ArticlePDF Available

A comparison between elemental and compound training of cues in retrospective revaluation

Authors:

Abstract and Figures

Associative learning theories assume that cue interaction and, specifically, retrospective revaluation occur only when the target cue is previously trained in compound with the to-be-revalued cue. However, there are recent demonstrations of retrospective revaluation in the absence of compound training (e.g., Matute & Pineño, 1998a, 1998b). Nevertheless, it seems reasonable to assume that cue interaction should be stronger when the cues are trained together than when they are trained apart. In two experiments with humans, we directly compared compound and elemental training of cues. The results showed that retrospective revaluation in the elemental condition can be as strong as and, sometimes, stronger than that in the compound condition. This suggests that within-compound associations are not necessary for retrospective revaluation to occur and that these effects can possibly be best understood in the framework of general interference theory.
Content may be subject to copyright.
Copyright 2002 Psychonomic Society, Inc. 228
Animal Learning & Behavior
2002, 30 (3), 228-238
In the literature of animal conditioning and human as-
sociative learning, it is well known that if a cue, X, is
consistently followed by an outcome, O (i.e., XO), X is
generally learned as a predictor of the occurrence of the
outcome. It is also well known that responding to X in a
subsequent test phase becomes altered if another cue, A,
is trained in compound with X as a predictor of the same
outcome. Some classic instances of these cue interaction
effects in the animal learning literature are overshadow-
ing (Pavlov, 1927), blocking (Kamin, 1968), conditioned
inhibition (Pavlov, 1927), and the relative stimulus va-
lidity effect (Wagner, Logan, Haberlandt, & Price, 1968).
These effects have also been demonstrated with human
participants (e.g., Chapman, 1991; Matute, Arcediano,
& Miller, 1996; Shanks, López, Darby, & Dickinson,
1996; Wasserman, Kao, Van Hamme, Katagari, & Young,
1996). They are important because they suggest that in
situations in which there are two potential predictors of
the same outcome, humans and other animals tend to se-
lectively respond to one of those predictors, with the
other one generally being behaviorally disregarded as a
reliable predictor of the outcome.
Despite the great relevance of cue interaction effects
to our understanding of the principles of learning, the
mechanisms underlying these effects are not yet entirely
clear. For example, there is disagreement concerning
whether cue interaction takes place at the moment of ac-
quisition (e.g., Rescorla & Wagner, 1972) or at the mo-
ment of retrieval (e.g., Miller & Matzel, 1988). Of more
relevance for our present purposes, there is disagreement
concerning whether compound training of the two cues
is necessary for cue interaction to take place. In the pres-
ent research, we look at the latter question, by comparing
cue interaction effects that take place after compound or
elemental training of two cues. (Importantly, the results
of the present experiments also speak to the former ques-
tion; however, it is not the purpose of this research to dis-
criminate between acquisition and retrieval theories of
cue interaction.)
Associative theories of acquisition (e.g., Rescorla &
Wagner, 1972; Wagner, 1981) and retrieval (e.g., Miller
& Matzel, 1988) assume that compound training is nec-
essary for the occurrence of cue interaction. That is, if X
and A are never trained in compound, the expectation of
the outcome when X is presented should be a result only
of the previous X–O pairings, regardless of whether or
not A was trained as a predictor of the same outcome.
Similarly, recent modifications of the Rescorla–Wagner
model (e.g., Van Hamme & Wasserman, 1994) and of
Wagner’s (1981) SOP model (e.g., Dickinson & Burke,
1996) suggest that the occurrence of cue interaction and
retrospective revaluation (a special type of cue inter-
action that could not be explained by their predecessor
theories) depends on the cues’ having acquired a within-
compound association prior to the revaluation treatment.
Retrospective revaluation is a particularly interesting
type of cue interaction because, contrary to other cue
interaction effects, it reflects changes in responding to a
cue (X) without further presentations (training) of that
cue. Rather, these changes in responding are brought
Support for this research was provided by Grant PI-2000-12 from the
Departamento de Educación, Universidades e Investigación of the Basque
Government to H.M. O.P. was supported by a F.P.I. fellowship from the
Spanish Ministry of Education (Ref. AP98, 44970323). M.E. was sup-
ported by a Dissertation Year Fellowship granted by Binghamton Univer-
sity. We thank Leyre Castro and Mirko Gerolin for assistance in data col-
lection and Francisco Arcediano, Leyre Castro, Nuria Ortega, and Sonia
Vegas for comments on an earlier version of this manuscript. We also
thank Jan De Houwer, Ralph Miller, and two anonymous reviewers for
their insightful comments on the manuscript. M.E. is now at Binghamton
University. Correspondence concerning this article should be addressed to
H. Matute, Departamento de Psicología, Universidad de Deusto, Apartado 1,
48080 Bilbao, Spain (e-mail: matute@orion.deusto.es) .
A comparison between elemental and compound
training of cues in retrospective revaluation
MARTHA ESCOBAR, OSKAR PINEÑO, and HELENA MATUTE
Universidad de Deusto, Bilbao, Spain
Associative learning theories assume that cue interaction and, specifically, retrospective revaluation
occur only when the target cue is previously trained in compound with the to-be-revalued cue. How-
ever, there are recent demonstrations of retrospective revaluation in the absence of compound train-
ing (e.g., Matute & Pineño, 1998a, 1998b). Nevertheless, it seems reasonable to assume that cue inter-
action should be stronger when the cues are trained together than when they are trained apart. In two
experiments with humans, we directly compared compound and elemental training of cues. The results
showed that retrospective revaluation in the elemental condition can be as strong as and, sometimes,
stronger than that in the compound condition. This suggests that within-compound associations are not
necessary for retrospective revaluation to occur and that these effects can possibly be best understood
in the framework of general interference theory.
CUE INTERACTION 229
about by training of an associate of X (e.g., Cue A) dur-
ing Phase 2 with which X was previously trained in com-
pound (Phase 1). Instances of retrospective revaluation
are backward blocking (e.g., Denniston, Miller, &
Matute, 1996; Miller & Matute, 1996; Shanks, 1985;
Wasserman & Berglan, 1998), recovery from overshadow-
ing (e.g., Kaufman & Bolles, 1981; Larkin, Aitken, &
Dickinson, 1998; Matzel, Schachtman, & Miller, 1985;
Wasserman et al., 1996), and backward conditioned in-
hibition (Chapman, 1991; Larkin et al., 1998). As with
other cue interaction effects, retrospective revaluation
effects have also been interpreted as either retrieval or
acquisition effects. The retrieval view posits that when
cue X is presented at test, its associative strength is com-
pared with that of A, and if As strength has been modi-
fied during Phase 2, this will be reflected during testing
as a change in responding to X (e.g., Miller & Matzel,
1988). In contrast, the acquisition view posits that the as-
sociative strength of X is modified during each of the tri-
als in which A is trained during Phase 2, even though X
is absent during those trials (e.g., Dickinson & Burke,
1996; Van Hamme & Wasserman, 1994).
It is with respect to retrospective revaluation effects
that the greatest controversy exists concerning the ne-
cessity of compound training of the two cues. Although
some studies suggest that compound training of cues is
a necessary condition for retrospective revaluation to
occur, others have presented evidence of retrospective
revaluation in the absence of compound training. For ex-
ample, there are several reports suggesting that retrospec-
tive revaluation can occur only when a within-compound
association between A and X is acquired during the first
phase of the study (e.g., Aitken, Larkin, & Dickinson,
2001; Dickinson & Burke, 1996; Wasserman & Berglan,
1998). In contrast, Matute and Pineño (1998a, 1998b;
see Escobar, Matute, & Miller, 2001, for a nonhuman
analogue) reported that human participants that received
XO training followed by AO training showed im-
paired responding to X in a subsequent test phase, rela-
tive to a group that did not receive the A–O training. Al-
though this observation is contrary to the predictions of
current theories of associative learning, it suggests that
the learning of a new cueoutcome association during
the second phase of training (i.e., AO) can affect the
expression of the previously acquired XO association,
even though X and A had never received compound
training. The decrease in responding to X was due nei-
ther to the interval elapsed between training and testing
with X nor to memory overload. Responding to X was
unaffected if (1) the second phase consisted of A–no-O
trials or mere exposure to the experimental context, with
no cues or outcomes being presented (Matute & Pineño,
1998b, Experiments 1 and 2), (2) the test of X was con-
ducted in a context different from that in which A was
trained (i.e., renewal effect; Matute & Pino, 1998b,
Experiment 3), (3) the training trials for A and X were
interspersed within a single phase, rather than conducted
sequentially (Pineño, Ortega, & Matute, 2000, Experi-
ment 2), (4) a retention interval was introduced between
training of A and testing on X (i.e., spontaneous recov-
ery; Pineño et al., 2000, Experiment 1), or (5) either re-
trieval cues for the X–O association or novel cues were
presented immediately before testing (Pineño et al.,
2000, Experiment 3).
Interaction between the X–O and the A–O associations
seems to occur during retrieval, rather than during acqui-
sition, because spontaneous recovery of the X–O associ-
ation is observed (Pineño et al., 2000, Experiment 1).
Moreover, such interaction seems to occur maximally
when the two cues have been associated to the same out-
come and the interfering association (AO) is of higher
availability than the target association (XO) at the time
(or in the context) of testing. Therefore, in addition to
within-compound associations, there seem to be other fac-
tors (e.g., recency and contextual manipulations) that de-
termine whether the A–O association will interfere with
retrieval of the XO association at testing, thereby en-
hancing the A–O association’s potential to alter respond-
ing to the target cue during testing (see also Escobar,
Arcediano, & Miller, 2001; Escobar, Matute, & Miller,
2001, for analogue findings with nonhuman subjects).
Other data in the literature can also be interpreted as
evidence of interaction between cues trained apart. For
example, several researchers have reported attenuated
responding to a target cue, X, if a series of X–O pairings
intermixed with A–no-O trials are followed by one or
more A–O trials (Lipp & Dal Santo, in press; Lipp, Sid-
dle, & Dall, 1993; Ortega & Matute, 2000; Packer & Sid-
dle, 1989; Siddle, 1985; Siddle, Broekhuizen, & Packer,
1990). In these studies, the outcome is miscued during
Phase 2 by the cue that predicts its absence (A). The mis-
cuing effect can occur with just one miscuing trial, and
it is more robust than retroactive interference in situa-
tions in which participants do not receive discrimination
training in the target phase of treatment (i.e., the X–O
trials; see Lipp & Dal Santo, in press, for a discussion).
A substantial body of evidence on interference be-
tween cues trained apart comes also from the classical
studies on retroactive interference in the verbal learning
tradition (e.g., Slamecka & Ceraso, 1960; Underwood,
1966). Although researchers in the interference tradition
more extensively studied retroactive interference with
the A–B, A– C paradigm (i.e., two different outcomes as-
sociated to a common cue; e.g., Bäuml, 1996, 1998;
Chandler, 1993; Chandler & Gargano, 1998), many ex-
periments have shown that interference can also take
place when two different cues are associated to a com-
mon outcome (i.e., the A–B, CB paradigm; see, e.g.,
Abra, 1967; Cheung & Goulet, 1968; Johnston, 1968;
Keppel, Bonge, Strand, & Parker, 1971; Schwartz, 1968).
The term retroactive interference is more frequently
used by memory researchers and seems to have been
more widely applied to conditions in which the two cues
are trained elementally. Presumably, retroactive interfer-
ence occurs at the time of testing (e.g., Chandler &
Gargano, 1998). In contrast, the term retrospective reval-
230 ESCOBAR, PINEÑO, AND MATUTE
uation is generally preferred by contemporary associa-
tive learning theorists and is frequently reserved for con-
ditions in which the two cues are trained in compound.
According to some theories, retrospective revaluation
occurs at the time of acquisition (e.g., Larkin et al.,
1998), whereas other theories assume that it occurs at the
time of testing (e.g., Miller & Matzel, 1988). Thus, the
only clear difference between the effects that have been
called retrospective revaluation and those called retro-
active interference seems to be that the former term refers
to cues trained in compound, whereas the latter refers to
cues trained elementally (but see Miller & Escobar,
2002). Because there is no reason to assume a priori that
these effects imply differential underlying mechanisms,
we will use the two terms interchangeably and will focus
on the question of whether compound training of cues is
necessary to retrospectively modify responding to a tar-
get cue without further training with that target cue (re-
gardless of whether this impairment is the result of ac-
quisition or retrieval processes). If compound training of
cues is not a critical factor in modifying responding to
the target cue, the additional question arises of whether
compound training results in more robust retrospective
revaluation than does elemental training, which would
possibly speak to whether within-compound associa-
tions play a facilitatory role in retrospective revaluation.
In Experiment 1, a cue interaction paradigm was used
in which elemental training has sometimes proven to be
as effective as, or even more effective than, compound
training. For example, in a study of conditioned inhibi-
tion with humans, O’Boyle and Bouton (1996) compared
compound (i.e., X–O1/AXO2) versus elemental (i.e.,
XO1/AO2) training of Cue A as a conditioned in-
hibitor of O1. Inhibition was assessed in terms of retar-
dation in the acquisition of an AO1 association by the
compound and elemental groups. According to the pre-
dictions of associative learning theories (e.g., Rescorla
& Wagner, 1972), learning that A prevents the occur-
rence of O1 should result in a stronger AO1 inhibitory
association in the compound condition than in the ele-
mental condition; hence, more retardation should have
been observed in the compound condition. However,
OBoyle and Bouton observed similar degrees of cue
interaction (inhibition) in both the elemental and the
compound conditions.
Although O’Boyle and Bouton (1996) did not exam-
ine retrospective revaluation of X after the Phase 2 pair-
ings were complete, with modification their paradigm
could be used to test for retrospective revaluation of the
training excitor (Cue X) after training of its associate
(Cue A) by simply testing for the predictive status of X
after AO1 pairings. Importantly, during Phase 1, all
participants would presumably acquire an X–O1 associ-
ation, but only participants in the compound group
would also acquire a within-compound association be-
tween A and X. Thus, if within-compound associations
are necessary for the occurrence of retrospective revalu-
ation, the A–O1 pairings should affect responding to X
at test only in those participants who had received com-
pound training of X and A during Phase 1. In Experi-
ment 1, we tested these predictions by using an in-
hibitory paradigm similar to that used by O’Boyle and
Bouton but testing our participants for responding to X
after the A–O1 pairings of Phase 2, rather than on their
acquisition of the A–O1 association. In Experiment 2,
elemental and compound training were compared, using
an excitatory paradigm that provided convergent evi-
dence for Experiment 1. In Experiment 2, either pairings
of the target cue and the outcome (i.e., X–O) or the com-
pound of these two cues and the outcome (i.e., AXO)
preceded pairings of the interfering cue and the outcome
(i.e., A–O).
EXPERIMENT 1
In Experiment 1, four groups of participants received
elemental or compound inhibitory training during Phase 1.
Groups Elemental-O1 and Compound-O1 received, dur-
ing Phase 1, XO1 and A–O2 interspersed training and
XO1 and AXO2 interspersed training, respectively.
During Phase 2, both groups received AO1 pairings.
The control groups, Group Elemental-O2 and Group
Compound-O2, received the same treatment as Groups
Elemental-O1 and Compound-O1 during Phase 1, respec-
tively, but were presented with AO2 trials instead of
A–O1 trials during Phase 2 (see Table 1). If interaction
occurs both between individually trained cues and be-
tween compound-trained cues in this procedure, the dif-
ference in responding to X between Groups Elemental-
O1 and Elemental-O2 should be similar to that between
Groups Compound-O1 and Compound-O2.
Method
Participants . Forty-eight undergraduate students from Deusto
University volunteered for the study. The participants were ran-
domly assigned to one of four groups (ns = 12).
Apparatus and Procedure. The experiment was conducted
using personal computers, with the experimental contingencie s
being presented on the computers monitor and participants re-
sponding through the keyboard. The computers used during the ex-
periment were located in a room with five individual cubicles. The
screen simulated an airplanes control panel, to be used in a ficti-
tious Air Force aptitude selection test. The instructions were shown
in two consecutive screens. An approximate translation from Span-
ish of the instructions used reads as follows:
Screen 1: You are going to take part on the Air Forces aptitude selec-
tion test. Pilots who obtain the maximum score will be accepted. Your
airplane is equipped with a special control panel that will indicat e
Table 1
Design of Experiment 1
Group Phase 1 Phase 2 Test
Compound-O1 X–O1/AX–O2 AO1 X
Compound-O2 X–O1/AX–O2 AO2 X
Elemental-O1 X–O1/A–O2 AO1 X
Elemental-O2 X–O1/A–O2 AO2 X
Note—A and X, yellow and blue lights, counterbalanced within groups;
O1, gaining points; O2, losing points; /, intermixed with.
CUE INTERACTION 231
whether or not you should shoot the airplane’s guns. Your main task
will be to learn how to interpret the lights in the panel. For each attack
you will have a maximum of 10 shots available. Regardless of how
many shots you use in a given attack, you will have 10 more shots avail-
able during the next attack. You have to decide how many shots you are
going to use each time. If you shoot when you should, according to the
information presented on the panel, you will gain points (the more
shots you use, the more points you will win) and by contrast, if you
shoot when you should not, you will lose points (the more shots you
use, the more points you will lose).
Screen 2: Each shot adds or subtracts one point from your overall score.
You will have to press < ENTER > to see the panels information for
each new attack. When the information is shown in the panel, you can
indicate the number of shots you decide to use according to that infor-
mation. To do this, you have to use the keyboards cursor keys to fill
the shots indicator bar presented at the right side of your screen. Once
you have made your choice, press < ENTER > to see how many points
you have won or lost.
These instructions were individually explained to each partici-
pant, using two demonstration trials with cues different from those
used in the actual experiment. After this explanation was com-
pleted, the experiment began. During the experiment, the partici-
pants advanced from one trial to the next one by pressing the Enter
key on the computer keyboard. No breaks were inserted between
the different phases of the experiment.
The control panel had six separate lights and was located below
the airplanes window, in two horizontal rows of three lights each.
Cues A and X were yellow and blue lights, counterbalanced within
groups. The six lights were “turned off (represented by a black
light) when no cues were presented. When one cue was presented,
one light turned on, showing the color corresponding to that cue,
and the other lights remained black (turned off). When a compound
of two cues was presented, two lights were turned on, and the other
lights remained off. The position in which the cues were presented
was randomly determined for each trial.
Outcome 1 (O1) consisted of (1) the message “You have gained:
[n] points (with [n] being the number of shots risked during the cue
presentation) and (2) gaining one point for each shot that was
risked. Outcome 2 (O2) consisted of (1) the message “You have
lost: [n] points and (2) losing one point for each shot that was
risked. In each trial, the participants were given a message in the
presence of the colored lights of the control panel, asking them to
decide the number of shots they would like to risk on that trial (i.e.,
“Indicate the number of shots). Presumably, the more certain the
participants were that they could earn points (O1), the greater num-
ber of shots they would risk in a given trial (i.e., up to 10), whereas
the more certain they were that they could lose points (O2), the
fewer shots they would risk in a given trial (i.e., down to 0). Thus,
the dependent variable assessed in the present experiment was the
number of shots risked by the participants during testing.
In this experiment, the different trial types of Phase 1 were pre-
sented following a pseudorandom sequence. This sequence was
XO1, X–O1, AX–O2, X–O1, AX–O2, AX–O2, XO1, AXO2,
XO1, AX–O2, AX–O2, AX–O2, XO1, AX–O2, XO1, XO1
for the compound groups (the elemental groups received an identi-
cal sequence, with AO2 trials replacing the AX–O2 trials).
Results and Discussion
The results of this experiment are presented in Fig-
ure 1. Retrospective revaluation was observed in Group
Elemental-O1, as compared with group Elemental-O2,
but not in Group Compound-O1, as compared with
Group Compound-O2. Moreover, there were no differ-
ences in responding to X between the two control groups
(Groups Compound-O2 and Elemental-O2). That is,
with a similar baseline for responding to X after com-
pound and elemental training, retroactive revaluation of
Cue X was greater when the cues were trained apart (el-
emental condition) than when they were trained together
(compound condition). These conclusions were sup-
ported by the following analyses.
A 2 (type of training, elemental vs. compound) 3 2
(Phase 2 outcome, O1 vs. O2) analysis of variance
(ANOVA) conducted on the ratings (i.e., number of shots
risked) of X at test revealed a main effect of Phase 2 out-
come, as well as an interaction [Fs(1,44) = 4.24 and 9.67,
respectively, MS
e
= 12.78, ps < .05]. Planned comparisons
using the error term from this ANOVA were conducted to
analyze the source of the interaction. Responding to X
in Group Elemental-O2 was greater than that in Group
Elemental-O1 [F(1,44) = 13.36, p < .001]. However, there
10
9
8
7
6
5
4
3
2
1
0
Compound Elemental
Type of Training
Mean Number of Responses
O1
O2
Figure 1. Mean number of responses during the test trial with X in Experi-
ment 1. Error bars represent the standard errors of the means.
232 ESCOBAR, PINEÑO, AND MATUTE
were no differences between Groups Compound-O1 and
Compound-O2 [F(1,44) < 1]. Thus, greater retrospective
revaluation was observed when Cues X and A had been
trained apart (elemental condition) than when these cues
were trained together (compound condition).
The results of this experiment extend the findings on
interaction between individually trained cues reported by
Matute and Pineño (1998a, 1998b), using a different
preparation and design, which speaks to the generality
of these effects. Most important, these results suggest
that, using the present preparation and parameters, ret-
rospective revaluation between individually trained cues
can be stronger than retrospective revaluation between
compound-trained cues. These results are contrary to
what could be expected on the basis of the postulates of
contemporary associative learning theories (e.g., Dick-
inson & Burke, 1996; Miller & Matzel, 1988; Rescorla
& Wagner, 1972; Van Hamme & Wasserman, 1994).
The present results do not speak to the issue of
whether the source of the interference is at the moment
of acquisition or at the moment of retrieval, but other re-
ports suggest that stimulus interference is probably due
to a retrieval failure that occurs when the interfering cue
is more available for retrieval during testing than is the
target cue. In the present case, however, it could be argued
that the participants in Group Elemental-O1 learned that
the contingencies changed between Phase 1 and Phase 2.
During Phase 1, they learned that X predicted O1 and A
predicted O2; then, during Phase 2, they learned that A
predicted O1, and this might have led them to expect
that, symmetrically, X would no longer predict O1. Al-
though this rule would also apply to Group Compound-
O1, in which A also predicted O2 during Phase 1 and O1
during Phase 2, the compound training might have some-
how attenuated reversal learning. In order to assess
whether reversal learning could have influenced the re-
sults of Experiment 1, we avoided pairing A (or AX)
with O2 during Phase 1 in Experiment 2.
It should be noted, however, that Groups Elemental-O1
and Compound-O1 were modeled directly from the stud-
ies in which inhibition with compound trials (commonly
known as Pavlov’s procedure) was compared with inhibi-
tion in situations that did not require compound trials (in
this case, the so-called differential procedure). To the best
of our knowledge, reversal learning has not been an issue
in those studies. Indeed, our results are consistent with sev-
eral other studies in which a similar design was used that
have also observed reliable cue interaction effects in the el-
emental (i.e., differential) inhibition condition (see, e.g.,
O’Boyle & Bouton, 1996, and Williams, Travis, & Over-
mier, 1986, for reviews). Moreover, these results are also
consistent with many other studies that have found no ev-
idence of interference (retrospective revaluation) after
compound training (e.g., Shanks, Darby, & Charles, 1998;
Wilson & Pearce, 1992). Most of those studies have attrib-
uted the resistance to interference that often occurs in the
compound conditions to compound training favoring con-
figural processing, unless special procedures are used to
favor elemental processing (see also Williams, Sagness, &
McPhee, 1994). For this reason, in Experiment 2, elemen-
tal training was defined as training in which the target and
the potentially interfering cue were not presented in com-
pound. However, in the elemental condition, the target cues
were presented in compound with cues irrelevant to the
revaluation treatment, so that configural processing would
equally affect the two conditions and similar degrees of cue
interaction could be observed between them.
EXPERIMENT 2
The main purpose of Experiment 2 was to extend the re-
sults of Experiment 1, using an excitatory paradigm. In
addition, for the reasons stated above, configural process-
ing was made equally likely in the elemental and the com-
pound conditions, so that equivalent degrees of retrospec-
tive revaluation could be observed in both cases. For this
purpose, the target cue, X, was trained in compound in all
groups during Phase 1, but the competing cue trained dur-
ing Phase 2 was either the Phase 1 companion cue (Con-
dition Together) or a different cue (Condition Apart).
Thus, Experiment 2 compared interaction between cues
trained together (analogous to the compound condition in
Experiment 1) and interaction between cues trained apart
(analogous to the elemental condition in Experiment 1) in
an excitatory preparation (see Table 2).
In addition, in this experiment, we chose to use a
preparation that had proven successful in obtaining ret-
rospective revaluation with individually trained cues
(Pineño et al., 2000) but that was different from that used
in Experiment 1 and in Matute and Pineño’s (1998b) stud-
ies (for a full description of the task used by Matute &
Pineño, 1998b, see Arcediano, Ortega, & Matute, 1996).
Introducing yet one more experimental preparation is not
without its problems, in that it reduces comparability
among different experiments. However, our reason for
doing so was twofold. First, using a different preparation
is necessary when one wants to assess the generality of the
results. Second, recent studies conducted in our labora-
tory have provided results indicating that the preparation
developed by Pineño et al. is more sensitive to (compound
and elemental) cue interaction effects than are the tasks
used in Experiment 1 and by Matute and Pineño (1998b).
1
Method
Participants . Eighty undergraduate students from Deusto Uni-
versity volunteered for the study. The participants were randomly
assigned to one of four groups (ns = 20).
Table 2
Design of Experiment 2
Group Phase 1 Phase 2 Test
Together-O1 AX–O1/CO2 AO1 X
Togetherno-O AX–O1/CO2 A–no-O X
Apart-O1 BXO1/C–O2 AO1 X
Apart–no-O BXO1/C–O2 A–no-O X
Note—A, B, and X, blue, red, and green lights, counterbalanced within
groups; C, yellow light; O1, gaining points; O2, losing points; no-O, no
outcome; /, intermixed with. C–O2 trials were inserted so that the par-
ticipants would not tend to respond indiscriminately to any stimulus
that appeared on the screen.
CUE INTERACTION 233
Apparatus and Procedure. The apparatus was the same as that
described in Experiment 1, with the exception of the task. In Ex-
periment 2, the participants were asked to imagine that they were to
rescue a group of refugees by helping them escape from a war zone
in several trucks (see Pineño et al., 2000). A translation of the in-
structions from Spanish reads as follows.
Screen 1: Imagine that you are a soldier for the United Nations. Your
mission consists of rescuing a group of refugees that are hidden in a
ramshackle building. The enemy has detected them and has sent forces
to destroy the building . . . . But, fortunately, they rely on your cunning
to escape the danger zone before that happens.
You have several trucks for rescuing the refugees, and you have to
help them get into those trucks. There are two ways of placing people
in the trucks:
Pressing the space bar repeatedly, so that one person per press is
placed in a truck.
Maintaining the space bar pressed down, so that you will be able to
load people very rapidly.
If you rescue a number of persons in a given trip, they will arrive to
their destination alive, and you will be rewarded with a point for each
person. You must gain as many points as possible!
Screen 2: But . . . your mission will not be as simple as it seems. The
enemy knows of your movements and could have placed deadly mines
on the road. If the truck hits a mine, it will explode, and the passengers
will die. Each dead passenger will count as one negative point for you.
Fortunately, the colored lights on the
SPY
-
RADIO
will tell you about the
state of the road. These lights can indicate that:
(a) The road will be free of mines. ® The occupants of the truck
will be liberated. ® You will gain points.
(b) The road will be mined. ® The occupants of the truck will die.
® You will lose points.
(c) There are no mines, but the road is closed. ® The occupants of
the truck will neither die nor be liberated. ® You will neither
gain nor lose points: You will maintain your previous score.
Screen 3: At first, you will not know what each color light of the
SPY
-
RADIO
means. However, as you gain experience with them, you will
learn to interpret what they mean.
Thus, we recommend that you:
(a) Place more people in the truck the more certain you are that the
road will be free of mines (keep the space bar continuousl y
pressed down
ONLY
if you are completely sure that there are no
mines, because in this way you will put a lot of people in the
truck . . . ).
(b) Introduce less people in the truck the more certain you are that
the road is mined.
After these instructions, the participants were shown a fourth
screen that gave instructions about contextual changes. Although
contextual changes were not used in the present experiment, in
order to avoid making more changes than necessary between dif-
ferent experimental series conducted with the same preparation, we
maintained a fourth instruction screen in the program for the task.
A translation of the fourth instruction screen reads as follows.
Screen 4: Finally, it is important to know that your mission may take
place in several different towns. The colors on the
SPY
-
RADIO
can mean
the same or a very different thing depending on the town in which you
are. Thus, it is important to pay attention to the message that indicates
the place in which you are. If you travel to another town, the message
indicating the name of the town will change. When a change of desti-
nation is occurring, you will read the message “Traveling to another
town,” so you will be continuously informed about such changes. Nev-
ertheless, sometimes you might end up returning to the same town even
if you have seen the message that indicates that you are traveling. Do
not worry if all this looks very complex at this point. Before we start,
you will have the opportunity to see the location of everything (radio,
town name, messages, scores, etc.) on the screen, and to ask the ex-
perimenter about anything that is unclear.
The cues were presented in the “spy-radio,which consisted of
six panels in which colored lights could be presented. Cues X, A, and
B were blue, red, and green lights, counterbalanced. Cue C was a
yellow light. All the cues were presented for 3 sec. During the inter-
trial intervals (ITIs), the lights were turned off (i.e., gray). The mean
ITI duration was 5 sec, ranging between 3 and 7 sec. During each
cue presentation, each response (i.e., pressing the space bar once)
placed 1 refugee in the truck, whereas holding the space bar down
placed up to 30 refugees per second in the truck. The termination
of the cue always coincided with the onset of the outcome.
Outcome 1 (O1) consisted of (1) the message [n] refugees safe
at home!!!(with [n] being the number of refugees introduced in
the truck during the cue presentation) and (2) gaining one point for
each refugee who was liberated. Outcome 2 (O2) consisted of
(1) the message “[n] refugees have died!!!and (2) losing one point
for each refugee who died in the truck. No-outcome (no-O) con-
sisted of (1) the message “Road closedand (2) maintaining the
previous score. Outcome messages were presented for 3 sec. The
number of refugees that the participants risked taking in each truck
was our dependent variable. Presumably, the more certain they were
that the trip would be successful (O1), the greater the number of
refugees they would take, whereas the more certain they were that
the truck would explode (O2), the fewer refugees they would risk
placing in the truck.
A score panel on the screen provided information about the num-
ber of refugees the participant was introducing into the truck on
each trial. Although pressing the space bar during the outcome mes-
sage had no consequences, this panel remained visible during the
presentation of the outcome and showed the number of people that
had boarded the truck while the cue was present. Upon outcome ter-
mination, the score panel was initialized to 0. Responses that oc-
curred during the ITIs had no consequence and were not reflected
in the panel.
During Phase 1, the participants were exposed to 4 pairings of the
AX or BX compounds with O1 (Conditions Together and Apart, re-
spectively), intermixed with four C–O2 trials. These CO2 trials
were used to prevent generalization of responding to cues other than
the cues paired with the outcome. The pseudorandom sequence of
Phase 1 trials generated for this experiment was AX–O1, AX–O1,
CO2, AX–O1, C–O2, C–O2, AXO1, CO2, for Condition To-
gether. An identical sequence was generated for Condition Apart,
except that BX–O1 replaced the AX–O1 trials. In Phase 2, the par-
ticipants in the O1 condition received 16 AO1 pairings, whereas
the participants in the no-O condition received 16 A–no-O pairings.
All the participants then received one test trial with X.
Results and Discussion
Figure 2 presents the results of this experiment. Ret-
rospective revaluation was observed in both Conditions
Apart and Together. That is, responding to the target cue,
X, was weaker in Group Apart-O1 than in Group
Apart–no-O and was also weaker in Group Together-O1
than in Group Together–no-O. Thus, with the present de-
sign and parameters, we observed retrospective revalua-
tion both between cues trained together and between
cues trained apart. The following analyses support our
conclusions .
A 2 (type of training, together vs. apart) 3 2 (Phase 2
outcome, O1 vs. no-O) ANOVA conducted on respond-
ing to X at test revealed a main effect of Phase 2 outcome
[F(1,76) = 10.34, MS
e
= 633.47, p < .005] but no effect of
type of training and no interaction ( ps > .27). Pairwise
comparisons revealed that responding to X in Group Apart-
O1 was weaker than in Group Apart–no-O [F(1,76) =
4.32, p < .05] and responding to X in Group Together-O1
was weaker than that in Group Together–no-O [F(1,76) =
234 ESCOBAR, PINEÑO, AND MATUTE
6.10, p < .05]. That is, the present design and parameters
resulted in similar degrees of cue interaction occurring be-
tween cues trained apart and between cues trained together.
GENERAL DISCUSSION
In two experiments, we compared the strength of ret-
rospective revaluation obtained between cues that had
been presented together at some point during training
with that obtained between cues that were always pre-
sented apart. In Experiment 1, we compared elemental
versus compound inhibitory training and observed ret-
rospective revaluation between cues trained apart, but
not between cues trained in compound (for similar re-
sults in inhibitory cue interaction, see O’Boyle & Bou-
ton, 1996; Williams et al., 1986). In Experiment 2, we
used an excitatory paradigm in which the target cue was
always presented in compound with another cue, either
the interfering cue trained during Phase 2 (Condition To-
gether, which was essentially backward blocking) or a
different cue irrelevant to the revaluation treatment
(Condition Apart). With this manipulation, we observed
retrospective revaluation both between cues trained apart
and between cues trained in compound, and the sizes of
these revaluation effects were similar. Thus, these results
support previous reports that compound training is not
necessary for retrospective revaluation to occur. More
important, these two experiments, taken together, sug-
gest that contrary to the predictions of most contempo-
rary associative learning theories, compound training
can sometimes hamper, rather than favor, the occurrence
of retrospective revaluation (see also Williams et al.,
1994).
Although our finding of retrospective revaluation (or
retroactive interference) with elementally trained cues
might seem surprising from the point of view of current
associative learning theories, it should be noted that our
findings are not as novel as they might seem. As was noted
in the introduction, additional examples of interference
between elementally trained cues can be found in what
was usually called the A–B, CB paradigm in the verbal
learning literature (e.g., Abra, 1967; Cheung & Goulet,
1968; Johnston, 1968; Keppel et al., 1971; Schwartz,
1968). Those studies on paired-associates learning have
generally been overlooked by current theories of asso-
ciative learning, perhaps because those studies were gen-
erally more concerned with retrieval of memory than
with the predictive value of the cues. However, the pres-
ent experiments are basically A–B, CB paradigms in
which A and C are different cues (colored lights) that
predict a common outcome, B (gaining points). The
finding that interference can also occur in predictive
learning, regardless of whether or not within-compound
associations are established, suggests the possibility of in-
tegrating the literature on elemental cues (called retroac-
tive interference in the paired associate tradition) with the
studies on compound cues (called retrospective revalua-
tion in the associative learning tradition).
Our results are also consistent with recent reports of
interaction between elementally trained cues (e.g., Es-
cobar, Arcediano, & Miller, 2001; Escobar, Matute, &
Miller, 2001; Matute & Pineño, 1998a, 1998b; Ortega &
Matute, 2000; Pineño et al., 2000; Pineño & Matute,
2000), as well as with several reports that show inter-
action between cues that have received differential (i.e.,
elemental) inhibition training (e.g., Lipp et al., 1993;
Packer & Siddle, 1989; Siddle, 1985; Siddle et al.,
1990). Moreover, some of the comparisons of the ele-
mental and compound inhibition procedures that have
been reported in the literature are also consistent with
the results of Experiment 1. In an animal study, Williams
et al. (1986) observed more robust inhibition with the el-
45
40
35
30
25
20
15
10
5
0
Together Apart
Type of Training
Mean Number of Responses
O1
no O
Figure 2. Mean number of responses during the test trial with X in Experi-
ment 2. Error bars represent the standard errors of the means.
no-O
CUE INTERACTION 235
emental (differential) procedure than with the compound
(Pavlovs) procedure. Similarly, OBoyle and Bouton
(1996, Experiment 1) reported similar levels of inhibi-
tion in a compound (Pavlov’s) procedure and in an ele-
mental (differential) procedure, using a predictive task
with human participants. Moreover, in their Experi-
ment 2, O’Boyle and Bouton observed that inhibition ob-
tained with the elemental procedure was greater than that
obtained with the compound procedure. The resistance
to interference in the compound condition has been gen-
erally attributed to compound presentation of cues fa-
voring configural, rather than elemental, processing
strategies (see Shanks et al., 1998; Williams et al., 1994;
Wilson & Pearce, 1992).
Our using different experimental tasks for the two ex-
periments deserves some comment. On the one hand,
comparisons between different experiments in the same
research area should benefit from the use of similar
preparations. On the other hand, it is also true that using
different preparations is necessary when one wants to
test the generality of the effects and to ensure that they
are not artifacts produced by the details of particular
preparations. In further support of the idea that inter-
action between cues trained apart is a general phenome-
non of associative learning, Escobar, Matute, and Miller
(2001; see also Escobar, Arcediano, & Miller, 2001) re-
ported analogous evidence of interference, using rats as
subjects in a conditioned lick suppression preparation.
The observation of similar effects of interaction between
elementally trained cues in different species (rats and hu-
mans), with different tasks (various computer prepara-
tions and paired-associates learning), using different
cues (different audiovisual stimuli), and with different
parameters (the present experiments and the miscuing
effect) suggests that these are not isolated results.
Our finding retrospective revaluation irrespective of
whether X was trained in compound with A or with a dif-
ferent cue can be interpreted as contrary to Dickinson
and Burke’s (1996; see also Aitken et al., 2001) sugges-
tion that retrospective revaluation takes place only if
there is a strong within-compound association between
the target and the interfering cues. It should be noted,
however, that there are too many differences between
their procedure and ours that could be responsible for
our observation of retrospective revaluation in both con-
ditions. Although careful parametric manipulations
should be performed to illuminate the basis of this dif-
ference, below we discuss some of the factors that might
influence the occurrence of retrospective revaluation in
the absence of strong within-compound associations.
A possible source for this discrepancy (i.e., the as-
sumption that retrospective revaluation requires strong
within-compound association vs. reports of retrospective
revaluation in the absence of compound training) is the
differential sensitivity of experimental tasks to detect
interaction between cues trained apart. Among the sev-
eral tasks we have used in our laboratory to obtain inter-
actions between individually trained cues, the task used
in Experiment 2 has proven to be the most sensitive,
whereas the task used by Dickinson and colleagues
(Aitken et al., 2001; Dickinson & Burke, 1996) has
proven to be the least sensitive. Dickinson and col-
leagues used a causal judgment task in which partici-
pants were requested to give their subjective rating of the
degree to which they believed that the cue (i.e., a partic-
ular food) was the cause of the outcome (i.e., an allergic
reaction). Despite our failure to obtain interference in
causal judgment tasks (see Matute & Pineño, 1998a, for
a review), some reports in the literature have suggested
that interaction between elementally trained cues could
be observed even in causal judgments tasks if the right
parameters were used. Wasserman and Berglan (1998)
reported a study in which participants received AW–O1,
BXO1, and CYO1 trials in Phase 1, followed by
A–O1 and Cno-O trials in Phase 2. Then, W, X, and Y
were presented at test. Wasserman and Berglan expected
that, as compared with the ratings at the end of Phase 1,
the training of Phase 2 would produce a decrease in the
ratings of W (i.e., backward blocking, because of the
within-compound association between A and W), an in-
crease in the ratings of Y (i.e., recovery from overshad-
owing, because of the within-compound association be-
tween C and Y), and no change in the ratings of X (i.e.,
because its companion cue, B, was not presented during
Phase 2). That is, they expected responding at test to be
W < X < Y. However, although the X < Y difference was
confirmed, the difference between X and W did not
reach statistical significance, because the rating for both
cues decreased after the Phase 2 training, relative to the
ratings at the end of Phase 1. As Wasserman and Berglan
noted, this decrease in the ratings of X could be viewed
as reflecting some degree of interaction between cues
not trained in compound. They reported that a difference
between X and W was observed in their study only when
the data from participants who, on a postexperimental
test, were not able to recall which cues had been pre-
sented together were eliminated from the analysis. Sim-
ilar results have been reported by Dickinson and his col-
leagues (e.g., Larkin et al., 1998, Experiments 1 and 2).
They used a within-subjects design analogous to that
used by Wasserman and Berglan and also observed the
X < Y effect, but not the W < X effect. According to
Larkin et al., the W < X effect is parameter dependent.
Although speculative, the results of Experiment 1 seem
to support this assumption, since retroactive interference
was not observed in the compound condition with the
specific parameters used.
Although the purpose of the present paper was not to
explore the mechanisms that might produce retrospec-
tive revaluation between cues trained together and be-
tween cues trained apart, the results of these experi-
ments, together with other data on cue competition and
interference, suggest that the two effects can be ex-
plained by a common mechanism. Retrospective revalu-
ation between both elementally trained and compound-
trained cues can be explained by Dickinson and Burkes
236 ESCOBAR, PINEÑO, AND MATUTE
(1996) revised SOP model if it is assumed that the absent
cue, X, can become associatively activated not only by
cues that have a within-compound association with X,
but also by the outcome itself (see Escobar, Matute, &
Miller, 2001; Matute and Pineño, 1998a). According to
this view, X and O1 would form inhibitory associations
during Phase 2 because X is absent during trials in which
the outcome is present. This approach can also account
for retrospective revaluation between cues trained apart
being sometimes stronger than retrospective revaluation
between cues trained in compound (Experiment 1). Po-
tentially, the compound (AXno-O/XO1) training re-
sults in the formation of a weaker X–O1 association than
the elemental (A–no-O/X–O1) training because, in the
former case, the A–X association could potentially over-
shadow the XO1 association. Consequently, the pre-
sentation of O1 during Phase 2 would better activate the
representation of X in the elemental than in the com-
pound groups, thus yielding a stronger X–O1 inhibitory
association (i.e., stronger retrospective revaluation of
X). By the same reasoning, elemental interference
should not be stronger than compound interference if the
elemental cue is trained in compound with an irrelevant
cue, as was the case in our Experiment 2 (Condition
Apart).
A problem with this approach is that it also predicts
that O1-alone presentations during Phase 2 should pro-
duce retrospective revaluation of the X–O1 association.
However, several studies have reported no effect of the
O1-alone presentations in situations in which the num-
ber of these O1-alone trials is maintained low enough to
prevent a degradation of the X–O1 contingency (e.g.,
Escobar, Arcediano, & Miller, 2001; Escobar, Matute, &
Miller, 2001, Experiments 2–4; Ortega & Matute, 2000,
Experiment 2). Quite possibly, this is due to the forma-
tion of a contextO1 association that could produce
some responding during testing to X. This responding,
produced by the contextual cues, could summate to the
response produced by X, thereby producing a final
stronger response.
2
Thus, in order to adequately assess
the possibility of retrospective revaluation caused by O1-
alone trials, testing on X should be performed in a dif-
ferent (physical or temporal) context from that in which
O1 was presented alone. However, it is not easy to test
this view, because contextual switches would produce
collateral effects here: Many experiments have shown
that contextual changes produce renewal of responding
to X following interference (e.g., Escobar, Matute, &
Miller, 2001, Experiment 3; Matute & Pino, 1998b,
Experiment 3; Pineño & Matute, 2000, Experiment 1).
Moreover, Dickinson and Burkes (1996) revised SOP
model cannot explain recovery of responding following
retrospective revaluation between elementally trained
cues owing to contextual manipulations (Matute and
Pineño, 1998b, Experiment 3), the introduction of a re-
tention interval (Pino et al., 2000, Experiment 1), or
the presentation of retrieval cues (Pineño et al., 2000,
Experiment 3; see Escobar, Matute, & Miller, 2001, for
similar manipulations in nonhuman subjects). These
studies suggest that for interference to take place, it is
also necessary that an interfering association be more
strongly primed at the time of testing than the target as-
sociation.
Thus, a potential way to incorporate most of the in-
terference results within the SOP framework would
imply the application of a mechanism by which excita-
tory and inhibitory associations could be selectively re-
trieved depending on the different contextual and tem-
poral manipulations performed at testing. According to
Matute and Pineño (1998a), this could be achieved by
integrating within the SOP framework the assumptions
of Boutons (1993) retrieval theory of interference be-
tween outcomes. According to Bouton’s theory, excita-
tory associations easily transfer to novel (physical
and/or temporal) contexts, but retrieval of inhibitory as-
sociations is modulated by the presence of (physical
and/or temporal) contextual cues that became associ-
ated with the memory of inhibition during inhibitory
training. Although Bouton developed his theory to ac-
count for interference between outcomes (e.g., extinc-
tion and counterconditioning), his assumptions on the
contextual specificity of the inhibitory associations could
be extended in order to account for interference between
cues. The present results add further support to the idea
that the integration of Bouton’s model with Dickinson and
Burke’s (1996) revision of Wagners (1981) SOP model
can account for most instances of retrospective revalua-
tion and retroactive interference, both between cues and
between outcomes.
Alternatively, Miller and Escobar (2002) have re-
cently proposed a dual-mechanism approach to account
for both cue competition and interference effects. Ac-
cording to this approach, two mechanisms would be si-
multaneously at work in all situations. The first is a
comparator mechanism (cf. Miller & Matzel, 1988), ac-
cording to which responding to a target cue is directly
related to the degree to which the target directly acti-
vates a representation of the outcome and inversely re-
lated to the degree to which the target indirectly acti-
vates a representation of the outcome through other cues
that have become associated to the target during train-
ing (so-called comparator cues). The second mechanism
is a priming mechanism, according to which responding
to the target would be directly related to the degree to
which the context of testing primes the target associa-
tion and inversely related to the degree to which the con-
text of testing primes other associations. Activation of
the comparator mechanism requires a within-compound
association between the target and the interfering cues;
activation of the priming mechanism requires that the
target and the interfering cues have different priming
stimuli. According to Miller and Escobar, when the tar-
get and the interfering cues are trained in compound, the
comparator mechanism will be maximally active (there
is a within-compound association between the cues), and
the priming mechanism will be minimally active (both
CUE INTERACTION 237
the target and the interfering cue would have the same
priming stimuli). In contrast, when the target and the in-
terfering cues are trained apart, the comparator mechanism
will be minimally active (there is no within-compound as-
sociation between the cues), and the priming mechanism
will be maximally active (the target and the interfering
cue each have their own, distinct priming stimuli).
At this point, we are not prepared to favor one interpre-
tation above the other. However, we strongly believe that
integrative strategies, such as those reviewed here, can ex-
plain both compound and elemental interference and ben-
efit from a common theoretical framework that should
prove more fruitful than the existing compartmentaliza-
tion of data from the two separate research designs.
REFERENCES
Abra, J. C. (1967). Time changes in the strength of forward and back-
ward associations. Journal of Verbal Learning & Verbal Behavior, 6,
640-645.
Aitken, M. R. F., Larkin,