Com bin ing CSs Associated w ith the Sam e or Differen t USs

Transcription

1 Ó T HE QUARTERLY JOURNAL OF EXPERIMENTAL PSYCHOLOGY, 1997, 50B (4), 350± 367 Com bin ing CSs Associated w ith the Sam e or Differen t USs Andrew Watt & R. C. Honey University of Wales, Cardiff, U.K. In three experiments, hungry rats received appetitive training with four stimuli, A, B, X, and Y. In each Experiment, A and B were paired with one unconditioned stimulus (US; e.g. food pellets) whereas X and Y were paired with a second US (e.g. sucrose). Subsequently, rats responded more vigorously to combinations of stimuli associated with different USs (A± Y & X± B) than to combinations of stimuli associated with the same US (A± B & X± Y; Experiments 1, 2, & 3). T his effect was observed when the stimuli were presented simultaneously and during the second elements of serial compounds (Experiments 2 & 3). Moreover, combining CSs associated with different USs resulted in a more marked CR than combining CSs that had each been paired with both US1 and US2 (Experiment 3). T hese results suggest that the sensory properties of appetitive reinforcers have an important in uence on performance. T he capacity of a conditioned stimulus (CS) to elicit a conditioned response (CR) following Pavlovian conditioning is often attributed to the ability of the CS to excite a representation of the unconditioned stimulus (US). T he nature of US representations and how their activation in uences performance has been the subject of much speculation. T hus, following Konorski (1967), it has been popular to distinguish between the sensory and emotive properties of a given US and to argue that these are represented separately and can have different effects on performance (see e.g. Wagner & Brandon, 1989). Several sources of evidence are consistent with this distinction. For example, the precise topography of conditioned responding varies as a function of the sensory characteristics of the US (see e.g. Jenkins & Moore, 1973). Moreover, acquisition of a discrimination, in which one response is reinforced during one stimulus and a different response is reinforced in the presence of a second stimulus, is facilitated when these responses are followed by outcomes with different sensory properties (see e.g. Kruse, Overmier, Konz, & Rokke, 1983; Peterson & Trapold, 1980; Trapold, 1970). Finally, following pairings of CS 1 with US1 (e.g. sucrose) and CS2 with US2 (e.g. food pellets), devaluing Requests for reprints should be sent to either author at the School of Psychology, University of Wales Cardiff, Park Place, Cardiff CF1 3YG, U.K. The research reported in this article was supported by a BBSRC studentship awarded to the rst author and was conducted when the second author was a Royal Society Research Fellow. We thank John Pearce for his comments on an earlier version of this manuscript and Mark Bouton for his constructive suggestions regarding the interpretation of Experiment The Experimental Psychology Society

2 STIMULUS COMBINATIONS 351 US1 by pairing it with lithium chloride selectively reduces the CR in the presence of CS 1 (Colwill & Motzkin, 1994; see also Colwill & Rescorla, 1985, 1988; Rescorla, 1994). Collectively, this evidence suggests that animals have a relatively rich encoding of appetitive reinforcers. However, little empirical effort has been directed to the issue of how these different representations interact with one another. Our research was undertaken in an effort to investigate this issue. One way to assess the nature of interactions between US representations is to compare the effect on performance of combining two CSs after they have separately signalled either the same US or different USs. Consider the case in which hungry rats receive independent pairings of different CSs either with the same US (e.g. food pellets) or with different USs (e.g. food pellets and sucrose); assume also that food pellets and sucrose support the CR of approaching the site of food delivery. A number of theoretical frameworks suggest that the effect on performance of combining CSs associated with different USs might differ from that of combining CSs associated with the same US. For example, Konorski (1948, pp. 109± 125) argued that there are excitatory and inhibitory connections between different ``unconditioned centres. Although his argument rested largely on the observation that conditioning is more rapid between certain pairs of stimuli than others (an observation subsequently con rmed by Garcia & Koelling, 1966; for a review, see LoLordo & Droungas, 1989), connections of the sort envisaged by Konorski could have a profound effect in the scenario we are considering. On the one hand, inhibitory connections between the representations of food pellets and sucrose would result in little responding to a compound whose components activate representations of both food and sucrose. On the other hand, excitatory links could, given certain assumptions (see Discussion of Experiment 1), result in more marked responding to such a compound. T he latter prediction can also be derived from the assumption that rats possess distinct representations of two appetitive reinforcers and that there is a limit on the extent to which each of them can be activated. Under these circumstances, combining two CS s that are independently capable of fully activating a given US representation might result in less responding than combining two CSs that are able to activate fully different US representations. Given that the issues considered above are of general theoretical signi cance and that little work has been conducted to examine them (see Ganesan & Pearce, 1988), we sought to compare the level of responding elicited by compounds as a function of whether their components had been paired with the same US or with different USs. Each of three experiments used food-deprived rats and an appetitive Pavlovian conditioning procedure in which food pellets and sucrose served as USs and magazine activity was the CR. EXPERIMENT 1 Experiment 1 employed the within- subjects design summarized in Table 1. Rats received pairings of A and B with US1 (e.g. food pellets) and X and Y with US2 (e.g. sucrose). Subsequently, rats received test trials with two sorts of serial compound: same and different. On one type of trial, designated ``same (A B and X Y), both components of the compound had signalled the same US. On the second sort of trial, designated ``different (A Y and X B), each component had signalled a different US. T he question of

3 352 WATT AND HONEY TABLE 1 Design of Experiment 1 Training Test A US1 A B (Same) B US1 A Y (Different) X US2 X Y (Same) Y US2 X B (Different) Note: During training, hungry rats received presentations of four stimuli, A, B, X, and Y. A and B were paired with US1 (e.g. food pellets) and X and Y were paired with US2 (e.g. sucrose). During the subsequent test, all rats received presentations of B and Y that were preceded either by stimuli that had signalled the same US during training (same trials, A B and X Y) or by stimuli that had signalled a different US (different trials, A Y and X B). interest was whether or not responding during B and Y would be in uenced by whether they had been preceded by a stimulus that had signalled the same US or a different US. Two features of this within-subjects design are noteworthy. First, it ensures that the components of the same and different compounds have equivalent associative strength; this would not have been possible had we used a between- subjects design. Second, the components of our test compounds were drawn from separate modalities and presented serially in order to reduce the likelihood of perceptual interactions between them (Kehoe, Horne, Horne, & Macrae, 1994; see also Rescorla & Coldwell, 1995). Method Subjects Eight male hooded lister rats supplied by Harlan-Olac (U.K.) served as subjects. T he rats were maintained at 85% of their free-feeding weights (mean = 457 g) by controlled administration of food at the end of each day. T he colony room, where rats were housed in pairs, was illuminated between the hours of 8.00 a.m. and 8.00 p.m. Training began at 10 a.m. Apparatus Training and testing were conducted in four standard Campden Instruments operant chambers that were housed in sound- and light-resistant shells. Each chamber was equipped with a diffusing, white, Plexiglass ceiling, above which was mounted a 12-inch striplight bulb, and a speaker mounted externally on the back wall of the chamber. Two visual CSs and two auditory CSs were used. One of the visual stimuli was the constant operation of the striplight; the second was ashing illumination of

4 STIMULUS COMBINATIONS 353 the same bulb (alternating 300- and 200-msec periods of light and dark, respectively). T he auditory stimuli, a 10-Hz clicker and white noise, were generated by a Campden Instruments audio generator (model no. 258) and a Campden Instruments white-noise generator (model no. 530). T hese stimuli were presented at an intensity of 75 db (A weighting). T he USs, a 45-mg Noyes pellet (Formula A) and 0.05 ml of 20% sucrose solution, were dispensed directly into a single food well, access to which could be gained by pushing a plastic door with a hinge at the top. A BBC Master computer with a Spider extension (Paul Fray Ltd., U.K.) controlled the apparatus and recorded the length of time that the door in front of the food well was open. T his was achieved using a single photo-electric cell activated by an aluminium shutter mounted on the top of the door to the food well. A movement of 2 mm at the base of this door was suf cient for the shutter to interrupt the photocell beam and start a timer. T he timer stopped either when the trial ended or when the door returned to the position where it no longer interrupted the photocell beam. T he response of central interest was the cumulative time within each 10-sec trial that rats spent in the food well. Procedure Pre-training. On the rst two days, rats were trained to retrieve the USs from the food well. On the rst day, the doors to the food wells were xed in a raised position; on the second day, rats had to move the doors in order to gain access to the reinforcers. In each session a total of 30 USs (15 food pellets, 15 sucrose) were presented in a random order on a variable-time 60-sec (VT 60) schedule. During these and all subsequent sessions, the boxes were in darkness. Conditioning. On each of the next 10 days, rats received training trials with four 10-sec stimuli, A, B, X, and Y. A randomly selected half of the trials with each stimulus were reinforced, and the remainder were nonreinforced. Two of the stimuli, A and B, were paired with US1, and two, X and Y, were paired with US2. US presentation immediately followed the offset of A and X, whereas there was a 10-sec trace interval between the offset of B and Y and US delivery. T he use of this trace-conditioning procedure was intended to avoid a ceiling effect at test when the in uence of presenting A or X prior to B or Y was assessed. However, the pattern of test results observed following this trace-conditioning procedure with B and Y was similar to that observed in Experiments 2 and 3, in which all stimuli were immediately followed by reinforcement during training. For all rats, A and X were auditory stimuli and B and Y were visual stimuli. T he identity of the auditory stimulus (noise or clicker) that served as A or X and that of the visual stimulus (constant or ashing light) that served as B or Y was counterbalanced. Similarly, for half of the rats in each of the four counterbalanced conditions, US1 was a food pellet and US2 was sucrose, and for the remainder this arrangement was reversed. T he 80 trials within each session were delivered at xed, 60-sec intervals. Sessions were divided into 20, four-trial blocks, containing one presentation of each of the four stimuli (A, B, X, and Y) in random sequence. Testing. On the next day, rats received a test session (Test 1). T his session began with a ``warm-up stage in which A, B, X, and Y were each presented on four occasions with the same sequencing and reinforcement contingencies in force as during training. Following the warm-up stage, there was a test stage. During this stage rats continued to receive presentations of A, B, X, and Y, in the same manner as during training and the warm-up stage. Rats also received nonreinforced presentations of two types of serial compound, same and different, the components of which had either signalled the same US (A B and X Y) or different USs (A Y and X B; see Table 1). T he offset of the rst element of these compounds was immediately followed by the onset of the second.

5 354 WATT AND HONEY More speci cally, during the test stage there were 4 sequences of 20 trials, each containing four presentations of A, B, X, and Yand one presentation of each of the serial compounds (A B, X Y, A Y, and X B). Each sequence contained 4 cyles in which presentations of A, B, X, and Y were followed by a different one of the four serial compounds. T he order in which the test compounds were presented within each of the four sequences was counterbalanced. For half of the rats, the sequence was same, different, same, different, and for the remainder it was different, same, different, same. T he identity of the particular same serial compound (A B or X Y) and that of the different serial compound (A Y or X B) that was presented rst or second within the test sequences was also counterbalanced. Test 1 failed to reveal any signi cant differences between the amount of food- well entry during the second, target elements of the same and different serial compounds. Consequently, after Test 1, rats received a further 10 days of training that were identical to the rst 10 days of training. T he rats then received a further two test sessions (Test 2), during which the trials were organized in an identical fashion to Test 1. As we have already mentioned, the CR measured was the cumulative amount of time that rats spent entering the food well during a given 10-sec trial. For the sake of simplicity we refer to this CR as the entry duration. T he rejection level adopted for all statistical analyses was p <.05. Results It is important to establish that our USsÐ food pellets and sucroseð support equivalent levels of conditioned responding for the following reasons. First, we have assumed that these two USs have the same emotive properties; to support this assumption, it is critical that the two USs support equivalent entry durations. Second, if the capacity of our two USs to support responding differed, then this could complicate interpretation of the test results. For example, higher levels of responding during different compounds might only re ect that these trials always included a CS that had been associated with the most potent US. However, we have observed, in a series of unpublished experiments, that food pellets and sucrose maintain equivalent levels of conditioned responding. T his observation proved to be reliable in each of the experiments that we report here. During the 10 days of training preceding Test 1, there was an increase in entry durations during the stimuli, A, B, X, and Y, that was of a similar magnitude for stimuli paired with food pellets or sucrose. On Day 1 the mean entry durations for the stimuli paired with food pellets and sucrose were 2.62 sec and 2.77 sec, respectively, and by Day 10 these scores had increased to 3.90 sec and 4.38 sec. An analysis of variance (ANOVA) with reinforcer (food pellets or sucrose) and day (1 or 10) as factors revealed no effect of reinforcer, F(1, 7) = 1.28, an effect of day, F(1, 7) = 8.68, and no interaction between these factors, F < 1. T he increase in the entry durations across training establishes that entry duration is a simple way to track the course of conditioning. T he entry durations were longer for the auditory stimuli (A and X) that were followed immediately by reinforcement than for the visual stimuli (B and Y) that were followed by reinforcement after a trace interval. On Day 10, the mean entry duration for A and X was 6.58 sec whereas for B and Y it was 1.70 sec. An ANOVA con rmed that entry durations were longer for A and X than for B and Y, F(1, 7) = After Test 1, during which there was no reliable difference between the entry durations for same (1.87 sec) and different (1.52 sec) test compounds, F(1, 7) = 2.16, rats received a further 10 days of training, on

6 STIMULUS COMBINATIONS 355 the nal day of which the mean entry duration during A and X was 6.49 sec and during B and Y was 2.07 sec. In this session, food pellets and sucrose maintained equivalent entry durations with means of 4.46 sec and 4.09 sec, respectively, F < 1. T he analysis that follows focuses on the results from the test compounds in the two sessions of Test 2; the differing treatments of single- element trials (reinforced) and compound trials (nonreinforced) during the test procedure precludes direct comparison of behaviour on these two types of trial. T he two test sessions were divided into two consecutive blocks each containing a total of 40 trials. Mean entry durations during target stimuli (B and Y) on same trials in Test 2 were not in uenced by the speci c US that these stimuli had been paired with throughout training. Over the four consecutive blocks of Test 2, mean entry durations were 1.69 sec, 1.56 sec, 1.46 sec, and 1.53 sec for targets paired with pellets, and 2.14 sec, 1.30 sec, 1.47 sec, and 1.92 sec for those paired with sucrose. A two-way ANOVA yielded no reliable effects of block or of reinforcer and no interaction between these factors (all Fs < 1). Accordingly, in order to simplify presentation of the results, the identity of the US with which the second component of the test compound was paired has been ignored. Figure 1 depicts performance during the second, target elements (B and Y) of the serial compounds as a function of whether the rst elements (A and X) had signalled the same US (on A B and X Y trials) or different USs (on A Y and X B trials). For the sake of simplicity, the results are pooled across the speci c exemplars of the same and different compounds and are presented in two- sequence blocks that separate the rst and second halves of each test session. On the rst block of trials, there was no difference in entry durations during B and Y as a function of whether the stimulus preceding them had been paired with the same or a different outcome. In contrast, for the next three blocks it is evident that rats consistently showed longer entry durations on different trials than on same trials. An ANOVA including the rst block revealed no effects of block, F < 1, or of stimulus, F(1, 7) = 3.10, and no interaction between these factors, F(3, 21) = However, in an analysis that excluded data from the rst block, there was an effect of trial type (same or different), F(1, 7) = 7.47, no effect of block, F < 1, and no interaction between these factors, F(2, 14) = Discussion In Experiment 1, hungry rats approached and entered the site of US delivery during stimuli, A or B, associated with one US (e.g. food pellets) and during those, X or Y, associated with a different US (sucrose). Subsequently, the CR elicited by B and Y was measured as a function of whether they were preceded by a stimulus that had signalled the same or a different US. After extensive training, the CR was more vigorous when the target stimuli were preceded by stimuli associated with a different US. T his nding is of interest in that it provides information about the nature of and interactions between US representations. Our results are inconsistent with the suggestion that the CR simply re ects the association of our CSs with the common emotive aspects of our USs; had this been the case, then the CR should have been equivalent during the two sorts of test compound (see Ganesan & Pearce, 1988). Instead, our results appear to support the suggestion that our rats have a richer encoding of the reinforcers and that this richer

7 356 WATT AND HONEY FIG. 1. Group mean entry durations for the target stimuli, B and Y, in Test 2 of Experiment 1. Following training in which A and B were separately paired with US1 and X and Y were separately paired with US2, rats received test trials with, for example, B. These test trials were either preceded by a stimulus that had been paired with the same US (A) or by one that had been paired with a different US (X). encoding affects performance. One way in which this richer encoding might in uence performance is based on the suggestion (see Konorski, 1948) that there are (excitatory) associations between US representationsð between the representations of food pellets and sucrose in Experiment 1. Provided that activity in the representation food pellets, for example, is more likely to provoke activity in the sucrose representation if the sucrose representation is also active, then combining stimuli that activate both representations will result in a stronger CR than combining stimuli that activate only one representation. Alternatively, if the extent to which a given US can be activated is limited, then combining stimuli that are associated with different US centres could result in a more vigorous CR than combining CSs associated with the same US. T here is, however, an alternative explanation for the effects observed in Experiment 1. It, too, relies on the notion that the representations of food pellets and sucrose are distinct, but it supposes that if a given US representation has recently been activated there is a refractory period during which it cannot be reactivated as readily (see e.g. Wagner, 1976; see also Wagner, 1981). If this is the case, then our use of serial compounds

8 STIMULUS COMBINATIONS 357 during testing in Experiment 1 might have resulted in stimulus B failing to activate the representation of US1 when it was preceded by a stimulus, A, that had recently primed US1. In Experiments 2 and 3 we examine whether the effect observed in Experiment 1 re ects our use of serial compounds, as an account based on the notion of priming predicts, or is a more general feature of compound stimuli, as Konorski s (1948) views and a limit on US activation would predict. EXPERIMENT 2 T he within-subjects design used in Experiment 2 (see Table 2) is formally identical to that of Experiment 1. Rats again received training in which A and B were paired with US1 and X and Y were paired with US2. Unlike Experiment 1, in Experiment 2 when reinforcement was scheduled it occurred immediately following the presentation of the four CSs. Subsequently, the CR to one stimulus, B, was measured as a function of whether it was preceded by a stimulus that had been paired with the same US (on A B trials) or a different US (on X B trials). Rats also received simultaneous compounds in which responding to the remaining target, Y, was assessed as a function of whether it was accompanied by a stimulus previously paired with the same US (on XY trials) or with a different US (on AY trials). T he issue of prime importance was whether or not the nature of the test combination (serial or simultaneous) would in uence the pattern of results obtained when CSs are combined that have been associated with the same or with different USs. TABLE 2 Design of Experiment 2 Training Test A US1 A B (Same) B US1 X B (Different) X US2 XY (Same) Y US2 AY (Different) Note: During training, hungry rats received presentations of four stimuli, A, B, X, and Y. A and B were paired with US1 (e.g. food pellets) whereas X and Y were paired with US 2 (e.g. sucrose). During the subsequent test, all rats received presentations of B that were preceded by a stimulus that had either signalled the same US during training (A) or a different US (X). Rats also received presentations of Y that were either accompanied by a stimulus that had signalled the same (X) or a different (A) US.

9 358 WATT AND HONEY Method Subjects and Apparatus Sixteen male, lister-hooded rats (mean free-feeding weight = 456 g) from the same source as Experiment 1 participated in Experiment 2. T he rats were maintained in the same way as in Experiment 1 and the apparatus was that used in Experiment 1. Procedure Experiment 2 employed the within- subjects design outlined in Table 2. After the rats had been trained to retrieve USs from the food well, there were 10 days of conditioning in which A and B were immediately followed by US1, whereas X and Y were immediately followed by US2. Following this, rats received a single test session. In this test, rats were given serial test compounds in which B was preceded by A or X and simultaneous compounds in which Y was accompanied by A or X. Details of the training and test procedure that have not been mentioned were identical to Experiment 1. Results and Discussion Over the 10 days of training, entry durations in A, B, X, and Y increased and were similar whether the stimuli were followed by food pellets or sucrose. T hus, on Day 1 mean entry durations for stimuli paired with food pellets and sucrose were 1.98 sec and 2.39 sec, respectively, and by Day 10 these scores had increased to 4.17 sec and 4.39 sec. ANOVA revealed no effect of reinforcer, F(1, 15) = 1.63, an effect of day, F(1, 15) = 14.58, and no interaction between these factors, F < 1. As in Experiment 1, by the end of training, entry durations were longer during auditory (A and X; 5.28 sec) than visual (B and Y; 3.29 sec) stimuli, F(1, 15) = Entry durations for the targets of same serial compounds (A B) were not in uenced by whether the targets had been paired with food pellets or sucrose: across the successive two blocks of the test session, mean entry durations for targets that had been paired with food pellets were 5.69 sec and 4.24 sec, respectively, whereas for the targets that had been paired with sucrose the corresponding means were 5.85 sec and 4.50 sec. An ANOVA con rmed that there were no effects of reinforcer or block, and no interaction between these factors, Fs < 1. Similarly, entry durations during the simultaneous same compounds on successive blocks were not in uenced by whether the components had been paired with food pellets, 4.03 sec and 3.63 sec, or sucrose, 4.25 sec and 2.94 sec. Again, ANOVA con rmed that there was no effect of reinforcer, F < 1, no effect of block, F < 1, and no interaction between these factors, F(1, 14) = T he mean entry durations in test trials in Experiment 2 are illustrated in Figure 2. Inspection of this gure reveals that entry durations tended to be longer during the second element, B, of the serial compounds (A B and X B) than during simultaneous compounds (XY and AY). T he simplest interpretation of this tendency is that presentation of the rst elements of the serial compounds resulted in the rats being in the vicinity of the food well when the second element, B, was presented. It is also clear that B elicited longer entry durations when preceded by X than by A. T his effect was

10 STIMULUS COMBINATIONS 359 observed in Test 2 of Experiment 1 and demonstrates that its occurrence does not rely on the use of a trace- conditioning procedure; indeed, given that in Experiment 1 this effect was observed only after more extensive training, it seems that the choice of the trace- conditioning procedure was misguided. Inspection of Figure 2 also suggests that entry durations during simultaneous compounds were longer when their components had signalled different USs (AY) than when they had been paired with the same US (XY). T his description of the results was supported by an ANOVA that revealed an effect of trial type (same or different), F(1, 15) = 4.90, no effect of compound type (serial or simultaneous), F(1, 15) = 3.19, and no interaction between these factors, F < 1. T here was also an effect of block, F(1, 15) = 5.74, re ecting the general decline in entry durations during testing, that did not interact with other factors, largest F(1, 15) = T he results of Experiment 2 con rm the nding from Experiment 1 that combining stimuli associated with different USs results in a more marked CR than combining those associated with the same US. Moreover, in Experiment 2 this difference was evident whether the test compounds were of a serial or a simultaneous nature. T hese results are consistent with the view that rats encode the sensory properties of food pellets and sucrose and, importantly, that this encoding in uences performance. Given the theore- FIG. 2. Group mean entry for the during test compounds in Experiment 2. The test compounds were either serial or simultaneous and comprised stimuli whose components had either been separately paired with the same US (serial A B and simultaneous XY) or with different US s (serial X B and simultaneous AY).

11 360 WATT AND HONEY tical signi cance of these observations and the numerically small size of the differences of primary importance, Experiment 3 attempted to replicate the results of Experiment 2 and to examine their nature in more detail. EXPERIMENT 3 One group of rats, Group Discrimination, received identical training and testing to those in Experiment 2 (See Table 2). T his group should allow us to replicate the results of Experiment 2. T he second group of rats, Group Ambiguous, received identical training and testing to Group Discrimination, with the exception that, on half of the reinforced training trials with each stimulus, US1 was presented and on the remainder US2 was presented. Following such training, the two types of test compound (same and different) will no longer differ since all stimuli have been paired equally often with US1 and US2. However, what is of interest is the comparison between the CR elicited by the compounds in Group Ambiguous and that elicited by the same and different compounds in Group Discrimination. We have assumed that different test compounds elicit a particularly marked CR because each component has been individually paired with a speci c US. T he test performance of Group Ambiguous will establish whether or not such discrimination training is necessary. Method Subjects and Apparatus T hirty-two male, lister-hooded rats (mean free-feeding weight = 432 g) from the same source as Experiment 1 served as subjects. T hey were maintained in the same way as Experiment 1, and the apparatus was that used in Experiment 1. Procedure Group Discrimination (n = 16) received the same training and testing as rats in Experiment 2 (see Table 2). T hus, following preliminary training, rats received presentations of A and B that were paired with US1 and presentations of X and Y that were paired with US2. Subsequently, performance to B was assessed as a function of whether it had been preceded by A or by X and performance to Y was assessed as a function of whether it was accompanied by A or by X. Group Ambiguous (n = 16) received identical training and testing to Group Discrimination, with the exception that on half of the reinforced trials with each stimulus, A, B, X, and Y, the reinforcer was US1 and on the rest it was US2. T he order in which US1 and US2 followed a given stimulus was random, with the constraint that on no more than three trials in succession was a given US presented. Other details of the procedure that have not been mentioned were the same as for Experiment 2.

12 STIMULUS COMBINATIONS 361 Results Entry durations for A, B, X, and Y increased over the ten days of training. For Group Discrimination, durations were of a similar magnitude whether the stimuli were paired with food pellets or sucrose: on Day 1, mean entry durations for stimuli paired with food pellets and sucrose were 1.10 sec and 1.50 sec, respectively, and by Day 10 these durations had increased to 3.20 sec and 3.80 sec. An ANOVA with reinforcer and day as factors revealed no effect of reinforcer, F(1, 7) = 2.58, an effect of day, F(1, 7) = 7.02, and no interaction between these factors, F < 1. By the end of training (Day 10) durations were longer for auditory (A and X) than visual (B and Y) stimuli for both groups. In Group Discrimination, the mean entry duration for auditory stimuli was 4.40 sec and for visual stimuli was 2.60 sec; in Group Ambiguous, the corrresponding mean durations were 4.70 sec and 3.60 sec. An ANOVA con rmed that there was an effect of modality, F(1, 14) = 10.55, no effect of group, and no interaction between these factors, Fs < 1. T he observation that responding during the various stimuli was similar in the two groups means that a comparison of their behaviour during the test compounds is potentially informative. As in Experiments 1 and 2, the capacity of the two USs to support the CR was evaluated; in this instance, this was achieved by comparing entry durations in the same test trials for Group Discrimination. Entry durations were similar for the targets of serial same compounds that had been paired with food pellets or sucrose. T he mean entry durations for targets that had been paired with food pellets were 6.02 sec and 4.66 sec, respectively, during the successive blocks of the test, and the corresponding means for the targets that had been paired with sucrose were 5.80 sec and 4.70 sec. An ANOVA con- rmed that there was no effect of reinforcer, no effect of block, and no interaction between these factors, Fs < 1. A similar pattern of results was observed for the simultaneous same compounds, with the compound comprising elements that had been paired with sucrose supporting mean entry durations on successive blocks of 5.80 sec and 5.01 sec, respectively, and the compound comprising elements paired with food pellets supporting mean entry durations of 5.82 sec and 5.04 sec. Statistical analysis con rmed that entry durations were equivalent for the two types of same test compound. ANOVA revealed that there were no effects of reinforcer or block and no interaction between these factors, Fs < 1. Figure 3 shows the mean entry durations for the two types of compound (same or different) in Group Discrimination pooled over the nature of the compounds (serial or simultaneous). As in Experiments 1 and 2, the entry durations were shorter during same compound trials than during different compounds trials. In fact, a more detailed analysis of rats behaviour during the various types of compound reveals a pattern similar to that observed in Experiment 2: same serial, 5.30 sec; different serial, 6.29 sec; same simultaneous, 5.42 sec; different simultaneous, 6.59 sec. Figure 3 also depicts the performance of Group Ambiguous during the compounds, again pooled over the nature of the compounds (serial, 4.78 sec; simultaneous, 4.44 sec). T hese entry durations are substantially shorter than those on the different trials and only marginally shorter than those on the same trials in Group Discrimination. Statistical analyses supported these impressions. An ANOVA conducted on the scores for Group Discrimination revealed an effect of trial type

13 362 WATT AND HONEY FIG. 3. Group mean entry during for the test compounds in Experiment 3. Prior to testing, Group Discrimination had received seperate pairings of A and B with US1 and pairings of X and Y with US2, whereas for Group Ambiguous each of the stimuli (A, B, X, and Y) was followed, on separate trials, by each of the US s. (same or different), F(1, 15) = 6.08, no effect of compound type (serial or simultaneous), F < 1, and no interaction between these factors, F < 1. T here was also an effect of block (1 or 2), F(1, 15) = 4.65, re ecting the general decline in entry durations during testing, but there were no interactions involving block, Fs < 1. Subsequent comparisons revealed that entry durations on the different trials for Group Discrimination differed from those on the compound trials for Group Ambiguous, F(1, 30) = 4.37, whereas entry durations on the same trials for Group Discrimination did not differ from those on the compound trials for Group Ambiguous, F < 1. Discussion Experiment 2 demonstrated that combining stimuli associated with different USs results in a more marked CR than combining those associated with the same US and that this effect does not interact with whether the CS s are presented sequentially or simultaneously. Experiment 3 replicated this pattern of results. Experiment 3 also demonstrated that the marked CR observed when two CSs associated with different USs are combined

14 STIMULUS COMBINATIONS 363 requires that each CS is uniquely associated with each US. T hus for Group Ambiguous, combining CS s that had each been paired with both USs did not elicit a particularly marked CR. Before we examine the implications of the performance of Group Ambiguous in more detail, we should comment on the failure of a previous study to reveal differences between the CR elicited by compounds whose components had signalled the same (e.g. food) or different outcomes (food and water; Ganesan & Pearce, 1988) and other results that are inconsistent with our own (Rescorla, 1991). Although the current study is formally equivalent to that reported by Ganesan and Pearce (1988), there are many procedural differences that might have contributed to their differing outcomes. For example, our within-subjects design might have been more sensitive, our stimuli more discriminable, and so on. Alternatively, it is possible that USs (food and water) from different appetitive systems interact in a different manner to those (food and sucrose) from the same appetitive system. Which of these differences turns out to be critical is a matter for future research. In many ways, the contrast between our own results and those reported by Rescorla (1991) is more dramatic and potentially more interesting. Rescorla (1991), using feature- positive and feature- negative pigeon autoshaping procedures, demonstrated that combining features associated with the same target resulted in greater transfer of their properties to a new target than combining features associated with different targets. T he procedures that we have employed are certainly quite different from those used by Rescorla (1991), and conclusions derived from the opposing patterns of results that they generate must be drawn with some caution. Nevertheless, the difference between the two sets of results does suggest that there might be an important distinction between the way modulatory features and simple CSs affect performance. GENERAL DISCUSSION Most theories of Pavlovian conditioning suppose that a CS comes to elicit conditioned responding because it activates a respresentation of the US (Mackintosh, 1975; Pearce & Hall, 1980; Rescorla & Wagner, 1972; Wagner, 1981). Given this analysis, it is surprising that relatively little evidence is available concerning the precise nature of US representations or how their activation affects performance. One can imagine a variety of ways that a US might be represented. For example, it is possible that the representation of a US might be relatively impoverishedð simply coding for its general affective value. According to this scheme, a hungry rat might represent two appetitive USs (e.g. food pellets and sucrose) in an equivalent fashion, for example, as an attractive event, and activation of this representation might determine performance; the fact that both USs support the same CR (approach) is certainly consistent with this possibility. In keeping with evidence outlined at the beginning of this paper, out experiments suggest that food pellets and sucrose are not represented in an equivalent fashion. More importantly, our experiments begin to reveal how the representations of food pellets and sucrose interact when they are associatively activated. We examined how rats respond when confronted with combinations of CS s that had either been paired with the same US (e.g. food pellets) or with different USs (food pellets and sucrose). In Experiment 3, for example, rats in Group Discrimination received two

15 364 WATT AND HONEY CSs, A and B, that were paired with US1 and two CSs, X and Y, that were paired with US2. Subsequently, compounds comprising CSs associated with different USs (X B and AY) elicited a more vigorous CR than compounds comprising CSs associated with the same US (A B and XY). T hese results suggest that conditioned responding is determined not only by the general emotive properties of appetitive USs but also by the (sensory) properties that distinguish them (see Konorski, 1967; see also Wagner & Brandon, 1989); if this were not so, then rats should respond no more vigorously to the combination of stimuli associated with different USs than to the combination of stimuli associated with the same US. One interpretation of these results is based on the notion that there is a limit on the extent that a given US representation can be activated. If we assume that with our training procedures this limit is achieved when a single CS is presented, then combining two stimuli associated with the same US should have little effect on performance. However, when stimuli are combined that are associated with different USs, each US would receive a full measure of activation and conditioned responding would be marked as a result. T he results of Group Ambiguous in Experiment 3 pose a problem for this analysis. In this group, on half of the reinforced training trials with each stimulus (A, B, X, and Y), food pellets were presented and on the remainder sucrose was presented. For this group, because a given stimulus is paired half as often with a given US as it is in Group Discrimination (Experiment 3), one simple and reasonable assumption is that this might result in each stimulus being only half as able to activate that US representation. However, given that each stimulus is paired with both USs, then the presentation of a compound should result in both US representations becoming fully active. Accordingly, this analysis predicts that the combination of stimuli that have each been paired with two USs should provoke a marked CR. In fact, the CR elicited by the test compounds in Group Ambiguous was relatively modest and statistically indistinguishable from the CR elicited by the same test compounds in Group Discrimination. T he results of Experiment 3 are broadly consistent with Konorski s (1948, pp. 109± 125) views regarding the connectivity between US representationsð speci cally, with his suggestion that ``unconditioned centres might be connected by excitatory links. Provided that we assume that activity in the US1 representation is more likely to provoke activity in the US2 representation when this representation is itself active, then combining stimuli that activate both representations will result in a stronger CR than combining stimuli that activate only one. Again the results of Group Ambiguous pose something of a problem for Konorski s analysis. One plausible way in which this analysis can be extended in order to account for the performance of Group Ambiguous is to allow that the value of the links between two US centres can be altered by experience. In general terms, during ambiguous training, but not during discrimination training, rats might learn that US1 and US2 are not presented on the same trial. In more formal terms, after initial ambiguous training, stimulus A will activate representations of US1 and US2. However, only one of these USs, say US1, will be presented on a given trial. T hese conditions are precisely those necessary for US1 to acquire the capacity to inhibit activity in the representation of US2. T his form of learning will mean that rats in Group Ambiguous have no reason to respond particularly vigorously in the presence of a compound stimulus: When the compound is presented, one US representation might become active, but the other will be

16 STIMULUS COMBINATIONS 365 unlikely to. Consequently, the level of responding to compound stimuli in Group Ambiguous should be akin to that elicited by the compounds composed of stimuli that have signalled the same US in Group Discrimination. Until this point, our discussion has focused entirely on possible interactions among US representations and how these interactions might directly in uence performance. However, the training procedures that we use might have affected test performance in a different fashion. It is now well established that the discriminability of stimuli can be in uenced by whether they have the same or different associatesð the so-called acquired equivalence and distinctiveness of cues (Honey & Hall, 1989, 1991). An account of our results can be derived from such changes in discriminability. For example, assume that our training procedure results in stimuli associated with the same US (e.g. A and B) acquiring equivalence with one another and distinctiveness from the stimuli (X and Y) associated with a different US. Now suppose that, other things being equal, a compound comprising highly discriminable elements elicits more responding than one whose elements are less discriminable (but see Pearce, 1994). According to this analysis, responding on different trials (X B and AY) would be expected to be more vigorous than on same trials (A B and XY) because the elements presented on different trials are more discriminable than those presented on same trials. Whichever of the accounts developed above one favours, it is clear that rats do not simply encode the general affective properties of appetitive reinforcers during a Pavlovian conditioning procedure. Rather, they have a more elaborate encoding, and this more elaborate encoding in uences performance. REFERENCES Colwill, R.M, & Motzkin, D.K. (1994). Encoding of the unconditione d stimulus in Pavlovian conditioning. Animal Learning & Behavior, 22, 384± 394. Colwill, R.M., & Rescorla, R.A. (1985). Post-conditionin g devaluation of a reinforcer affects instrumental responding. Journal of Experimental Psychology: Animal Behavior Processes, 11, 120± 132. Colwill, R.M., & Rescorla, R.A. (1988). Associations between the discriminative stimulus and the reinforcer in instrumental learning. Journal of Experimental Psychology: Animal Behavior Processes, 16, 40± 47. Ganesan, R., & Pearce, J.M. (1988). Interactions between conditioned stimuli for food and water in the rat. Quarterly Journal of Experimental Psychology, 40B, 229± 241. Garcia, J., & Koelling, R.A. (1966). Relation of cue to consequence in avoidance learning. Psychonomic Science, 4, 123± 124. Honey, R.C., & Hall, G. (1989). T he acquired equivalence and distinctivene ss of cues. Journal of Experimental Psychology: Animal Behavior Processes, 15, 388± 346. Honey, R.C., & Hall, G. (1991). Acquired equivalence and distinctiveness of cues using a sensorypreconditionin g procedure. Quarterly Journal of Experimental Psychology, 43B, 121± 135. Jenkins, H.M., & Moore, B.R. (1973). T he form of the auto-shaped response with food or water reinforcers. Journal of the Experimental Analysis of Behavior, 20, 163± 181. Kehoe, E.J., Horne, A.J., Horne, P.S., & Macrae, M. (1994). Summation and con guration between and within sensory modalities in classical conditionin g of the rabbit. Animal Learning & Behavior, 22, 19± 26. Konorski, J. (1948). Conditioned re exes and neuron organization. Cambridge: Cambridge University Press. Konorski, J. (1967). Integrative activity of the brain. Chicago, IL: University of Chicago Press. Kruse, J.M., Overmier, J.B., Konz, W.A., & Rokke, E. (1983). Pavlovian conditioned stimulus effects upon instrumental choice behavior are reinforcer speci c. Learning and Motivation, 14, 165± 181.

17 366 WATT AND HONEY LoLordo, V.M., & Droungas, A. (1989). Selective associations and adaptive specializations: Taste aversions and phobias. In S.B. Klein & R.R. Mowrer (Eds.), Contemporary learning theories: Instrumental conditioning theory and the impact of biological constraints on learning (pp. 145± 179). Hove, U.K. Lawrence Erlbaum Associates Ltd. Mackintosh, N.J. (1975). A theory of attention: Variations in the associability of stimuli with reinforcement. Psychological Review, 82, 276± 298. 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 conditioning : Variations in the effectiveness of conditioned but not of unconditione d stimuli. Psychological Review, 87, 532± 552. Peterson, G.B., & Trapold, M.A. (1980). Effects of altering outcome expectancies on pigeon s delayed conditional discrimination performance. Learning and Motivation, 11, 267± 288. Rescorla, R.A. (1991). Combinations of modulators trained with the same and different target stimuli. Animal Learning & Behavior, 19, 355± 360. Rescorla, R.A. (1994). A note on depression of instrumental responding after one trial of outcome devaluation. Quarterly Journal of Experimental Psychology, 47B (1), 27± 37. Rescorla, R.A., & Coldwell, S.E. (1995). Summation in autoshaping. Animal Learning & Behavior, 23, 314± 326. Rescorla, R.A., & Wagner, A.R. (1972). A theory of Pavlovian conditioning : Variations in the effectiveness of reinforcement and non-reinforcement. In A.H. Black & V.W. Prokasy (Eds.), Classical conditioning II: Current research and theory (pp. 64± 69). New York: Appleton-Century-C rofts. Trapold, M.A. (1970). Are expectancies based upon different positive reinforcing events discriminably different? Learning and Motivation, 1, 129± 140. Wagner, A.R. (1976). Priming in ST M: An information-processing mechanism for self-generated or retrieval-generated depression in performance. In T.J. T ighe & R.N. Leaton (Eds.), Habituation: Perspectives from child development, animal behavior, and neurophysiology (pp. 95± 128). Hillsdale, NJ: Lawrence Erlbaum Associates, Inc. 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., & Brandon, S.E. (1989). Evolution of a structured connectionist model of Pavlovian conditioning. In S.B. Klein & R.R. Mowrer (Eds.), Contemporary learning theories: Pavlovian conditioning and the status of traditional learning theory (pp. 149± 190). Hillsdale, NJ: Lawrence Erlbaum Associates, Inc. Combinaison de SCs associeâ s avec des SIs identiques ou diffeâ rents Dans une seâ rie de trois expeâ riences, des rats affameâ s ont recë u un entraõ à nement appeâ titif avec quatre stimuli, A, B, X, et Y. Dans chaque expeâ rience, A et B eâ taient accompanieâ s d un stimulus inconditionneâ (SI; e.g. grains de nourriture) et X et Y eâ taient accompanieâ s d un second SI (e.g. sucrose). Par la suite, les rats reâ pondaient de facë on plus vigoureuse en preâ sence de combinaisons de stimuli associeâ s avec diffeâ rent SIs (A± Y et X± B) qu en preâ sence de combinaisons de stimuli associeâ s avec le meã me SI (A± B et X± Y; Expe riences 1, 2 et 3). Cet effet fut observeâ quand le stimuli furent preâ senteâ s simultaneâ ment et durant les deuxieá mes eâ leâ ments des composeâ s en seâ ries (Experiences 2 et 3). De plus, les combinaisons de stimuli conditionneâ s (SC) associeâ s avec diffeâ rents SIs ont produit des reâ ponses conditionneâ s plus forte que les combinaisions de SCs qui avaient eâ teâ associeâ s avec chacun des deux SIs (Expe rience 3). Ces reâ sultats suggeá rent que les proprieâ teâ s sensorielles des reâ compenses appeâ titives in uencent la performance de facë on importante.