CONCEPTUAL BEHAVIOR IN RATS: CROSS MODALITY NON-MATCHING-TO- SAMPLE USING THREE DIMENSIONAL AND OLFACTORY STIMULI. Rachel A. Eure

Transcription

1 CONCEPTUAL BEHAVIOR IN RATS: CROSS MODALITY NON-MATCHING-TO- SAMPLE USING THREE DIMENSIONAL AND OLFACTORY STIMULI Rachel A. Eure A Thesis Submitted to the University of North Carolina Wilmington in Partial Fulfillment of the Requirements for the Degree of Master of Arts Department of Psychology University of North Carolina Wilmington 2012 Approved by Advisory Committee Ray Pitts Mark Galizio Kate Bruce Chair Accepted by Dean, Graduate School

2 TABLE OF CONTENTS ABSTRACT... iii ACKNOWLEDGEMENTS...iv LIST OF TABLES v LIST OF FIGURES....vi INTRODUCTION METHOD Subjects...22 Materials and Apparatus Procedure Design Data Analysis RESULTS.. 31 Subject X Subject A Subject D Subject Z Subjects Dropped from Study Inter-rater Reliability Discussion.38 References.45 Appendix...56 ii

3 ABSTRACT This study examined conceptual behavior in rats using a non-match-to-sample (NMTS) procedure. Previous research suggests that nonhuman conceptual behavior examined to date may be explained by generalization of alternative sources of control, rather than transfer of control by the relation between stimuli. The subjects, eight male rats, were trained on the NMTS procedure with either visual or olfactory stimuli and then tested using novel stimuli of the alternate modality. Using stimuli from different modalities allowed for tests of generalization of stimulus control across stimulus modality. Because transfer across modalities cannot be explained by generalization, evidence of this transfer would be a clear example of conceptual behavior in rats. Transfer to new modalities was assessed by comparing performance on probes using the training modality with performance on probes using a novel modality. Results showed that the behavior of rats can come under relational control with olfactory stimuli. However, training with visual stimuli may not always result in relational control. Only one of the five rats trained with visual stimuli were able to reach criterion to receive novel probes. Training with olfactory stimuli may have aided in transfer of relational control to visual stimuli. While they do not show initial transfer of relational control, all three rats trained with olfactory stimuli showed relational control of behavior with visual stimuli and required fewer training sessions to do so. iii

4 ACKNOWLEDGEMENTS I would like to thank Dr. Kate Bruce for her countless hours of assistance as I worked to prepare my thesis. Without her guidance and advice, this project would not have come to fruition. Dr. Mark Galizio also played an integral part in the design and implementation of this project. I am grateful for all of his time and guidance. My appreciation also goes out to all my colleagues in the Animal Cognition Lab who helped run subjects and were always available to share ideas. This project was also molded under the input of Dr. Ray Pitts. His participation and suggestions have proved invaluable. I am eternally grateful to my family and friends, without whom I would not be where I am today. Thank you to everyone involved in the making of this project, no matter how large or small the assistance may have seemed. iv

5 LIST OF TABLES TABLE PAGE 1. The number of training sessions required between probes for each rat.55 v

6 LIST OF FIGURES FIGURE PAGE 1. SCHEMATIC OF THE EXPERIMENTAL PHASES SUBJECT X5 S PERFORMANCE SUBJECT A5 S PERFORMANCE SUBJECT D4 S PERFORMANCE SUBJECT Z11 S PERFORMANCE.54 vi

7 INTRODUCTION The use of conceptual behavior, or the ability to differentially respond to groups of stimuli, has long been seen as an important component in the evolution of human intelligence. Nissen (1958) argued that humans ability to reason resulted from an underlying ability to develop concepts based on similarities and differences between stimuli. He suggests that before humans were able to make decisions about their environment, they first had to learn to group stimuli into classes based on their properties. The properties of a stimulus occasion responses based on previous experience with similar stimuli. All stimuli that occasion the same response are said then to be members of the same stimulus class (Pierce & Cheney, 2008). It does seem logical that having an ability to group stimuli together based on their properties, and respond to them appropriately, would give an individual an advantage over others who did not have that ability. Any animal that was able to generalize its responding to novel stimuli without requiring firsthand experience would learn faster, allowing that animal to survive and reproduce better than those who were not able to respond to novel stimuli in this way. There is a great deal of debate, however, over whether or not nonhumans can show conceptual behavior. There are two factors contributing to this debate. The first is that conceptual behavior has proven difficult to operationalize and even more difficult to compare across species (Katz, Wright, & Bodily, 2007). The goal of most studies to date has been to test whether certain species have the ability to show conceptual behavior by administering an experimental test in a pass-fail fashion. If the organism fails the test, it must not possess the ability to show conceptual behavior. Species differ, though, on predispositions such as sensory limitations and the properties of stimuli to which they are more likely to attend (Slotnick, 2001). It is possible that these predispositions are an underlying factor which controls their ability to perform a task.

8 Thus, a more important experimental undertaking is to design experiments that allow for expression of these abilities in different species (Katz et al., 2007). Another factor contributing to the debate over nonhumans ability to show conceptual behavior is that the use of terms such as concept, generalization, discrimination, and categorization can often be misconstrued. This confusion arises because these terms are often used interchangeably to explain very similar instances of behavior. Differential responding to groups of stimuli is sometimes referred to as conceptual behavior and sometimes called categorization. Categorization is also sometimes considered the behavior that is used when concepts are formed. (Lazareva & Wasserman, 2008). If advances are to be made in the comparative study of conceptual behavior, each of these terms must be separately and clearly defined. Keller and Schoenfeld (1950) were the first to suggest that a concept is not something that an organism possesses. Rather, a concept should be thought of as a specific pattern of responses. These researchers operationally define conceptual behavior as generalization within classes and discrimination between classes (Keller and Schoenfeld, 1950: p. 155). Thus, there must be multiple stimuli involved and the organism must show differential responding to these stimuli. Discrimination is a label given to instances of behavior where an organism shows differential responding between multiple stimuli (Pierce & Cheney, 2008). Differential responding simply means that an organism responds in opposing ways to stimuli. Organisms learn to respond in certain ways through experience with stimuli. As the organism acts on his environment, his behavior will produce reinforcers and punishers for responding to stimuli in certain ways. Stimuli that share similar characteristics as those previously experienced then come to occasion generalization. The term generalization, when referring to a behavioral 2

9 process, is used to label those instances of behavior where an organism emits the same response for all stimuli that share common properties. (Keller & Schoenfeld, 1950). Behavior can come under the stimulus control of many different properties of stimuli. Thus, conceptual behavior is normally broken down into two types. These two types of conceptual behavior are differentiated based on the properties of stimuli that control responses. Behavior which is controlled by constant physical or function-based properties of an individual stimulus is characterized as a natural concept. For example, when an organism shows discrimination between a class of red stimuli and a class of green stimuli, the behavior would be an example of a natural concept (i.e. color). Another example would be when a person responds to a voic from one colleague and an from another colleague in the same way. The two types of stimuli, and voic , are physically different but occasion the same response, communication between individuals. In this type of conceptual responding each member of the class is evaluated independently and the organism responds based on whether or not the current stimulus has the properties of the class (i.e. red or the function of communication). This type of learning can be accomplished using individual stimuli and requires no comparison between multiple stimuli or simultaneous stimulus presentation. The second type of conceptual behavior is abstract conceptual behavior. Abstract concepts are important because they are often seen as being an example of higher-order intelligence (Wright, Rivera, Katz, & Bachevalier, 2003). In abstract responding, behavior is said to be under the stimulus control of the relation between two or more stimuli, rather than the properties of any one class member. These relations are said to address higher-order abilities than natural concepts because they must be applied to multiple stimuli at any one time. It must be the relation between the stimuli that controls responding and not the features of the stimuli 3

10 themselves. Like natural concepts, abstract concepts may be based on the physical properties of stimuli, but the important distinction is that in abstract concepts two or more stimuli must show a relation based on that physical property and it must be this relation which controls responding. Abstract conceptual behavior is also different from natural conceptual behavior because abstract conceptual behavior cannot be explained by simple stimulus generalization or some nonspecific artifact of the experimental procedure (Zentall & Hogan, 1974). Examples of this type of conceptual behavior include responding based on size, position, quantity, identity, etc; such as when an individual receives more food for picking the larger of two options. One of the most important abstract concepts is that of oddity, or, the ability to respond based on the relation of sameness/difference. Nissen (1958) even suggested that all conceptual responding is a product of comparing similarities and differences between stimuli. An operational definition that uses only differential responding to class members may not, however, give a complete picture of abstract conceptual behavior. In order for it to be said that an organism has accurately shown conceptual behavior, the differential responding across classes and similar responding within classes must go beyond those stimuli with which the organism was first trained (Lazareva & Wasserman, 2008). That is, the generalization and discrimination shown with training stimuli must transfer to stimuli which are different from the stimuli used in training and to which the organism has never been exposed (Wasserman, Kiedinger, & Bhatt, 1988). Wright and Katz (2009) suggested that the transfer of control of behavior by the relation between stimuli to novel exemplars is the primary feature of abstract concept learning. Trials in which these novel stimuli are presented are called novel probe trials or transfer test trials. The use of these novel probe trials, in addition to the differential responding put forth by Keller and Schoenfeld (1950), helps to rule out alternative sources of control that 4

11 may be incompatible with the explanation that behavior is under the stimulus control of the relation between stimuli. One commonly used procedure for testing the oddity concept is the non-match-to-sample (NMTS) procedure (e.g. Lazarowski, 2010). In this procedure a sample is presented to the subject and, after an observing response to that sample, two comparison stimuli are presented. One of these stimuli will share the properties of the sample and the other will have differing properties. Because the procedure is NMTS, responses to the stimulus that is different from the sample will be reinforced. A variation of this procedure called match-to-sample (MTS) is also sometimes used. Here, responses to the comparison stimulus that is identical to the sample are reinforced (e.g. Pena, Pitts, & Galizio, 2006). In both of these procedures, presentation of the comparisons can either be simultaneous with the sample, as in Pena et al., or delayed (Mumby, Pinel, & Wood, 1990; Rothblat & Hayes, 1987). The delay may be as small as 0 s, where, following an observing response to the sample, the sample is removed and the comparisons are immediately presented, as in Lazarowki (2010). The delay may also be quite large. Using a delay procedure, the sample is removed and after the specified delay the comparisons are presented (Rothblat & Hayes, 1987). This delayed NMTS procedure is often used to study working memory in animals (Mumby et al, 1990; Rothblat & Hayes, 1987). All of these procedural variations would still require the organism to respond differentially based on the relation between the sample and comparison stimuli. In order to produce reinforcement, a different response is required for each comparison presented. Depending on the identity of the sample, the organism must either respond to the comparison he first observed or make a response on the alternate comparison. It is possible though, that the organism has come under an opposing form of stimulus control that would lead him to high 5

12 accuracy levels with training stimuli but would not transfer to novel stimuli. This problem is especially prevalent in experiments where only a small number of stimuli are used. In these procedures, it is possible that behavior may come under alternative sources of control, such as responding based on previous reinforcement history, configurations, if-then strategies, novelty, or engaging in stimulus pairing (Carter & Werner, 1978). Responding based on previous reinforcement history is simply when the organism learns to either respond to or reject a stimulus because it has been most recently reinforced. If the organism learns to respond to configurations, he is responding to the group of stimuli (the sample and two comparisons) as if they are one discriminative stimulus that occasions either a response on the left or right. Assuming the sample occurs in the same position on each trial, each sampleincorrect comparison pair has two possible configurations. Therefore, the more stimuli used in an experiment, the less efficient configural learning becomes. If-then strategies are similar to configurations in that the organism learns to respond to pairs of stimuli together. As discussed by Katz, Wright, & Bodily (2008), the subject learns the sample and correct comparison as a pair of stimuli and responds as such. For instance, if an experimenter used a NMTS design to teach pigeons the oddity relation with colored stimuli and always paired a red sample with a blue comparison, the pigeon could learn to respond to blue if red is the sample, rather than learning to respond to the relation between stimuli. This type of responding is sample-specific and does not require the individual to respond based on the relation between stimuli. He merely learns to respond to the pair of stimuli, red and blue. For this reason it is important for all stimuli to occur equally as the sample and the incorrect stimulus, for all stimuli to be paired together, and that an adequate number of stimulus pairs are used. Katz et al. (2008) used a MTS procedure with three picture stimuli (a 6

13 bunch of grapes, an apple, and a duck) in the training set to test learning strategies in pigeons. Using these three stimuli, there are twelve possible configurations. The pigeons in this experiment were first trained using six of these twelve configurations. After reaching a criterion of 70% correct for two consecutive days for all three training pairs, transfer tests were given. These transfer tests consisted of baseline trails, untrained display trials, and trials using all novel stimuli. The untrained display trials were the other six possible configurations that were not used in training. This type of test was used to determine whether the pigeons were responding based on if-then strategies. For instance, if the sample is the picture of the duck, they would respond again to the picture of the duck and accuracy would remain high. This would resemble abstract conceptual behavior but would not actually be so. These untrained display tests are also important because all the configurations that could have been learned in training are changed. If the pigeons were responding based on configurations, accuracy should drop to chance levels. Results on the novel trials showed no evidence of abstract concept learning. Accuracy levels on the untrained display trials were significantly lower than baseline trials but significantly better than chance. Since performance is at an intermediate level between baseline and chance, the authors suggest that this is evidence that pigeons were responding based on configurations and if-then strategies (Katz et al., 2008). Carter and Werner (1978) discuss at length possible sources of control that may masquerade as conceptual responding, but are not actually responses that are controlled by a difference relation. One such example was that an organism may respond to a novel stimulus simply because it is novel. This would yield accurate performance on novel probe trials (if using a NMTS procedure) but clearly is not abstract concept learning. However, if all of the training stimuli occur multiple times, responding based on novelty would not be an effective method and 7

14 should decrease as training continues. Kastak & Schusterman (1994) discuss the fact that responding in this way would cause inaccurate results and suggest that multiple novel trials should be presented. Another example discussed by Carter and Werner (1978) is what they termed stimulus pairing. Here, the organism does in fact show conceptual responding, but it is based on properties of individual stimuli rather than any relation. The problem with stimulus pairing is that (in a nonmatch to sample trial) the sample always occurs with a matching, incorrect stimulus. For example, it is possible that the organism could learn not to respond to a red key light comparison, given a red key light sample, merely because in the presence of red samples, responses to red comparisons are never reinforced. It is possible that the organism could learn this type of responding for multiple sample stimuli, but it is not the relation between the sample and correct comparison that is controlling responding. When the organism is presented with a novel sample, responding will be at chance levels. These different strategies are generally cited as an explanation for high performance levels on baseline stimuli when transfer to novel stimuli does not occur (Carter & Werner, 1978). Without the use of novel transfer trials, high performance on baseline trials could be credited to any of these alternative explanations. To combat this, the organism must not only be exposed to novel probe trials, he must also respond correctly to a percentage of these trials that is equal to or better than the accuracy level achieved on baseline training trials (Katz & Wright, 2006). If the accuracy on transfer tests using novel stimuli is equivalent to those achieved in training, the organism can be said to have shown conceptual behavior. This type of conceptual behavior has been shown in sea lions [Kastak & Schusterman, 1994), dolphins (Herman, Hovancik, Gory, & Bradshaw, 1989), capuchin and rhesus monkeys (Wright et al., 2003), rats 8

15 (Pena, Pitts, & Galizio, 2006), humans (Weinstein, 1941), chimpanzees (Oden, Thompson, & Premack, 1988), and pigeons (Wright, 1997). The successful concept learning shown in these studies may have been facilitated by the use of stimuli from species specific modalities, meaning that the modalities of the stimuli that were used were in accordance with stimuli that would occur and require discrimination in the organism s evolutionary environment. As noted by Katz et al. (2007) and Slotnick (2001), species rely on different sensory receptors to solve complex problems. In each of these studies, modalities appropriate to the species in question were used (i.e. sea lions-visual (Kastak & Schusterman, 1994); dolphins-auditory/visual (Herman et al., 1989); monkeys-visual (Wright et al., 2003); rats-olfactory (Pena et al., 2006); humans-visual, three dimensional (Weinstein, 1941); chimpanzees-visual/tactile, three dimensional (Oden et al., 1988); and pigeons-visual (Wright, 1997)]. The sea lion experiment is unique in that after pretraining, but before novel transfer tests, reshuffling occurred. The two sea lions were first trained on a simultaneous MTS procedure with a large number of stimuli. During training, the stimuli always occurred in the same pairs. However, after criterion was met on this phase, experimenters implemented a reshuffling phase in which all stimuli were paired with all other stimuli, both as the correct and incorrect stimulus. This rules out configural learning strategies before novel tests are given. It does not, however, rule out control of responding by an if-then relation. It is still possible that, after repeated training, the sea lions learned to respond to comparison A in the presence of sample A. This is not control by the relation between stimuli, and so, is not abstract concept learning. Thus, both sea lions underwent a series of novel transfer tests. These transfer tests were separated by days of baseline training in which the novel stimuli were integrated into baseline sessions. By the end of 9

16 the experiment, both subjects were showing abstract conceptual behavior, controlled by the relation between sample and comparison (Kastak & Schusterman, 1994). In an experiment with preverbal children, three dimensional visual stimuli were used (Weinstein, 1941). The children were trained using a simultaneous MTS procedure where they were allowed to look at both the sample and comparison stimuli at any time during a trial. While the stimuli were three dimensional objects, the authors report that the children looked at, but did not handle, the stimuli while making a choice. It was shown that when new stimuli were introduced, the children were able to respond at high levels of accuracy very quickly and needed far less training to generalize performance than the rhesus monkeys to which they were being compared. One human subject even met criterion with novel stimuli in the minimum numbers of trials (Weinstein, 1941). Chimpanzees were also taught to match to sample using three dimensional objects (Oden et al., 1988). Here, though, the subjects did get the opportunity to handle the objects. Each trial began when the experimenter handed the sample object to the subject. In this experiment only two training stimuli were used. Three different types of transfer tests were given with both the sample and comparison stimuli being novel. There were two trials of each type: objects, fabrics, and food. One important aspect of this study is that the subjects had encountered the transfer stimuli before, though not in the experimental context. Results of this experiment showed that percent correct on novel transfer trials involving new objects and fabrics was not significantly different from baseline, nor were they different from each other. However, the tests involving food yielded a significantly lower accuracy level. Experimenters attribute this drop in accuracy to preference of one food type over the other and argue that the high accuracy levels on the other two transfer test types are evidence for generalized matching (Oden et al., 1988). 10

17 Even in the face of high accuracy levels on novel probe trials, one issue still remains with studying conceptual behavior in this way. To date, the vast majority of experiments studying conceptual behavior have used stimuli on novel transfer tests are from the same modality (e.g. visual, olfactory, auditory, etc.) as the training stimuli. Because the training and testing stimuli are from the same modality, it is possible that generalization is controlling responding in these experiments, rather than transfer of relational control. When an organism encounters a previously unseen stimulus that shares properties with previously seen stimuli, the organism is likely to respond to the new stimulus in a manner consistent with what he learned using the previous stimuli, based on generalization. MacIntosh (2000) argues that stimuli from the same modality will share common properties. Thus, probe trials from the same modality used in training cannot be said to be truly novel. Since the testing stimuli share properties with those stimuli used in training, generalization of an alternative source of control is controlling responding, not the relation between stimuli. For example, if the pigeon learned to respond to a red key light in training and was probed with an orange key light, his responding will be similar for the two situations, not because of the transfer of relational control but simply because he has generalized responding in one familiar situation to responding in a novel but somewhat similar situation. Responding in this way cannot be viewed as an example of conceptual behavior because no truly novel probe test has been given. Although MacIntosh was disputing data from previous experiments with pigeons and greater attempts at demonstrating conceptual behavior have been made since the posting of his argument (e.g. rats-pena et al., 2006; pigeons-wright et al., 2003), it may still hold true today. A meta-analysis done by Wright and Katz (2007) suggests that the generalization hypothesis is unlikely. These authors point out that increasing the number of stimuli used in training only 11

18 compounds the possibility that testing stimuli will not be truly novel. This makes proving that an animal is showing conceptual behavior even more difficult, since it usually takes a large number of stimuli to obtain relational control of behavior in nonhumans. Wright and Katz (2007) discredit the generalization hypothesis, pointing out that there is no way to know how to design an experiment with enough stimuli to teach the relation and yet not so many stimuli that generalization can be used to explain responding. They also offer their prediction model, based on previous studies of abstract concept learning, as evidence against the generalization hypothesis. The fact remains, however, that modality may cause a problem for the experiments done in this field to date. In most studies of abstract concept learning, training and testing stimuli do not span multiple modalities. Thus, generalized responding between training and testing stimuli could be another alternative explanation for successful same/different concept learning. If organisms can respond to testing stimuli in the same manner as training stimuli, based on generalization, it could be argued that these stimuli are not novel. Thus, a critical test of conceptual behavior in nonhumans must demonstrate that subjects behavior has come under the stimulus control of the relation between stimuli and further that they can generalize this stimulus control to transfer test trials that involve stimuli that are truly novel (Lazareva & Wasserman, 2008). If training and testing stimuli from two separate modalities are used, there will presumably be no shared feature of the stimuli that could be controlling responding. Responding then, is due only to the relation between sample and comparison. To date, attempts to study cross modal concept learning are limited. One example studied dolphins, and it required intense pretraining (Herman, Gory, Hovancik, & Bradshaw, 1989). In this experiment, one female bottlenosed dolphin first learned to associate a series of underwater whistle sounds with objects located in her tank, actions, and locations. Next, she underwent 12

19 training to learn to associate a visual replica of the object held by the trainer with an object in her tank. Training occurred in four stages. First, only the object in question was in the tank. The trainer held up the object and an action sound for the object was played. Performing the correct action with the object resulted in reinforcement. In training stage two, two objects were placed in the tank, one that matched the sample object and one that did not. Again, the trainer held up the object and performing the appropriate action with the matching object was reinforced. Because accuracy levels were lower than experimenters expected them to be, the dolphin was moved to training stage three. In this stage, the previously learned auditory cue that matched the sample object was played while the sample was presented. This improved performance and the dolphin was moved on to stage four. In stage four, the procedure was switched from simultaneous matching to zero second delay in order to gradually fade out the need for an auditory cue. This was done by experimenters lessoning the probability that an auditory cue would occur on any given trial. After this probability reached zero, the subject moved on to the testing phase. Before testing, the dolphin was given 36 refresher training trials, 18 with an auditory cue and 18 without. During the testing phase, no auditory cues were given. Over 564 testing trials the subject achieved accuracy levels of 80% or higher, and, overall accuracy improved as a function of number of trials. After completion of this experiment, the dolphin moved on to Experiment 2, where novel transfer tests were applied. There were three different types of tests: (1) objects that had previously been trained with an acoustic cue, (2) objects that were familiar to the dolphin but did not have an acoustic cue, and (3) objects that were completely novel. During all types of probes trials, two objects were placed in the tank. Again, the dolphin was shown the object, heard an action sound, and was reinforced after performing the appropriate object/action pair. She was 13

20 able to show similar high levels of accuracy for all twelve probe trials. Experimenters argue that this is evidence for cross modal matching because the visual object and auditory action cue were used to create one command that included elements from two modalities (Herman et al., 1989). However, the extensive pretraining with objects and sounds that the subject was familiar with makes a case that this may not be true generalized matching. While the dolphin does learn to transition from using auditory samples to visual samples, the comparison stimuli are always of the same modality. It can be argued that what may have happened in this experiment is that the subject learned two different behavioral processes (i.e. pairing sounds with objects then pairing objects with objects), rather than generalizing stimulus control from one modality to another. Perhaps the best example of an attempt to study cross modal concept learning is Zentall and Hogan s (1974) experiment using pigeons. In Experiment II of this study, the authors trained one group of pigeons on a MTS procedure and the other on a NMTS procedure. For this experiment, training stimuli were keys lit with four different brightness levels for both groups. After 30 sessions, both groups received transfer tests using red and green keys that were matched for brightness level. Performance on the first session of transfer tests was at chance for all pigeons. However, after six sessions with the new stimuli, all rats were responding at a level greater than 95%. Zentall and Hogan (1974) argue that this savings is evidence for concept learning since the pigeons were able respond to the trials using new stimuli at high accuracy levels faster than if they were learning a new task. They also suggest that this transfer cannot be due to stimulus generalization, since the testing stimuli were balanced for brightness level. While the training and testing stimuli used in this experiment do differ in physical characteristics, they are both types of visual stimuli. An interesting question remains then of whether nonhumans can transfer conceptual responding across modalities that use different senses. In addition, readers 14

21 are left to wonder whether nonhumans can show generalization across modalities on the initial testing session. Historically, the two types of stimuli most commonly used to study rodent oddity concept learning are olfactory (Lazarowski, 2010; Lu, Slotnick, & Silberberg, 1993; Pena et al., 2006; Thomas & Noble, 1988) and visual (Iversen, 1993, 1997; Mumby, Pinel, & Wood, 1990; Nakagawa, 1993; Rothblat & Hayes, 1987; Wodinsky & Bitterman, 1953) stimuli. Results from these experiments seem to suggest that rats are more likely to show abstract conceptual behavior when olfactory stimuli are used. Indeed, of the studies cited, all were able to show that rats could learn the matching to sample task with olfactory stimuli and two of the four were able to show high accuracy levels on transfer test trials. Pena et al. (2006) conducted an experiment using a simultaneous MTS procedure where rats were trained to dig in cups of sand that had been scented with household spices. Rats in this study started with two stimuli and encountered two additional, novel stimuli each time they met criterion. Because the rats in this study met the first criterion (90% on two consecutive days) rather quickly, and did well on their first set of transfer tests, alternative probe tests were constructed to ensure that responding was due to the control of the relation between the sample and comparison stimuli. In the first novel probe phase, the novel stimulus appeared as the sample and a familiar stimulus was the incorrect comparison. Each stimulus was novel only once, after that it was incorporated into the set of training stimuli. Each time the subject met a criterion of two consecutive days at 90% or better, two more novel stimuli were introduced. During novel probe phase one, the sample stimulus was always baited and only the correct comparison contained reinforcement. To rule out responding due to previous reinforcement history and 15

22 pellet tracking, novel probe phase two contained probe trials where the sample never contained reinforcement and both comparison stimuli contained reinforcement. After meeting criterion on this phase, subjects moved to novel probe phase three which included trials that were set up to be novel-familiar (novel sample and familiar comparisons) or familiar-novel (familiar samples and novel comparisons). This was done to rule out responding to either only novel or only familiar comparisons on probe trials. Control procedures such as reducing sample reinforcement, placing a pellet in both comparison cups, pairing novel and familiar stimuli, and changing configurations helped to rule out alternative explanations in this experiment. All rats in the study met the initial criterion with two training stimuli and went on to achieve high accuracy levels on novel stimuli in all phases. By the end of the experiment, the rats encountered between 21 and 35 total stimuli. Previous evidence has shown that rats are difficult to train on the match-to-sample procedure using visual stimuli (Iversen, 1993, 1997). In these experiments, rats were trained with two stimuli on a fully automated MTS procedure. The stimuli were lighted nose poke keys, either steady or blinking. With only two stimuli, and thus four possible configurations, it was presumably more efficient for the rats to learn to respond to these configurations rather than any relational stimulus control. During the first experiment, no novel probe trials were administered. However, when analyzing data from each configuration, it was concluded that each of the four separate configurations may have been learned on its own and this might be an explanation for the high accuracy levels that were observed overall (Iversen, 1993). In the follow up experiment, configurations were changed to see if the rats could then change their behavior to responding based on the relation between stimuli rather than configurations. First, instead of the sample always being in the center, it occurred randomly on any of the three keys. Next, the sample 16

23 occurred always on the left most key. It was shown that, despite the intervention attempts, once configural learning had occurred, generalized matching-to-sample may never occur (Iversen, 1997). In both of these studies, no control by the relation between stimuli was found. There is evidence, however, that rats can show high accuracy levels on a non-match to sample task with visual stimuli when three dimensional objects are used (Rothblat & Hayes, 1987; Mumby et al., 1990). Rothblat and Hayes (1987) trained rats to respond to a delayed NMTS task with trial unique stimuli. The use of trial unique stimuli is important because it eliminates the need for transfer tests. In an experiment of this nature, the stimulus set grows with each trial because new stimuli are used every time. Therefore, alternative explanations such as configural learning or if-then strategies are not a problem because the stimuli are constantly changing. On each trial, both the sample and comparison stimuli are novel. The entire set of stimuli is used, usually over several sessions, before stimuli occur in a trial again. A down side to the use on trial-unique stimuli, as observed in this experiment, is that performance levels do not usually increase from the first to the last session. If control by the relation between stimuli was to be an explanation for responding, performance should improve as testing continues. However, in this experiment, the delayed NMTS procedure was being used to test for working memory in rats, not concept learning. Experiments of this nature usually show a decrease in accuracy as the delay increases. However, it is important to note that the subject must still compare sample and comparison stimuli and make a choice based on which comparison is different from the sample. If the subject was responding arbitrarily, performance would be at chance levels. Results of this experiment though, showed that performance was significantly higher than chance. The stimuli that Rothblat and Hayes used for this experiment were three dimensional objects which the rats were required to displace in order to achieve reinforcement. While the use 17

24 of duplicate stimuli eliminates scent marking as an explanation for performance, the material from which the stimuli were created is also an issue. The stimuli were constructed from wood, plastic, and metal. This creates a problem because while it may seem that the task involves a visual discrimination, the animal may actually be responding based on olfactory cues from the different stimulus materials (Rothblat & Hayes, 1987). In 1990, Mumby et al. followed up the Rothblat & Hayes (1987) experiment by extending delays up to 600 seconds. This experiment also used three dimensional, trial-unique stimuli constructed from varying materials. However, these experimenters were able to show an increase in performance as training increased. Significant differences in performance were seen from the first session of a delay to the last at several of the different delay levels (i.e. 4 s, 30 s, 60 s, and 120 s). Also, as expected, as the delay increased, performance decreased. Results of this experiment are interesting because they suggest that animals can respond based on the relation between stimuli across sessions and trials but that their memory for a particular sample may be strained at higher delays (Mumby et al., 1990). This comparison of short- and long-term stimulus control may have ecological ramifications. Experiments of this nature suggest that nonhumans are able to learn to respond based on relations between stimuli even when delays are implemented. There are two explanations for the varying results with visual stimuli. Rats in the Rothblat and Hayes (1987) and Mumby et al. (1990) studies were exposed to many more stimuli than rats in the Iversen (1993) experiment. The number of stimuli an organism is exposed to in training is referred to as the set size. There is evidence that increasing the set size makes abstract concept learning more likely. One study showed that rats trained using two olfactory stimuli performed significantly worse on novel transfer tests than rats trained using a set size of ten 18

25 stimuli. Additionally, the same study showed that if the set size was increased from two to ten stimuli, after the first set of novel probe tests, the same rats could learn to respond based on the relation between stimuli (Lazarowski, 2010). Lazarowski (2010) used a zero s delay procedure to train rats on either MTS or NMTS tasks. The rats in this study were initially trained using either a set of two or ten olfactory stimuli. The stimuli were plastic Plexiglas lids that had been scented with household spice. Rats were required to push back the lid and dig in the cup of plain sand underneath in order to retrieve a food reward. Once the subjects met criterion of two consecutive days at 90% or better, the rats were given a probe using 10 novel olfactory stimuli. After this probe they received further training with the novel stimulus set until meeting the same criterion again. This meant that the group that previously had been trained with two stimuli was now being trained with 10 stimuli. Upon meeting criterion, they got a second probe with 10 more novel stimuli. The results did show differences between rats trained with two stimuli and those trained with 10 stimuli. The group trained with 10 stimuli did significantly better on the first probe than the group trained with two stimuli. Interestingly though, when the set size was expanded for the rats in the two stimuli group, they did significantly better on the second probe, so much so that no significant differences were found of the second probe between the group initially trained with two stimuli and the group initially trained with 10 stimuli (Lazarowski, 2010). Set size also played a crucial role in another experiment done with capuchin and rhesus monkeys (Wright et al., 2003). This study used a same/different procedure where the monkeys were shown two pictures and a white rectangle on a computer screen. If the pictures were the same, the response of touching the lower picture was reinforced. If they were different, responses to the white rectangle were reinforced. The same/different task is another commonly used 19

26 experimental manipulation for studying oddity/identity concept learning. Presumably, as in match-to-sample, the organism must observe both stimuli and respond in one way if they are the same and in another way if they are different. The monkeys in this experiment were trained using a set of eight stimuli and then given transfer tests. Performance for all three monkeys on this first set of transfer tests was not above chance. Size of the training set was then doubled four times for all of the subjects. Additional transfer tests were administered after criterion was met for the 32-, 64-, and 128-item training phases (Wright et al., 2003). Results of this experiment showed that as the size of the training set increased, so did accuracy levels on the transfer tests. Using a set size of only eight stimuli, the subjects were not able to generalize stimulus control to novel exemplars. However, as training set size increased, attending to specific features of individual stimuli became less effective and was replaced by comparison of the two stimuli. With the increased set sizes, all subjects were able to show transfer performance equivalent to their baseline accuracy levels (Wright et al., 2003). Also of note here is that increases in set size also seem to bring the organism s behavior under control of the relation between stimuli more rapidly. As set size increased, the number of trials needed to meet criterion for all three monkeys decreased. One subject was even able to meet criterion on the new set size training (64 & 128) in as little as one session (Wright et al., 2003). The studies cited suggest that the use of three dimensional and olfactory stimuli produce less variability in results than do experiments using two dimensional visual stimuli. This finding may be a result of three dimensional and olfactory stimuli being more ecologically relevant than two dimensional visual stimuli. Also, the use of digging as a response may more closely resemble scavenging behavior of wild rodents than nose poking a key. Wright and Delius (1994) showed that experiments that more closely resemble an organism s evolutionary environment 20

27 may yield higher accuracy levels, using fewer stimuli and requiring fewer trials. In this experiment, pigeons were trained to dig in gravel pots to obtain food pellets using MTS and NMTS procedures. High levels of accuracy were achieved quite rapidly, using only two different training stimuli. However, when transfer tests were administered, accuracy levels were not significantly different than chance. The authors postulate, though, that if a larger set size were used, higher levels of transfer accuracy would be seen. As Katz and Wright (2007) suggest, using a procedure that is better suited to the predispositions of the animal may be a more valid way to study conceptual behavior. The present study incorporated ecologically relevant procedures (i.e. displacing stimuli and digging in sand to retrieve reinforcers, as in Pena et al., 2006; Wright & Delius, 1994) and larger set sizes (i.e., ten stimuli as in Lazarowski, 2010) to study cross modality nonmatching to sample in rats. Rats were trained using one modality and then given novel transfer tests with stimuli from another modality. Since the evidence presented above suggests that rats rely heavily on olfaction, it is predicted that rats trained with olfactory stimuli will require fewer sessions to reach their first probe compared to the number of sessions required for rats trained initially on visual stimuli. However, no difference on probe performance is expected between olfactory and visual stimuli. It is expected that, after sufficient training, the rats will show transfer of relational control to their training modality and to the new modality. Rats should be able to transfer relational control to the new modality, regardless training stimuli, as evidenced by above chance accuracy on novel probes. 21

28 Method Subjects Eight male Harlan Sprague Dawley rats (Harlan Laboratories, Indianapolis, IN) approximately days old at the start of testing were tested in this experiment. The rats were housed in individual home cages as part of a colony that is maintained on a 12 hr reverse light/dark cycle, with the dark cycle beginning at 7 a.m. EST. Rats were tested five days a week during the dark cycle. Outside of the experimental session, the rats were given free access to water. They were maintained at a weight that was approximately 85% of their free feeding body weight. The subjects received their daily food ration, in the form of Purina Lab Diet grain pellets, approximately half an hour after the end of the experimental session. 45 mg sucrose pellets used for reinforcement in this experiment were manufactured by Bioserve. Materials and Apparatus A webcam connected to a home computer using the Windows XP operating system was used to video record all sessions. This webcam was mounted to the wall directly above the experimental chamber. The experimental chamber itself was a modified operant chamber (30 cm x cm x 25 cm); the top, front, and back of the operant chamber were clear Plexiglas, while the left side, right side, and floor were metal. The floor consists of 14 metal rods spaced 1.25 cm from each other and the walls of the chamber. A waste pan 3.75 cm in height was located under the floor of the chamber. Thus, the dimensions of the actual testing space for the rat are 30 cm x 27.5 cm x 25 cm. There is a 10 cm x 30 cm gap at the bottom of the front wall of the chamber to allow stimuli to be inserted into the chamber. The stimuli were provided by the experimenter on two trays, the sample tray and the comparison tray. The trays were constructed of clear Plexiglas. Both trays were 25 cm x 20 cm x 22

29 3.75 cm. A hole was cut in the sample tray 9.5 cm from the sides of the tray, 3.25 cm from the front, and cm from the back of the tray. The comparison tray contained two holes; these holes are located 3.25 cm from the front, cm from the back, and 3.25 cm from the sides of the tray, with 7.8 cm between holes. All holes were circular with a diameter of 5.5 cm. Each hole was filled by a plastic 2 oz condiment cup (5.5 cm in diameter), filled approximately 75% full with sand. Surrounding each hole are four metal screws,.6 cm from both sides of hole. The first set of screws was positioned cm from the front of the hole and the second set was 2.5 cm behind the first. These screws are in place to guide the lid straight back as it is pushed by the rat. Two types of stimuli were used for this experiment, olfactory and three dimensional visual. The olfactory stimuli were clear, plastic lids, 5 cm x 7.5 cm, that have been scented with common household spices obtained from Great American Spice Company (e.g. cumin, rosemary, oregano, raspberry, etc.). To ensure that the lids absorbed the scent, they were housed in containers containing the spice any time they were not in use. The visual stimuli were 7.5 cm tall and 5 cm wide. All were constructed from Legos, and vary in color and shape. The visual objects were mounted on 5 cm x 7.5 cm clear, plastic lids using Velcro. Duplicates of all stimuli were used throughout the study to control for scent marking cues. Within each experimental session, each individual in a pair of duplicates served as both the sample and comparison stimulus. Forty total stimuli of both types were constructed and then divided into four sets of 10 stimuli. Procedure Throughout the study, a response was defined as movement of the plastic lid with either the paws or nose such that the lid was pushed back over the cup past the first set of screws on the tray. Pellets were located near the rear of the cup, buried just below the surface of the sand. In 23