You are on page 1of 19

JOURNAL OF THE EXPERIMENTAL ANALYSIS OF BEHAVIOR 1999, 72, 279–297 NUMBER 3 (NOVEMBER)

SAMPLE-DURATION EFFECTS ON PIGEONS’


DELAYED MATCHING AS A FUNCTION OF
PREDICTABILITY OF DURATION
P ETER J. U RCUIOLI , T HOMAS B. D E M ARSE , AND K AREN M. L IONELLO
PURDUE UNIVERSIT Y

Three experiments assessed the impact of sample duration on pigeons’ delayed matching as a func-
tion of whether or not the samples themselves signaled how long they would remain on. When
duration was uncorrelated with the sample appearing on each matching trial, the typical effect of
duration was observed: Choice accuracy was higher with long (15-s) than with short (5-s) durations.
By contrast, this difference either disappeared or reversed when the 5- and 15-s durations were
correlated with the sample stimuli. Sample duration itself cued comparison choice by some birds in
the latter (predictable) condition when duration was also correlated with the reinforced choice
alternatives. However, even when duration could not provide a cue for choice, pigeons matched
predictably short-duration samples as accurately as, or more accurately than, predictably long-dura-
tion samples. Moreover, this result was observed independently of whether the contextual conditions
of the retention interval were the same as, or different from, those of the intertrial interval. These
results strongly support the view that conditional stimulus control by the samples is partly a function
of their conditioned reinforcing properties, as determined by the relative reduction in overall delay
to reinforcement that they signal.
Key words: sample duration, duration predictability, delayed matching, delay-reduction hypothesis,
conditioned reinforcement, key peck, pigeons

It has long been known that delayed havior will come under control of that stim-
matching by pigeons is sensitive to the dura- ulus.
tion of the sample stimuli. For example, Rob- Observing and conditional stimulus con-
erts (1972) and others (Grant, 1976; Nelson trol by the sample stimuli can also be influ-
& Wasserman, 1978; Roberts & Grant, 1974) enced, however, by variables other than the
demonstrated that the longer pigeons view length of sample exposure. For example,
the sample stimulus prior to the retention in- Hartl, Dougherty, and Wixted (1996, Experi-
terval, the more accurate they are in choos- ment 2) recently reported the interesting
ing the correct comparison on the retention finding that the usual performance advan-
test. This finding is often interpreted to mean tage accruing to long- as opposed to short-
that increasing duration strengthens the me- duration samples can be eliminated if the du-
morial trace established by the sample stim- ration of the sample on each matching trial
ulus. Consequently, its relevant characteristics is predictable. In their study, pigeons per-
formed two-alternative delayed matching with
or those of its coded representation (Roitblat,
samples either 5 or 15 s in duration. In one
1980; Urcuioli & Zentall, 1986) are more like-
condition, each sample appeared equally of-
ly to exert control over comparison choice. ten for 5 and 15 s in a session (i.e., duration
Simply put, the greater the exposure that pi- was uncorrelated with the sample stimuli).
geons have to a sample on a given trial, the Here, pigeons showed the typical duration ef-
more likely it will be that its subsequent be- fect: They matched more accurately with 15-
s samples than with 5-s samples. By contrast,
the pigeons matched short-duration samples
This research was partially supported by National Sci- as accurately as long-duration samples when
ence Foundation Grant IBN-94-19929 and by National In-
stitutes of Mental Health Grant MH 56487 to Peter Ur- one sample stimulus always appeared for 5 s
cuioli. We thank Lisa Huff, Michelle McIntosh, Leslie in a session and the other sample always ap-
Peck, Jada Pierce, and Jessica Rightsell for their assistance peared for 15 s. In other words, when dura-
in conducting the experiments. tion was correlated with (predictable from)
Correspondence concerning this article should be ad- the samples themselves, delayed matching ac-
dressed to Peter J. Urcuioli, Department of Psychological
Sciences, Purdue University, 1364 Psychology Building, curacy was unaffected by the length of time
West Lafayette, Indiana 47907-1364 (E-mail: uche@psych. pigeons were exposed to the sample on each
purdue.edu). trial.

279
280 PETER J. URCUIOLI et al.

The finding from this correlated condition ference observed when duration is uncorre-
indicates that although a relatively short du- lated with the sample stimuli). This is pre-
ration affords less opportunity to observe a cisely what Hartl et al. (1996, Experiment 2)
sample stimulus, the observing occasioned by observed. Note, by the way, that in an uncor-
such a short-duration sample is apparently related condition, all samples would have the
enhanced by its predictability. One way this same conditioned reinforcing value because
might arise is if samples of predictably differ- all would be associated with the same average
ent durations have different conditioned re- reduction in delay to reinforcement. Thus,
inforcing properties, as outlined below. In- under these conditions, observing should be
deed, Hartl et al. (1996) explicitly postulated primarily (if not solely) a function of length
differences in the conditioned reinforcing of sample exposure.
properties of the samples in their correlated Although the findings of Hartl et al. (1996,
and uncorrelated conditions in order to ac- Experiment 2) are consistent with the predic-
count for their findings, using the postulates tions of the delay-reduction hypothesis, they
of delay-reduction theory (Fantino, 1977; are (as the authors themselves acknowl-
Wixted, 1989) to derive those differences. edged) open to an alternative interpretation
According to the delay-reduction hypothe- that requires no appeal to variation in the
sis, conditioned reinforcement is a direct conditioned reinforcing properties of the
function of the size of the reduction in the samples. Specifically, sample duration itself
overall wait (delay) to reinforcement signaled may have provided a cue for choice in Hartl
by a stimulus. In a delayed matching para- et al.’s correlated condition (cf. Stubbs,
digm, it is computed by subtracting how long 1968). This possibility arose because the
the pigeon must wait for reinforcement given matching contingencies in that condition al-
the appearance of a sample stimulus from the ways reinforced choice of one comparison
average interreinforcement interval in a ses- following the short-duration sample and
sion. Thus, samples that are predictably short choice of the other comparison following the
in duration will have larger delay-reduction long-duration sample. If duration did exert
values than samples that are predictably long conditional stimulus control in this situation
in duration (assuming, of course, that the fre- and if the duration cue was at least as potent
quency of reinforcement on each trial type is as the samples themselves, then the fact that
approximately the same). Consequently, delayed matching accuracy did not vary as a
short-duration samples should be more pow- function of duration in their correlated con-
erful conditioned reinforcers than long-du- dition is not surprising. Indeed, it would be
ration samples and would thus be expected expected. One hint that duration had provid-
to occasion more vigorous observing behav- ed a redundant cue was the additional find-
ior (cf. Case & Fantino, 1981; McMillan, ing that overall accuracy was higher in Hartl
1974). As a result, predictably short-duration et al.’s correlated condition than in their un-
samples should exert relatively strong control correlated condition. This effect is analogous
over choice, although this would not neces- to other findings in the discrimination liter-
sarily entail higher matching accuracy than ature showing that redundant relevant cues
with predictably long-duration samples. The often yield faster rates of learning (Miles &
reason for this qualification is that any en- Jenkins, 1973; Warren, 1953) and higher lev-
hanced level of observing generated by stron- els of discrimination accuracy (Urcuioli,
ger conditioned reinforcing properties of 1985, 1990a) than a just single cue.
predictably short-duration samples should be The present experiments were thus con-
offset to some degree by less opportunity to ducted in order to clarify the sources of the
observe those samples than long-duration interesting effects reported by Hartl et al.
samples. (1996; Experiment 2) and to expand upon
Nevertheless, the net result of the interac- them. More specifically, we designed our ex-
tion between the effects of conditioned re- periments to assess whether the performance
inforcement and the length of exposure to differences in their correlated versus uncor-
the sample stimuli should be a reduction in related conditions were entirely attributable
the size of the performance difference on to the presence versus absence of a duration
long- versus short-duration trials (i.e., the dif- cue, respectively. We were especially interest-
DURATION PREDICTABILITY 281

ed in determining if the reduced effect of Table 1


sample duration when it is correlated with The conditions of Experiment 1.
the sample stimuli also occurs when duration
Uncorrelated Correlated
cannot, by design, provide a cue for choice.
The latter assessment obviously provides a 5/15 S1 → C11 5 S1 → C11
more discriminating test of the delay-reduc- 5/15 S2 → C21 15 S2 → C21
5/15 S3 → C11 15 S3 → C11
tion hypothesis of the sample-duration effect 5/15 S4 → C21 5 S4 → C21
because it avoids the potential confounding
effect in the Hartl et al. correlated proce- Note. S1 through S4 denote the four sample stimuli, C1
and C2 denote the correct (1) comparison alternatives,
dure. We also recorded pecking to the sample and 5 and 15 refer to sample durations (in seconds).
stimuli in the correlated condition as an an- Counterbalancing of correct choices and duration by
alogue measure of observing behavior in or- sample in the correlated condition are not shown.
der to see if it might be sensitive to duration
predictability in a way that is suggestive of dif-
ferent conditioned reinforcing properties. trained on many-to-one matching to sample
Toward these ends, we modified the de- (e.g., Santi & Roberts, 1985; Urcuioli et al.,
layed matching tasks used by Hartl et al. 1995) with four samples and two compari-
(1996, Experiment 2) to include two addi- sons. In the uncorrelated condition, each
tional samples. In other words, our pigeons sample appeared equally often for 5 and 15
were trained on four-sample, two-comparison s in a session. In the correlated condition, two
(many-to-one) delayed matching (e.g., Ur- of the samples appeared for 5 s and the re-
cuioli, Zentall, & DeMarse, 1995). In this pro- maining two samples appeared for 15 s. To
cedure, it is possible to present samples for a prevent sample duration from becoming a
predictably short (or long) duration in such cue for choice in the latter condition, one
a way that duration itself is unrelated to the comparison choice was reinforced after one
subsequently reinforced comparison choice. of the 5-s samples, whereas the alternative
Experiment 1 used this procedure to deter- comparison choice was reinforced after the
mine the effect of predictable versus unpre- other 5-s sample. Similar contingencies were
dictable sample durations on delayed match- in effect for the two 15-s samples.
ing. In Experiment 2, we reproduced Hartl et If comparable levels of accuracy observed
al.’s correlated contingencies in the many-to- in Hartl et al.’s (1996, Experiment 2) corre-
one paradigm and included independent lated condition simply reflect control over
tests to see if duration in fact became a cue choice by duration per se, then removing that
for choice under conditions that mimicked cue via the many-to-one contingencies of the
theirs. Experiment 3 returned to the proce- present experiment should yield higher
dure of Experiment 1 but used intertrial in- matching accuracy with long- than with short-
tervals that were visually distinct from the de- duration samples. In other words, a perfor-
layed matching trials (and, hence, from the mance advantage on trials with 15-s samples
retention intervals). Here, we were interested should occur independently of whether du-
in determining if the effect of duration pre- ration is correlated or uncorrelated with the
dictability under these contextual conditions samples themselves. On the other hand, if
would be similar to, or different from, that Hartl et al.’s results reflect stronger condi-
occurring when the background conditions tioned reinforcing properties of predictably
of the trial and intertrial periods are identi- short-duration samples, then the correlated
cal. condition data from our experiment should
resemble theirs; that is, the long- versus short-
duration accuracy difference should be re-
EXPERIMENT 1 duced or eliminated in the correlated as op-
Experiment 1 compared the effects of pre- posed to the uncorrelated condition.
dictable versus unpredictable sample dura- M ETHOD
tions on delayed matching accuracy in a pro-
cedure that insured that sample duration Subjects
would not provide an additional cue for com- Eight White Carneau pigeons (retired
parison choice (see Table 1). Birds were breeders) from the Palmetto Pigeon Plant
282 PETER J. URCUIOLI et al.

served as subjects. Throughout the experi- samples (S1 and S3) was to the circle (C1)
ment, the pigeons were maintained at 80% of comparison, whereas the reinforced choice
their free-feeding body weights by restricted following the green and horizontal-line sam-
feeding in their home cages. Birds were ples (S2 and S4) was to the triangle (C2)
housed individually in stainless steel cages in comparison. These relations were reversed
a colony room with a 14:10 hr light/dark cy- for the remaining birds.
cle. Grit and water were available at all times. Each matching trial began with a white ho-
All 8 birds had previously participated in a mogeneous field on the center key. A single
study involving zero-delay identity and sym- peck to this ‘‘trial-ready’’ cue immediately
bolic matching tasks with red and green hues turned it off and, 500 ms later, produced one
and vertical and horizontal lines as stimuli. of the four samples on the same key. The
sample remained on for 10 s, after which it
Apparatus went off automatically as the two comparisons
The experiment was conducted in two appeared on the adjacent side keys. A single
BRS/LVE sound-attenuating enclosures peck to either comparison then turned both
(Model SEC-002) each of which contained a off and produced either reinforcement or an
panel (Model PIP-016) with three response equivalent timeout period depending on
keys, a houselight, and a feeder. The response whether the choice was correct or incorrect,
keys were located behind 2.5-cm circular respectively. Reinforcement for correct com-
openings positioned in a horizontal row 25.5 parison choice was constant for each bird
cm above the chamber floor. The center-to- within each session but varied between 2 and
center distance between the keys was 8.3 cm. 6 s across sessions in a manner that main-
Behind each key was a 12-stimulus inline pro- tained body weights as close to the 80% val-
jector (Model IC-901-IDD) that could display ues as possible. Following reinforcement or
red, green, and white homogeneous fields, a timeout, a dark 20-s intertrial interval (ITI)
single white vertical or horizontal line on a ensued.
dark background, and a white open circle or In each session, the eight possible trial
open triangle on a dark background (BRS/ types (four samples 3 two left-right compar-
LVE Pattern 696). The rear-mounted feeder ison configurations) occurred equally often
could be accessed through an opening (5.0 and in random order in successive blocks of
cm by 5.8 cm) centered approximately 13 cm 48 trials with the constraint that no trial type
below the center response key. A houselight occur more than three times in a row. Each
located 7.6 cm above the center key was not bird was trained on the zero-delay task until
used at any time in this experiment. A blower it matched correctly on 90% or more of all
fan mounted on the outside of each enclo- trials for five of six consecutive sessions. One
sure provided continuous ventilation and bird did not meet this criterion after 80 train-
masking noise, and a single Zenith PC-AT mi- ing sessions but was nonetheless advanced to
crocomputer interfaced to both chambers the next phase of preliminary training be-
scheduled and recorded all experimental cause its matching accuracy was consistently
events. high (about 87% correct).
Next, all birds received mixed-delay train-
Procedure ing with RIs of 1 and 6 s separating offset of
Preliminar y training. All pigeons began the 10-s samples from onset of the compari-
training on many-to-one matching to sample sons. The two intervals occurred equally of-
with four samples (red and green hues, and ten on each of the eight trial types in a ses-
vertical and horizontal lines), two compari- sion. Each bird received a minimum of 10
sons (circle and triangle forms), and a 0-s re- mixed-delay training sessions and continued
tention interval (RI). Each 96-trial session until it met the following three performance
consisted of 24 trials with each sample (S1 stability criteria: (a) overall matching accura-
through S4 in Table 1) and with the two com- cy of at least 75% correct over five consecu-
parison alternatives (C1 and C2 in Table 1) tive sessions, (b) at least 70% choice accuracy
appearing equally often on the left and right at each RI over those same sessions, and (c)
side keys. For half of the birds, the reinforced overall accuracy for each session of the five-
(1) choice following the red and vertical-line session block within 65% of the mean. One
DURATION PREDICTABILITY 283

Fig. 1. Average proportion of correct choice responses by retention interval on trials with 5-s and 15-s samples in
the uncorrelated and correlated conditions of Experiment 1.

bird did not meet these criteria after consid- ously. At that point, it was advanced to the
erable training and was dropped from the ex- next condition in its sequence.
periment. Its data are not included in any of Statistical analyses. For all statistical analyses
the analyses reported below. reported in this experiment and those that
Correlated versus uncorrelated training. Once follow, Type I error rate was set at .05 using
each bird met the aforementioned criteria the tabled F values reported by Rodger
with 10-s sample durations, 5- and 15-s dura- (1975).
tions were introduced. In the correlated con-
dition, two samples always appeared for 5 s R ESULTS
and the other two samples always appeared Despite some variability in the rates at
for 15 s in a session. For half of the birds, the which birds acquired zero-delay many-to-one
5-s samples were the red hue and the hori- matching and reached the performance cri-
zontal line (S1 and S4 in Table 1) and the 15- teria on the initial mixed-delay task, terminal
s samples were the green hue and the vertical matching performances prior to the correlat-
line (S2 and S3 in Table 1). These relations ed and uncorrelated manipulation were very
were reversed for the remaining birds. In the accurate. For example, mean accuracy over
uncorrelated condition, all four samples were the last five mixed-delay sessions with the 10-
presented equally often for 5 s and 15 s in s sample duration averaged 85.2% correct
each session. Otherwise, all procedural de- (range, 81.0% to 87.9%). Accuracy on trials
tails were identical to those described previ- with the 1- and 6-s RIs averaged 91.5% correct
ously. Every bird was exposed twice to the cor- (range, 84.6% to 96.7%) and 79.0% correct
related and uncorrelated conditions. For 4 (range, 77.5% to 84.6%), respectively.
birds, the sequence of conditions was corre- The left and right panels of Figure 1 show
lated, uncorrelated, correlated, uncorrelated; mean accuracy by sample duration and by RI
for the remaining 3 birds, it was the reverse. in the uncorrelated and correlated condi-
A bird remained in each condition for a min- tions, respectively. The data from the two rep-
imum of 10 sessions and until it met the same lications of each condition have been com-
three performance criteria mentioned previ- bined because the results from each repli-
284 PETER J. URCUIOLI et al.

Table 2
Delayed matching accuracy by sample duration and retention interval for each pigeon in the
uncorrelated and correlated conditions of Experiment 1.

Uncorrelated Correlated
Bird Duration (s) 1s 6s Average 1s 6s Average

DR1 5 91.67 89.17 90.42 88.75 89.16 88.96


15 96.46 93.33 94.90 95.21 90.83 93.02
DR2 5 77.50 69.79 73.64 91.04 79.58 85.31
15 84.17 80.62 82.40 80.42 78.75 79.58
DR3 5 87.92 76.04 81.98 93.33 83.54 88.44
15 94.38 87.29 90.83 96.66 89.79 93.22
DR4 5 79.58 71.88 75.58 91.88 80.62 86.25
15 93.96 83.33 88.64 87.50 77.08 82.29
DR5 5 76.04 74.58 75.31 91.04 75.42 83.23
15 81.88 81.67 81.78 87.29 79.38 83.33
DR6 5 81.67 70.42 76.04 85.42 73.75 80.21
15 88.96 82.30 85.63 92.92 87.29 90.10
DR7 5 88.12 84.58 86.35 97.92 87.08 92.50
15 88.96 87.50 88.23 91.46 90.42 90.94

cation were very similar. Each data point rep- The only noteworthy difference between the
resents the average of each bird’s perfor- data shown in Figure 1 and those from each
mance over its last 10 sessions in a condition. replication involved this interaction. In the
The results from the uncorrelated condi- second replication (but not in the first), the
tion show that delayed matching was more crossover in performance as a function of RI
accurate at both RIs with 15-s than with 5-s was significant, F(1, 6) 5 10.38.
samples. This finding replicates Hartl et al.’s Table 2 lists each bird’s matching accuracy
(1996) uncorrelated results and is consistent by sample duration and by RI in the corre-
with the typical effects of sample duration on lated and uncorrelated conditions. As in Fig-
pigeons’ matching performances (Grant, ure 1, the data have been averaged over rep-
1976; Nelson & Wasserman, 1978; Roberts & lications. In the uncorrelated condition,
Grant, 1974). By contrast, the long- versus every bird matched more accurately overall
short-duration difference virtually disap- with 15-s than with 5-s samples. By contrast,
peared in the correlated condition. Indeed, only 4 of the 7 birds (DR1, DR3, DR5, and
at the 1-s RI, there was a slight performance DR6) showed the same advantage in the cor-
advantage in favor of the short-duration (5-s) related condition and, for 2 of them (DR3
samples. and DR5), this effect was noticeably reduced.
Separate analyses of variance (ANOVAs) For the remaining 3 birds (DR2, DR4, and
were performed on the uncorrelated and cor- DR7), overall accuracy in the correlated con-
related data to assess the impact of sample dition was actually higher with 5-s than with 15-
duration and RI on performance. ANOVA on s samples.
the uncorrelated data confirmed that delayed Interestingly, duration predictability in the
matching was more accurate with 15-s than correlated condition did appear to affect the
with 5-s samples, F(1, 6) 5 30.44. There was rate at which pigeons pecked the samples,
also a clear effect of RI with overall accuracy even though the 5- and 15-s durations had no
lower at 6 s than at 1 s, F(1, 6) 5 16.81, but overall effect on matching accuracy. Mean
no Sample Duration 3 RI interaction, F(1, 6) peck rates averaged over the last 10 correlat-
5 3.01. By contrast, ANOVA on the correlat- ed sessions from each replication were 1.8
ed data showed that sample duration had no versus 0.9 pecks per second for the 5-s and
overall effect on matching accuracy, F(1, 6) 15-s hue samples, respectively, and 2.4 versus
5 0.30. There was an effect of RI, F(1, 6) 5 1.2 pecks per second for the 5-s and 15-s line
38.71, but again this variable did not signifi- samples, respectively. All 7 birds had higher
cantly interact with duration, F(1, 6) 5 5.23. peck rates to the 5-s than to the 15-s hue sam-
DURATION PREDICTABILITY 285

matched with long- and short-duration sam-


ples in the correlated versus uncorrelated
condition. In the uncorrelated condition, in
which all four samples randomly appeared
for 5 s and 15 s in a session, pigeons showed
the usual effect of sample duration: They
consistently chose the correct comparison on
the retention test more often following long
(15-s) than short (5-s) samples. By contrast,
the usual performance advantage of long-
over short-duration samples disappeared in
the correlated condition, in which duration
was predictable on the basis of what sample
stimulus appeared on each trial. Here, pi-
geons were equally accurate in their delayed
Fig. 2. Mean sample-response rates (in pecks per sec- matching choices with both 5- and 15-s sam-
ond) to the 5- and 15-s hue and line samples on the first ples.
and the last sessions of the correlated conditions of Ex- This pattern of results is consistent with the
periment 1. basic premise of the delay-reduction hypoth-
esis, which states that the conditioned rein-
forcing properties of a stimulus are a direct
ple, and 6 of the 7 birds had higher rates to function of the size of the reduction in over-
the 5-s than to the 15-s line sample. all delay-to-reinforcement that it signals. If
Although these findings cannot be com- the likelihood of sample observing behavior
pared to the rates in the uncorrelated con- increases with a sample’s conditioned rein-
dition (because we recorded peck rates by forcing properties (cf. Case & Fantino, 1981;
sample and not by duration), it nonetheless Dinsmoor, 1983), then we would expect that
seems clear that the ability to predict the du- a predictably short-duration sample would
ration of each sample in the correlated con- engender a higher rate of observing than a
dition affected how pigeons responded to predictably long-duration sample. This, in
those samples. For example, Figure 2 shows turn, should counteract some of the advan-
that the differences in the sample peck rates tage normally associated with long-duration
to the short- versus long-duration samples in- samples (i.e., with the opportunity to view
creased from the first to the last session of such samples for a longer period of time).
exposure to this condition. At the outset of Indeed, our observation that peck rates to
correlated training, there was a relatively the predictably short-duration samples in-
small difference in peck rates to the 5- and creased over sessions in the correlated con-
15-s samples. Moreover, this initial-session dif- dition can be viewed as evidence for en-
ference was not even apparent in the first rep- hanced observing generated by the stronger
lication of the correlated condition (data not conditioned reinforcing properties associated
shown). By the end of correlated training, with them.
however, the difference in peck rates had The hypothesized differences in the con-
grown substantially and, as shown in Figure ditioned reinforcing properties of the sam-
2, was entirely attributable to the increasing ples in the correlated and uncorrelated con-
rates at which pigeons pecked the short-du- ditions of this experiment can be derived
ration (5-s) samples. from the delay-reduction hypothesis in the
following manner. Assuming perfect match-
D ISCUSSION ing accuracy, the average interreinforcement
Experiment 1 showed that predictability of interval in both types of sessions was 33.5 s
sample duration affects conditional stimulus (20-s ITI 1 10-s average sample duration 1
control by the samples even when duration 3.5-s average RI). In the uncorrelated condi-
itself cannot provide a redundant cue for tion, in which the duration of the sample on
choice. The predictability effect can be seen each trial was unpredictable, the time to re-
by comparing how accurately pigeons inforcement signaled by all four samples was
286 PETER J. URCUIOLI et al.

the same at 13.5 s (10-s average duration 1 should be above the level expected by
3.5-s average RI). Consequently, all four sam- chance.
ples shared the same delay-reduction value of We also ran the same transfer test after un-
20 s (33.5 s 2 13.5 s). In the correlated con- correlated training (i.e., after training in
dition, the signaled time to reinforcement for which all four samples appeared equally often
the predictably short-duration samples was for 5 and 15 s). Here, the expectation was
only 8.5 s (5 s 1 3.5-s average RI) versus 18.5 that matching accuracy with the novel 5- and
s for the predictably long-duration samples 15-s samples should be at chance and, most
(15 s 1 3.5-s average RI). Consequently, the certainly, lower than accuracy in the test that
delay-reduction value was larger for the short- followed correlated training.
than for the long-duration samples (25 s vs.
15 s, respectively) in this condition. M ETHOD
In short, the hypothesized conditioned re- Subjects and Apparatus
inforcing strengths of the samples in this ex-
periment are ordered as follows: predictable Eight White Carneau pigeons of the same
5-s samples . uncorrelated samples . pre- age and obtained from the same source as
dictable 15-s samples. Comparing the right those used in Experiment 1 served in this ex-
and left panels of Figure 1 indicates that, in periment. Their housing and maintenance
line with this ordering, accuracy with predict- conditions were identical to those described
ably short-duration samples was higher than in Experiment 1. All birds had previous ex-
the averaged performance in the uncorrelat- perience in two-choice identity and symbolic
ed condition. Accuracy with predictably long- matching tasks involving red and green hues
duration samples was not lower, however, and vertical and horizontal lines (BRS/LVE
than the average uncorrelated accuracy, al- Pattern No. 696) but none had any experi-
though the greater opportunity to observe ence on the delayed matching contingencies
the 15-s samples in the correlated condition used here. Prior to the start of this experi-
might have compensated for any weakening ment, they were randomly divided into two
of observing arising from their lower condi- equal groups.
tioned reinforcing properties. The apparatus was identical to that previ-
ously described with the exception that the
inline projector behind the center response
EXPERIMENT 2 key was equipped to display two additional vi-
Experiment 2 assessed the possibility that sual stimuli: a white X on a black background
the correlated contingencies in Hartl et al. and a blue 1 also on a black background.
(1996, Experiment 2) introduced a duration The latter stimulus consisted of the simulta-
cue for choice. In other words, might their neous illumination of a uniform blue hue
pigeons have chosen one comparison when a and an otherwise white 1 on a black back-
sample appeared for only 5 s and the other ground.
comparison when a sample appeared for 15
Procedure
s? To answer this question, we initially trained
pigeons on many-to-one tasks similar to those The design of this experiment was similar
of Experiment 1. However, unlike Experi- to that of Experiment 1 with the addition of
ment 1, the matching contingencies in the two matching-to-sample transfer tests, one af-
correlated phase of this experiment arranged ter initial correlated (or uncorrelated) train-
consistent reinforcement for one comparison ing with 5- and 15-s samples and the other
choice following each short-duration sample after subsequent training in the opposite con-
and for the alternative comparison choice fol- dition. One group of pigeons received cor-
lowing each long-duration sample. Later, we related training first and uncorrelated train-
tested for transfer of those choice responses ing second (Group Corr-Uncorr), whereas
to novel samples that were presented for one the other group was trained in the opposite
of the two durations used in training (i.e., 5 order (Group Uncorr-Corr).
and 15 s). If duration per se had acquired Preliminary training. Preliminary training in-
control over choice in the correlated phase, volved zero- and mixed-delay many-to-one
then matching accuracy in the transfer test matching to sample similar to the corre-
DURATION PREDICTABILITY 287

Table 3
The two training conditions and the transfer-test contingencies in Experiment 2.

Training
Group Phase 1 Phase 2 Transfer test

Corr-Uncorr 5 S1 → C11
15 S2 → C21 5 S5 → food 5 S5 → C11
5 S3 → C11 15 S6 → food 15 S6 → C21
15 S4 → C21
Uncorr-Corr 5/15 S1 → C11
5/15 S2 → C21 5 S5 → food 5 S5 → C11
5/15 S3 → C11 15 S6 → food 15 S6 → C21
5/15 S4 → C21
Note. S1 through S6 denote the six sample stimuli, C1 and C2 denote the correct (1) comparison alternatives, and
5 and 15 refer to sample durations (in seconds). Counterbalancing of correct choices and duration by sample in the
correlated condition are not shown.

sponding tasks in Experiment 1. All pigeons the samples themselves provided a reliable
initially learned to match 10-s red and green cue for choice. The sample–correct choice re-
sample hues and vertical and horizontal sam- lations in this group were matched to those
ple lines to circle and triangle comparisons in Group Corr-Uncorr.
to the 90% accuracy criterion previously de- All pigeons were trained on their respective
scribed. Once each bird reached criterion, Phase 1 tasks until they met the three afore-
mixed-delay training with RIs of 1 and 6 s be- mentioned performance criteria for mixed-
gan and continued until the three perfor- delay training.
mance criteria previously described for this Training Phase 2. In preparation for the first
preliminary training phase were met. All oth- transfer test, pigeons in both groups received
er details of the tasks, including counterbal- reinforced off-baseline training with two new
ancing of the sample–correct choice rela- center-key stimuli, a blue 1 and a white X (S5
tions, were identical to those of Experiment 1. and S6 in Table 3). Each trial of the 60-trial
Training Phase 1. Next, each bird was Phase 2 sessions began with the white trial-
trained on mixed-delay, many-to-one match- ready stimulus on the center key. A single
ing with 5- and 15-s sample durations. For peck to the white center key turned it off and,
Group Corr-Uncorr, one hue and one line 500 ms later, produced either the blue 1 or
sample always appeared for 5 s in each ses- the white X on the same key for 5 or 15 s,
sion, and the other hue and line sample al- respectively (i.e., each stimulus was presented
ways appeared for 15 s. For these birds, then, at only one duration). After the scheduled in-
samples were correlated with duration during terval had elapsed, the stimulus went off in-
Phase 1 training. The sample–duration and dependently of responding, and food was de-
sample–correct choice relations were coun- livered. The two new stimuli appeared equally
terbalanced across birds. Unlike Experiment often in each Phase 2 session, with the restric-
1, however, the reinforced (correct) compar- tion that neither appeared more than three
ison choice was the same on all trials with a times in a row.
5-s sample and likewise for all trials with a 15- Most pigeons consistently pecked the blue
s sample (see Table 3). Thus, when trained 1 and white X stimuli (S5 and S6) on their
in this condition, birds had two potential cues initial Phase 2 sessions and were thus given a
to guide their choices: the sample stimulus total of five sessions of reinforced key-peck
appearing on the center key (red, green, ver- training with them. Two birds (one in each
tical, or horizontal) and its duration (5 or 15 group), however, seldom pecked these stim-
s). uli during initial exposure. Consequently,
For Group Uncorr-Corr, all four samples pecking to S5 and S6 was explicitly shaped by
(S1 through S4 in Table 3) appeared equally the method of successive approximations
often for 5 and 15 s during each Phase 1 over the course of two or three separate ses-
training session. For these birds, then, only sions. Once pecking was established, these
288 PETER J. URCUIOLI et al.

birds then received two or three additional teria were identical to those for Training
training sessions with each stimulus appear- Phase 1.
ing for its respective duration. Transfer Test 2. Once the aforementioned
Upon completion of Phase 2, each bird re- criteria had been met, each bird was returned
ceived refresher training on its Phase 1 to the off-baseline task with the blue 1 and
mixed-delay matching task. Refreshers were white X center-key stimuli for one or two ses-
given until overall matching accuracy was at sions. On the following session, they were
least 75% correct and accuracy at each RI was tested again with these stimuli serving as sam-
at least 70% correct for a single session. This ples for the circle and form comparisons.
required one to four sessions (M ø two ses- This test session was identical in every respect
sions). to the first transfer test session.
Transfer Test 1. A single 80-trial transfer test
with the blue 1 and white X samples (S5 and R ESULTS
S6) and the circle and triangle comparisons Despite relatively slow acquisition of zero-
(C1 and C2) was given on the day immedi- delay matching with 10-s sample durations
ately following the last refresher. The dura- (range, 26 to 80 sessions) and progress to-
tions of the blue 1 and white X samples con- ward meeting the performance criteria for
tinued to be 5 and 15 s, respectively, and each mixed-delay training (range, 11 to 56 ses-
sample was preceded by the white trial-ready sions), delayed matching performances at the
cue on the center response key. Each of the end of preliminary training were quite accu-
four possible trial types (two samples 3 two rate and did not differ appreciably across
left-right configurations of the comparisons) groups. Mean overall accuracy for Groups
was presented equally often and in random Corr-Uncorr and Uncorr-Corr over the last
order during the test session. Likewise, the five mixed-delay sessions with 10-s samples
two RIs (1 and 6 s) occurred with equal fre- was 82.0% versus 83.0% correct, respectively.
quency on each trial type. For Group Corr- At the 1-s RI, accuracy averaged 86.2% and
Uncorr, the reinforced choice following the 89.6%, respectively, for the two groups. The
5-s blue 1 sample (S5) was to the same com- corresponding figures for the 6-s RI were
parison alternative that was correct following 77.9% and 76.4%, respectively. There were
the two 5-s samples in its preceding many-to- no significant between-group differences in
one task. Similarly, the reinforced choice fol- any of these performance measures, all Fs(1,
lowing the 15-s white X sample (S6) was to 6) , 1.50.
the alternative comparison (i.e., the compar- Figure 3 shows the retention functions by
ison correct following the two 15-s samples in sample duration for the last 10 sessions of un-
the many-to-one task). For Group Uncorr- correlated (left panel) and correlated (right
Corr, the reinforced sample–choice relations panel) training with the 5- and 15-s samples
in this test were matched on a bird-by-bird averaged across all birds. When each sample
basis to the reinforced relations represented appeared equally often for 5 and 15 s in a
in Group Corr-Uncorr. session (uncorrelated training), delayed
Training Phase 3. Following the transfer test, matching accuracy was higher at each RI with
each bird was returned to mixed-delay many- the long than with the short duration. By con-
to-one matching but was trained in the con- trast, when duration was correlated with sam-
dition opposite of that they had initially ex- ple stimulus, the usual advantage of the long
perienced. In other words, for Group over short duration reversed. A separate AN-
Corr-Uncorr, each sample stimulus now ap- OVA on the uncorrelated data indicated an
peared equally often for 5 and 15 s in each overall effect of duration, F(1, 7) 5 57.32, but
session (i.e., they received uncorrelated train- neither the RI effect nor the Duration 3 RI
ing). Conversely, for Group Uncorr-Corr, two interaction were significant, Fs(1, 7) 5 2.61
samples (one hue and one line) always ap- and 0.24, respectively. ANOVA on the corre-
peared for 5 s, whereas the other two samples lated data showed no significant effect of ei-
(the remaining hue and line) always ap- ther variable or its interaction with the other,
peared for 15 s (i.e., they received correlated all Fs(1, 7) , 5.40.
training). Otherwise, all procedural details Table 4 shows delayed matching accuracy
including the performance and stability cri- for each subject over its last 10 sessions of
DURATION PREDICTABILITY 289

Fig. 3. Average proportion of correct choice responses by retention interval on trials with 5-s and 15-s samples in
the uncorrelated and correlated conditions of Experiment 2.

uncorrelated and correlated training. For the long- versus short-duration difference for the
uncorrelated condition, all 8 subjects showed remaining 2 subjects (C1 and U4) was smaller
higher overall accuracy with 15-s than with 5- than that obser ved in the uncorrelated
s samples. For the correlated condition, 6 of phase. On average, the mean difference in
the 8 subjects showed the reverse effect; the accuracy between the 15- and 5-s samples was

Table 4
Delayed matching accuracy by sample duration and retention interval for each pigeon in the
uncorrelated and correlated conditions of Experiment 2.

Uncorrelated Correlated
Bird Duration (s) 1s 6s Average 1s 6s Average

C1 5 83.33 77.50 80.42 84.17 79.17 81.67


15 85.83 86.67 86.25 88.33 82.08 85.21
C2 5 81.25 82.50 81.88 90.42 84.58 87.50
15 85.00 82.50 83.75 80.42 80.42 80.42
C3 5 75.42 84.17 79.80 87.92 88.75 88.34
15 86.25 91.25 88.75 80.83 93.75 87.29
C4 5 75.00 69.58 72.29 96.67 81.67 89.17
15 82.08 73.33 77.71 91.25 72.50 81.88
U1 5 67.08 70.42 68.75 82.50 79.17 80.84
15 76.25 80.00 78.12 67.92 65.42 66.67
U2 5 89.17 70.42 79.80 97.92 90.42 94.17
15 95.83 82.92 89.38 89.58 90.83 90.21
U3 5 79.58 65.00 72.29 88.75 82.92 85.84
15 84.58 75.00 79.79 86.25 81.25 85.63
U4 5 85.83 78.75 82.29 91.67 76.25 83.96
15 92.92 85.00 88.96 87.08 81.25 84.17
290 PETER J. URCUIOLI et al.

Fig. 4. Left panel: mean percentage of correct choice responses averaged over retention interval for the two
mixed-delay transfer tests in Experiment 2. Right panel: mean percentage of correct choice responses for the two
zero-delay transfer tests in Experiment 2. Group Corr-Uncorr received the first of its two tests after correlated training,
and the second after uncorrelated training, in each delay condition. The training conditions preceding each test
were reversed for Group Uncorr-Corr.

6.9% in the uncorrelated condition and Averaged over RIs and subjects, choice accu-
23.8% in the correlated condition, F(1, 7) 5 racy on the test that immediately followed
22.86. correlated training was 59.1% correct versus
The left panel of Figure 4 shows each 50.0% correct on the test following uncorre-
group’s accuracy averaged over RIs on its first lated training, F(1, 7) 5 7.02, p , .05.
and second transfer tests in which the blue 1 Nonetheless, the averaged data indicate
and white X samples were substituted for the only very weak transfer of matching to the
samples used during many-to-one training. novel samples. Moreover, there was consid-
For Group Uncorr-Corr, the first test depicts erable variability in test performances across
performance following uncorrelated train- individual subjects. As shown in the leftmost
ing, whereas the second test depicts perfor- columns of Table 5, duration was most likely
mance following correlated training. For a cue for choice for C4. It may have also ex-
Group Corr-Uncorr, the first test shows trans- erted some weak conditional stimulus control
fer performance following correlated train- for C1, U2, U3, and U4, although their ab-
ing, and the second test shows transfer follow- solute levels of performance are not high, es-
ing uncorrelated training. pecially considering that the results for the U
Delayed matching with the two novel sam- birds came from their second test session.
ples was more accurate on the first test than After obtaining these results, we were con-
on the second for Group Corr-Uncorr, but cerned that testing with delays may have con-
the reverse was true for Group Uncorr-Corr. tributed to the relatively poor overall accu-
In other words, for both groups, matching ac- racy. Consequently, we conducted two
curacy with the novel samples was higher fol- additional transfer tests without delays. In
lowing correlated than uncorrelated training. preparation for them, each bird was first re-
DURATION PREDICTABILITY 291

Table 5 (1996, Experiment 2) were replicable in


Matching accuracy for each subject on the transfer tests many-to-one matching with contingencies
following uncorrelated and correlated training in Exper- that mimicked those in their correlated con-
iment 2.
dition. In other words, we found that long
Transfer with delays Transfer without delays sample durations produced higher levels of
Bird Uncorrelated Correlated Uncorrelated Correlated
delayed matching accuracy than short sample
durations when duration was unpredictable
C1 50.00 61.25 72.50 65.00 (i.e., uncorrelated with the sample stimuli).
C2 52.50 45.00 32.50 48.75 We also found that this effect disappeared
C3 46.25 43.75 57.50 75.00
C4 66.25 80.00 88.75 91.25 when duration was predictable (i.e., correlat-
U1 43.75 51.25 58.75 58.75 ed with the sample stimuli).
U2 42.50 61.25 62.50 63.75 Second, our transfer results indicated that
U3 41.25 61.25 51.25 62.50 duration exerted some conditional stimulus
U4 57.50 68.75 65.00 77.50
control after training in the correlated con-
Note. The order of the tests for the C birds was corre- dition. That control was certainly ‘‘spotty’’ in
lated-uncorrelated (the reverse of the order of the two the sense that only the results of a few pi-
sets of columns), whereas for the U birds it was uncor-
related-correlated. geons showed evidence of choices being cued
by the 5- versus 15-s duration. Furthermore,
the level of transfer was not particularly high.
turned to its Phase 1 task with 5- and 15-s Despite this, it is noteworthy that in each set
samples for a minimum of 10 sessions and of tests, 2 of the 4 birds receiving their first
until the three aforementioned performance transfer test after correlated training were less
criteria were once again met. They then re- accurate when subsequently tested again fol-
ceived two additional off-baseline sessions lowing uncorrelated training. This is the op-
with the blue 1 and white X stimuli, after posite of what would be expected on the basis
which they were given a third transfer test of practice.
with 0-s delays separating sample offset from In any event, any control by a duration cue
comparison onset. After this third test, each in the correlated condition makes the be-
bird was returned for a minimum of 10 ses- tween-condition comparisons and interpreta-
sions to its Phase 2 task and until it again met tions of delayed matching accuracy compli-
the performance criteria. This was followed cated, if not impossible. Thus, the procedure
by two additional off-baseline sessions with used in Experiment 1 becomes all the more
blue 1 and white X stimuli and, finally, a important for evaluating the predictions of
fourth zero-delay transfer test. the delay-reduction hypothesis. Our final ex-
The right panel of Figure 4 and the right periment returned to that procedure modi-
columns of Table 5 show the results of this fied in such a way that the delayed matching
second set of transfer tests. Once again, the trials were visually distinct from the intertrial
overall pattern of results showed more accu- interval.
rate matching of the ‘‘novel’’ samples to the
familiar comparisons following correlated
than following uncorrelated training. In Fig- EXPERIMENT 3
ure 4, this can be seen by the higher first- Experiment 3 was designed and conducted
session data point for Group Corr-Uncorr in exactly the same fashion as Experiment 1
and the higher second-session data point for except that the houselight was illuminated
Group Uncorr-Corr. Six of the 8 birds showed throughout each delayed matching trial and
this effect. The data in Table 5 suggest that was turned off during the ITI. Thus, the trial
duration was most probably a cue for choice and intertrial periods were distinct, a proce-
following correlated training for C3, C4, U3, dure commonly used in our laboratory for
and U4. studies of animal working memory (e.g., Ur-
cuioli & DeMarse, 1997; Urcuioli & Zentall,
D ISCUSSION 1986, 1990). At issue was whether the re-
There are two noteworthy findings from duced sample-duration effect observed with
Experiment 2. First, the correlated versus un- predictable sample durations also occurs un-
correlated differences reported by Hartl et al. der these training conditions.
292 PETER J. URCUIOLI et al.

A priori, there is no principled reason ac- until an incorrect comparison choice re-
cording to delay-reduction theory to expect sponse was made. The first 19 s of the ensu-
that the results should be any different from ing ITI was spent in darkness, after which the
what we found in Experiment 1 ( John Wix- houselight was again turned on for the sub-
ted, personal communication, May 19, 1998; sequent trial.
Edmund Fantino, personal communication, Four birds, 2 in each group, were dropped
May 30, 1998). Nevertheless, in some delayed from the experiment prior to completing any
matching tasks, performances can vary sub- of the major phases. Two birds did not reach
stantially depending upon the correspon- criterion on the initial zero-delay matching
dence (or lack thereof) between the condi- task after 70 training sessions, 1 did not meet
tions of the RI and those of the ITI (e.g., the performance criteria on the preliminary
Sherburne, Zentall, & Kaiser, 1998; Spetch & mixed-delay task after 50 sessions, and the re-
Rusak, 1992; see also Fetterman & MacEwen, maining bird did not meet these criteria after
1989; Zentall, 1997). If the same is true here, 30 sessions in its first uncorrelated condition.
then differentiating the trial from the inter-
trial interval could affect delay-reduction val- R ESULTS AND D ISCUSSION
ues in such a way that predictably short-du- Figure 5 shows delayed matching accuracy
ration samples no longer have a conditioned by sample duration and RI in the uncorrelat-
reinforcement advantage. For instance, seg- ed and correlated conditions averaged over
menting a session into distinct trial and in- pigeons and replications. Table 6 provides
tertrial periods might significantly alter the the individual-subject data. The results closely
bird’s within-session discrimination of the av- mirror those obtained in Experiment 1. Spe-
erage interreinforcement interval. If so, long- cifically, matching was more accurate with
duration samples might support higher levels long- than with short-duration samples when
of accuracy than short-duration samples in- duration was uncorrelated with the sample
dependently of duration predictability. stimuli, F(1, 7) 5 11.40. Seven of the 8 pi-
geons showed this effect. The overall effect
M ETHOD of RI and its interaction with duration in this
Subjects and Apparatus condition were not statistically significant,
Fs(1, 7) 5 4.99 and 1.33, respectively. By con-
Twelve White Carneau retired breeders trast, matching was more accurate overall
with similar histories to those used in the pre- with short- than with long-duration samples
ceding experiments were used. They were when duration was correlated with the sam-
housed and maintained in the same fashion ple stimuli, F(1, 7) 5 6.69. Seven of the 8
as the pigeons in Experiments 1 and 2. One pigeons showed this ‘‘reversed’’ effect. AN-
half of the birds were randomly assigned to a OVA also revealed significant effects of RI
group in which the alternating correlated and its interaction with sample duration in
and uncorrelated delayed matching sequence the correlated condition, Fs(1, 7) 5 31.78
began with the correlated condition. The oth- and 19.19, respectively.
er half were assigned to a group in which the As in Experiment 1, birds pecked the short-
sequence began with the uncorrelated con- duration samples at higher rates than the
dition. long samples, with the effect most pro-
The apparatus was identical to that previ- nounced at the end of correlated training.
ously described. Figure 6 shows mean rates of sample respond-
ing to the 5- and 15-s samples from each di-
Procedure
mension on the first and last sessions of cor-
The procedure for this experiment and the related training (averaged over the two
performance criteria for each phase were replications of that condition). On the first
identical to those of Experiment 1. The only session, the mean peck rate to the 5-s line
procedural difference was that the houselight sample was higher than to the 15-s sample,
was illuminated 1 s prior to trial onset (i.e., but the corresponding rates for the hue sam-
prior to the center-key presentation of the ples were virtually identical. With continued
white trial-ready stimulus) and remained on training in the correlated condition, however,
until the end of the reinforcement period or peck rates to the short-duration (5-s) samples
DURATION PREDICTABILITY 293

Fig. 5. Average proportion of correct choice responses by retention interval on trials with 5-s and 15-s samples in
the uncorrelated and correlated conditions of Experiment 3.

increased substantially, whereas rates to the but 1 bird (UNC1 on line-sample trials) show-
long-duration (15-s) samples decreased slight- ing this effect.
ly. Consequently, by the last correlated ses- In sum, this experiment indicates that the
sion, mean peck rates were considerably high- conditions associated with a delayed match-
er to the 5-s than to the 15-s samples, with all ing trial vis-à-vis those associated with the ITI

Table 6
Delayed matching accuracy by sample duration and retention interval for each pigeon in the
uncorrelated and correlated conditions of Experiment 3.

Uncorrelated Correlated
Bird Duration (s) 1s 6s Average 1s 6s Average

CU1 5 75.21 71.88 73.54 92.30 86.67 89.49


15 82.30 83.96 83.13 83.12 77.50 80.31
CU2 5 87.71 71.04 79.38 93.75 73.75 83.75
15 90.62 79.17 84.90 89.38 75.62 82.50
CU4 5 92.50 86.46 89.48 96.88 86.25 91.56
15 96.67 89.79 93.23 93.12 89.38 91.25
CU6 5 75.41 72.50 73.96 96.04 84.38 90.21
15 91.67 88.12 89.90 92.30 85.42 88.86
UNC1 5 86.25 80.00 83.12 95.62 87.71 91.67
15 87.71 85.62 86.67 90.42 88.54 89.48
UNC2 5 80.83 73.96 77.40 95.62 87.29 91.46
15 92.29 85.94 89.17 84.17 82.29 83.54
UNC3 5 66.25 75.42 70.83 88.96 82.08 85.52
15 80.83 83.54 82.18 88.34 83.33 85.83
UNC4 5 81.25 70.42 75.84 87.50 71.67 79.58
15 73.75 71.25 72.50 70.42 66.67 68.54
294 PETER J. URCUIOLI et al.

could not provide a cue for choice. This is an


important finding because the results of
Hartl et al. (1996, Experiment 2) could be
interpreted in an ‘‘apples versus oranges’’
manner. In other words, their results were po-
tentially, and most likely, contaminated by the
introduction of a duration cue in their cor-
related (predictable) condition, as suggested
by the transfer results from our second ex-
periment.
Given the indication from Experiment 2
that duration itself may at least partly cue
choice when it is correlated with the correct
comparison alternative, Experiments 1 and 3
provide better tests of how the purported
Fig. 6. Mean sample-response rates (in pecks per sec- conditioned reinforcing properties of the
ond) to the 5- and 15-s hue and line samples on the first samples affect their conditional stimulus con-
and the last sessions of the correlated conditions of Ex- trol. According to the delay-reduction hy-
periment 3.
pothesis, predictably short-duration samples
have higher conditioned reinforcing values
do not affect the effects of predictable versus than predictably long-duration samples, and
unpredictable sample durations on perfor- this difference in value should diminish the
mance. The implication, of course, is that de- usual effect of sample exposure time on
lay-reduction values associated with the vari- matching accuracy. The results of Experi-
ous sample stimuli, or at least the relative ments 1 and 3 clearly confirm this prediction:
ordering of those values, are not affected by Samples of different but predictable dura-
the segmentation of a session into distinct tri- tions had more similar effects on delayed
al and intertrial periods. matching accuracy than samples of different
but unpredictable durations.
Other predictability effects in the delayed
GENERAL DISCUSSION matching literature include signaling the
The results from the present experiments forthcoming RI (MacDonald & Grant, 1987;
add to a long list of findings showing that Wasserman, Grosch, & Nevin, 1982), the par-
conditional discrimination learning and per- ticular outcome following correct choices
formance are affected by predictable rela- (e.g., Urcuioli, 1990b), and common correct
tions besides those specifically comprising the choice responses (e.g., Urcuioli, Zentall, Jack-
matching contingencies themselves. Here, we son-Smith, & Steirn, 1989). For example,
showed, as Hartl et al. (1996) did before us, when the length of the RI is signaled by pre-
that the duration of a sample stimulus has a senting a cue along with or following the sam-
much smaller (and even the opposite) effect ple stimulus, delayed matching accuracy
on delayed matching accuracy when it is pre- tends to be higher at short intervals and low-
dictable than when it is not. Specifically, al- er at long intervals, relative to performance
though pigeons matched more accurately at those same intervals when they are unsig-
with long- than with short-duration samples naled. The full interpretation of this repro-
when duration was uncorrelated with the ducible finding (cf. MacDonald & Grant,
sample stimuli, they showed either no such 1987; Wasserman et al., 1982) is likely to in-
effect (Experiments 1 and 2) or the opposite clude heightened and lowered attention to
effect (Experiment 3) when duration was cor- samples preceding short and long RIs, re-
related with (predictable from) the samples. spectively, as well as differential delays to or
Our major contribution was to show that the rates of reinforcement associated with those
diminution (or reversal) of the typical dura- samples. Attentional factors, perhaps in the
tion effect (Grant, 1976; Nelson & Wasser- form of differential observing, are likely to
man, 1978; Roberts, 1972; Roberts & Grant, operate when the signal for the upcoming RI
1974) occurred even when duration itself is presented prior to or along with the sam-
DURATION PREDICTABILITY 295

ple. On the other hand, when the signal is samples will immediately transfer to the re-
presented following the sample, the predict- maining samples (Urcuioli et al., 1989, Ex-
ability effect must involve other (e.g., rein- periment 2; Wasserman, deVolder, & Cop-
forcement) variables (MacDonald & Grant, page, 1992). Although the precise
1987). mechanisms underlying this effect are not ful-
A better known predictability effect in the ly understood (but see Urcuioli, 1996), the
delayed matching literature involves differ- effect itself clearly depends upon the within-
ential outcomes (Trapold, 1970). In this par- session predictability of a common choice re-
adigm, each sample stimulus signals a differ- sponse by multiple samples (Urcuioli et al.,
ent reinforcing outcome for choosing the 1995).
correct comparison that follows it. These con- The processes responsible for the duration
tingencies typically generate faster acquisition predictability effect reported here are
and higher levels of delayed matching accu- thought to be related to the conditioned re-
racy than when the different outcomes are inforcing properties of each sample stimulus.
random (unpredictable) with respect to the The translation of predictable and unpredict-
samples (e.g., Peterson, 1984; Urcuioli, able sample durations into different condi-
1990b). The usual, and independently verifi- tioned reinforcing values has been described
able, interpretation of this effect is that when elsewhere in this paper and in Hartl et al.
the samples uniquely signal end-of-trial out- (1996). An important link in the argument
comes, they generate an additional cue for involves how those values produce variation
comparison choice, namely, an outcome expec- in delayed matching accuracy. One sugges-
tancy (Honig, Matheson, & Dodd, 1984; Pe- tion that has already been described is that
terson, 1984). Although there is some dispute samples with stronger conditioned reinforc-
regarding the nature of outcome expectan- ing properties support more vigorous observ-
cies (e.g., whether or not they are differential ing behavior. This connection seems quite
behavior conditioned to the samples; Urcuio-
plausible for two reasons. First, in concurrent-
li & DeMarse, 1994; cf. Urcuioli & Honig,
chains studies of conditioned reinforcement,
1980), there is much less dispute regarding
observing behavior occurs more frequently
how they enhance performance. It is gener-
when it produces stimuli associated with larg-
ally agreed that outcome expectancies pro-
er reductions in delay to reinforcement (i.e.,
vide an additional (redundant) cue for
choice (Urcuioli, 1991; Urcuioli & DeMarse, stronger conditioned reinforcers; e.g., Fanti-
1996). no & Moore, 1980). Second, Experiments 1
The many-to-one matching procedure itself and 3 showed that pigeons pecked predict-
also contains elements of predictability. In its ably short-duration samples more rapidly
simplest version (four samples and two com- than predictably long-duration samples. In
parisons), two perceptually different samples conjunction with previous findings that de-
signal a common comparison-choice re- layed matching accuracy increases the more
sponse. As a consequence of this common re- often pigeons peck the samples (Roberts,
lation, an additional cue for comparison 1972), these behavioral effects could easily re-
choice (besides the samples themselves) ap- duce the ‘‘normal’’ effect of sample duration
pears to develop—namely, what Urcuioli et on accuracy.
al. (1989) and others (e.g., Grant, 1982) have Nonetheless, this account of our data and
called a common code. The common code re- those of Hartl et al. (1996) is supported only
fers to the finding that the pigeon’s choices indirectly. The delay-reduction hypothesis of
in the many-to-one task are governed not conditioned reinforcement as it applies to the
only by the physical characteristics of the sam- samples in delayed matching would be but-
ple stimuli but also by their associative attri- tressed considerably by independent tests
butes (viz., the common comparison associa- (i.e., those conducted outside of the match-
tion the samples share with each other). This ing context) showing that pigeons will re-
can be demonstrated by a transfer-of-control spond more to produce samples with large
test. Specifically, after training on many-to- than with small delay-reduction values. Such
one matching, new comparison choices sub- a demonstration would solidify the explana-
sequently learned to a subset of the original tions for all of these results.
296 PETER J. URCUIOLI et al.

REFERENCES Santi, A., & Roberts, W. A. (1985). Prospective represen-


tation: The effects of varied mapping of sample stim-
Case, D. A., & Fantino, E. (1981). The delay-reduction uli to comparison stimuli and differential trial out-
hypothesis of conditioned reinforcement and punish- comes on pigeons’ working memory. Animal Learning
ment: Observing behavior. Journal of the Experimental & Behavior, 13, 103–108.
Analysis of Behavior, 35, 93–108. Sherburne, L. M., Zentall, T. R., & Kaiser, D. H. (1998).
Dinsmoor, J. A. (1983). Observing and conditioned re- Timing in pigeons: The choose-short effect may result
inforcement. Behavioral and Brain Sciences, 6, 693–728. from pigeons’ ‘‘confusion’’ between delay and inter-
Fantino, E. (1977). Conditioned reinforcement: Choice trial intervals. Psychonomic Bulletin & Review, 5, 516–
and information. In W. K. Honig & J. E. R. Staddon 522.
(Eds.), Handbook of operant behavior (pp. 313–339). En- Spetch, M. L., & Rusak, B. (1992). Time present and
glewood Cliffs, NJ: Prentice Hall. time past. In W. K. Honig & J. G. Fetterman (Eds.),
Fantino, E., & Moore, J. (1980). Uncertainty reduction, Cognitive aspects of stimulus control (pp. 47–67). Hills-
conditioned reinforcement, and observing. Journal of dale, NJ: Erlbaum.
the Experimental Analysis of Behavior, 33, 3–13. Stubbs, D. A. (1968). The discrimination of stimulus du-
Fetterman, J. G., & MacEwen, D. (1989). Short-term ration by pigeons. Journal of the Experimental Analysis of
memory for responses: The ‘‘choose-small’’ effect. Behavior, 11, 223–238.
Journal of the Experimental Analysis of Behavior, 52, 311– Trapold, M. A. (1970). Are expectancies based upon dif-
324. ferent reinforcing events discriminably different?
Grant, D. S. (1976). Effect of sample presentation time Learning and Motivation, 1, 129–140.
on long-delay matching in the pigeon. Learning and Urcuioli, P. J. (1985). On the role of differential sample
Motivation, 7, 580–590. behaviors in matching-to-sample. Journal of Experimen-
Grant, D. S. (1982). Prospective and retrospective cod- tal Psychology: Animal Behavior Processes, 11, 502–519.
ing of samples of stimuli, responses, and reinforcers Urcuioli, P. J. (1990a). Differential outcomes and many-
in delayed matching with pigeons. Learning and Mo- to-one matching: Effects of correlation with correct
tivation, 13, 265–280. choice. Animal Learning & Behavior, 18, 410–422.
Hartl, J. A., Dougherty, D. H., & Wixted, J. T. (1996). Urcuioli, P. J. (1990b). Some relationships between out-
Separating the effects of trial-specific and average come expectancies and sample stimuli in pigeons’ de-
sample-stimulus duration in delayed matching to sam- layed matching. Animal Learning & Behavior, 18, 302–
ple in pigeons. Journal of the Experimental Analysis of 314.
Behavior, 66, 231–242. Urcuioli, P. J. (1991). Retardation and facilitation of
Honig, W. K., Matheson, W. R., & Dodd, P. W. D. (1984). matching acquisition by differential outcomes. Animal
Outcome expectancies as mediators for discriminative Learning & Behavior, 19, 29–36.
responding. Canadian Journal of Psychology, 38, 196– Urcuioli, P. J. (1996). Acquired equivalences and medi-
217. ated generalization in pigeons’ matching-to-sample.
MacDonald, S. E., & Grant, D. S. (1987). Effects of sig- In T. R. Zentall & P. M. Smeets (Eds.), Stimulus class
naling retention interval length on delayed matching- formation in humans and animals (pp. 55–70). New
to-sample in pigeons. Journal of Experimental Psychology: York: Elsevier.
Animal Behavior Processes, 13, 116–125. Urcuioli, P. J., & DeMarse, T. B. (1994). On the rela-
McMillan, J. C. (1974). Average uncertainty as a deter- tionship between differential outcomes and differen-
minant of observing behavior. Journal of the Experimen- tial sample responding in matching-to-sample. Journal
tal Analysis of Behavior, 22, 401–408. of the Experimental Psychology: Animal Behavior Processes,
Miles, C. G., & Jenkins, H. M. (1973). Overshadowing 20, 249–263.
in operant conditioning as a function of discrimina- Urcuioli, P. J., & DeMarse, T. B. (1996). Associative pro-
bility. Learning and Motivation, 4, 11–27. cesses in differential outcome discriminations. Journal
Nelson, K. R., & Wasserman, E. A. (1978). Temporal of Experimental Psychology: Animal Behavior Processes, 22,
factors influencing the pigeons’ successive matching- 192–204.
to-sample performance: Sample duration, intertrial Urcuioli, P. J., & DeMarse, T. B. (1997). Memory pro-
interval, and retention interval. Journal of the Experi- cesses in delayed spatial discriminations: Response in-
mental Analysis of Behavior, 30, 153–162. tentions or response mediation? Journal of the Experi-
Peterson, G. B. (1984). How expectancies guide behav- mental Analysis of Behavior, 67, 323–336.
ior. In H. L. Roitblat, T. G. Bever, & H. S. Terrace Urcuioli, P. J., & Honig, W. K. (1980). Control of choice
(Eds.), Animal cognition (pp. 135–148). Hillsdale, NJ: in conditional discriminations by sample-specific be-
Erlbaum. haviors. Journal of Experimental Psychology: Animal Be-
Roberts, W. A. (1972). Short-term memory in the pi- havior Processes, 6, 251–277.
geon: Effects of repetition and spacing. Journal of Ex- Urcuioli, P. J., & Zentall, T. R. (1986). Retrospective cod-
perimental Psychology, 94, 74–83. ing in pigeons’ delayed matching. Journal of Experimen-
Roberts, W. A., & Grant, D. S. (1974). Short-term mem- tal Psychology: Animal Behavior Processes, 12, 69–77.
ory in the pigeon with presentation time precisely Urcuioli, P. J., & Zentall, T. R. (1990). On the role of
controlled. Learning and Motivation, 5, 393–408. trial outcomes in delayed discriminations. Animal
Rodger, R. S. (1975). The number of non-zero, post-hoc Learning & Behavior, 18, 141–150.
contrasts from ANOVA and error rate. I. British Jour- Urcuioli, P. J., Zentall, T. R., & DeMarse, T. (1995).
nal of Mathematical & Statistical Psychology, 28, 71–78. Transfer to derived sample-comparison relations by
Roitblat, H. (1980). Codes and coding processes in pi- pigeons following many-to-one and one-to-many
geon short-term memory. Animal Learning & Behavior, matching with identical training relations. Quarterly
8, 341–351. Journal of Experimental Psychology, 48B, 158–178.
DURATION PREDICTABILITY 297

Urcuioli, P. J., Zentall, T. R., Jackson-Smith, P., & Steirn, Wasserman, E. A., Grosch, J., & Nevin, J. A. (1982). Ef-
J. N. (1989). Evidence for common coding in many- fects of signaled retention intervals on pigeon short-
to-one matching: Retention, intertrial interference, term memory. Animal Learning & Behavior, 10, 330–
and transfer. Journal of Experimental Psychology: Animal 338.
Behavior Processes, 15, 264–273. Wixted, J. T. (1989). Nonhuman short-term memory: A
Warren, J. M. (1953). Additivity of cues in visual pattern quantitative reanalysis of selected findings. Journal of
the Experimental Analysis of Behavior, 52, 409–426.
discrimination by monkeys. Journal of Comparative and
Zentall, T. R. (1997). Animal memory: The role of ‘‘in-
Physiological Psychology, 46, 484–486. structions.’’ Learning and Motivation, 28, 280–308.
Wasserman, E. A., deVolder, C. L., & Coppage, D. J.
(1992). Nonsimilarity-based conceptualization in pi- Received February 4, 1999
geons. Psychological Science, 3, 374–379. Final acceptance July 7, 1999

You might also like