THE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY, 2001, 54B (2), 127–144

The role of nonreinforcement in the learning

of honeybees
P.A. Couvillon, C.D. Ablan, T.P. Ferreira, and M.E. Bitterman
University of Hawaii, Honolulu, USA

Two series of experiments with honeybees were designed to test the assumption that inhibition is
generated by nonreinforcement as a function of the excitatory value of the context. In the first
series (Experiments 1–3), summation tests with B were made after A+/C–/AB– as compared to
A+/C–/CB– training, with precautions taken to minimize the possibility of a masking effect of
excitatory within-compound conditioning on AB trials; responding to B did not vary with train-
ing procedure. In the second series (Experiments 4–5), retardation tests rather than summation
tests were used, in the belief that they might be more sensitive; after A+/AB–/CD– training,
acquisition in a B+/D– problem was found to be no less rapid than in a D+/B– problem. A third
series of experiments (Experiments 6–9) was designed to test the more general assumption that
the effectiveness of nonreinforcement increases with the excitatory value of the context; response
to B was found to be no different after A+/B+/C– training followed by A+/AB– training than
after A+/B+/C– training followed by A+/CB– training. The results are compatible with the
view that the role of nonreinforcement in honeybees is not to generate inhibition, but only to re-
duce excitation in a manner independent of the excitatory value of the context.

Learning in honeybees is similar in many respects to learning in vertebrates (Bitterman, 1988,

1996), but there are indications of difference as well that invite more detailed consideration.
One of them has appeared in an extensive series of recent experiments on the role of
nonreinforcement that failed to give evidence of inhibitory conditioning in honeybees
(Couvillon, Ablan, & Bitterman, 1999).
The best evidence of inhibitory conditioning in the vertebrate literature (Papini &
Bitterman, 1993) comes from the comparison of training procedures designed to develop what
Pavlov (1927) called “conditioned” inhibition (A+/AB– or CI training) and “differential”
inhibition (A+/B– or S– training). In experiments by Rescorla (1979) and by Pearce and Kaye
(1985) that reflected the “two-test” strategy of Rescorla (1969), a putative inhibitor (B) was
found to produce greater suppression of response to a known excitor (summation test) after CI
training than after S– training, and to condition less rapidly when subsequently reinforced

Requests for reprints should be sent to M.E. Bitterman, Békésy Laboratory of Neurobiology, 1993 East-West
Road, Honolulu, HI 96822, USA. Email: [email protected]
This research was supported by Grant IBN-9601991 from the U.S. National Science Foundation and by RCMI
Grant RR03061, MARC Grant GMO7684, and MBRS Grant GM56930 from the National Institutes of Health.

Ó 2001 The Experimental Psychology Society

http://www.tandf.co.uk/journals/pp/02724987.html DOI:10.1080/02724990042000092
128 COUVILLON ET AL.

(retardation test). Although it would be well to have further evidence on which to rest such
important conclusions, the results of these experiments encourage the view (1) that inhibitory
conditioning exists, and (2) that a stimulus becomes inhibitory as a function, not of
nonreinforcement alone (as Pavlov thought), but of the excitatory value of the context in
which it is nonreinforced. The second proposition—derived from the theory of Rescorla and
Wagner (1972; see, especially, Wagner & Rescorla, 1972) and subsequently from a number of
other theories (Kehoe, 1988; Pearce, 1994; Pearce & Hall, 1980; Wagner, 1981)—is referred to
here as the Rescorla–Wagner proposition. If the two training procedures were not differen-
tially effective, neither proposition would be supported.
In contemplation of analogous experiments designed to look for evidence of inhibitory con-
ditioning in honeybees, Couvillon et al. (1999) did some pilot work on CI versus S– training
using only a summation test, which for free-flying foragers could be implemented more easily
than could a retardation test. The plan was to go on to a formal investigation involving tests of
both kinds if the results of an initial experiment were positive, but they were negative, and so
also were the results obtained in six subsequent experiments of varying designs. Although
positive summation results would not in themselves have been conclusive (leaving open the
possibility of an attentional interpretation), the negative results give no reason to suppose that
inhibition in honeybees is generated by nonreinforcement as an increasing function of the
excitatory value of the context, or even that inhibition is generated at all.
The role of nonreinforcement in the learning of honeybees was studied further in the three
sets of experiments reported here. In the first set, we attempted to minimize the possibility
that the inhibitory conditioning of B is masked by excitatory within-compound association on
AB trials in CI training. In the second set, on the supposition that retardation tests would
afford a more sensitive measure of inhibition than would summation tests, we looked for
retarded acquisition after CI as compared with S– training. In the third set, we went beyond
the search for evidence of inhibition to consider whether the effect of nonreinforcement in
honeybees is at all dependent on the excitatory value of the context.

EXPERIMENTS 1–3
Within-compound association
In these experiments, the possibility was explored that B does indeed become more inhibitory
in CI than in S– training, but that the difference is masked under some circumstances by excit-
atory within-compound conditioning—by the association of B with A on AB trials (Rescorla,
1981). In an effort to minimize the influence of within-compound conditioning, which has
been found in honeybees (Couvillon & Bitterman, 1982) as well as in vertebrates, we scheduled
nonreinforced trials with A before summation tests with B, a strategy used previously in work
both with vertebrates (Durlach & Rescorla, 1980; Rescorla & Colwill, 1983) and with honey-
bees (Funayama, Couvillon, & Bitterman, 1995),
The animals were trained with gray targets (x) that could be labelled with odours or colours
(A, B, C). The experimental design is summarized in Table 1. In Stage 1 of each experiment,
there was differential reinforcement of A and C (A+/C– training). In Stage 2, that training
continued, but with added AB– trials for a conditioned inhibition (CI) group, and added CB–
trials for a differential conditioning (S–) group; that is, for the CI group, the new stimulus
(B) was nonreinforced in compound with an excitor, whereas for the S– group it was
 NONREINFORCEMENT IN HONEYBEES 129

TABLE 1
Experimental design

Exp. Group Stage 1 Stage 2 Stage 3 Test

a
1–3 CI Ax+/Cx– Ax+/Cx–/ABx–/x+ Ax–/Cx–/x+ Bx–
S– Ax+/Cx– Ax+/Cx–/CBx–/x+ Ax–/Cx–/x+ Bx–

4–5 CI Ax+/ABx–/CDx– Bx+/Dx–

Control Ax+/ABx–/CDx– Dx+/Bx–
b
Supplementary Ax+/ABx–/CDx– Ax–/ABx–

6–9 RW Ax+/Bx+/Cx– Ax+/ABx– Bx–

Control Ax+/Bx+/Cx– Ax+/CBx– Bx–
Note: A, B, C, and D were colours and odours used to label x, a gray, unscented target; + and – indicate
reinforcement and nonreinforcement.
a
The design of Experiment 3 was like that of Experiments 1 and 2 except that the test was a choice between Bx– and x–
b
presented as a pair. The design of Experiment 5 was like that of Experiment 4 except that there was no supplementary
group.

nonreinforced in compound with an equally familiar stimulus that was never reinforced. In
Stage 3, for both groups, there was nonreinforced training with both A and C that was
intended to reduce any excitatory value of B attributable to association with A in the prior AB–
training of the CI group. Then there was a test with B designed to look for differential suppres-
sion of response to x, the unlabelled gray target that had been separately reinforced in the
training. The Rescorla–Wagner proposition suggests that responding in the test with B should
be less in the CI than in the S– group.

Method
Subjects
The subjects were 49 foraging honeybees, all experimentally naïve, from our own hives situated near
the laboratory. They were randomly assigned to the CI and S– groups of the three experiments, each of
which had eight subjects except for the CI group of Experiment 2, to which a ninth animal was inadver-
tently assigned.

Procedure
The subjects of these and all subsequent experiments were trained individually, each in a single ses-
sion lasting several hours. In the pretraining, a forager was selected upon arrival at an established feeding
station providing 10–12% sucrose solution, picked up in a match box, carried to a laboratory window,
and set down at a large (100-m l) drop of 50% sucrose solution on a pretraining target (Stimulus A
described below) centred on the deep sill of the window. The animal was marked with a spot of coloured
lacquer as it fed to repletion (taking 50 m l of the solution on average), after which it left for the hive.
Typically, the animal would return to the laboratory for more sucrose a few minutes later, continuing to
fly back and forth between the hive and the window as long as sucrose was available there. If the marked
animal did not return after its first placement, it was picked up at the feeding station (where it could
usually be found) and carried to the pretraining target, where it was permitted again to feed to repletion.
More than two placements were required only rarely. The pretraining ended with the animal’s second
return to the window of its own accord.
130 COUVILLON ET AL.

The targets were plastic petri dishes, 5.5 cm in diameter, with gray covers (x). Drilled in each cover,
6 mm from its outer circumference, was a circle of eight equally spaced holes, 5 mm in diameter. The
dishes contained pieces of cotton batting that could be impregnated with the odours of geraniol, pepper-
mint, or hexanol, which served (their roles balanced over subjects) as A, B, and C in Experiment 1, and as
A and C in Experiments 2–3. (Where a dish was labelled with two odours, they were not mixed, but
applied to the batting in separate drops at opposite sides of the dish.) Some of the covers were labelled
with a disk of yellow plastic, 2.5 cm in diameter, which served as B in Experiments 2–3. A target used on
any trial of a visit was drawn at random from a set of identical targets to which it was returned after wash-
ing of its cover at the end of the visit; the purpose of this procedure (used routinely in our laboratory) is to
randomize irrelevant stimuli.

Experiments 1–2. In Stage 1 of each of these experiments, there were 12 choice trials with A+ and
C– (different odours), A containing a 10-m l drop of 50% sucrose solution as reward, and C containing a
10-m l drop of water; the function of the water, unacceptable to the animals and distinguishable from the
sucrose only by taste, was to rule out the possibility of differential responding to the positive target on the
basis of the perceived presence of the drop of sucrose solution. On each trial, the two targets were set 10
cm apart in a horizontal arrangement (parallel to the edge of the window sill), their positions balanced
quasi-randomly over trials, and the initial choice of the animal (landing on one or the other of the targets)
was recorded. If the subject went first to C, it was permitted to correct its choice, and while it was ingest-
ing the 10 m l of sucrose solution provided by A, which required about 10 s, C was removed by the experi-
menter. After the animal had finished the sucrose, it would fly up from the target and hover over the
window sill while the experimenter removed the empty target and set out two fresh targets for the next
trial. The animal was observed through a pair of sliding panels, separating it from the experimenter,
which could be displaced quickly to permit manipulation of the targets by the experimenter. When the
animal was replete (after about five reinforcements), it would leave for its hive, returning to the window
several minutes later for another series of trials.
In Stage 2, there were 12 additional choice trials with A+ and C– that were interspersed quasi-ran-
domly with 6 x+ trials and 12 non-reinforced compound trials (AB– for the CI group and CB– for the S–
group); B was a third odour in Experiment 1 and a yellow disk in Experiment 2. The procedure on each
compound trial was to present the compound with a 10-m l drop of water for 1 min and to record all con-
tacts with it in that period.
When, in work with free-flying foragers, multiple trials with small reward are scheduled in a series of
visits, the number of trials on any given visit is determined by the animal; the training continues on each
visit until the animal is replete (has taken, on average, about 50 m l of sucrose solution) and leaves for the
hive of its own accord to deposit the sucrose. The series of visits ends only once the animal has had at least
the scheduled number of trials of each kind, which means that the actual number of trials of each kind
may be different from the scheduled number, but the approximation is close. In Stage 2 of Experiment 1,
there were on average 13.4 (rather than 12) choice trials, 6.2 (rather than 6) x+ trials, and 12.7 (rather
than 12) compound trials. The corresponding frequencies in Experiment 2 were 13.1, 6.1, and 12.3.
In Stage 3, there were four visits, each beginning with a 1-min choice trial with A– and C– (both tar-
gets containing water), during which all responses to each target were recorded. At the conclusion of the
trial, both targets were removed and replaced by a single unlabelled target (x alone) containing a 100-m l
drop of 50% sucrose solution, from which there was feeding to repletion. The function of the feeding to
repletion on x was two-fold—first, to encourage the return of the animal to the experimental situation
(unsatiated, it would have gone off to forage elsewhere) and, second, to enhance the excitatory strength of
x, which was to be used in the subsequent summation test with B.
On the visit following the last visit of Stage 3, there was a nonreinforced presentation of Bx—a gray
target labelled in Experiment 1 with the odour that served as B, and in Experiment 2 with the yellow disk
that served as B. The target contained a 100-m l drop of water, and all responses to it in a 10-min period
 NONREINFORCEMENT IN HONEYBEES 131

were recorded. The expectation based on the Rescorla–Wagner proposition was that suppression of
responding to x by B would be greater in the CI group than in the S– group.

Experiment 3. The training in this experiment (with a colour as B and odours as A and C) was the
same as in the prior experiments, except that Stages 1 and 2 were defined in terms of number of visits (2
and 6, respectively), as a result of which there were fewer trials in Stage 1 (7) and more trials in Stage 2: on
average, 16.5 choice trials with A+ and C–, 8.0 x+ trials, and 16.3 nonreinforced compound trials (AB–
for the CI group and CB– for the S– group). The main difference between Experiment 3 and the previ-
ous ones was that, instead of a test with Bx, there was a 10-min nonreinforced choice test with x and Bx—
an unlabelled gray target and one that was labelled with the yellow disk that served as B; each target
contained a drop of water, and responses to each were recorded. The expectation based on the Rescorla–
Wagner proposition was that the preference for x over Bx would be greater in the CI group than in the
S– group.

Results
Experiments 1–2. Having been pretrained with A+, the animals of each experiment came
to Stage 1 with a preference for A over C, and the preference was maintained as the A+/C–
training continued. The mean probability of choosing A in Stage 1 was 0.91 in Experiment 1
and 0.95 in Experiment 2. With the introduction in Stage 2 of AB– trials for each CI group and
CB– trials for each S– group, the preference for A over C remained much the same in both
groups of Experiment 1, and in the S– group of Experiment 2, but declined in the CI group of
Experiment 2. In Figure 1, performance on choice trials of Stage 2 is plotted in terms of the
mean probability of choosing A over successive 2-trial blocks (left panel, Experiment 1; right
panel, Experiment 2). ANOVA shows a significant difference between the CI and S– groups
of Experiment 2, F(1, 15) = 11.96, p < .05 (the a -level used throughout), although not of
Experiment 1 (F < 1)—that is, the previously established A+/C– discrimination was
impaired by the addition of AB– trials when (in Experiment 2) B was a colour, but not when (in
Experiment 1) B was another odour, perhaps because of a difference in salience. That odours
of the intensities routinely employed in our experiments—not odours in general (Bitterman &
Couvillon, 1991)—are more salient than the colours is suggested by the results of some of our
earliest experiments with honeybees in which odour overshadowed colour but was not over-

Figure 1. Performance in Stage 2 of Experiments 1 (left panel) and 2 (right panel). The curves are plotted in terms
of the mean choice of A rather than C in successive 2-trial blocks.
132 COUVILLON ET AL.

shadowed by colour, and in which response to odour alone was much greater than to colour
alone in a choice test given after training with the compound (Couvillon & Bitterman, 1980,
1982).
On nonreinforced trials with the compounds in Stage 2, there was more responding to AB
by the CI groups than to CB by the S– groups, which was to be expected on the basis of the
greater excitatory value of A as compared with C. As training continued, responding to AB
declined progressively from 5.1 to 1.9 in Experiment 1 and from 5.8 to 3.3 in Experiment 2
(which provides an indication of the effectiveness of the training technique). ANOVA based
on two-trial blocks yields significant block effects, F(5, 35) = 2.58 for Experiment 1, and F(5,
40) = 4.99 for Experiment 2. Responding on CB– trials was low throughout, 1.6 in Experiment
1 and 2.0 in Experiment 2. If inhibition develops as a function of nonreinforcement apart from
the excitatory value of the context, these differences in frequency of nonreinforced responding
to the compounds might alone lead us to expect less responding by the CI groups than by the
S– groups when tested with B, and by Rescorla–Wagner theory they should increase the likeli-
hood of getting such a result. Unfortunately, there is no way in work with free-flying honey-
bees to exactly equate experience with the stimuli.
Mean responding to A and C on the nonreinforced trials of Stage 3 in each experiment is
plotted in Figure 2. The initial preference for A in all four groups was not entirely eliminated

Figure 2. Performance of the CI and S– groups in Stage 3 of Experiments 1 (upper panel) and 2 (lower panel). The
curves are plotted in terms of the mean number of responses to A (formerly reinforced) and C (formerly
nonreinforced) during nonreinforced exposures of the two targets.
 NONREINFORCEMENT IN HONEYBEES 133

in the course of the training, and it is difficult to say how much further training would have
been necessary for that purpose, but it is clear that whatever masking effect might be attrib-
uted to within-compound association of B with A in the CI groups must at least have been sub-
stantially reduced. In each experiment, there was a significant decline in responding to A, F(3,
42) = 11.59 in Experiment 1 and F(3, 45) = 7.44 in Experiment 2, with no significant group
effect in either experiment, F < 1 in Experiment 1 and F(1, 15) = 2.83 in Experiment 2.
In Figure 3, the test results are plotted in terms of the mean cumulative number of
responses to Bx in successive 30-s intervals of the 10-min test period. In each experiment,
there was as much responding in the CI group as in the S– group (F < 1), an outcome that fails
to support the Rescorla–Wagner proposition. A striking feature of the results is the far greater
frequency of responding in Experiment 2 than in Experiment 1. The difference may be
explainable on the assumption of greater generalization of excitation from x (the unscented
gray target, which was independently reinforced in training) to the same target labelled with a
colour rather than with an odour. Another possibility is that B became inhibitory in both
experiments, but odour more so than colour, perhaps because of greater salience; similar per-
formance of CI and S– groups in these experiments may speak against the Rescorla–Wagner
proposition, but not necessarily against the reality of inhibition.

Experiment 3. The training in this experiment was very much like that in Experiment 2,
with A and C as odours and B as a colour, and the results were very much the same. There was a

Figure 3. Test performance of the CI and S– groups of Experiments 1 (upper panel) and 2 (lower panel) plotted in
terms of mean cumulative number of responses in successive 30-s intervals.
134 COUVILLON ET AL.

0.87 preference for A in Stage 1. With the introduction in Stage 2 of AB– trials for the CI
group and CB– trials for the S– group, the preference for A over C remained much the same in
the S– group (0.88) and worsened in the CI group (0.56). ANOVA shows a significant differ-
ence between the CI and S– groups, F(1, 14) = 45.87. On nonreinforced trials with the com-
pounds in Stage 2, as in the previous experiments, there was significantly more responding to
AB by the CI groups than to CB by the S– group. The means were 4.8 for the CI group and 1.1
for the S– group, F(1, 14) = 96.32. On the nonreinforced trials with A and C in Stage 3,
responding to A declined progressively from 5.3 to 3.3, F(3, 42) = 6.01, with no significant
group effect, F < 1. The curves look very much like those for Experiment 2 in Figure 2.
In Figure 4, the test results for each group are plotted in terms of the mean cumulative
number of responses to x and Bx. ANOVA shows significantly more responding to x than to
Bx, F(1, 14) = 19.31. The difference is actually somewhat greater for the CI group than for the
S– group, although the Group ´ Stimulus interaction is statistically unreliable (F < 1).
Whether greater responding to x than to Bx is explained in terms of afferent interaction and
generalization decrement or on the assumption that B is inhibitory, there is no evidence here
for the proposition that the development of inhibition is supported by an excitatory context.
The more frequent responding to x by both groups does, however, have a bearing on the inter-
pretation of the results of Experiment 2, where the A+/C– discrimination was impaired in the
CI group by the introduction of AB– trials, and where there was substantial responding by
both groups to B x in the test. The possibility that B (as a colour) was simply not well

Figure 4. Test performance of the CI (upper panel) and S– (lower panel) groups of Experiment 3 plotted in terms
of the mean cumulative number of responses to x and Bx in successive 30-s intervals.
 NONREINFORCEMENT IN HONEYBEES 135

discriminated from x in the course of training with A and C (the relatively more salient odours)
can be discounted on the basis of the results of Experiment 3, in which the training procedure
was very much the same as in Experiment 2. There was clearly a good deal of learning about B,
evidence again of the effectiveness of the training technique.

EXPERIMENTS 4–5
Retardation tests
In these experiments, with a design like one of those employed by Couvillon et al. (1999), we
used retardation tests rather than summation tests, on the assumption that they might prove to
be more sensitive. Of the four two-test vertebrate experiments reviewed by Papini and
Bitterman (1993) that they considered to be of acceptable design, two showed slower acquisi-
tion of response to B after CI than after S– training, but their summation results were negative
(Cunningham, 1979; Hoffman & Fitzgerald, 1982). Given that the general purpose of summa-
tion testing in the two-test strategy is to rule out an explanation of retardation on the basis of
decreased attention to the putative inhibitor—an explanation difficult to credit in the case of
successful A+/AB– training—Papini and Bitterman suggested that the retardation results
alone favoured an inhibitory interpretation and that the summation tests were simply less sen-
sitive. In any case, even if the results of our summation tests had been positive, retardation
tests would still be necessary to rule out an explanation in terms of the enhancement of atten-
tion to B by A+/AB– training.
In each of the present experiments, two groups of animals—a CI group and a control
group—were trained A+/AB–/CD–, after which they were given differentially reinforced
choice training with B and D, B positive for the CI group and D positive for the control group.
The experimental design is summarized in Table 1. The Rescorla–Wagner proposition sug-
gests that the CI group should learn less rapidly than the control group. As the negative results
of Experiments 1–3 did not encourage the suspicion that a differential effect of CI and S–
training had been masked by excitatory within-compound association in previous experi-
ments (the results of which also were negative), no further attention was given to the possibil-
ity in subsequent experiments.

Method
Subjects
The subjects were 40 foraging honeybees, all experimentally naïve, from our own hives situated near
the laboratory. They were randomly assigned to the CI and control groups of each of the two experiments
and to a supplementary group in Experiment 4—8 subjects in each group.

Procedure
The training situation and the training materials were much the same as in the preceding experi-
ments; the gray targets (x) could be labelled with peppermint, geraniol, yellow, blue, or one of the four
possible colour–odour pairs. A and C were peppermint and geraniol (balanced over subjects) in Experi-
ment 4, and B and D were yellow and blue (balanced over subjects); in Experiment 5, the roles of the two
sets of stimuli were reversed (A and C the colours, B and D the odours). The recruiting and pretraining
procedures also were very much the same as in the previous experiments.
136 COUVILLON ET AL.

In the first stage of each experiment, there were 12 visits, on each of them a single trial terminating
with a large reward (a 100-m l drop of 50% sucrose solution, from which there was feeding to repletion).
Of the 12 trials, 2 were A+ trials, and the rest (in quasi-random order) were differential conditioning
trials, 5 of them beginning with AB– and 5 with CD–. Each compound, containing a 100-m l drop of
water, was presented for 60 s on the first trial with that compound, 90 s on the second, and 120 s on subse-
quent trials, during which time all contacts with the target were recorded. Then the compound was
replaced with a pair of targets—A+/AB– on half the trials, A+/CD– on the rest—and choice was mea-
sured. The two targets were set 10 cm apart in a horizontal arrangement (parallel to the edge of the win-
dow sill), their positions balanced quasi-randomly over trials, the positive target containing a 100-m l
drop of the sucrose solution and the negative target containing a 100-m l drop of water. If the subject went
first to the negative target, it was permitted to correct its choice.
In the testing stage, there were 8 further training visits for the CI and control groups, on each of which
there was a differentially reinforced choice trial with B and D. For the CI group of each experiment, B
was positive (B+/D– training); for the control group, D was positive (D+/B– training). The supple-
mentary group of Experiment 4, which had been trained like the CI and S– groups in Stage 1 of that
experiment, now had a nonreinforced choice test with A and AB—the A odour and the B colour—set as
usual 10 cm apart on the window sill; each target contained a 100-m l drop of water, and all contacts with
each in a 10-min period were recorded. The purpose of the supplementary group was to provide inde-
pendent evidence of the effectiveness of the discrimination training in Stage 1.

Results
In Stage 1 of each experiment, there was significantly more responding to AB– than to CD– on
the unreinforced presentations of those compounds,3.2 versus 0.6 in Experiment 4, F(1, 15) =
66.59, and 1.7 versus 0.6 in Experiment 5, F(1, 15) = 92.37. The choice data show better-than-
chance performance on the choice trials with A+ and AB–, 0.76 in Experiment 4, t(15) = 7.00
(two-tailed), and 0.83 in Experiment 5, t(15) = 8.09. There was also better-than-chance per-
formance on choice trials with A+ and CD–, 0.95 in Experiment 4, t(15) = 15.59, and 0.96 in
Experiment 5, t(15) = 22.95. Performance on the A+/CD– trials was significantly better than
on the A+/AB– trials, t(15) = 3.76 in Experiment 4 and t(15) = 3.10 in Experiment 5.
In Figure 5, the performance of the supplementary group in the extinction test with A and
AB is plotted in terms of the mean cumulative frequency of responding to each of the

Figure 5. Test performance of the supplementary group of Experiment 4 plotted in terms of the mean cumulative
number of responses to A and AB in successive 30-s intervals.
 NONREINFORCEMENT IN HONEYBEES 137

Figure 6. Performance in acquisition of the CI and control groups in Stage 2 of Experiments 4 (left panel) and 5
(right panel) plotted in terms of the mean proportion of correct choice in successive two-trial blocks.

alternatives in successive 30-s intervals of the 10-min test period. ANOVA shows a significant
stimulus (A versus AB) effect, F(1, 7) = 40.32, and a significant Stimulus × 2.5-min Block
interaction, F(3, 21) = 20.65. Whether the difference is to be understood in terms of afferent
interaction or on the assumption that B acquired inhibitory properties, there can be no doubt
about the effectiveness of the discrimination training in Stage 1.
In Figure 6,the course of acquisition in the retardation test of each experiment is plotted in
terms of the mean proportion of correct choice in successive two-trial blocks. The discrimina-
tion between the two odours in Experiment 5 developed more rapidly than did the discrimina-
tion between the two colours in Experiment 4, perhaps again because of greater salience of the
odours. In neither case, however, was the performance of the CI group (for which B was posi-
tive) poorer than that of the control group (for which D was positive), as the Rescorla–Wagner
proposition suggests it should be; both small mean differences are, in fact, in the opposite
direction. In neither case was there a significant group effect (F < 1) or a significant Group ×
Block interaction (F < 1). The retardation results, like those of our previous summation exper-
iments, fail to support the proposition that a stimulus becomes inhibitory as a function of the
excitatory value of the context in which it is nonreinforced.

EXPERIMENTS 6–9
The decremental hypothesis
In these experiments, we tested the more general hypothesis, implicit in the Rescorla–Wagner
proposition, that the effectiveness of nonreinforcement varies directly with the excitatory
value of the context. Couvillon et al. (1999) dubbed it the “decremental hypothesis” to distin-
guish it from the complementary “incremental hypothesis” (tested in blocking experiments)
that the effectiveness of reinforcement varies inversely with the excitatory value of the con-
text. The decremental hypothesis is meaningful quite apart from the assumption of inhibi-
tion—see, for example, the purely excitatory model of generalization by Blough (1975), which
incorporates the Rescorla–Wagner principle of shared associative strength—and it is not con-
tradicted by the null results of our summation and retardation experiments. If the only func-
tion of nonreinforcement in honeybees is to reduce associative strength (assumed always to be
138 COUVILLON ET AL.

³ 0), the decremental hypothesis does not predict differential responding to B after CI as com-
pared with S– training because the associative strength of B is negligible to begin with. It does,
however, predict less responding to B where the two kinds of training are preceded by rein-
forced trials with B, a procedure suggested by Wagner, Saavedra, and Lehman (Wagner &
Rescorla, 1972, pp. 311–312). Their results for rabbits are in accord with the prediction, and so
also are some results for rats (Rescorla, 2000), although not for pigeons (Pearce & Wilson,
1991).
Each of the present experiments began with A+/B+/C– training, after which an RW
group had training with A+ and AB– (B nonreinforced in compound with the other excitor),
whereas a control group had training with A+ and CB– (B nonreinforced in compound with
the nonreinforced alternative). Then for both groups there was a test with B. The experimen-
tal design is summarized in Table 1. The decremental hypothesis suggests that there should be
less responding to B in the RW group than in the control group.

Method
Subjects
The subjects were 64 foraging honeybees, all experimentally naïve, from our own hives situated near
the laboratory. They were randomly assigned to the RW and control groups of the four experiments, 8
subjects in each group.

Procedure
The training situation and the training materials were much the same as in the preceding experi-
ments. The gray targets were labelled, as in Experiments 4 and 5, with peppermint, geraniol, yellow, and
blue, but here an adjacent visual landmark was also used in two of the experiments. The landmark was a
white wooden block, 11 cm long and 4 cm in each of its other dimensions, lying on its long side, parallel to
the outer edge of the window sill, at a distance of 1 cm directly behind the target; the salience and
discriminability of the landmark had been established in prior work on blocking (Couvillon, Arakaki, &
Bitterman, 1997). In Experiments 6 and 9, A and C were colours and B was the landmark; in Experiment
7, A and C were colours and B was an odour; in Experiment 8, A and C were odours, and B was a colour.
The recruiting and pretraining procedures also were very much the same as in the previous experiments.

Experiments 6–8. In Stage 1 of each of these experiments, there were 8 visits, on each of which there
was a single trial terminating in large reward (feeding to repletion from a 100-m l drop of 50% sucrose
solution). Two of the visits (in balanced order) were to A+ and two were to B+. On the four remaining
visits, the animal found C– (containing a 100-m l drop of water), which was removed after the first
response to it and replaced either with A+ (on two visits) or with B+ (on two visits). In Stage 2, there were
again 8 visits. For the RW group, there were 2 visits to A+ and 6 to AB–, which was replaced with A+
after the first response to it. For the control group, the training was the same, except that CB– trials were
substituted for the AB– trials. On returning from the hive after the last training visit, each animal was
tested with B– for a 10-min period during which all contacts with the target were recorded.

Experiment 9. The training here was with multiple trials per visit, including choice trials, and a
smaller reward (10 m l of the 50% sucrose solution). In Stage 1, 10 A+ trials, 10 B+ trials, and 10 C– trials
(containing a 10-m l drop of water and withdrawn after the first response to it) were scheduled in quasi-
random order. As usual, the scheduled trials continued on each visit until the animal was replete (on
 NONREINFORCEMENT IN HONEYBEES 139

average, after about 5 rewarded trials) and returned to the hive of its own accord to deposit the sucrose.
The actual frequencies of the trials of the three kinds averaged 11.2, 10.8, and 10.9, respectively. In
Stage 2, for the RW group, there were 30 scheduled AB– trials and 30 scheduled choice trials with A+
and AB–, the two targets set 10 cm apart in the usual way; if the animal went first to AB, it was free at once
to correct its choice. For the control group, the nonreinforced compound was CB rather than AB. For
both groups, the choice trials provided a measure of discrimination between A and the nonreinforced
compound. The actual mean frequencies in six visits were 30.3 compound trials and 31.5 choice trials.
On returning from the hive after the last training visit, each animal was tested with B– for a 10-min
period during which all contacts with the target were recorded.

Results
Experiments 6–8. In Figures 7–9, the results of Experiments 6–8 are plotted in terms of
the mean cumulative frequency of responding in successive 30-s intervals of the 10-min test
with B. As the plots suggest and ANOVA confirms, there was no significant difference in the
performance of the RW and control groups in Experiment 6 (F < 1), where B was the land-
mark, or in Experiment 8 (F < 1), where B was a colour—no support in either case for the
hypothesis that the effectiveness of nonreinforcement increases with the excitatory value of
the context. As the level of responding in each case was substantial, there is no basis for sus-
pecting the operation of a floor effect. In Experiment 7, where B was an odour, there was a sig-
nificant group effect, F(1, 15) = 6.23, with a significant Group × 2.5-min Interval interaction,
F(3, 45) = 5.29, but the direction of the difference is opposite to that expected on the basis of
the decremental hypothesis. What the difference means is not clear.

Experiment 9. In Figure 10, the performance of the two groups on the choice trials of
Stage 2 is plotted in terms of the mean proportion of correct choice in successive visits. As
might be expected, the A+/CB– performance of the control group was significantly better
than the A+/AB– performance of the RW group, F(1, 14) = 21.5—clearly, the control ani-
mals were affected by the differential reinforcement of A and C in Stage 2—although the A+/
AB– performance was significantly better than chance, t(7) = 2.93 (two-tailed). Despite the
fact that both components of the nonreinforced compound must have been strongly excitatory
at the outset of Stage 2, it is not clear why the A+/AB– discrimination should have been quite

Figure 7. Test performance of the RW and control groups of Experiment 6 plotted in terms of the mean cumulative
number of responses in successive 30-s intervals.
Figure 8. Test performance of the RW and control groups of Experiment 7 plotted in terms of the mean cumulative
number of responses in successive 30-s intervals.

Figure 9. Test performance of the RW and control groups of Experiment 8 plotted in terms of the mean cumulative
number of responses in successive 30-s intervals.

Figure 10. Performance of the RW and control groups on the choice trials in Stage 2 of Experiment 9 plotted in
terms of the mean proportion of correct choice in successive visits.

140
 NONREINFORCEMENT IN HONEYBEES 141

Figure 11. Test performance of the RW and control groups of Experiment 9 plotted in terms of the mean cumula-
tive number of responses in successive 30-s intervals.

so poor. In any case, performance in the terminal extinction test, which is plotted in Figure 11,
showed that their different training experiences reduced the excitatory value of B to the same
low level for both groups (F < 1).

DISCUSSION
On the proposition of Rescorla and Wagner (1972) that inhibition is generated by
nonreinforcement in proportion to the excitatory value of the context, B is expected to be more
inhibitory after CI than after S– training. The positive results obtained in a few such experi-
ments with vertebrates point at the same time, not only to the correctness of the proposition,
but also to the reality of inhibitory conditioning. For honeybees, the results obtained previ-
ously by Couvillon et al. (1999) and here in Experiments 1–5 are entirely negative, from which
it might be concluded either that our tests of the proposition have been insufficiently sensitive
or that the proposition does not hold for honeybees.
As to the sensitivity of our tests, it should be understood that the comparison of CI and S–
training is a comparison of two potentially inhibitory treatments, because S– is expected to
become inhibitory to the extent that the training background becomes excitatory. To increase
the differential effectiveness of the two training procedures, ways might be sought of minimiz-
ing the excitatory conditioning of the background. Other paradigms might be explored as well.
In recent work on proboscis-extension conditioning in harnessed honeybees, Hellstern,
Malaka, and Hammer (1998) examined the possibility (which follows also from the Rescorla–
Wagner proposition) that backward conditioning is inhibitory. Their only dependable finding
was of poorer forward conditioning after a single backward trial with a US–CS interval of 15 s
as compared with 6 s, which might be attributed to less excitation in the 15-s group rather than
to inhibition in the 6-s group; better conditioning after training with a longer as compared with
a shorter backward interval would be a more promising indication of inhibition. Unfortu-
nately, rigorous experiments on backward conditioning in free-flying honeybees are unfeasi-
ble, but further work with the proboscis-extension technique, which affords better control
of stimulation, might be useful. A limitation of the technique is that the response can be
142 COUVILLON ET AL.

conditioned only to a narrow range of stimuli (olfactory and mechanical), although there is
some recent indication that the response can at least be modulated by light (Gerber & Smith,
1998).
As to the validity of the Rescorla–Wagner proposition, the conclusion that it does not hold
for honeybees is encouraged by the negative results of Experiments 6–9, which were designed
to test the broader decremental hypothesis that the effectiveness of nonreinforcement in
reducing excitation increases with the excitatory value of the context. It is interesting to note
that Rescorla (2000) recently claimed to find evidence of inhibition in one such experiment
with rats, but the claim was based on a summation test alone. The complementary incremental
hypothesis—that the effectiveness of reinforcement is inversely related to the excitatory value
of the context—might be invoked to account for overshadowing and blocking where they
occur in honeybees, except that the same hypothesis leads us to expect them also under cir-
cumstances in which they fail to occur (Couvillon, Mateo, & Bitterman, 1996; Couvillon et al.,
1997). In honeybees, overshadowing and blocking experiments point, not to competition for
associative strength, but to competition for attention, evidence of which has also been found in
experiments on dimensional transfer (Shapiro & Bitterman, 1998).
The negative results of our comparisons of CI and S– training do not, of course, speak
against the reality of inhibition in honeybees, which may be generated by nonreinforcement
independently of the excitatory value of the context, although it is not clear how the hypothesis
might be tested directly. The difficulty lies in the choice of a suitable control group. In a two-
test experiment with rats, for example, Cole, Barnet, and Miller (1997) reported positive
results using a novel-stimulus control instead of an S– control; after A+/AB– training, exper-
imental groups were tested with B and control groups with C (an untreated stimulus). As
Papini and Bitterman (1993) cautioned, however, the novel-stimulus control is open to the
objection that B may simply be less excitatory than C, generalized excitation from A both to B
and to C being differentially reduced by nonreinforced encounters with B in the A+/AB–
training. Cole et al., who set out to meet other criticisms by Papini and Bitterman of previous
work, discounted the interpretation in terms of reduced generalized excitation as less parsimo-
nious than the inhibitory interpretation, although quite the opposite is true. There is abundant
independent evidence of the generalization of excitation in vertebrates, but relatively little
definitive evidence of inhibitory conditioning; the purpose of the work by Cole and colleagues
was, after all, to demonstrate more unambiguously than before that there is such a thing. It
might be reasonable, nevertheless, instead of dismissing the novel-stimulus control out of
hand, to try to find ways of testing the sufficiency of the Papini–Bitterman interpretation
where positive results are obtained.
It has long been apparent that the phenomena taken by Pavlov (1927) as prima facie evi-
dence of inhibition generated by nonreinforcement—spontaneous recovery, disinhibition,
and extinction below zero, as well as conditioned inhibition—can readily be explained without
reference to inhibition. For example, the development of a sharp preference for A over AB in
CI training such as shown in Figure 5 can be accounted for on the assumption that afferent
interaction with B either changes A, with the result that there is less generalized excitation
from A alone to A in compound with B (Hull, 1943), or—as traditional Gestalt psychology
would have it—generates a unique AB configuration (say, Q), which is distinguished from A
as such (Pearce, 1987, 1994). Instead of turning elsewhere in the face of such possibilities to
look for evidence of inhibition in honeybees, it may be reasonable here again to try to test
 NONREINFORCEMENT IN HONEYBEES 143

alternative interpretations. A readily testable implication of the configurational view is that

performance in A+/AB– (A+/Q–) training should be as good as in AB+/A– (Q+/A–) train-
ing, unless perhaps A and Q are unequally salient—a possibility that would call for further
investigation if the problems proved to be of unequal difficulty. In contrast, the assumption
that A is merely changed by the interaction rather than subsumed in a new whole suggests that
the AB+/A– problem will be the easier. It is interesting to note that the concept of inhibition
(albeit in different forms) finds a place both in Hull’s theory and in Pearce’s, although each
opens the door to a non-inhibitory account of CI training.
For the present, in any case, there is no good reason to reject the implication of our consis-
tently negative results that the role of nonreinforcement in honeybees is simply to reduce exci-
tation in a manner independent of the excitatory value of the context. The prevalence in the
vertebrate literature and the intuitive validity of the concept of inhibition make it difficult,
however, to accept that implication and suggest that it might be useful to continue to look for
more powerful experiments. At the same time, it seems fair to suggest that the evidence of
inhibitory conditioning in vertebrates is somewhat less than overwhelming and that research
might well go forward on both fronts.

REFERENCES
Bitterman, M.E. (1988). Vertebrate–invertebrate comparisons. In H.J. & I.L. Jerison (Eds.), Intelligence and evolu-
tionary biology (pp. 251–276). Berlin: Springer.
Bitterman, M.E. (1996).Comparativeanalysis of learning in honeybees. Animal Learning & Behavior, 24, 123–141.
Bitterman, M.E., & Couvillon, P.A. (1991). Failures to find evidence of adaptive specialization in the learning of
honeybees. In J.L. Goodman & R.C. Fischer (Eds.), The behaviour and physiology of bees (pp. 288–395).
Wallingford, UK: CAB International.
Blough, D.S. (1975).Steady state data and a quantitative model of operant generalization and discrimination. Journal
of Experimental Psychology: Animal Behavior Processes, 1 (104), 3–21.
Cole, R.P., Barnet, R.C., & Miller, R.R. (1997).An evaluation of conditioned inhibition as defined by Rescorla’s two-
test strategy. Learning and Motivation, 28, 323–342..
Couvillon, P.A., Ablan, C.D., & Bitterman, M.E. (1999).Exploratorystudies of inhibitory conditioning in honeybees
(Apis mellifera). Journal of Experimental Psychology: Animal Behavior Processes, 25, 103–112.
Couvillon, P.A., Arakaki, L., & Bitterman, M.E. (1997). Intramodal blocking in honeybees. Animal Learning &
Behavior, 25, 277–282.
Couvillon, P.A., & Bitterman, M.E. (1980). Some phenomena of associative learning in honeybees. Journal of Com-
parative & Physiological Psychology, 94, 878–885.
Couvillon, P.A., & Bitterman, M.E. (1982).Compound conditioning in honeybees. Journalof Comparative & Physio-
logical Psychology, 96, 192–199.
Couvillon, P.A., Mateo, E.T., & Bitterman, M.E. (1996).Reward and learning in honeybees: Analysis of an overshad-
owing effect. Animal Learning & Behavior, 24, 19–37.
Cunningham, C.L. (1979). Alcohol as a cue for extinction: State dependency produced by conditioned inhibition.
Animal Learning & Behavior, 7, 45–52.
Durlach, P.J., & Rescorla, R.A. (1980).Potentiation rather than overshadowingin flavor-aversionlearning. Journalof
Experimental Psychology: Animal Behavior Processes, 6, 175–187.
Funayama,E.S., Couvillon, P.A., & Bitterman, M.E. (1995).Compoundconditioning in honeybees: Blocking tests of
the independence assumption. Animal Learning & Behavior, 23, 429–437.
Gerber, B., & Smith, B.H. (1998). Visual modulation of olfactory learning in honeybees. Journal of Experimental
Biology, 201, 2213–2217.
Hellstern, F., Malaka, R., & Hammer, M. (1998).Backward inhibitory learning in honeybees: A behavioral analysis of
reinforcement processing. Learning & Memory, 4, 429–444.
144 COUVILLON ET AL.

Hoffman, J.W., & Fitzgerald, R.D. (1982). Bidirectional heart rate responses in rats associated with excitatory and
inhibitory stimuli. Animal Learning & Behavior, 10, 77–82.
Hull, C.L. (1943). Principles of behavior. New York: Appleton-Century-Crofts.
Kehoe, E.J. (1988). A layered network model of associative learning: Learning to learn and configuration. Psycho-
logical Review, 95, 411–433.
Papini, M.R., & Bitterman, M.E. (1993). The two-test strategy in the study of inhibitory conditioning. Journal of
Experimental Psychology: Animal Behavior Processes, 97, 396–403.
Pavlov, I.P. (1927). Conditioned reflexes. Oxford, UK: Oxford University Press.
Pearce, J.M. (1987).A model for stimulus generalization in Pavlovian conditioning.Psychological Review, 94, 61–73.
Pearce, J.M. (1994). Similarity and discrimination: A selective review and a connectionist model. Psychological
Review, 101, 587–607.
Pearce, J.M., & Hall, G. (1980).A model for Pavlovian learning: Variations in the effectiveness of conditioned but not
of unconditioned stimuli. Psychological Review, 87, 532–552.
Pearce, J.M., & Kaye, H. (1985).Strength of the orienting response during inhibitory conditioning. Journalof Experi-
mental Psychology: Animal Behavior Processes, 11, 405–420,
Pearce, J.M., & Wilson, P.N. (1991). Effects of extinction with a compound conditioned stimulus. Journal of Experi-
mental Psychology: Animal Behavior Processes, 17, 151–162.
Rescorla, R.A. (1969). Pavlovian conditioned inhibition. Psychological Bulletin, 72, 77–94.
Rescorla, R.A. (1979). Conditioned inhibition and extinction. In A. Dickinson & R.A. Boakes (Eds.), Mechanisms of
learning and motivation (pp. 83–110). Hillsdale, NJ: Lawrence Erlbaum Associates, Inc.
Rescorla, R.A. (1981). Within-signal learning in autoshaping. Animal learning & Behavior, 9, 245–252.
Rescorla, R.A. (2000). Extinction can be enhanced by a concurrent excitor. Journal of Experimental Psychology:
Animal Behavior Processes, 26, 251–260.
Rescorla, R.A., & Colwill, R.M. (1983). Within-compound associations in unblocking. Journal of Experimental
Psychology: Animal Behavior Processes, 9, 390–400.
Rescorla, R.A., & Wagner, A.R. (1972). A theory of Pavlovian conditioning: Variations in the effectiveness of rein-
forcement and nonreinforcement. In A.H. Black & W.F. Prokasy (Eds.), Classical conditioning II. Theory and
research (pp. 64–99). New York: Appleton-Century-Crofts.
Shapiro, M.S., & Bitterman, M.E. (1998). Intramodal competition for attention in honeybees. Psychonomic Bulletin
& Review, 5, 334–338.
Wagner, A.R. (1981). SOP: A model of automatic memory processing in animal behavior. In N.E. Spear & R.R.
Miller (Eds.), Information processing in animals: Memory mechanisms (pp. 5–47). Hillsdale, NJ: Lawrence Erlbaum
Associates, Inc.
Wagner, A.R., & Rescorla, R.A. (1972).Inhibition in Pavlovian conditioning: Application of a theory. In R.A. Boakes
& M.S. Halliday (Eds.), Inhibition and learning (pp. 301–336). London: Academic Press.

Manuscript received 7 June 2000

Accepted revision received 12 September 2000