History

My research area is Comparative Learning and Cognition. In some ways what we now call comparative learning is what we used to call Behaviorism. My very first experiments, conducted when I was an undergraduate at Stetson University, were designed to address one of the fundamental assumptions of the theory of Clark L. Hull as explained in his classic statement of behaviorist theory, Principles of Behavior (1943). Those early experiments of mine were done with rats as were virtually all behaviorist experiments at that time. There was not much comparative psychology in the sense that comparative psychology involves comparing different species of animals.

As a graduate student, I had the good fortune to work with one of the most important comparative psychologists of recent time, Professor M. E. Bitterman. There the work was theoretical and behavioristic, as it had been in my earlier experiments, but it was also comparative. I worked with crows, pigeons, turtles, and three species of fishes. I have also studied armadillos, feral rats, gerbils, lizards, shrimp, monkeys, and introductory psychology students. Lots of animals.

For a number of years, I worked outside the traditional domain of comparative in the area of instrumental conditioning of autonomic behavior, what is now called Biofeedback. I began this work with professor H. D. Kimmel who was the first psychologist to show that involuntary behavior may be altered by rewards and punishments, the basis of biofeedback therapy.

In the early 1980's, my interests shifted to the study of serial learning in animals. Serial learning had been the basis of the first important experiments on human memory conducted by Hermann Ebbinghaus beginning in 1879, the year of the founding of modern psychology, and published later in 1885. It was not until the 1970's and 80's that psychologists who studied animals began the study of serial learning in earnest. The serial learning research with animals soon gave very clear evidence that these animals have a well-developed ability to behave appropriately in problems that require learning the characteristics of a series of events regularly presented to them. This ability having been established, our task now is to develop our theoretical understanding of how they do it. So, as always, my work is concerned with evaluating theories.

In the late 1980's, my interest in serial learning took an interesting turn. A theory that I had been developing to explain serial learning led me to propose that animals in some sense are capable of counting the events they experience in series. We designed a large number of studies with rats that showed these animals are sensitive, though we do know quite how they it, to number as a cue in problems that require numerical solutions. Rats and other animals count things. Counting seems to be a rather basic element of the intelligence of animals.

Current Interests

All the research that I have done over the years has been done with the collaboration of undergraduate students working with me in my laboratory. Teaching and working with students is the most important element of my research. Lately the students and I have been doing experiments designed to establish the role played by counting in the serial learning of animals. Shown below is an abstract of our most recent paper.

Attempts to understand the serial learning of animals have taken a theoretical course that tracks theory in human serial learning. Capaldi and his associates have developed contemporary theories of animal serial learning that resemble the tradition in human learning established by Ebbinghaus. This view emphasizes the role of representational memories of events in a series and the associations of those representations. A few animal investigators, the present authors included, have developed theories which emphasize the role played by cues associated with the position of an item in a serial list. This minority tradition is represented in human research by Bower and Ebenholtz. The research presented here was an attempt to separate the Capaldi view of animal serial learning, to the extent that this view emphasizes cognitive representations, from our own, which emphasized S-R bonds established by reward. Rats were trained in an L-shaped runway for 55 days with two series of rewards. On one of the 3-trial series, the first and last trials were rewarded with qualitatively differing rewards, .045-g Noyes pellets on the first trial and a unit of breakfast cereal, Coco Roos, on the third trial. We will refer to this series as XNY allowing the X and Y to represent the two rewards and N to represent nonreward, 30-sec confinement in an unbaited goal box. On the other series, ZNN, only the first trial was rewarded, with a unit of Apple Jacks cereal. The interval between trials within a series was about 15 sec whereas the interval between series was about 20 min. All animals received both series each day, order determined randomly, but one group (n=6) had one series always with either a smooth black runway floor or a rough white floor, counterbalanced. A second group (n=6) had the floor cues, but they were uncorrelated with series. This arrangement meant the group for which the floor cue was a relevant series cue had two, intentionally confounded, series cues because the memory of the first series trial (either X or Z) could also serve as a series cue. The other group had only the memory cue because the floor cue was uncorrelated with series. Results showed that although the group with a correlated floor cue clearly discriminated between the two series by running slower on the third trial of the ZNN series than on the third trial of the XNY series, the group relying solely on the memory cue did not differentiate the two series. The results imply that memory cues are not necessary in this serial procedure, a conclusion that recommends a position-cue interpretation of the findings.

This paper was prepared for presentation at the Eighty-ninth Annual Meeting of the Southern Society for Philosophy and Psychology, Atlanta, Georgia, March 27-29, 1997

The group allowed only the memory of X and Z as a series cue failed, over the course of 55 training days, to develop a significant discrimination between the nonreward of Trial 3 in the ZNN series and the reward of Trial 3 on the XNY series. This result certainly implies that the memory of Trial 1 reward was an ineffective series cue contrasted to the floor cue provided the other group in addition to memory. The group with the compound memory\floor cue showed significant Trial 3 differences on the two series. That the memory cue alone was inadequate suggests the discrimination in the group provided the floor cue was not dependent on memory, the other relevant cue. Although Trial 2 was a non-rewarded trial in both series and in both groups, there was clear differential performance on that trial in the group with the floor cue. The rats ran faster on Trial 2 of XNY than on Trial 2 of ZNN. This effect has been termed interevent anticipation, though we prefer to call it intertrial generalization, and it has appeared in numerous studies employing this general serial procedure. Differential Trial 2 behavior also was statistically reliable for the group provided only memory cues, however, the magnitude of the effect was clearly diminished in that group.

In one previous study (Detloff & Burns, 1996), we trained rats on ZNY and ZNN with both memory and floor cues relevant, just as in the one group reported here. After 31 day of training the floor cues were reversed so that rats used to receiving the ZNY series in smooth\black, for example, not got it in rough\white. The consequence of this transfer test was that the animals ran on XNY as if it were ZNN and vice versa. That earlier result strongly suggested that memory of the Trial 1 reward was not the cue controlling the differential series discriminations. The floor cue controlled the discriminations. Neither that earlier study nor the present study which shows the relative ineffectiveness of memory cues argues that rats do not employ memory cues at all. The literature on single-alternation patterning alone confirms that memory processes are important in the analysis of rat learning. What the results do show is that associations among memorial representations of reward are not necessary for the appearance of "interevent anticipation."

It must be mentioned that a previous study (Capaldi & Miller, 1988) examined the XNY/ZNN problem and reported significant Trial 3 and Trial 2 discriminations in a group for which memory of the reward on Trial 1 was the only possible series cue. In that study memory cues were at work. Although we have no idea at present why our procedure failed to replicate the Capaldi & Miller (1988) result, the failure to replicate is not the important result. The important result is that the set of discriminations which appear is these problems are not necessarily tied to memory representations.

The view that Interevent Anticipation occurs in serial learning, and is responsible for the differential performance on Trial 2 of the XNY and ZNN series, is a view articulated by E. J. Capaldi in several influential papers. This view assumes that the memorial representations of reward events become linked in a manner like that imagined by Ebbinghaus (1885) to allow for anticipation of not only immediate but also remote events. A bundle (gestalt) of associations among reward memories is assumed to mediate the slower running on Trial 2 of the ZNN series than on Trial 2 of the XNY series because, although Trial 2 is never rewarded, remote anticipations are of reward in XNY and of nonreward in ZNN.

Our view is that the assumptions of Capaldi are not necessary to explain the data of all but two published experiments in this area. They are certainly not necessary to explain the data of the current experiment because the differential Trial 2 performance appeared in a context that was revealed by transfer tests reported earlier and by the group differences reported here not to involve control of responding by reward memories. The discriminations in the present study were controlled by the floor cues, not by memory cues, yet Trial 2 performance clearly differed as did performance on the differentially reinforced Trial 3. Another very important point about our results is that the so-called anticipation effect diminishes with training at the same time that the discrimination of the different Trial 3 outcomes improved with training. If the Trial 2 performance is the result of anticipating the outcome of the third trial, which is differentially reinforced, then that anticipation should increase, not decrease, as the Trial 3 discrimination improves. Described below is an account that better fits the evidence.

We have contended that stimuli arising from trial to trial which differ depending on the trial (Trial Stimuli) may become conditioned to approach following a simple S-R reinforcement principle. These trial stimuli may arise from a number of sources including unintentional variation in procedure from one trial to another, the number of trials preceding a given trial, the passage of time, and cues inserted intentionally to differentiate trials. Assume that the tendency to respond on any given trial, why not call it Reaction Potential, is a function of the simple summation of excitation (E) and inhibition (I) brought about respectively, by having response to Trial Stimuli followed by reinforcement or non-reinforcement. Assume further that both E and I generalize so that as many as six sources sum to produce Reaction on any given trial: Primary and Generalized E, and Primary and Generalized I. The Generalized E and I are assumed to come from other Trial Stimuli that are adjacent. Although the model can, of course, be made more complicated, we will consider generalization only from immediately adjacent trials. We will further assume, with some justification from the literature and our current data, that the trial stimuli from the first trial in both series are distinct more so than those of other trials and are, therefore unaffected by generalization and do not generalize.

To illustrate this simple model, assume that Primary E and Primary I have values of 5 and -5. Generalized E and I from either a different trial in the same series or the same trial in a different series have values of 2 and -2. Generalized E and I from a different trial in a different series have values of 1 and -1.

XNY Series

Trial 1 would have Primary E (5).

Trial 1 Reaction = 5

Trial 2 would be expected to have Primary I (-5) because that trial stimulus is never reinforced. I should generalize from the second trial in the ZNN series (-2) and from the third trial of the ZNN series (-1). E should generalize from the third trial of the XNY series (2).

Trial 2 Reaction = (-5)+(-2)+(-1)+(2) = -6

Trial 3 would be expected to have Primary E (5) because that trial stimulus is always reinforced. That excitation is countered by Generalized I from Trial 2 of the XNY series (-2), Trial 3 of the ZNN series (-2), and Trial 2 of the ZNN series (-1).

Trial 3 Reaction = (5)+(-2)+(-2)+(-1) = 0

ZNN Series

Trial 1 would have Primary E (5).

Trial 1 Reaction = 5

Trial 2 would be expected to have Primary i (-5) and Generalized I from Trial 2 of XNY (-2), Trial 3 of ZNN (-2). Generalized E should occur from Trial 3 of XNY (1).

Trial 3 Reaction = (-5)+(-2)+(-2)+(1) = -8

Trial 3 would have Primary I (-5) and Generalized I from Trial 2 of both ZNN (-2) and XNY (-1). It would have Generalized E from Trial 3 of XNY (2).

Trial 3 Reaction = (-5)+(-2)+(-1)+(2) = -6

Figure 6 depicts performance under the model. This kind of predicted performance fits the actual data well and it predicts features of the data that are not predicted by the event-memory model. Note that the performance predicted by the trial-stimulus model on Trial 2 of the two series is indeed different with running times being slower on Trial 2 of ZNN than on Trial 2 of XNY. Differential performance on Trial 3 is also predicted. Both of these predictions are shared with the event-memory model. Notice, also, what has been a very consistent finding in these experiments: Running on Trial 3 of ZNN is faster than on Trial 2 of ZNN. This effect has not been considered directly in discussions of the event-memory model, but the assumption would be that Trial 2 of ZNN involves the anticipation of both Trial 2 and Trial 3 nonrewards whereas Trial 3 involves anticipation only of the nonreward of that trial. The trial-stimulus model accounts for the same difference primarily by reference to Generalized E from Trial 3 of XNY.

The trial-stimulus model assumes that training sharpens generalization gradients. These changes due to training are not a component of the simple additive model presented here, but the changes in generalization, though not modeled here, engender predictions that separate the event-memory and trial-stimulus models. Here are three notable differences in the models: (1) Extended training should reduce the differences seen on Trial 2 of the two series according to the trial-stimulus model, but it should increase the differences according to the event-memory model. The data, especially in the current results, with relatively extensive training, clearly favor the trial-stimulus model. (2) Extended training should, according to the trial-stimulus model, increase the differentiation seen on Trial 3 of the two series, and that increase should parallel a decrease seen on Trial 2. This is what happens, and that finding runs counter to the assumptions of the event-memory model according to which better Trial 3 differentiation should increase, not decrease, interevent anticipation. (3) The trial-stimulus model predicts that extended training should increase the difference between Trials 2 and 3 in the XNY series, and it does, but should decrease the difference between Trials 2 and 3 in the ZNN series, and it does. Neither of these effects are predicted by the event-memory model.