pp. 207–262). New York, NY: Academic Press.
Raaijmakers, J. G. W., & Shiffrin, R. M. (1981). Search of associative
memory. Psychological Review, 88, 93–134. doi:10.1037/0033-
295X.88.2.93
Ranganath, C. (2010). Binding items and contexts: The cognitive neuro-
science of episodic memory. Current Directions in Psychological Sci-
ence, 19, 131–137. doi:10.1177/0963721410368805
Reips, U.-D. (2002). Internet-based psychological experimenting: Five dos
and five don’ts. Social Science Computer Review, 20, 241–249.
Rescorla, R. A., & Wagner, A. R. (1972). A theory of Pavlovian condi-
tioning: Variations in the effectiveness of reinforcement and nonrein-
forcement. In A. H. Black & W. F. Prokasy (Eds.), Classical condition-
ing: II. Current research and theory (pp. 64–99). New York, NY:
Appleton-Century-Crofts.
Rosas, J. M., & Callejas-Aguilera, J. E. (2006). Context switch effects on
acquisition and extinction in human predictive learning. Journal of
Experimental Psychology: Learning, Memory, and Cognition, 32, 461–
474. doi:10.1037/0278-7393.32.3.461
Rosas, J. M., Callejas-Aguilera, J. E., Ramos-A
´lvarez, M. M., &
Ferna´ndez-Abad, M. J. (2006). Revision of retrieval theory of forgetting:
What does make information context-specific? International Journal of
Psychology and Psychological Therapy, 6, 147–166.
Schacter, D. L., Addis, D. R., & Buckne, R. L. (2007). Remembering the
672 MATUTE, LIPP, VADILLO, AND HUMPHREYS
past to imagine the future: The prospective brain. Nature Reviews
Neuroscience, 8, 657–661. doi:10.1038/nrn2213
Schwartz, G., Howard, M. W., Jing, B., & Kahana, M. J. (2005). Shadows
of the past: Temporal retrieval effects in recognition memory. Psycho-
logical Science, 16, 898–904. doi:10.1111/j.1467-9280.2005.01634.x
Sederberg, P. B., Howard, M. W., & Kahana, M. J. (2008). A context-
based theory of recency and contiguity in free recall. Psychological
Review, 115, 893–912. doi:10.1037/a0013396
Shanks, D. R., & Dickinson, A. (1987). Associative accounts of causality
judgment. In G. H. Bower (Ed.), The psychology of learning and
motivation (Vol. 21, pp. 229–261). San Diego, CA: Academic Press.
Suddendorf, T., & Corballis, M. C. (1997). Mental time travel and the
evolution of the human mind. Genetic, Social, and General Psychology
Monographs, 123, 133–167. Retrieved from http://cogprints.org/725/
Suddendorf, T., & Corballis, M. C. (2007). The evolution of foresight:
What is mental time travel and is it unique to humans? Behavioral and
Brain Sciences, 30, 299–313. doi:10.1017/S0140525X07001975
Tulving, E. (2005). Episodic memory and autonoesis: Uniquely human? In
H. S. Terrace & J. Metcalfe (Eds.), The missing link in cognition:
Origins of self-reflective consciousness (pp. 3–56). New York, NY:
Oxford University Press.
Usher, M., & McClelland, J. L. (2001). The time course of perceptual
choice: The leaky, competing accumulator model. Psychological Re-
view, 108, 550–592. doi:10.1037/0033-295X.108.3.550
Vadillo, M. A., Ba´rcena, R., & Matute, H. (2006). The Internet as a
research tool in the study of associative learning: An example from
overshadowing. Behavioural Processes, 73, 3640. doi:10.1016/
j.beproc.2006.01.014
Vadillo, M. A., & Matute, H. (2011). Further evidence on the validity of
Web-based research on associative learning: Augmentation in a predic-
tive learning task. Computers in Human Behavior, 27, 750–754. doi:
10.1016/j.chb.2010.10.020
Vadillo, M. A., Vegas, S., & Matute, H. (2004). Frequency of judgment as
a context-like determinant of predictive judgments. Memory & Cogni-
tion, 32, 1065–1075. doi:10.3758/BF03196882
Vila, J., Romero, M., & Rosas, J. M. (2002). Retroactive interference after
discrimination reversal decreases following temporal and physical con-
text changes in human subjects. Behavioural Processes, 59, 47–54.
doi:10.1016/S0376-6357(02)00063-3
Wasserman, E. A. (1990). Detecting response–outcome relations: Toward
an understanding of the causal texture of the environment. In G. H.
Bower (Ed.), The psychology of learning and motivation (Vol. 26, pp.
27–82). San Diego, CA: Academic Press.
Wasserman, E. A., DeVolder, C. L., & Coppage, D. J. (1992). Non-
similarity-based conceptualization in pigeons via secondary or mediated
generalization. Psychological Science, 3, 374–379. doi:10.1111/j.1467-
9280.1992.tb00050.x
Received August 31, 2010
Revision received March 30, 2011
Accepted April 1, 2011
673
EPISODIC MEMORY AND ASSOCIATIVE LEARNING
... This conceptualization is consistent with mechanisms proposed in perceptual and memory research, where it is suggested that the continuous flow of information is automatically segmented and structured into discrete events (Zacks et al., 2007). Matute, Lipp, Vadillo, and Humphreys (2011) have similarly invoked the concept of temporal contexts in research on associative learning. They found that participants spontaneously (i.e., without instructions) structure learning procedures by creating temporal contexts. ...
... We created different contexts for the first and second phase to standardize participants' temporal organization of the learning history (Matute et al., 2011) and facilitate later reference to each phase in targeted questions about particular portions of the learning procedure. The context features-background color and CS position-were randomly assigned to the first or second phase for each participant. ...
... Hence, the standard end-of-study evaluation assessment may act as context change, affect judgment strategies, and produce renewal effects. Moreover, we assume that in the absence of explicitly induced context changes, participants spontaneously generate and use temporal contexts to structure the incoming information and that these contexts can affect behavior (Matute et al., 2011;Zacks et al., 2007). Due to the use of external contexts, contextualization in our study may have been more pronounced but-we assume-not qualitatively different from previous studies. ...
Article
Full-text available
Evaluative conditioning (EC) is a change in liking of neutral conditioned stimuli (CS) following pairings with positive or negative stimuli (unconditioned stimulus, US). A dissociation has been reported between US expectancy and CS evaluation in extinction learning: When CSs are presented alone subsequent to CS-US pairings, participants cease to expect USs but continue to exhibit EC effects. This dissociation is typically interpreted as demonstration that EC is resistant to extinction, and consequently, that EC is driven by a distinct learning process. We tested whether expectancy-liking dissociations are instead caused by different judgment strategies afforded by the dependent measures: CS evaluations are by default integrative judgments-summaries of large portions of the learning history-whereas US expectancy reflects momentary judgments that focus on recent events. In a counterconditioning and two extinction experiments, we eliminated the expectancy-liking dissociation by inducing nondefault momentary evaluative judgments, and demonstrated a reversed dissociation when we additionally induced nondefault integrative expectancy judgments. Our findings corroborated a priori predictions derived from the formal memory model MINERVA 2. Hence, dissociations between US expectancy and CS evaluation are consistent with a single-process learning model; they reflect different summaries of the learning history. (PsycINFO Database Record
... The contents of memory constantly inform cognition and behavior, but different affordances of dependent measures may affect how information is retrieved and integrated in a specific situation and for a specific purpose (e.g., Matute, Lipp, Vadillo, & Humphreys, 2011). This section addresses dissociations between dependent measures that may arise when comparing direct versus indirect measures (and model parameters, as in the processdissociation approach), or evaluative versus expectancy judgments. ...
... On the other hand, US expectancy judgments are typically understood as momentary predictions of an imminent event (e.g., "given this CS, do you expect a US to be presented?") and should therefore more strongly rely on recent trends (Collins & Shanks, 2002;Matute et al., 2011). As a consequence of these defaults, a CS that has signaled negative events in the past is still evaluated negatively overall, even if recent encounters were neutral and negative events are no longer expected to occur in its presence (Lipp & Purkis, 2006). ...
Article
Full-text available
The article proposes a view of evaluative conditioning (EC) as resulting from judgments based on learning instances stored in memory. It is based on the formal episodic memory model MINERVA 2. Additional assumptions specify how the information retrieved from memory is used to inform specific evaluative dependent measures. The present approach goes beyond previous accounts in that it uses a well-specified formal model of episodic memory; it is however more limited in scope as it aims to explain EC phenomena that do not involve reasoning processes. The article illustrates how the memory-based-judgment view accounts for several empirical findings in the EC literature that are often discussed as evidence for dual-process models of attitude learning. It sketches novel predictions, discusses limitations of the present approach, and identifies challenges and opportunities for its future development.
... The contents of memory constantly inform cognition and behavior, but different affordances of dependent measures may affect how information is retrieved and integrated in a specific situation and for a specific purpose (e.g., Matute, Lipp, Vadillo, & Humphreys, 2011). This section addresses dissociations between dependent measures that may arise when comparing direct versus indirect measures (and model parameters, as in the process-dissociation approach), or evaluative versus expectancy judgments. ...
... On the other hand, US expectancy judgments are typically understood as momentary predictions of an imminent event (e.g., 'given this CS, do you expect a US to be presented?') and should therefore more strongly rely on recent trends (Collins & Shanks, 2002;Matute et al., 2011). As a consequence of these defaults, a CS that has been signaling negative events in the past is still evaluated negatively overall, even if recent encounters were neutral and negative events are no longer expected to occur in its presence (Lipp & Purkis, 2006). ...
Preprint
Full-text available
The article proposes a view of evaluative conditioning (EC) as resulting from judgments based on learning instances stored in memory. It is based on the formal episodic memory model MINERVA 2. Additional assumptions specify how the information retrieved from memory is used to inform specific evaluative dependent measures. The present approach goes beyond previous accounts in that it uses a well-specified formal model of episodic memory; it is however more limited in scope as it aims at explaining EC phenomena that do not involve reasoning processes. The article illustrates how the memory-based-judgment view accounts for several empirical findings in the EC literature that are often discussed as evidence for dual-process models of attitude learning. It sketches novel predictions, discusses limitations of the present approach, and identifies challenges and opportunities for its future development.
... Participants were asked to provide their 'best guess' if unsure how to evaluate the displayed trigram, as a trial would not progress until the slider had been interacted with. Intervals between evaluations were jittered at 150-350 ms to minimize temporal conditioning artefacts (Matute, Lipp, Vadillo, & Humphreys, 2011 Completion of 80 conditioning trials was followed by a second round of can preference tests and CS evaluations (Figure 1, Phases 5 and 6). ...
... Occasionally reinforced extinction training not only enables the formation of a CS-noUS association, but also supports learning that non-reinforced trials can follow reinforced trials, thus slowing down reacquisition and providing potential protection against relapse. Dunsmoor et al. (2018) suggested a mechanism derived from research on episodic memory -event segmentation -that may account for the effects of additional US presentations during extinction (for a similar account see Matute, Lipp, Vadillo, & Humphreys, 2011). Event segmentation has been proposed as a mechanism that protects memory items against interference from other items encountered in close temporal proximity by chunking items within one episode together and including boundaries that separate successive episodes (Ezzyat & Davachi, 2011). ...
Article
Since Watson and Rayner's (1920) initial demonstration that human fear can be learned by means of Pavlovian conditioning, neuroscientific and behavioral studies have provided a thorough understanding of fear acquisition. Less is known about the manner in which we can harness insights from Pavlovian conditioning research to reduce fears and, most importantly, make the reduction of fear lasting and resistant against relapse. The current paper reviews three manipulations that have shown promise in achieving a reduction of conditional fear that is more resistant to relapse than is the reduction of conditional fear after standard extinction: novelty-facilitated extinction training, presentation of conditional-unconditional stimulus pairings or of unpaired unconditional stimuli during extinction, and extinction with additional stimuli that are similar to the original conditional stimuli. It summarizes past research involving human and non-human animal subjects and highlights knowledge gaps in the current literature. Moreover, it discusses potential mechanisms that mediate the reduction of fear seen as a result of these manipulations in an attempt to enhance our understanding of what renders fear extinction less vulnerable to the known pathways to fear relapse. It is hoped that this review will contribute to the achievement of the goal that was denied to Watson and Rayner, the development of experimental techniques that can be utilized to remove conditioned emotional responses permanently.
... Subjects continued evaluating CS twice a week for 6 weeks. Earlier proof-of-concept studies revealed that varying the interval between CS evaluations reduced the likelihood of control by temporal cues, as subjects could not predict which days they would be probed for CS evaluations (e.g., Matute et al., 2011). After the final evaluation, subjects returned to the lab to complete a set of card-sorting tasks that were unrelated to the present investigation. ...
Article
Full-text available
A meaningless symbol that repeatedly co-occurs with emotionally salient faces (US) can transform into a valenced symbol (CS). US-to-CS valence transformations have been observed for CS that have been directly (US>CS0) and indirectly (US>CS0>CS1>CS2) linked with face US. The structure of a US>CS0>CS1>CS2 series may be conceptualized in terms of ‘nodal distance’, where CS0, CS1 and CS2 are 0, 1 and 2 nodes from the US respectively. Increasing nodal distance between an evaluated CS and it’s linked US can reduce magnitude of observed CS valence transformations. We explored currently whether nodal distance can influence CS valence extinction, which describes reductions in CS valence following repeated exposures to CS without any accompanying US. In our study, faces with happy/neutral/sad expressions (US) were directly linked with nonsense words (US>CS0). The directly linked CS0 were concurrently linked with other words (CS0>CS1, CS1>CS2). Subjects evaluated all stimuli before and after conditioning, then continued to provide CS evaluations twice a week for six weeks. Bayesian factors provided credible evidence for the transformation and extinction of CS valences that were 0 and 1 nodes from US (all BF10’s > 100). The variability across post-conditioning CS evaluations provides indirect evidence for context-sensitive/propositional and structural/associative operations during CS evaluations.
... This relates to the issue of complete contextual detachmentthe assertion that human mental time travel can go above and beyond the cues in the environment (Suddendorf and Corballis, 2007). One should also question whether such detachment is possible, given the importance of the environmental cues in evoking episodic memories (Matute et al. 2011;Nairne, 2002;Osvath, 2016). ...
Article
A growing literature has sought to combine fear conditioning paradigms with the trauma film paradigm to study the associative learning properties of intrusive re-experiencing in PTSD. We review this innovative approach and the recent findings by highlighting their relevance to cognitive and conditioning theories of PTSD. We also conduct a meta-analysis of the available studies to demonstrate that, for most outcome measures, fear learning using a traumatic film clip unconditional stimulus yields results similar to those seen with an electro-tactile unconditional stimulus, which implies that the combined paradigm shares at least some properties of more standard fear conditioning paradigms. We argue that careful use of this combined paradigm will provide important new insights into the mechanisms underlying memory symptoms of PTSD and will allow rigorous testing of cognitive theories of the disorder.
Article
Measuring the quality of classroom-based interactive shared book reading within the early childhood classroom represents a specific dimension of teacher–child interactions that is of great interest to researchers. This interest reflects decades of research demonstrating the benefit of reading to young children in both the home and the classroom. The purpose of this validation study was to conduct a systematic investigation of the psychometric characteristics of a revised version of the Systematic Assessment of Book Reading (SABR-2.2), a measure that examines qualities of teacher talk through observation of classroom shared book reading. Validation of the SABR involved video-based coding of book-reading sessions in pre-k and Kindergarten classrooms (n = 286 teachers and 877 children). Teacher race/ethnicities included: Caucasian (64%), African American (27%), Latino (19%), American Indian/Alaska Native (1%), Asian (4%), and 4% reported their ethnicity as ‘other’. Children’s race/ethnicity was Caucasian (52.3%), African American (39.7%), Latino (30.6%), American Indian/Alaska Native (1.2%), Asian (5.3%), and 1.5% reported their ethnicity as ‘other’. Using the SABR, we coded the frequency of 11 teacher behaviors that relied on explicitly defined behaviors and sets of keywords that indicated the presence of a behavior. Factor analyses yielded a single factor with item loadings between .41 and .86. Inter-rater reliability estimates were high (i.e., 95%–99%). Test–retest reliability revealed that the SABR factor scores were strongly correlated (r = .71) at Fall and Spring. We established concurrent validity with moderate relations with the ELLCO (rs = .31–.50). Predictive validity of the SABR was investigated for 877 children tested at baseline and follow-up. Findings revealed a significant and small association between teachers’ SABR scores and gains in children’s vocabulary and print knowledge scores. Additional analyses were conducted to examine psychometrics by race/ethnicity.
Article
In associative learning, if stimulus A is presented in the same temporal context as the conditional stimulus (CS) - outcome association (but not in a way that allows an A-CS association to form) it becomes a temporal context cue, acquiring the ability to activate this context and retrieve the CS-outcome association. We examined whether a CS- presented during acquisition or extinction that predicted the absence of the unconditional stimulus (US) could act as a temporal context cue, reducing or enhancing responding, in differential fear conditioning. Two groups received acquisition (CSx-US, CSa-noUS) in phase 1 and extinction (CSx-noUS; CSe-noUS) in phase 2 (AE groups), and two groups received extinction in phase 1 and acquisition in phase 2 (EA groups). After a delay, participants were presented with either CSa (AEa and EAa groups) or CSe (AEe and EAe groups). Responding to CSx was enhanced after presentation of CSa but reduced after presentation of CSe, suggesting that training was segmented into two learning episodes and that the unreinforced CS present during an episode retrieved the CSx-US or CSx-noUS association. These findings suggest that temporal context cues may enhance or reduce fear responding, providing an exciting new avenue for relapse prevention research.
Article
Full-text available
Five experiments were conducted to explore trial order and retention interval effects upon causal predictive judgments. Experiment 1 found that participants show a strong effect of trial order when a stimulus was sequentially paired with two different outcomes compared to a condition where both outcomes were presented intermixed. Experiment 2 found that a 48-h retention interval eliminates the trial order effect, so that participants gave a global judgment about the relationship between the stimulus and the two outcomes equivalent to the one given by participants that received the two phases intermixed. This result was replicated in Experiment 3 in a situation in which the probability of the outcome in the presence of the cue was changed from .5 for both outcomes (Experiments 1, 2, 4, and 5) to .75 and .25 for outcomes 1 and 2, respectively. Experiment 4 found that retention intervals ranging from 45min to 48h eliminated the trial order effect similarly. Experiment 5 found that a 10-min retention interval replicated the effect of time upon sequential training found in precedent experiments, regardless of whether participants remained within the laboratory during the retention interval or spent it outside. The combined results of this experimental series suggest that retention intervals reduce retroactive interference in causal learning by allowing participants to use all the information presented across phases, rather than differentially increasing or decreasing retrieval of information about each of them.
Chapter
The chapter tackles the placement of self-reflective consciousness amongst the numberless gradations by Darwin. Discussions of self-consciousness inevitably lead to Descartes' dictum, "I think, therefore I am". The goal is a rapprochement between this view and the Cartesian view, emphasizing this kind of consciousness applicable only to humans. Descartes maintained that animals are unable to engage in self-reflection. Negative results of various ape language projects and broad advances in animal cognition suggest that Descartes was right about the uniqueness of language but that he was wrong about animal's capacity for thought and self-reflection. There is abundant evidence that nonhuman pirates can form representations and use them to solve problems. The concept of autonoetic consciousness, as Tulving calls it, seemed close to the construct of self-reflective consciousness and metacognition which was the concern. Thus, instead of focusing on language, more fundamental capabilities are considered-the origins of self-reflective consciousness.
Chapter
This chapter describes the potential explanatory power of a specific response rule and its implications for models of acquisition. This response rule is called the “comparator hypothesis.” It was originally inspired by Rescorla's contingency theory. Rescorla noted that if the number and frequency of conditioned stimulus–unconditioned stimulus (CS–US) pairings are held constant, unsignaled presentations of the US during training attenuate conditioned responding. This observation complemented the long recognized fact that the delivery of nonreinforced presentations of the CS during training also attenuates conditioned responding. The symmetry of the two findings prompted Rescorla to propose that during training, subjects inferred both the probability of the US in the presence of the CS and the probability of the US in the absence of the CS and they then established a CS–US association based upon a comparison of these quantities. The comparator hypothesis is a qualitative response rule, which, in principle, can complement any model of acquisition.
Article
The development of the sequential approach to instrumental learning from about 1958 to the present is described. The sequential model began as an attempt to explain a particular class of neglected partial reward phenomena, those in which performance in acquisition and extinction is influenced by the particular sequence in which rewarded and nonrewarded trials occur in acquisition, and it was subsequently applied to a variety of other phenomena. Over time, the sequential model grew, sometimes through the replacement of older assumptions by novel ones, as when retrieved memories replaced stimulus traces, and sometimes simply through the addition of novel assumptions, such as that animals are capable of remembering retrospectively one, two, three or more prior nonrewarded outcomes-the N-length assumption. The most recent assumption added to the sequential model is that on a given trial the animal may utilize its memory of prior reward outcomes to anticipate both the current reward outcome and one or more subsequent reward outcomes. One way to view the sequential model is to say that it is a specific theory in various degrees of competition with other specific theories. Several examples of this are provided. Another way to view the sequential model, a more important way in my opinion, is to see it as a representative of a general theoretical approach, intertrial theory, which differs in fundamental respects from another much more generally utilized theoretical approach, intra-trial theory. I suggest that there is a substantial body of data that can be explained by inter-trial mechanisms but not by intratrial mechanisms. The future may well reveal that the inter-trial mechanisms have greater explanatory potential than the currently more popular intratrial mechanisms.
Article
In order to remember a past event, the brain must not only encode the specific aspects of an event but also bind them in a manner that can later specify the spatiotemporal context in which event occurred. Here, I describe recent research aimed at characterizing the functional organization of two brain regions—the medial temporal lobes and the prefrontal cortex—that allow us to accomplish this task. Converging evidence indicates that different regions of the medial temporal lobes may form representations of items, contexts, and item-context bindings and that areas in the prefrontal cortex may implement working-memory control processes that allow us to build meaningful relationships between items that are encountered over time. The results are compatible with an emerging model that generates novel predictions at both the behavioral and neural levels.
Article
We used a three-step procedure to produce and disclose non-similarity-based conceptualization in pigeons. Merely by being associated with the same response in the first step, classes of perceptually dissimilar stimuli, like cars and chairs, appear to amalgamate into a new category of functionally equivalent stimuli. Thus, requiring a new response to be made to only one of these two stimulus classes in the second step transfers to the other stimulus class in the third step. This case of non-similarity-based conceptualization in pigeons is relevant to the construction of superordinate categories and to the role of language in secondary or mediated generalization.
Article
This chapter discusses that experimental psychology is no longer a unified field of scholarship. The most obvious sign of disintegration is the division of the Journal of Experimental Psychology into specialized periodicals. Many forces propel this fractionation. First, the explosion of interest in many small spheres of inquiry has made it extremely difficult for an individual to master more than one. Second, the recent popularity of interdisciplinary research has lured many workers away from the central issues of experimental psychology. Third, there is a growing division between researchers of human and animal behavior; this division has been primarily driven by contemporary cognitive psychologists, who see little reason to refer to the behavior of animals or to inquire into the generality of behavioral principles. The chapter considers the study of causal perception. This area is certainly at the core of experimental psychology. Although recent research in animal cognition has taken the tack of bringing human paradigms into the animal laboratory, the experimental research is described has adopted the reverse strategy of bringing animal paradigms into the human laboratory. A further unfortunate fact is that today's experimental psychologists are receiving little or no training in the history and philosophy of psychology. This neglected aspect means that investigations of a problem area are often undertaken without a full understanding of the analytical issues that would help guide empirical inquiry.