Sometimes the best way to understand current affairs is to examine them from a historical perspective. When police canine tracking began on a large scale during the late 1800s and early 1900s, it was widely believed trained police tracking dogs would scent match an item handled by a perpetrator to his track and/or that individual, respond to (follow) human scent, and discriminate between human scents while tracking. However, early Prussian tests revealed most police tracking dogs tested did not scent match human scent from strangers, respond to human scent, or discriminate between human scents while tracking. Prior to testing, police canine tracking solutions had not been divided into categories; problems that could be solved without the use of a scent sample and others that could only be solved by matching the odor on a scent sample to a matching human scent trail. When trials began with the presentation of a scent sample but training exclusively involved tracking problems that could be solved without use of the scent sample to signal the correct choice from one trial to the next (pseudo matching-to-sample; pseudo-MTS), subsequent tests that controlled against the simpler solutions revealed dogs did not learn to scent match even though each track began with the presentation of a scent sample. In addition, training typically involved track layers with prior reinforcement history, whereas the controlled tests involved strangers. Here, evidence from canine tracking and animal learning science is contrasted. Evidence indicates, much of the unsatisfactory test results are not because dogs are incapable of scent matching novel human scent, detecting human scent along a track, or discriminating between human scents while tracking, but rather, dogs do not learn the task requirement to scent match from arrangements of training that do not control against alternative solutions. Due to training and testing procedures that lack appropriate controls, many of the same misconceptions existing when police canine tracking began, also persist today. Although many police tracking problems can be very accurately solved with tracking dogs that do not to scent match, when circumstances require human scent matching canine teams, dogs should be control tested prior to deployment in order to determine whether they scent match. For human scent matching/nonmatching test criterion see “Canine Human Scent Matching/Nonmatching Conditional Discrimination” at https://k9scentdiscrimination.com/canine-human-scent-matchingnonmatching-conditional-discrimination/.
The inception of comparative psychology
George J. Romanes’ book “Animal Intelligence”, published in 1882, marked the beginning of the field of comparative psychology. Romanes, a close acquaintance of Charles Darwin, not only accepted Darwin’s evolutionary theory but also adopted his inductive approach in order to infer mind in others. Romanes systematically compiled a collection of anecdotal stories that portrayed exceptional feats of animal intelligence, from which he built an elaborate theory of the evolution of intelligence. For Romanes, in an effort to demonstrate psychological continuity of man in other species, the aim of comparative psychology was to find evidence of mind in animals other than man. The comparative study of animal learning was important to Romanes because if the occurrence of flexible adaptive behavior could be shown to be a consequence of individual experience rather than innate endowment, it would provide indisputable evidence for the attribution of mind. However, Romanes and his contemporaries did not distinguish what they observed from what they subjectively inferred.
Lloyd Morgan (1894), often recognized for his delivery of comparative psychology from clutches of subjectivity, argued that the apparent complexities of animal behavior is not a reliable guide to the complexity of the underlying processes involved. The challenge is to objectively identify adaptive behavior that cannot be accounted for by simpler means. To that end Morgan wrote his famous canon, which remains is a fundamental precept of comparative psychology. It states, “In no case is an animal activity to be interpreted in terms of higher psychological processes if it can be fairly interpreted in terms of processes which stand lower in the scale of psychological evolution and development.” However, Morgan’s canon does not mean higher psychological processes can never be attributed. Rather, in the absence of clear objective evidence the canon admonishes us against interpreting an animal’s actions in terms of our own higher cognitive processes when lower processes are sufficient to account for behavior observed. In his 2nd edition of “An Introduction to Comparative Psychology” (1903, p. 59) Morgan added, “To this, however, it should be added, lest the range of the principle be misunderstood, that the canon by no means excludes the interpretation of a particular activity in terms of higher processes, if we already have independent evidence of the occurrence of these higher processes in the animal under observation”. Morgan cautions, we must take an objective approach by devising experiments with adequate controls in order to rule out alternative explanations. Unless objective evidence can be found demonstrating clearly that more complex processes are involved, it must supposed that a simpler process is responsible for the behavior, at least until clear objective evidence to the contrary is provided.
Following Morgan, the American psychologist Edward L. Thorndike (1898, 1911) agreed. It was largely in reaction to the anecdotal tradition during the last decades of the nineteenth century that Thorndike took a more analytical approach. He opposed the uncritical acceptance of stories about exceptional feats of animal intelligence. For Thorndike, continuity of mental capacity involving unsubstantiated anecdotes left too many unanswered questions concerning the nature of the processes underlying adaptive behavior. He believed that investigation of the actual ways in which animals set about solving problems required the development of controlled experimental techniques. Hence, Thorndike shares with Pavlov the honor of founding the controlled experimental analysis of animal learning. However, on the basis of his own experimental findings, and probably in part due to his opposition to the extremes of the anecdotal tradition, Thorndike came to believe the way animals went about solving problems was actually quite simple, involving trial and error learning in which stimulus-response connections were strengthened or weakened in accordance with his law of effect. Thus, although comparative psychology began at one extreme in which animals were credited with extraordinary powers of intelligence, beliefs of animal acumen quickly swung to an opposite extreme in which animals were thought only capable of trial and error learning through random attempts until they hit upon the correct solution by chance. Not only was associative learning limited to automatic stimulus-response habit formation, animals were thought incapable of the acquisition of knowledge about the causal or predictive relationship between a response (or stimulus) and reinforcement, observational learning, insight, reasoning, perceptual learning, comparative relational learning, analytical reasoning, generalized rule learning, concept formation, numerical processing, or any forms of language acquisition. Although it was Watson (1913, 1914, 1916) who advanced the Behavioristic Revolution in America, during which the study of animal learning took center stage of all psychology, Thorndike’s work marks the beginning of the behaviorist tradition in which private unobservable events were not accepted as instrumental to an organism’s behavior.
About the time Thorndike began his studies on instrumental conditioning as a young research student in America, Ivan Pavlov, already a prominent research scientist, was starting his work on classical conditioning in Russia. Pavlov was the first to show how it is possible to objectively and quantitatively measure the processes of perception and learning in animals. During his work on the digestive system of dogs, Pavlov noticed that dogs learn about the stimulus precursors of food. Moreover , Pavlov saw that it could be used as a tool for the objective study of animal learning. From around 1898 to 1930, he occupied himself with the study of classical conditioning and his theory of stimulus substitution (conditioned reflexes). Pavlov (1927) developed experimental procedures, acquired data, and formulated theories to explain the data.
While distinguishing between two types of conditioning, classical and instrumental, the American psychologist and radical behaviorist B. F. Skinner (1938) acknowledged Pavlov’s theory of stimulus substitution. Yet, Skinner thought classical conditioning was limited to reflex responses. Thus, Skinner assumed the responses elicited by stimuli form just a small part of an organism’s potential repertoire of behavior and were therefore of little benefit. Skinner thought the function of conditioning was to adapt animals to their environment. He thought all adaptive behavior was modified by its consequences according to Thorndike’s law of effect. Actions with beneficial consequences were repeated and actions with detrimental consequences were not. Furthermore, behaviorists thought all changes in behavior, instrumental and classical, occurred not because animals acquired knowledge about the causal or predictive relationship between events, but were the result of the gradual strengthening of unconscious automatic stimulus-response habits.
However, by the late 1960s early 70s behaviorism began to fall out of favor, in part, when it was shown through purely classical conditioning experiments that Pavlovian conditioning does involve adaptive behavior, a much wider range of behavior than Skinner and others had supposed. Classical conditioning does not simply involve the establishment of discrete reflexes and is not confined to automatic responses. Both autoshaping (e.g. Brown & Jenkins, 1968) and sign-tracking experiments (e.g. Hearst & Jenkins, 1974) have shown that purely classical conditioning procedures can modify an animal’s movements in space towards stimuli associated with (predictive of) appetitive outcomes. The phenomenon is termed “sign-tracking” because in the absence of instrumental conditioning, animals adapt their behavior so as to contact or be in close physical proximity with the predictive stimuli. Yet behaviorists did not believe animals acquired knowledge (learned) about the predictive relationship between events in which a change in behavior reflected such knowledge. For them, a change in behavior was learning; the automatic strengthening of stimulus-response connections in which reinforcement served only as a catalyst to strengthen the link between a stimulus and response in order to form stimulus-response habits.
In order to account for autoshaping and sign-tracking experimental findings, a radical reappraisal of the importance of classical conditioning, and conditioning in general, was required. For, if after purely classical conditioning animals modify their behavior adaptively in order to approach stimuli associated with appetitive events and avoid stimuli associated with aversive events, the change in behavior from reflexive responses (such as salivation) indicates animals acquire knowledge about the predictive relationship between a CS and subsequent reinforcer during conditioning. Resulting from a contingent relationship between a stimulus and reinforcer, animals acquire knowledge about the causal (predictive) relationship between them, and the change in behavior is a measure of such learning. The reappraisal of classical conditioning was important because much of the behavior animal learning psychologists study involves approaching or avoiding places of benefit or harm. Moreover, from an evolutionary perspective, in order to respond adaptively and survive in the real world, much of what animals must learn about are the signals predicting benefit or harm so that appropriate approach or withdrawal can occur in anticipation of such outcomes.
In response to mounting experimental evidence disputing traditional views, Great Britain comparative psychologists Stuart Sutherland and Nickolas Mackintosh laid the foundation for contemporary associative learning and conditioning with the publication of “Mechanisms of Animal Discrimination Learning” (Sutherland & Mackintosh, 1971) and “The Psychology of Animal Learning” (Mackintosh, 1974; see also Mackintosh, 1983). The new field of contemporary associative learning and conditioning took a radical departure from the simplistic views of behaviorism while preserving controlled objective experimental analysis of animal learning. Since the second half of the 20th century, there has been a renewed interest in classical conditioning. Today, it is well established that classical conditioning plays a very large part in adaptive behavior (see e.g. Rescorla). Researchers now know that animals actively process stimulus information. Classical conditioning is now regarded as the acquisition of new knowledge about the relationship between stimulus events, rather than the acquisition of conditioned reflexes. It is no longer hazardous or even out of the ordinary for CS-US and response and outcome associations to be referred to in terms of central representations, rather than in behaviorist automaton stimulus-response connection terms. Today, the study of causal observational learning, insight, reasoning, perceptual learning, relative relational learning, analytical reasoning, generalized rule learning, concept formation, numerical processing, and language learning in animals are all possible thanks to experimental discoveries involving classical conditioning.
During experimental investigation of animal learning, experimenters employ reinforcement contingencies to expose subjects to certain causal relationships between a particular to-be-conditioned event and reinforcement with the assumption representations will be formed between them. As a consequence of learning the causal relationship between events, behavior changes in certain specifiable and measurable ways. Some of the ways learning is measured is by changes in behavior, error rate, transfer of a solution strategy to a new task involving different response requirements, enhanced learning, retarded learning, and response probability, ease, intensity, speed, or latency. For the purposes of objectivity, alternative solutions in which subjects can earn reward are identified and controlled against during conditioning according to specified procedural constraints defined within the conditioning procedure. Experimenters assume subjects have learned about the causal relationship if behavior changes as a consequence of exposure to the contingent relationship.
However, in order to determine whether or not behavior would have changed similarly without exposure to the reinforcement contingency, one of two types of experimental designs is employed; a between-subjects or within-subject design. In between-subjects experiments, there are two or more groups of experimental subjects; experimental and control groups. The experimental group is exposed to the contingent relationship, whereas the control group is not. After conditioning, differences in learning between the groups are compared and analyzed. The task of conditioning analysis is to determine the critical aspects of training responsible for the changes in behavior observed, to specify the relationships learned, to discern how events are represented, to elucidate rules that influence the formation of associations, and finally to predict how future behavior will change as a consequence of exposure to certain relationships. If, after careful detailed analysis it is determined that the experimental group’s behavior changed as a consequence of exposure to the reinforcement contingency, learning about the contingent relationship is presumed to have occurred. In the within-subject design, a subject’s behavior is recorded before, during, and after the introduction of a reinforcement contingency and response behavior before, during, and after is compared. If the subject’s behavior changes as a consequence of exposure to the reinforcement contingency, learning is assumed to have occurred about the contingent relationship.
The inception of large scale police canine tracking
The founders of comparative psychology were Romanes, Morgan, Pavlov, and Thorndike. Noteworthy is that Romanes’ writings not only marked the beginning of the field of comparative psychology in 1882, a report he wrote about tracking experiments he conducted with his setter, published in Nature in 1887, marked the beginning of both canine tracking on a large scale and (though not by modern standards) the scientific study of canine tracking (also published in Revue Scientifique and Journal of the Linnean Society).
Following Romanes 1887 report in Nature, police canine tracking gained rapid popularity. For example, by 1897, the Association of bloodhound breeders was founded in England with trailing as one of its objectives. Schäferhundverein (the study of tracking dogs) was founded in Germany in 1899, which played an important part in the study of tracking dog training. Around 1899, reports of tracking dogs trained with methods used by police in Ghent Belgium brought worldwide recognition, resulting in some dogs being exported to Germany, Russia, Britain, and France. Brough, who was chairman of the Association of bloodhound breeders, published an article in 1902 on the use of bloodhounds in tracking criminals. In 1908 a German police dog training school was opened in Karlshort Berlin where German handlers and others from around the country attended to learn their training methods (later moved to Grünheide). Zell published a book on police canine tracking in 1909, in which he called for scientific studies. The same year the Verein für deutsche Schäferhunde (German Shepherd Dog Society) offered a reward of 25 marks to every police dog handler who successfully solved a homicide case with a German Shepherd, resulting in 18 payments during the first eight months of the offer. In Saarbrücken Germany, Konrad Most, who was once director of the Grünheide training school for police dogs in Germany and canine specialist in the German army, began controlled investigation of tracking dogs in 1909. By 1911, Friedo Schmidt brought popularity to the use of police dogs as detectives with his accounts of a number of extraordinary scent matching tracking and lineup criminal cases. His accounts of criminal cases were analogous to Romanes’ reports of exceptional feats of animal intelligence. In those cases, police dog teams identified known suspects, which resulted in some teams receiving the Verein für deutsche Schäferhunde reward.
By 1913 and 1914 however, the popularity and credibility of police canine tracking began to give way to much controversy when highly controlled official Prussian tracking and scent matching tests held near Berlin Germany found that when unknown suspects were involved, police dogs from around the country did not scent match and discriminate between stranger laid tracks. During the official Prussian tracking tests, all dogs tested on tracks laid by strangers failed to discriminate between target and decoy tracks of the same age or fresher and were diverted by decoy tracks (Böttger, 1936). That is, after presentation of a scent sample from a stranger, they failed to respond only to the corresponding track when given a choice between a matching track and similar stranger laid decoy tracks of the same age. At junctures in which a target track turned and a decoy track continued straight, dogs failed to change direction. No tracking could be undertaken on dry sandy ground without undergrowth, dry stone, or asphalt. None could follow a track older than two hours. None could sample the human scent on an item and then choose from among alternatives another item that had been handled by the same individual. Similarly, after sampling the odor on a scent sample, none could correctly choose from among others, the individual who had handled the scent sample. The findings from official Prussian tests that police dogs from around the country did not scent match and discriminate between stranger laid tracks, lead the Prussian Government to forbid tracking dogs to be used as detectives. Konrad Most also reported similar results in tests he conducted.
On the basis of anecdotal evidence, Friedo Schmidt (1910) and others believed police tracking dogs used individual unique human scent information in order to scent match, follow tracks, and discriminate between tracks. The Menzels, who under well controlled testing established dogs can scent match, argued response behavior must be guided solely by individual human odor. If this was so, it was supposed tracking dogs should be able to match the individual odor on a scent sample to a corresponding track, overcome difficulties caused by various types of terrain, and discriminate between diversionary tracks. Alternatively, experimental evidence gathered from dogs trained to track on soft surfaces, led Most (1954) to believe the guiding factor is almost entirely the scent of the track (such as physical changes to downtrodden vegetation and ground surfaces), in which human odor may sometime play a part, but never individual (discriminative) human scent elements. Hansmann (1929), who after Most was regarded as the most prominent of German experts of that time and who was also once director of Grunheide Training School for Police Dogs, also contended that human odor or individual odor is rarely, if ever, the guiding factor. He theorized, due to biochemical changes continually occurring to foot printed tracks, individual human odor is destroyed while a specific distinguishable track odor is produced that dogs can detect and discriminate from other tracks. Thus, Hansmann maintained human laid tracks differ not because of differences in individual human scents but because every track has its own specific track odor and because in many cases, tracks also differ in age. Böttger, a colleague of Hansmann, was another highly regarded expert of the time who published a canine tracking training booklet that the Berlin police used. From tracking tests involving Berlin police tracking dogs trained on vegetative surfaces, Böttger (1936) concluded it is the odor of the track that is significant and it is the odor of the track dogs should be trained to follow. Thus, according to Most, Hansmann, and Böttger, dogs should not be able to overcome difficulties caused by certain types of terrain and they should not be able to discriminate between tracks on hard surfaces. However, much of the evidence they obtained was from dogs trained on soft surfaces, which followed the threat to discontinue the use of police tracking dogs all together, rather than just their use as detectives.
In a memorandum to the Prussian Minister of Interior in 1925, Hansmann stated, among other things, it is a false assumption that every human being transmits to his track his special individual scent to which dogs can smell, follow, and discriminate between. Although, criminal experience had confirmed adequately trained police tracking dogs can follow human laid tracks hours old without changing to fresher or older diversionary tracks, they achieve such feats, not by the scent of the track layer, but by the scent of the track. He argued, tracking dogs are not criminal investigators and their function is not to identify criminals. Therefore, they should never be allowed to halt and bark at a suspected individual in order to identify them. Later, at an official conference in Munich in 1931, Hansmann’s memorandum was accepted and rules for deployment and prevention of the misuse of police dogs tracking were drawn up. As a result of the compromise, the controversy over whether dogs scent match and follow individual human odor or respond to track scent was quelled, but not resolved. Although the range of applicability that had initially been applied was reduced by excluding scent dogs as detectives, the threat to the use of police tracking dogs was eliminated. Following the changes, police tracking dogs achieved great popularity and credibility throughout Germany.
Although much of the controversy stemming from tracking and lineup tests was over individual odor theory and whether dogs were capable of scent matching, some believed inadequate test results were, in part, due to faulty training procedures (Blunk, 1926; Menzel & Menzel, 1930). There was good evidence to indicate when training arrangements were changed, the solution strategies dogs learned changed in turn. For example, Blunk successfully trained a dog to sample the scent from a stranger and then choose from among other stranger scented items, the item that had been handled by the scent sample stranger. Böttger (1936) reported that the Menzels exhibited two boxers that could also choose from among all novel alternatives, the human scented alternative that matched a previously smelled human scent sample. Of even greater significance is in 1928 Schäferhund Verein published reports of tests they conducted in which dogs trained by the Menzels not only scent matched human scent from strangers, even when Rudolfina Menzel’s scent was among nonmatching alternatives, but moreover, refused to select an item from among stranger scented alternatives when there was no matching human scent present. The significance of the nonmatching test trials is that correct response indicates the Menzels dogs learned same-as/different-form relational rules to solve human scent matching/nonmatching problems, which involves higher-order cognitive processes than alternative solutions that can be used to solve matching only tests.
Assuming training methods have changed, it would be easy to dismiss early test results as unimportant. However, we must first consider the 1913/14 standard of testing. By taking into account that during active service suspects and members of the public are strangers, during testing dogs were not only required to remain on a target track while discriminating between competing tracks of the same saliency, moreover, they were tasked to use the scent sample as a cue to signal which novel discriminative stimulus was the correct choice on any given trial, which is a more complex achievement than simple discrimination between stimuli that have come to signal important outcomes from prior reinforcement history. If contemporary tests are not conducted with the same high standard of controls against simpler alternative solutions, we cannot know if new or somewhat changed training methods improve accuracy and reliability of scent matching dogs. Because, in the absence of controls against simpler solutions, dogs can achieve 100% accuracy during training and testing, but fail miserably when the only way to solve multiple choice novel human scent problems is with a scent matching solution strategy that can transcend training stimuli. Although many real world tracking problems can be solved with simpler solutions, such as simple discrimination between the presence and absence of a track, simple discrimination between track saliencies, or simple discrimination between olfactory information that has come to predict reinforcement from prior reinforcement history and similar odors, unless good evidence is provided that a dog in question has been tested with arrangements that control against simpler solutions, it must be supposed that a dog in question has not learned a more complex scent matching solution and response performance is controlled by simpler processes. And thus, the dog is not qualified for operations in which a scent sample must be used to solve the problem.
Discrimination training is the most important part of any kind of scent work training. Despite its importance, it often seems the various discrimination training procedures and what they produce are not as well understood as they should be. For instance, it is not uncommon for people to think they can establish reliable discrimination between similar stimuli, such as human scents, with single stimulus conditioning. Single stimulus conditioning involves discrimination between the presence and absence of a to-be-discriminated stimulus. Training is arranged by pairing a to-be-discriminated stimulus (or a to-be-discriminated stimulus followed by a response) with reinforcement, while the context alone (in the absence of the to-be-discriminated stimulus) is paired with the omission of reinforcement. Although single stimulus conditioning involves discrimination in that subjects must discriminate between the presence and absence of a stimulus, it does not produce reliable discrimination between similar stimuli.
Pavlov (1927, pg. 117) originally thought that if he paired a single stimulus with food often enough, his dogs would ultimately respond only to that stimulus (conditional stimulus, CS) and no other when given a choice between the conditional stimulus and similar test stimuli. However, in some cases even after over a thousand pairings of a single conditional stimulus with food, his dogs never mastered complete discrimination between the conditional stimulus and other similar test stimuli. In other words, accuracy reverted to chance when similar discriminative stimuli were added; his dogs not only responded to the conditional stimulus, they also responded to the test stimuli.
In Pavlov’s experiments, involving single stimulus conditioning, failure to discriminate between the conditional stimulus and a similar test stimulus was not because the dogs were incapable of discriminating between them. Pavlov reported that on the first occasion that the test stimulus was presented, it would elicit little or no response. No response to the test stimulus can only occur if it is discriminated from the conditional stimulus. In order to not respond, Pavlov’s dogs must have detected the test stimulus was different from the conditional stimulus. Therefore, when the dogs eventually started responding to similar stimuli on subsequent test trials, it could not have been due to an inability to discriminate between the conditional stimulus and similar test stimuli. Pavlov’s observation, that on the first occasion a test stimulus was presented it elicited little or no response, indicates the dogs were capable of discriminating between the conditional and test stimulus, but they had not learned from single stimulus conditioning the task requirement to discriminate between the conditional and test stimuli. Single stimulus conditioning does not produce reliable discrimination because it is not specific. There is nothing in the arrangement of single stimulus conditioning to inform subjects the task requirement to discriminate between the similar stimuli.
Likewise, when dogs are exclusively trained on a single laid track, with no competing tracks, there is nothing in the arrangement of training to inform them the task requirement to discriminate between the target track and similar decoy tracks. Thus, when dogs are tested with competing decoy tracks to see if they will continue following the target track, the results show indiscriminant performance over initial trials. When dogs are routinely trained one way — to generalize — and are then tested another way — to discriminate, they typically respond in the manner in which they were trained, show generalization or chance performance during discrimination tests between similar stimuli. Moreover, this is the case regardless of whether each training trial begins with the presentation of a scent sample.
During single stimulus conditioning, Pavlov trained his dogs on a discrimination between the presence and absence of a given stimulus and then tested them not between the presence and absence of the stimulus, but along a specific stimulus dimension.*1 For example, following one experiment Pavlov (1927, pg. 113) reported that if a 1000-hertz tone was established as a conditioned stimulus, many other tones acquired similar properties. He conditioned dogs to a tone of 1000 Hz frequency followed by food. Once the dogs were reliably salivating after the presentation of the tone, Pavlov tested them in extinction (without reinforcement) on tones varying from 1000 Hz. What Pavlov discovered was that there was an orderly relationship between the strength of response and frequency of the tones; response strength diminished proportionally, with frequencies nearest the training stimulus eliciting the strongest responses to frequencies farther removed from a 1000 Hz eliciting little or no response. In other words, response to 950 and 1100 Hz tones was stronger than response to 850 and 1200 Hz, with 300 and 3500 Hz tones eliciting the least response. Pavlov termed this phenomenon generalization.
Stimulus generalization refers to the tendency of an animal conditioned to respond in a certain way to one stimulus to respond in the same way, but to a lesser extent, to another similar stimulus. Pavlov thought generalization was the consequence of excitation from one particular cortical point that spread out or irradiated across the cortex, diminishing as it progressed further, and generalization experiments were taken to support his account. Alternatively, Konorski (1948 pg. 128) argued that generalization was not the consequence of excitation or irradiation from one particular cortical point, but the result of distributed overlapping activation. He wrote, “It must be assumed that the cortical centers of particular stimuli represent complex and widely dispersed formations, that they can partially overlap, and that this overlapping is the cause of generalization.” Subsequent experimental research supported Konorski’s analysis over Pavlov’s.
The phenomenon of generalization can be more easily understood when it is considered that the sensory event called a stimulus is a nonspecific term of convenience, which is actually comprised of a complex set of elements involving both absolute physical properties and stimulus dimensions. Although a single element is also a stimulus, training stimuli are always comprised of a set of elements that may or may not vary during training. For example, even the simple 1000 Hz tone Pavlov used during single stimulus conditioning was comprised not only of frequency, but also amplitude, presentation duration, and site of origin. If none of those elements vary during conditioning, all of them may be presumed capable of entering into association and thus, contributing to some extent to the response elicited by the tone after conditioning. As far as their correlation with reinforcement is concerned, they would be as relevant as frequency. Thus, to the extent that another stimulus has elements in common with the conditioned stimulus, the subject may respond to it similarly after single stimulus conditioning. For example, during and after single stimulus conditioning, the conditioned and test stimuli may each have certain unique distinguishing elements, symbolized as A and B, but also have some overlapping elements in common with one another, symbolized as X. Therefore, the conditioned stimulus would consist of a compound of AX elements and the test stimulus would consist of BX elements. Generalization to the test stimulus B will occur after conditioning to A because X elements are present during both AX conditioning and BX testing. The greater the proportion of unchanged X elements controlling the subject’s behavior, the less will variations in the test dimension (such as discrimination tests between different Hz) effect change in behavior.
Konorski’s analysis predicts that if two stimuli, A and B, share no features in common, there will be no generalization between them. But that the addition of common elements, X, will result in substantial generalization. To test Konorski’s analysis that generalization arises by virtue of elements shared in common with training and test stimuli, researchers explicitly constructed stimulus compounds with independently controllable common elements. In one example, Mackintosh, Kaye, and Bennett (1991) confirmed these simple predictions by using dissimilar flavor stimuli; saline, sucrose, and lemon. An aversion conditioned to saline (A) did not generalize to sucrose (B) when tested, but compound conditioning to saline-lemon (AX) generalized strongly to a test compound consisting of sucrose-lemon (BX).
Subsequently, Bennett, Wills, Wells, and Mackintosh (1994) showed that generalization between AX and BX is dependent on the strength of conditioning to X. After conditioning an aversion to AX, a single extinction trial, involving presentation of X followed by the omission of the expected aversive reinforcer (illness), was enough to reduce a generalized aversion to BX during testing. However, after an AX aversion conditioned to a control group of subjects (a comparison group), an unreinforced trial to B did not have any effect on a generalized aversion to BX during testing.
Note, according to modern methods a reinforcer can be either appetitive or aversive as long as it is motivationally significant; the responses observed are simply opposite in sign. Modern thinking defines an appetitive reinforcer as one that will usually increase the probability of a response and an aversive reinforcer as one that will usually decrease the probability of a response it follows. By more broadly defining reinforcers as events that support a change in behavior, rather than the more limited behaviorist view that an event is a reinforcer only if its presentation strengthens a response, a reinforcer for classical conditioning can also be a reinforcer for instrumental conditioning. Thus, whether a stimulus or response is associtated with a particular reinforcer depends on how well an event is correlated with that reinforcer relative to other events.
Bennett et al. (1994) and Mackintosh et al. (1991) also showed that generalization can be reduced by unreinforced exposure to X prior to AX conditioning. Unreinforced preexposure to X causes latent inhibition to X (it causes subjects to learn to ignore X because it predicts nothing of importance), thus when an aversion is conditioned to the compound AX, latent inhibition to X insures that the aversion is conditioned to A alone, rather than X.
By regarding stimuli as being comprised of a complex set of stimulus elements, experimenters have found the degree to which either generalization or discrimination occurs between similar stimuli can be modified by prior experience and arrangements of training. Some arrangements, such as single stimulus conditioning (discrimination between the presence and absence of a stimulus), can increase generalization, while other training arrangements can reduce generalization; increase the ease to which similar stimuli are discriminated between.
*1. A stimulus dimension can be any class of stimuli comprised of a spectrum of stimuli in which each new adjacent stimulus on the spectrum is produced by the addition or elimination of some perceptual component of its neighbor and replacing the eliminated component with a new component not present in the prior stimulus, thus allowing the stimulus dimension to be ordered by similarity from one end of a spectrum to another. For example, the color red not only has its own absolute physical properties, but is also part of a stimulus dimension, color.
Likewise, if we regard track scent as being comprised of a complex set of stimulus elements (olfactory information such as crushed vegetation, disturbed ground, foot wear odors, relative saliency, and information about the track layer – gender, hormonal, dietary, perhaps disease and medication, hygiene, hygiene products, individual unique genetic information, or tobacco), then to the extent a competing track has unchanged controlling elements in common with previously reinforced single laid tracks, the less will discriminative elements affect performance; that is, dogs can be expected to respond indiscriminately. The greater the proportion of unchanged elements controlling a dog’s response performance (such as crushed vegetation, disturbed ground scent and perhaps elements common to track layers), the less will discriminative elements unique to a target track affect choice performance, such as choosing the target track at a choice point over a decoy track.
Simple Discrimination between Similar Stimuli
Eventually Pavlov discovered that if he exposed his dogs to two or more similar stimuli and correlated them with different schedules of reinforcement, his dogs would rapidly learn the discrimination task. In this procedure, a positive discriminative stimulus (CS+) was randomly presented and always followed with food, a reinforcer (US). The other stimulus, a negative discriminative stimulus (CS-), was also randomly presented but never followed with the reinforcer. Randomly presented means, from one trial to the next there is no predictable order or relative presentation position (in the case of simultaneous presentation of discriminative stimuli) in which the positive and negative stimuli are presented. Thus, over trials stimulus presentation order or position was not reliably followed with reinforcement or its omission, whereas CS+ was reliably followed with reinforcement and CS- was reliably followed with the omission of reinforcement. After just a few random presentations of the positive and negative stimuli, Pavlov’s dogs responded only to the positive stimulus during testing. Due to its rapid learning effect, the procedure of randomly presenting one stimulus that is always followed with reinforcement and another that is never followed with the reinforcer has come to be known as the discrimination learning procedure.
In simple stimulus discrimination learning procedures, subjects initially respond similarly in the presence of both the positive (CS+ or S+) and negative stimulus (CS- or S-). However, with sufficient exposure to the discrimination procedure, responding in the presence of the positive stimulus (S+) persists and responding in the presence of the negative stimulus (S-) declines. Thus, the effect of differential reinforcement, reinforcement in the presence of the S+ and the omission of reinforcement in the presence of the S-, is that subjects comes to respond differently in the presence of the S+ than in the presence of the S-.*3 For example, by virtue of their association with the motivationally significant outcome or reinforcer, over the course of conditioning the S+, signaling reinforcement, comes to elicit responses such as approach and contact while the S-, signaling the omission of that reinforcement, comes to elicit avoidance. In their simplest form, discrimination learning procedures, involving differential reinforcement, establish stimulus control when the S+ and S- (unique, discriminative elements) come to signal when a particular outcome will and will not be available better than other events (such as, contextual and common elements) present during conditioning. Once the S+ and S- have gained control over behavior, both predicting an important outcome, they are both called discriminative stimuli. Since they are both informative, S+ signaling reinforcement and S- signaling the omission of reinforcement, they are both relevant stimuli.*4
A more recent experiment demonstrates the difference between single stimulus conditioning and simple discrimination training between similar stimuli. In this experiment, designed to determine if exhaled breath can be used for early diagnosis of lung cancer, dogs were initially trained over trials to discriminate between blank unscented vapor sampling tubes, a tube that contained exhaled breath from a lung cancer patient (Ehmann et.al. 2012; see McCulloch, 2006 for the training procedure). After conditioning in which the dogs were required to discriminate between the presence and absence of exhaled breath, dogs were tested to see if they would reliably discriminate between tubes containing healthy breath, the tubes that contained exhaled breath from lung cancer patients. The results correspond with Pavlov’s (1927) early experimental findings in which dogs were trained one way, single stimulus conditioning, and tested another way, discrimination between similar stimuli. Ehmann et.al. (2012) found, dogs that had been previously trained to discriminate between the presence and absence of cancerous breath, initially responded at chance when tested to see if they would reliably discriminate between healthy and cancerous breath, but as discrimination testing continued, lung cancer detection improved. Over testing, as discrimination improved, the experimenters found dogs could not only reliably discriminate between healthy and cancerous breath, they could also reliably discriminate cancer from tobacco smoke, food odors, and potential drug metabolites. Although the experimenters attributed the gradual improvement over testing to a need for more single stimulus conditioning prior to testing, the gradual improvement over testing more appropriately reflects the difference between training and testing procedures. During initial training, the dogs had not learned from single stimulus conditioning the task requirement to discriminate between healthy and cancerous breath. Initial training, in which dogs were required to discriminate between the presence and absence of exhaled cancerous breath, was not specific. There was nothing in the arrangement of training to inform the dogs the task requirement to discriminate between similar test stimuli. However, when the dogs were subsequently tested using a simple discrimination between similar stimuli procedure, dogs could learn the task requirement to discriminate between healthy and cancerous breath and as a consequence, discrimination improved.
*3. Differential reinforcement involves different schedules of reinforcement, in either classical or instrumental conditioning. During differentially reinforcement trials, subjects are reinforced following a select stimulus (behavior, or both) according to one schedule of reinforcement and following a different stimulus (behavior, or both) a different schedule is implemented. For example, in the simplest case, differential reinforcement involves reinforcement after the presentation or occurrence of stimulus and/or behavior A, and the omission of that particular reinforcer after the occurrence of stimulus and/or behavior B. However, differential reinforcement does not necessarily have to involve reinforcement and the omission of a particular reinforcer. Discriminative stimuli may come to signal different probabilities or magnitudes of an outcome. For instance, presentation of the positive stimulus (S+) may be on a continuous reinforcement schedule, always followed with reinforcement, whereas presentation of the negative stimulus (S-) may be on a variable reinforcement schedule, only sometimes followed with reinforcement. Presentation of S+ may be followed with a large or highly motivational reinforcer, whereas S- may be followed with a smaller less significant reinforcer or a reinforcer of opposite motivational value.
*4. Relevant stimuli in a discrimination problem are stimuli that predict the occurrence or omission of reinforcement better than any other stimuli. Irrelevant stimuli, presented in conjunction with predictive relevant stimuli, are stimuli that are uncorrelated or not as well correlated with the delivery of reinforcement or its omission than relevant stimuli.
Simple discrimination training procedures
During simple discrimination training, the to-be-discriminated positive and negative stimuli may be presented either successively or simultaneously. When the positive and negative discriminative stimuli are presented successively or one at a time, the procedure is termed a successive discrimination procedure or successive go/no-go in order to distinguish it from successive conditional discrimination (see successive conditional discrimination procedure below, under Conditional Discrimination). During successive discrimination conditioning, response is reinforced in the presence of the positive stimulus and extinguished in the presence of the negative stimulus. Alternatively, when both the positive and negative stimuli are presented simultaneously or at the same time during a single trial, the procedure is termed a simultaneous discrimination procedure. Because the positive and negative stimuli are presented at the same time, simultaneous discrimination involves a choice between alternatives with response to only the positive discriminative stimulus being reinforced.
In their simplest form, both successive and simultaneous discrimination procedures involve exposing subjects to two or more to-be-discriminated stimuli that are presented in random order over trials. One stimulus is designated the positive stimulus (S+) and exposure or a response to it is always followed with reinforcement, whereas reinforcement is always omitted after exposure to the negative stimulus (S-). In other words, in simple discrimination training procedures, S+ is always positive and S– is always negative throughout training, they are not reversed from positive to negative and vice versa, as they are in either pseudo discrimination procedures or can be during conditional discrimination training (reviewed below). By always pairing S+ with reinforcement and S– with the omission of reinforcement, over the course of discrimination learning S+ comes to signal (predict) reinforcement and S– comes to signal the omission of that particular reinforce.
The reason S+ and S- are presented in a random or unpredictable order over trials is to prevent subjects from learning to predict reinforcement on the basis of some systematic order in which the to-be-discriminated stimuli are presented; such as, learning a presentation or position discrimination at the expense of learning the task to discriminate between the discriminative stimuli. The goal in simple discrimination procedures is for subjects to learn that both S+ and S- predict an important outcome — that S+ predicts a particular reinforcer and that S- predicts the omission of that particular reinforce — better than any other potential cues. Therefore, other potential cues must be controlled against by less reliably correlating them with reinforcement.
Simple successive and simultaneous discrimination may be programed during both classical and instrumental conditioning. The defining criterions of the two types of conditioning are the operational descriptions or rules for the contingent delivery of reinforcement. In classical conditioning, there is a relationship between a CS and reinforcer regardless of the subject’s behavior; whereas in instrumental conditioning the occurrence of reinforcement is contingent upon some aspect of the subject’s behavior.
During the early years following the rise of police canine tracking, investigation of tracking took place primarily around Germany. Although not necessarily meeting today’s standards of experimental investigation, initially German investigators, such as Belleville, (1938); Blunk, (1926); Böttger, (1926, 1932, 1936, 1937); Hansmann, (1929, 1931, 1949-51, 1955); Löhner, (1924, 1926a, 1926b, 1926c, 1927, 1931); Menzel, (1933); Menzel & Menzel, (1928a, 1928b, 1929a, 1929b, 1929c, 1930, 1932, 1935); Most, (1926, 1928, 1951, 1952, 1954); Most & Brückner, (1935, 1936); and Schmid, (1933), were responsible for much knowledge of canine tracking. Early investigators developed hypotheses of what tracking dogs must be capable of learning in order to respond successfully during active service. They developed training methods in an effort to achieve those goals, formulated tests, and then developed theoretical explanations on the basis of their findings. For example, from the assumption dogs can detect and discriminate between individual unique human scents during real world operations, it was hypothesized;
1) Dogs should be able to apply their discriminative capabilities, not only to their master or trainer, but also to strangers.
2) From the scent on an object or along a track, dogs should be capable of responding to the person from whom the scent was transmitted at the exclusion of all others.
3) As a control precaution, dogs should demonstrate the ability to use scent on an object or along a track to signal the correct choice even when all alternatives were from strangers and were of equal scent strength.
4) Conversely, dogs should be capable of taking scent from an individual and respond only to the corresponding scent on an object or a track.
5) They should be able to ignore nonmatching scents when the scent to be matched is absent.
Then on the basis of the early hypotheses, methods of training were developed, tests were devised to determine what was learned, data was acquired, and theoretical explanations were formulated, which in turn made predictions that could be used for future investigation.
Konrad Most began his investigation of tracking dogs in 1909, asking what odor information does track scent consist of and what information controls response performance. Similar to the term ‘stimulus’ as a term of convenience, composed of many elements, Most knew the term track was also a term of convenience, involving a variety of elements or stimulus information. He supposed track scent was composed of human species odor, personal or individual human odor that penetrated through footwear, odors from materials used in the tanning of shoe leather, shoe polish or any substance that may adhere to footwear, organic material such as scent from crushed soil and plants, subsidiary smells from insects or dung, volatile chemical molecules, bacteria, enzyme action or fermentation, moisture content, and varying scent strengths, quality from age, and meteorological conditions. Most (1954) reasoned that the track of a human is composed of various scents; some of which may undergo considerable and abrupt changes, while others remain constant. He was aware that under certain circumstances a dog may recognize distinctive scents while tracking and use that information to discriminate one track from others. Like modern animal learning investigators interested in how stimulus control is affected by the interaction between other stimuli and reinforcement, Most questioned what stimulus information controlled response performance and under what circumstances it came to do so.
In view of scent matching evidence from dogs trained by the Menzels and the assumption that dogs could learn to follow and discriminate between human scents along a track, Most conducted experiments to find evidence in support or against the individual unique human scent tracking theory, some of which involved attempts to train scent matching. After several years of investigation, he succeeded in training some dogs to discriminate between the track of their handler when decoy tracks of equal saliency were laid nearby or crossed their handler’s track. Most used the term track-fidelity for dogs that could reliably choose their handler’s track when discrimination was between the handler’s and decoy tracks of equal saliency. However, those dogs were not track-sure when they were subsequently tasked to discriminate between stranger laid target and decoy tracks. Although training trials originally began with the presentation of a scent sample, transfer trials from track-fidelity to discrimination between all stranger laid tracks showed dogs had not learned to use the scent sample as a cue to signal which track was correct and had not learned about the relationship between the scent sample and matching scent along the track.
Eventually, by randomly varying the saliency of target and decoy tracks over trials, Most succeeded in training dogs to reliably stay on a stranger laid track when decoy stranger laid tracks differing in saliency were laid nearby or crossed the target track, provided the dogs had enough time to become familiar with the specific saliency of the target track before encountering competing tracks. Most termed dogs that could reliably stay on the stranger laid track in which they were started when decoy tracks differed in saliency, track-sure.
It is important to understand the significance of the track-sure solution strategy. Dogs were not simply solving discrimination problems by learning to predict reinforcement from prior reinforcement history. For example, by learning to always choose the freshest track. Instead, provided dogs had enough time to become familiar with target track saliency, the track-sure solution involved comparing the saliency of the target track with the saliency of competing tracks and choosing according to the relationship between them; less-salient-than or more-salient-than the target track. However, although the solution is more complex than the fresher-than solution involving prior reinforcement history, the dogs used track saliency cues rather human scent to solve discrimination problems at choice points. Most never learned why dogs did not use human scent to solve discrimination problems. Although he conducted elaborate tests to determine whether response performance was controlled by human scent and found that it was not, he never knew whether it was because the dogs were incapable of discriminating between human scents along tracks or if dogs learned to ignore human scent due to the training arrangement.
Track-happy to discrimination between track saliency training
During his research, Most learned that although dogs may switch indiscriminately from one human laid track to another, it was not because they could not discriminate between tracks, but is more a case that dogs must be trained to discriminate between tracks. He argued (1954) that even on a track of its own master or trainer, a dog does not necessarily follow the personal or individual scent of the man he knows [at the exclusion of others] without first being specifically trained to develop this capacity; [without being trained the task requirement to discriminate between scent from its master’s track and others]. Consistent with Pavlov’s experimental observations (1927), Most found if dogs switch from one human laid track to another, it was not necessarily because they failed to recognize variations in scent. In other words, it was not because they were incapable of discriminating between different human laid tracks. According to Most, the mistaken opinion that dogs will follow the track of their master at the exclusion of all others arose from the commonly observed ease with which dogs recognize their own master by his individual body-scent. However, Most discovered that although track-fidelity on the track of the handler could be relied on after discrimination training between the handler’s track (S+) and decoy tracks of the same age (S-), no such reliance could be counted on from the dog’s own experience without discrimination training between the similar stimuli. Contrary to some modern day ill-advised arguments that dogs learn naturally without any scent discrimination training, Most also learned through investigation that training should involve scent discrimination between similar stimuli in order for dogs to learn the task requirement to discriminate between tracks.
However, Most thought training should proceed in stages, from the easiest to the more difficult, not vice versa. Also, on the basis of some of his findings, Most felt the most favorable tracking conditions are on ground overgrown with grass and vegetation, with moist air, moderate winds, and no sunshine. Unsuitable training conditions are on dry stone or asphalt pavements. Therefore, he insisted dogs should initially be trained only on their master’s/handler’s strongly scented fresh track in country as far as possible from other tracks, on only one grassy terrain with track scent blowing downwind along the track to reach the dog’s nose. As dogs gradually became more proficient at following their handler’s track, he thought cross tracks should be introduced, but initially care should be taken that cross tracks be at least half an hour older than the handler’s track, rather than fresher or the same age. As performance during this initial stage improved, Most found when the difference between the ages of the handler’s and decoy cross tracks were gradually brought closer together, dogs would ultimately discriminate the handler’s track from decoy tracks of the same age or fresher. He argued that unless the animal is given practice in not switching from the handler’s track to strangers’ tracks of the same age, it will never develop track-fidelity.
Thus, training initially involved single stimulus conditioning; discrimination between the presence of the handler’s track (the handler’s track plus the training context, S+) and the absence of the handler’s track (the context alone, S-), the product of which Most termed track-happy. In common with scientific evidence, this arrangement of training did not produce reliable discrimination between similar stimuli. Instead, in keeping with experimental evidence, discrimination between the presence and absence of a single track produced stimulus generalization. To the extent that an added stimulus excites some of the same representational elements as a conditioned stimulus (common elements), subjects can be expected to respond similarly (show stimulus generalization) and the less will variations between them bring about a change in behavior (show discrimination). In talking about the track-happy dog, Most (1954, p. 173) said they are guided by the most outstanding odors of the track, which as a rule is downtrodden vegetation and footprinted earth, not odors of human type and still less by individual human odors [the distinguishing elements of human scent that enable discrimination between human scents, such as genetic cues]. He said, “they change over from the initial track to others as fresh or fresher. These animals do not feel bound to follow any particular scent-component of the track. On the contrary, any of the usual scent-components of a human track will, as a rule, suffice to initiate pursuit. This will even occur when there is no characteristic or personal human scent.”(p. 160). In other words, dogs trained on single laid tracks did not discriminate between tracks; accuracy reverted to chance when competing tracks were subsequently introduced. However, it was not because they could not discriminate between tracks. Rather, it was because single stimulus conditioning is not specific. There was nothing in the training arrangement to inform dogs the task requirement to discriminate between tracks.
Although he probably did not know it, Most was talking about the robust phenomena of stimulus generalization, first discovered by Pavlov in Russia; the tendency to respond to one stimulus in the same way, but to a lesser extent to another similar stimulus. Because all of the track elements were paired with reinforcement during single stimulus conditioning, all of the track elements were capable of entering into association with reinforcement and thus, eliciting response. Due to the arrangement of training involving discrimination between the presence and absence of a track, as far as their correlation with reinforcement was concerned all of the track elements were relevant. Thus, to the extent that a subsequent decoy track had elements in common with controlling elements from the handler’s track, dogs may respond to it similarly. Furthermore, the greater the proportion of unchanged track elements controlling the dogs behavior, the less would discriminative elements between the handler’s and decoy tracks, such as differing track saliencies or individual unique odors from the track layers, bring about a change in behavior.
As training proceeded, the arrangement of training was changed from discrimination between the presence and absence of a single track, to discrimination between similar stimuli; discrimination between two or more tracks that differed, among other things, in age or saliency. This stage involved a compound discrimination between the handler’s/fresher track always positive (S+) and decoy/older tracks always negative (S-). Thus, due to differential reinforcement between similar stimuli, dogs could learn to use the distinctive elements from variations in track saliencies and the track layers to predict reinforcement during this new arrangement of training. That is, when training was subsequently arranged so discrimination was between similar stimuli, discriminative elements from the handler’s fresher track, were always paired with reinforcement (S+) and discriminative elements from the decoy’s older track, were always paired with the omission of reinforcement.*5 Whereas, elements common to both target and decoy tracks were correlated with both reinforcement and the omission of reinforcement over training trials and could not be used to reliably predict reinforcement. Therefore, common elements, which previously produced stimulus generalization, were now irrelevant and the unique distinguishing cues that could be used to discriminate between tracks were now relevant. Furthermore, with this new arrangement of training, dogs could not only learn the task requirement to discriminate between tracks, but with sufficient conditioning, attention and associability could be increased to the stimulus dimension that the dogs used to discriminate between tracks because variations in the dimension used for discrimination, such as differences in track saliencies or genetic cues from the track layers, signaled a change in reinforcement, both reinforcement and the omission of reinforcement (see e.g. Mackintosh, 1975).
However, the stimulus dimension that Most’s dogs used to solve discrimination problems were not genetic cues from human scent. Most observed his dogs were guided by the most outstanding odors of the track, which as a rule is downtrodden vegetation and footprinted earth, not odors of human type and still less by individual human odors. Either, his dogs were incapable of discriminating between variations in the track layer’s scents or downtrodden vegetation and footprinted earth overshadowed the human scent during soft surface tracking training. Recall, Most believed training should proceed in stages from the easiest to the more difficult and thus, insisted tracking training should begin only on ground overgrown with grass and vegetation. If elements from the track, such as vegetative odors, were more salient than the human scent component during initial training, the more salient elements could overshadow human scent and consequently, dogs could learn to ignore human scent. By now, there good evidence to suppose human scent was overshadowed by a more salient element during soft surface tracking training. Although it is now well established that dogs can reliably trail over hard surfaces and discriminate between competing human scents at choice points, Most found that no tracking could be undertaken on dry stone or asphalt surfaces in experiments he conducted with track-sure dogs who had been trained on soft vegetative surfaces and the same applied to dry sandy ground with no undergrowth. Once again, it appears the problem stemmed from the training arrangement, rather than the dogs.
Track saliency discrimination to track-fidelity training
As discriminative performance between tracks improved, the difference between the ages (saliencies) of the handler’s and decoy tracks was gradually faded out, until ultimately dogs would discriminate the handler’s track from decoy tracks of the same age. Thus, provided dogs were not using something like weight differences to discriminate between track saliencies, the dimension of track saliency (variations of which had initially enabled discrimination) was now irrelevant, whereas some other discriminative cue remained relevant. However, although it would be nice to think those dogs used individual unique human scent information, such as genetic cues, to discriminate between their handler’s track and other human laid tracks, barefoot tracking experiments reported by Romanes (1887), Budgett (1933), Böttger (1936), and Most (1954), in which dogs had difficulty following barefoot tracks, indicate that individual unique human scent was not the information those dogs learned to use when discriminating between their handlers track and decoys.
By changing the arrangement of training from single stimulus conditioning to simple discrimination between the handler’s track always positive (S+) and decoy tracks always negative (S-), which were subsequently both laid at the same time, dogs could not only learn the task requirement to discriminate between the handler’s track and decoy tracks, but also that track saliency was irrelevant. Once tracks were equal in saliency, track saliency could not be used to solve discrimination problems at choice points. To reliably stay on their handler’s track when decoy tracks of the same age were laid nearby or intersected the handler’s track, some other distinctive information enabled the dogs to discriminate between the handler’s track always positive and tracks laid by someone else always negative. Thus, due to the change in the training arrangement, Most succeeded in training track-fidelity.
However, although the change in training enabled dogs to learn track-fidelity (the task to discriminate between the handler’s and decoy tracks of equal saliency), transfer trials showed the solution learned only involved simple discrimination. Because tracks laid by the handler were reliably paired with reinforcement (S+) and decoy tracks of equal saliency were reliably paired with the omission of reinforcement (S-), to reliably solve track-fidelity discrimination problems dogs only needed learn from reinforcement history about the discriminative information that reliably predicted reinforcement. Dogs did not need to learn to use a scent sample or scent along the track as a cue to signal which alternative is correct at choice points. And transfer trials showed dogs only learned the simple discrimination solution. Once track-fidelity dogs reached criterion of reliable discrimination performance, subsequent chance performance during transfer trials involving stranger laid tracks of equal saliency showed dogs did not learn to use a cue to signal which discriminative stimulus is correct from one trial to the next.
Between 1927 and 1929, Berlin police dogs were tested under similar conditions as the 1913-1914 tests. Although most dogs failed, some dogs did succeed in discriminating between stranger laid tracks. This occurred when the dogs were given an opportunity to become familiar with the target track before encountering decoy cross tracks that differed in laying times. In order to familiarize the dogs with the target track, the track layer created a scent pad by trampling for three minutes, a half square meter surface at the start of the track where dogs were induced to sniff (Böttger, 1936). Later, in his final edition of Die Abrichtung des Hundes (1951), Most argued that an essential condition for success in practical tracking work is the presence of enough olfactory information at the start of the track to enable dogs to become familiar with the specific track scent left behind by the maker of the track. To that end, he recommended starting with a scent pad, followed with a target track free of cross tracks for a certain length in order for dogs to become familiar with the scent to be followed. Most also insisted training should proceed in stages, from the easiest to the more difficult, which involved training on ground overgrown with grass and vegetation, but not stone or asphalt. Thus, during the various initial stages of training, scent pads and tracks were on ground overgrown with grass and vegetation.
Arguing that persistence on an initial track is a primary objective of training, to familiarize dogs with the track to be followed, Most advised the track-layer should make a scent pad by walking to and fro at the start of the track in an area about half a square yard, without scraping up dirt. Then, upon leaving the start, the track layer should walk in very short strides for the first ten steps before resuming normal stride. Furthermore, Most maintained that the discriminative information produced from a series of foot steps can only be held by a track-sure dog if the start of the track runs for a certain length of time into the surrounding territory in isolation of other human laid tracks. As training proceeded, dogs were tasked to learn to track from a single foot print when only a small quantity of olfactory information was available at the start. However, if dogs were started on a single foot print, in order to pick up the quantity of olfactory information necessary to become familiar with the track, they needed a long stretch before encountering competing tracks, between thirty to sixty yards depending on the nature of the ground. With the addition of scent pads and an opportunity to become familiar with the target track before encountering competing tracks, Most and others succeeded in training track-sure dogs that would stay on the stranger laid track in which they were put and discriminate between other stranger laid tracks.
Yet, when all strangers laid tracks were laid at the same time, accuracy rate fell to chance. Thus, the findings indicate that track-sure dogs learned a relative saliency discrimination solution strategy, in which they became familiar with the saliency of the target track before encountering competing tracks that differed in saliency. In other words, track-sure dogs did not learn a relative familiarity discrimination solution strategy, in which they became familiar with the novel human scent from the target track layer and then solved discrimination problems at choice points by discriminating between more and less familiar novel human scents (see e.g. Premack, 1978). Nor did they learn a rule to choose between alternatives the individual unique human scent information that is the same-as the individual unique human scent information on the scent pad. If track-sure dogs had learned to use human scent to solve all stranger discrimination problems, as some have supposed, the dogs would have been able to solve equal saliency, all stranger discrimination problems. However, evidence indicates, the solution strategy track-sure dogs learned involved becoming familiar with the saliency of the track they were started on before encountering another track differing in saliency and then discriminating between the saliency of the initial track and subsequent tracks laid at different times. When dogs had enough time to become familiar with the scent strength of the track they were started on, they could discriminate between competing tracks that differed in scent strengths regardless of whether the decoys tracks were fresher or older. And thus, accurately respond provided stranger laid target and decoy tracks differed in laying times. With this solution, track-sure dogs could be used as investigative tools for much police work, but could not be used as detectives to identify suspects.
For Most, track-sure training began with ten to fifteen minute old single laid tracks that could be laid by the handler until the dog reached a stage in which the [track-happy] dog would reliably follow tracks laid people unknown to the dog. In other words, track-sure training began with track-happy training, involving discrimination between the presence and absence of a single track regardless of who laid the track. Who laid the track did not matter. As track-sure training continued, cross tracks were introduced. Initially they were not to be more than a half hour older than the target track. Then gradually the difference in age between newer target and older decoy tracks was decreased to a few minutes. Thus, in Most’s method, at this stage track-sure training involved simple discrimination between fresher tracks (S+) always paired with reinforcement and older decoy tracks (S-) always paired with the omission of reinforcement. Once dogs learned the simple discrimination, that reinforcement was contingent upon always choosing the freshest track, dogs were next required to disregard the freshest track solution strategy when the training arrangement was changed by unpredictably varying the saliency of target and decoy tracks over trials so that sometimes the decoy track was fresher than the target and other times it was older. With this new arrangement, the freshest track no longer reliably predicted reinforcement. However, according to Most, cross tracks of the same age were to be avoided as much as possible during training.
On the basis of evidence that Most acquired from experiments conducted with tracking dogs trained on soft vegetative surfaces, he did not advise track-sure training should involve discrimination between tracks of the same age. Most theorized track-sure dogs follow a specific scent given off by a track of a certain age. Although he said it was not certain which scent was decisive for track-sure dogs, he had no doubt variations of scent caused by time differences had something to do with it. Most reported time intervals between tracks as short as three minutes may secure persistence on a target track, but a ten minute difference as a rule did assure persistence.
Most was not the only investigator to report accuracy on stranger laid tracks that were solved by discriminating between track saliencies. Hansmann (1931), reported that success decreased when the age difference between stranger laid target and decoy tracks was less than three to five minutes. Provided there was sufficient age difference between track saleincy, Hansmann maintained well trained tracking dogs could hold fresh and old target tracks while ignoring old and fresh diversionary tracks. Böttger (1936, 1937) reported that well trained Berlin police dogs could discriminate stranger laid target tracks from stranger laid older and fresher diversionary tracks provided they were first allowed time to become familiar with the target track and the age differences between tracks was sufficient. Accuracy increased as the difference between the ages of tracks increased. He found the smallest age difference between tracks in which dogs could accurately respond was three minutes (see also Belleville, 1938 and Honhon, 1967). However, Böttger also reported that Berlin police dogs would not discriminate between strangers laid tracks of the same age and at the end of the track would not reliably choose the target track layer from among others when all choice alternatives were strangers to the dog.
*5. In Most’s arrangement response to a decoy track was not only paired with the omission of reinforcement, it was also, at some point, followed with compulsive inducements. Most advised at cross tracks, novice dogs should be given a chance to smell and compare the difference between tracks. Thus, a dogs initial response to a cross track did not indicate it made an error. Since correct response at cross tracks involved discrimination which necessitated comparison, the handler was advised to freely permit the smelling and testing of cross tracks. However, Most argued novice dogs also need be trained to decline strange tracks. Therefore, at some point, if they did not abandon the cross track on their own accord, Most instructed the handler must apply compulsive inducements.
Simple successive and simultaneous discrimination procedures can be made more complex by adding a conditional cue. When reinforcement is contingent upon use of a conditional cue to signal which discriminative stimulus is correct on any given trial, the procedure is termed a conditional discrimination procedure.
Matching-to-sample – MTS
A conditional discrimination procedure used to study memory and transfer of generalized rules in animals (concept learning) that can also be used to train human scent matching service canines is termed matching-to-sample (MTS). In the standard matching-to-sample procedure, subjects are first trained to make a particular response. During response training, a sample stimulus is not presented at the beginning of every trial. It is not until after response training that trials begin with the presentation of a single sample stimulus. Over MTS trials, presentation of the set of sample stimuli used for training are varied randomly or unpredictably. In other words, with the exception of correction trials (rerunning the previous trial without altering the sample stimulus or positions of the comparison stimuli after an incorrect response) , which are sometimes employed, the sample stimulus in a subsequent trial is randomly alternated with the set of training stimuli used, to control against learning some systematic or predictable order solution. After response to the sample stimulus, it is removed and two or more choice alternatives (comparison stimuli, meant to be compared with memory of the sample stimulus) are presented, one of which is the same as the sample. The subject’s task is to choose the comparison stimulus that matches the sample. The reinforcement contingency in the MTS procedure is, if the subject responds to the comparison that matches the sample the subject is rewarded, but if the subject chooses a nonmatching comparison the trial is terminated without reward. To ensure training involves conditional discrimination (use of the sample stimulus as a conditional cue to signal the correct discriminative stimulus on any given trial), simple discrimination solutions are controlled against. For instance, in the standard MTS procedure the comparison stimuli and the positions of the comparison stimuli relative to one another are also randomly or unpredictably varied from trial to trial. Over trials, the sample stimulus, the comparison stimuli, and the relative positions of the comparison stimuli are varied randomly or unpredictably to prevent subjects from learning to solve the problem on the basis of some systematic pattern that does not involve matching; for instance, leaning an order or presentation discrimination and thus, not learning about the informative significance of the sample stimulus (such as the matching relationship between the sample and matching comparison) or not learning to discriminate between the comparison stimuli.
Unlike simple successive and simultaneous discrimination procedures in which S+ remains positive over trials and S- remains negative over trials, during MTS training, unless novel stimuli are used in each trial, the set of stimuli used for training serve both as sample stimulus/comparison correct and comparison incorrect (sometimes matching and sometimes nonmatching) depending on which conditional cue (sample stimulus) is randomly presented on any given trial over training. For example, when sample stimulus A is presented, response to comparison stimulus A is correct (matching) and response to B is incorrect (nonmatching), but when sample stimulus B is presented, response to comparison stimulus B is correct and response to A is incorrect.
MTS is a scientific procedure; controls against alternative solutions (i.e. solutions other than using the conditional cue to signal the correct choice on any given trial) enable the random control MTS conditional discrimination procedure to meet scientific standards of objectivity.
Systematic pseudo matching-to-sample – pseudo-MTS
When trials begin with the presentation of a sample stimulus but simple discrimination solutions are not controlled against, the arrangement involves systematic pseudo-MTS. Although trials begin with the presentation of a sample stimulus and the subject’s task is choose between alternatives the comparison that matches the sample, systematic pseudo-MTS arrangements do not involve MTS because reinforcement is not conditional upon use of the sample stimulus as a cue to signal which discriminative stimulus is correct on any given trial. Systematic pseudo-MTS arrangements lack scientific rigor. Thus, rather than learning a more complex matching solution, subjects can learn the more readily acquired simple discrimination solutions not controlled against, at the exclusion of learning a matching solution. For example, when tracking training begins with the presentation of a scent sample and the simpler solution of discriminating between the presence and absence of a single track is not controlled against, the arrangement involves pseudo-MTS, during which the scent sample does not predict anything more than the easily acquired simple discrimination solution perfectly predicts. Furthermore, following training with pseudo-MTS single track arrangements, in which dogs achieve a high degree of accuracy, transfer trials involving presentation of a scent sample and two or more tracks show accuracy rate reverts to chance; dogs do not learn a matching solution from the pseudo-MTS arrangement. Simple discrimination between fresher S+ tracks and older S- tracks is another example. Discrimination between the owner/handler S+ (or who ever is systematically used as a track layer during training) and members of the public S- (strangers), is also an example of pseudo-MTS when trials begin with the presentation of a scent sample.
Because systematic pseudo-MTS arrangements do not control against more readily acquired alternative solutions, they do not meet scientific standards of objectivity. As such, in a critical review of the validity of experimental results from matching studies that did not adequately specify controls against alternative solutions, Lashley (1940) argued that “in view of the lack of any adequate control of extraneous cues and the failure of other investigators to obtain remotely similar results, the achievements of the monkey of Liége must be classed with those of the horses of Elberfeld and the dog of Mannheim”. Applying Morgan’s Canon, unless good evidence is provided that subjects have learned a more complex matching solution, it must be supposed that more readily acquired simpler solutions are responsible for reports of high accuracy.
A conditional discrimination procedure used to investigate complex learning in animals, which can also be used to train service canines same as and different from relationships, is the same/different (S/D) procedure. In this procedure the relational cues or concepts same as and different from serve as conditional cues. In the same/different procedure, ‘same’ and ‘different’ trials occur separately and are equally but randomly distributed over trials. During a given trial, subjects are simultaneously presented two comparison stimuli that are either the same or different from one another. If the to-be-compared stimuli are the same, subjects are required to make an arbitrarily designated ‘same’ response to obtain reward, such as respond to one of the matching stimuli. If they are different, a response to stimulus X, present on both ‘same’ and ‘different’ trials is required for reinforcement, which indicates they are different. Thus, in this procedure subjects can obtain reinforcement on both ‘same’ and ‘different’ trials. In addition, the stimuli to-be-compared are presented simultaneously, rather than successively as in MTS when a sample stimulus is first presented and then the comparison stimuli.
Successive conditional discrimination procedure
A similar procedural arrangement to the S/D procedure that has been used to investigate how animals solve discrimination tasks is the successive conditional discrimination procedure. In the successive conditional discrimination procedure subjects are exposed, usually in a T-maze or on a display screen, to two pairs of conditional stimuli on successive trials. For example, on some trials the pair of stimuli may both be black and on other trials the pair of stimuli may be white. Over trials, the two different pairs of stimuli are randomly presented. When both stimuli are black, a response to the left may be reinforced whereas a response to the right is reinforced when both stimuli are white. Thus, successful solution of the successive conditional discrimination problem requires response to be controlled simultaneously by two cues, here visual and spatial cues.
A procedural difference between successive go/no-go and successive conditional discrimination is that in simple successive go/no-go discrimination problems only one response is conditioned, and that response is either reinforced or not, depending on the stimulus presented on that trial. On go trials, responding is reinforced in the presence of S+ and on no-go trials responding is extinguished in the presence of S-. Thus, subjects can only obtain reinforcement on go trials. Whereas, in successive conditional discrimination problems, subjects are given a choice between two responses, one of which is reinforced in the presence of one stimulus pair and the other in the presence of another stimulus pair. Thus, provided they choose correctly subjects can obtain reinforcement on all successive conditional discrimination trials.
MTS go/no-go conditional discrimination procedure
An arguably inadequate procedure used to train and test service canines both human scent matching during matching (go) trials and indication of no matching human scent present during nonmatching (no-go) trials will be termed here MTS go/no-go conditional discrimination, to distinguish it from the matching/nonmatching relational learning procedure described next. In the MTS go/no-go procedure, matching and nonmatching trials are successively presented in random order. Matching trials are the same as or similar to MTS.*7 A trial begins with the presentation of a scent sample. After response to the sample, it is removed and dogs are next required to choose between alternatives the individual unique human scent that matches the individual unique human scent on the scent sample. If dogs choose correctly the matching odor, they are rewarded, but if they choose incorrectly a nonmatching alternative, reinforcement is omitted. Thus, matching (go) trials involve a reinforcement contingency; reinforcement is contingent upon response to the individual unique human scent matching alternative. Over trials, sample stimuli, choice alternatives, and the presentation positions of the alternative stimuli are also randomly varied.
On nonmatching (no-go) trials however, in which there is no matching individual unique human scent present, there is no reinforcement contingency; reinforcement is always omitted regardless of whether dogs respond correctly or not. Nonmatching trials are randomly interspersed with matching trails to control against learning a systematic pattern in which to suppress response. A nonmatching trial begins with the presentation of a scent sample. After response to the scent sample, it is removed and two or more nonmatching human scent alternatives are presented, typically in a lineup or on trails laid near one another. Part of the dog’s task is to suppress response on nonmatching no-go trials. When they correctly do so, reinforcement is omitted. On nonmatching no-go trials, reinforcement is omitted regardless of whether the dog correctly suppresses response or incorrectly responds to a nonmatching alternative. There is no reinforcement contingency on no-go trials to inform dogs the task requirement to suppress response. Furthermore, the responses used to measure learning on go and no-go trails are unequal; response is used to measure matching learning on go trials and suppression of response is used to measure nonmatching learning on no-go trials.
Although the MTS go/no-go conditional discrimination procedure involves a conditional cue (the scent sample), whereas the successive go/no-go procedure does not, both procedures are similar in that subjects can only obtain reinforcement on go trials. On no-go trials, reinforcement is always omitted regardless of response.
Matching/nonmatching relational learning procedure
In another procedure introduced here, termed the matching/nonmatching relational learning procedure, there is a reinforcement contingency on both matching and nonmatching trials and the responses used to measure matching and nonmatching are equal. In this procedure designed to train human scent matching/nonmatching service canines, matching and nonmatching trials are also successively presented in random order. Subjects are initially trained to make a particular response, just as in MTS. Following response training, subjects are first exposed to a sample stimulus. After response to the sample stimulus, two or more comparison stimuli, plus an additional X stimulus, are presented. The subject’s task on matching trials is to choose the comparison stimulus that matches the sample. The reinforcement contingency on matching trials is, if the subject responds to the comparison that matches the sample the subject is rewarded, but if the subject chooses either a nonmatching comparison or stimulus X the trial is terminated without reward. On nonmatching trials all comparisons are different from the sample stimulus. Similarly to matching trials, after response to the sample stimulus it is removed and two or more comparison stimuli are presented in addition to stimulus X. The subject’s task on nonmatching trials is to respond to stimulus X, which indicates all comparisons are different from the sample. The reinforcement contingency on nonmatching trials is, if the subject responds to stimulus X the subject is rewarded, but if the subject chooses a nonmatching comparison the trial is terminated without reward. In addition to random presentation of matching and nonmatching trials, the sample stimuli, comparison stimuli, and the presentation positions of the comparison stimuli relative to one another are also randomly varied from trial to trial.
The matching/nonmatching relational learning procedure has similarities to the same/different procedure. In the two procedures, there is a reinforcement contingency on both same as and different from trials in which subjects can earn reward. When the to-be-compared stimuli are the same, reinforcement is contingent upon response to a matching comparison. When the to-be-compared stimuli are all different, reinforcement is contingent upon response to stimulus X.
*7. Rather than applying a MTS procedure used in animal learning experiments, which has been experimentally tested, dog trainers tend to use improvised forms of MTS in the interest of training from the easiest to the more difficult, which do not take stimulus selection into consideration and actually may retard learning rather than enhance it (see Kaldenbach, 1998; Schoon, Haak, 2002 for examples of improvised MTS versions).
During training, it is the subject’s task to determine which stimuli most reliably signal the important outcomes relative to other stimuli present and assign the predictive value relevant to each. A task all mammals and birds are evolutionarily capable of, if they were not, discrimination would not be possible. Likewise, it is the trainer’s task to arrange training so the to-be-conditioned stimuli are the most reliable predictors of important outcomes relative to other potential signals for those outcomes. Discrimination training requires the arrangement of a cause and effect relationship between the to-be-conditioned event and reinforcement. When discrimination is between similar stimuli, trainers must distinguish from the multitude of stimulus information present during training the stimulus information that must exert control over behavior in order to perform the task accurately and reliably. After specifying the stimulus information important to the task (such as genetic human scent information), in order to increase attention and associability to that information at the expense of other stimuli, trainers must arrange training so that information is the most reliable predictor of reinforcement (both reinforcement and the omission of reinforcement with regard to genetic information) relative to other stimuli present during training. However, in order to arrange training so the to-be-discriminated stimulus information is the most reliable predictor of reinforcement, trainers should know the circumstances under which animal come to select the most reliable predictors of important outcomes at the expense other stimuli present in its compound.
Numerous animal discrimination experiments, such as relative validity, overshadowing, blocking, presolution reversal, overtraining reversal, latent inhibition, and learned irrelevance have shown animals are capable of selectively controlling the attention they pay to a particular stimulus or stimulus dimension and not others. Causally speaking, relative validity, overshadowing, and blocking experiments reveal that animals learn to attend to stimuli that are more predictive of an important outcome at the expense of the less valid predictors of that particular outcome.
During the reign of behaviorism, most classical and instrumental experimental procedures involved single stimulus or non-specific conditioning. Typically, a single stimulus — a conditional stimulus (CS) or ess-dee (SD) — was paired with reinforcement and, presumably, there was no discrimination training involved; no differential reinforcement. However, all conditioning procedures, involving a reinforcement contingency specifying when reinforcement will and will not be available (which also includes naturally occurring cause and effect events), involve differential reinforcement or discrimination training. Since it is impossible to present a single stimulus in the absence of other stimuli, the subject is arguably exposed to two sets of stimuli during single stimulus conditioning, one comprising the to-be-conditioned stimulus plus the contextual stimuli present during training (AX) and the other comprising the contextual stimuli without the to-be-conditioned stimulus (X alone) that are present during inter trial intervals (ITI — the time between trials).*8 With sufficient exposure to the two sets of stimuli and their different outcomes, the subject comes to respond differently in the presence of one set of stimuli than in the presence of the other set of stimuli. Therefore, it is reasonable to argue that both classical and instrumental conditioning procedures can also be regarded as discrimination training procedures.
One problem of single stimulus conditioning or discrimination between the presence and absence of a stimulus is to explain why with sufficient exposure to the two sets of stimuli paired with different outcomes, subjects come to respond differently in the presence of one set of stimuli (AX) than in the presence of the other set of stimuli (X). By explicitly arranging stimulus compounds, researchers have been able to investigate this question. Consider a discrimination experiment conducted by Jenkins and Harrison (1960) involving differential reinforcement , in which S+ involved a tone-light compound and S- was light alone. In this experiment Jenkins and Harrison first trained pigeons, in the presence of a tone, to peck a response key that was illuminated by a white light. Once responding in the presence of the tone-light (TL) was established, they randomly alternated reinforced trials in which both the tone was sounded and key illuminated (TL+) with unreinforced trials in which the key alone was illuminated without the tone (L-) so that ultimately there was an equal number of reinforced and unreinforced trials over conditioning. Light was reinforced on 50 per cent of the trials and not reinforced on 50 per cent of the trials. Successful discrimination required responding on all trials in which the tone was sounded and refraining from response in the absence of the tone. All subjects easily learned the discrimination between the presence and absence of the tone.
The question is, if light was reinforced on 50 per cent of the trials, why didn’t subjects continue to respond to it on light alone trials? What is the means by which single stimulus conditioning comes to suppress control of contextual stimuli? For behaviorists the answer was simple. Behaviorist conditioning-extinction theory or excitatory-inhibitory conditioning theory (Hull, 1952 and Spence, 1956) accounts for how associative changes are translated into performance by assuming that the probability of a response occurring at a given moment is determined by the net excitatory values of all stimuli present at that moment. Thus, the theory predicts that equal numbers of reinforced and unreinforced trials would produce equal amounts of excitation and inhibition, leaving the light with a net excitatory value of zero; association to light would be neutralized. On the face of it, this explanation sounds so plausible that it hardly seems to require further consideration. However, by the 1950’s it was already well established that a 50 per cent reinforcement schedule did not result in a net excitatory value of zero. For example, a random 50 per cent reinforcement schedule often, but not always, yielded and maintained high levels of response rather than low levels or none at all. Therefore, conditioning-extinction theory cannot be correct in explaining suppression of contextual control; suppression of incidental or irrelevant stimuli.
Eventually, Wagner, Logan, Haberlandt, and Price (1968) found the cause of the conflicting results. Their experiment showed that a 50 per cent reinforcement schedule can produce and maintain high levels of responding only when there is no better predictor of reinforcement. In the Jenkins and Harrison experiment, tone was a better predictor of reinforcement than light. Therefore, the answer to the question of why subjects did not continue to respond on light alone trials in the Jenkins and Harrison experiment must be that although subjects would continue to respond if there was no better signal of reinforcement, the presence of the tone changed the informative significance of the light during conditioning. Recall, the tone was always correlated with reinforcement, whereas the light was correlated with both reinforcement and the omission of reinforcement. Relatively, the tone was a better predictor of reinforcement than the light. The better correlation of the tone with reinforcement prevented the light from gaining the predictive or associative value that it would have done without the tone. Subjects selectively associated the tone with reinforcement at the expense of the light. In the Jenkins and Harrison experiment, subjects could learn over trials that the light was an irrelevant stimulus because presentation of the tone was always followed with reinforcement whereas presentation of the light was only some of the time followed with reinforcement.
Wagner et al. (1968) provided clear evidence that excitatory-inhibitory conditioning theory is incorrect, which assumes that incidental stimuli gain and lose associative value and that the presence of stimuli better correlated with reinforcement should have no effect on this process. Wagner et al. showed that animals selectively associate events better correlated with reinforcement at the expense of events that are less well correlated with that reinforcer.
Before reviewing the typical relative validity experiment, it is necessary to first describe a pseudo-discrimination procedure. In this pseudo-discrimination procedure, in compound with a common stimulus (X), two discriminable stimuli of the same dimension are both equally paired with reinforcement and the omission of reinforcement (A1X and A2X). For example, the dimension might be color where A1 represents red, A2 represents blue, and X represents a triangle. The compound A1X (red triangle) is paired with reinforcement in 50 per cent of the trials within a training session and the omission of reinforcement in 50 per cent of the trials within the same session, and so is the compound A2X (blue triangle). Although a 50 per cent reinforcement schedule, such as this, slows down the course of conditioning, it is quite sufficient to yield high levels of conditioning provided there is no better predictor of reinforcement and the omission of reinforcement. In other words, in pseudo-discrimination procedures, provided A1, A2, and X are all equally salient and there has been no prior learning about any of them, any of the events can become conditioned or associated with the reinforcer. A1 and A2 are no better correlated with the outcome of each trial than X; A1, A2, and X are all valid predictors of reinforcement and the omission of reinforcement.
However, in true discrimination training procedures, there is a better or more valid predictor of reinforcement and a more valid predictor of the omission of reinforcement relative to X. A1+ is presented in compound with X on all reinforced trials (A1X+) and A2– is presented in compound with X on all unreinforced trials (A2X-). Thus, the outcome of each trial is perfectly predicted by the nature of A occurring on that trial, which significantly reduces any control by X that is present on both reinforced and unreinforced trials.
For example, A1+ might be the discriminative or individual unique information of Fred’s scent, A2– might be Mike’s individual unique information, and X might not only be the training context but also the information common to both Fred and Mike’s scent. In true discrimination training, Fred’s discriminative information (S+) would be a more valid predictor of reinforcement relative to the training context and common elements and Mike’s discriminative information (S-) would be a more valid predictor of the omission of reinforcement relative to the training context and common elements; whereas in pseudo-discrimination, Fred’s discriminative information, Mike’s discriminative information, the training context, and odors common to both Fred and Mike are all valid predictors of reinforcement and the omission of reinforcement. None of them predict reinforcement or the omission or reinforcement any better than the other.
In a typical relative validity experiment, one group of subjects is conditioned using a true discrimination procedure and the other group is conditioned using a pseudo-discrimination procedure. Both groups receive the same number of reinforced and unreinforced trials. Thus, X is equally paired with reinforcement and the omission of reinforcement in both the true discrimination and pseudo-discrimination groups. After conditioning, both of the groups are tested with X in isolation of A1 and A2. The tests show that stimulus control by X in the true discrimination group is much lower than the pseudo-discrimination group, despite the fact that X is equally paired with reinforcement and the omission of reinforcement in both the true discrimination and pseudo-discrimination groups. Relative validity experiments show that in the true discrimination group, when the outcome of each trial is perfectly predicted by the nature of A occurring on that trial, conditioning to X is almost completely eliminated; whereas in the pseudo-discrimination group, where X predicts the outcome of each trial as well as A1 and A2, it is readily associated with reinforcement and produces reliable conditioning. Thus, relative validity experiments reveal that the degree of conditioning to X is significantly affected by the relative validities of A and X. In other words, it is not the absolute validity so much as it is the relative validity of an event signaling reinforcement that produces conditioning. The strength of conditioning to a particular to-be-conditioned stimulus depends not only on its relationship with reinforcement, but also on whether that reinforcer is signaled by other events, stimulus or behavioral events, and their correlation with reinforcement.
Because it is the relative validity of a stimulus signaling reinforcement that produces conditioning, an irrelevant stimulus or set of stimuli need not be paired with equal numbers of reinforced and unreinforced trails for subjects to learn that they are irrelevant. Provided subjects can detect the to-be-conditioned stimulus, successful conditioning requires that it be better correlated with the reinforcer than other events. Over trials, subjects selectively associate stimuli better correlated with reinforcement at the expense of irrelevant stimuli less well correlated with reinforcement. However, initially a stimulus or a set of stimuli are not irrelevant as far as the subject knows until the subject learns through differential reinforcement that they are less well correlated with reinforcement than the to-be-conditioned stimulus.
Likewise, if response to and discrimination between individual unique human scents is your goal, successful conditioning requires individual unique human scent information be correlated with reinforcement (S+) and its omission (S-) better than other cues present in its compound during training, such as gender, race, smoking, cosmetics, diet, medication, or disease; all of which can be explicitly arranged to be less well correlated with reinforcement and its omission than individual unique human scent information, which can not only enhance learning to discriminate between individual unique human scent information, but increase attention and associability to that particular bit of information.
*8. The context is the array of all uncontrolled stimuli that comprise the situation in which training is conducted.
When two or more stimuli are presented in compound and one is more salient than the other, the more salient stimulus will overshadow leaning about the less salient stimulus. Pavlov (1927) was the first to report that after compound conditioning with a more salient stimulus and a less noticeable stimulus, dogs would respond to the more salient stimulus and very little or not at all to the less noticeable stimulus when the stimuli were tested in isolation of one another.
Similarly, Jenkins and Harrison (1960) found that if they simply required pigeons to peck the key-light in the presence of the tone (LT), without unreinforced trials involving just the light and no tone (L), control acquired by the key-light completely overshadowed any control by the tone. Successful conditioning to the tone required either that the tone be better correlated with reinforcement or that there be no key-light to overshadow conditioning to the tone.
In overshadowing experiments, a group of experimental subjects is conditioned to respond to a compound of two stimuli, Ab, that are presented simultaneously. One stimulus, A, is more salient or more noticeable than the other, b. When the response to the Ab compound has been fully conditioned, the strength of response to each of the components is measured by presenting each stimulus alone and recording the response. The results typically show that the more salient stimulus, A, elicits a strong response, whereas the less salient stimulus, b, elicits only a weak response or none at all. As a control condition, a separate group of subjects (called a control group) is conditioned to only the less salient of the two stimuli. When the control group is subsequently tested and the results are compared to the experimental group, the response conditioned to the weaker stimulus b, in isolation of the stronger stimulus A, is substantially stronger for subjects in the control group than the experimental group that received compound Ab conditioning. Thus, overshadowing experiments show that a stronger stimulus will overshadow conditioning to an otherwise conditionable weaker stimulus when the two stimuli are conditioned in compound, but response to the weaker stimulus can be conditioned when the weaker stimulus is conditioned in isolation of the more salient stimulus. Overshadowing experiments reveal that a stimulus that is otherwise sufficient to establish an association with a reinforcer will fail to do so unless that stimulus is also better correlated with the reinforcer than other stimuli present in compound.
Behaviorist conditioning-extinction theory or excitatory-inhibitory theory (Hull, 1952 and Spence, 1956) accounts for how associative changes are translated into performance by assuming that the probability of a response occurring at a given moment is determined by the net excitatory values of all stimuli present at that moment. Thus, the theory predicts that equal numbers of reinforced and unreinforced trials would produce equal amounts of excitation and inhibition, leaving the light with a net excitatory value of zero; association to light would be neutralized.
While reviewing relative validity it was argued that behaviorists thought animals learn simultaneously and automatically about every stimulus or stimulus dimension present during conditioning that is detectable to them. Traditionally theorists thought all stimuli, including their elements, present when a response was reinforced acquired excitatory potential and all stimuli present when a response was not reinforced acquired inhibitory potential.
According to conditioning-extinction theory, behaviorists assumed the set of stimuli present when a response was reinforced came to elicit (excite) that response, whereas the set of stimuli present when a response was not reinforced came to inhibit that response. Additionally, those stimuli present during equally reinforced and unreinforced trials acquired both excitatory and inhibitory potential, which resulted in equal excitatory and inhibitory potentials that neutralized one another. Thus, they neither excited nor inhibited the response over the course of conditioning. According to conditioning-extinction theory, for a particular to-be-conditioned stimulus to gain stimulus control over behavior, all of the other stimuli present when a response was reinforced would, at some point, have to also acquire inhibitory potential so that ultimately the particular to-be-conditioned stimulus would have more excitatory potential than the other stimuli. However, overshadowing experiments show that conditioning is not an automatic consequence of pairing stimuli with reinforcement and the omission of reinforcement. In overshadowing experiments, both the salient and less salient stimuli are presented simultaneously or equally often in compound. Thus, responses following their presentation receive equal amounts of reinforcement. Conditioning-extinction theory predicts that both the salient and less salient stimuli should acquire equal amounts of excitatory potential, but overshadowing experiments reveal that they do not. The more salient stimulus (A) elicits a strong response whereas the less salient stimulus (b) elicits only a weak response or none at all. Overshadowing experiments show that the presence of a more salient or valid predictor of a particular reinforcer will interfere with conditioning of a stimulus that is otherwise sufficient to establish an association with that reinforcer.
Moreover, some overshadowing experiments have shown that during compound conditioning with a more and less salient stimulus, subjects selectively learn to attend to the more salient stimulus and ignore the less salient stimulus, which in turn retards subsequent conditioning to the less salient stimulus (e.g. Seraganian 1979). They reveal that animals actively process stimulus information and choose to attend to the more salient stimuli at the expense of the less salient stimuli. In Seraganian’s experiments, there were three groups of subjects; a compound group in which more salient color and less salient line orientation discriminative stimuli were presented simultaneously during stage one, a less salient alone group in which the less salient line orientation discriminative stimuli were presented in isolation of color during stage one, and a more salient alone group that was conditioned with the more salient color discriminative stimuli alone during stage one. In stage two, all subjects were tested with the less salient discriminative stimuli presented in isolation of the more salient stimuli. The tests showed that when the more salient discriminative stimuli were removed in the compound group, discriminative performance immediately reverted to chance even though both the more and less salient discriminative stimuli were equally relevant to the discrimination solution during compound conditioning. Furthermore, when discrimination training with the lines alone was continued in all groups (30o off vertical line positive (S+) and a 60o line off vertical negative (S-)), the compound group not only continued to perform less accurately than the less salient alone group, they also learned significantly more slowly than both the less salient alone and the more salient alone groups. Even though both color and line orientation were relevant to the discriminative solution during compound conditioning, acquisition of control by the lines was significantly retarded if they had been overshadowed by the presence of the more salient color stimuli.
These experiments reveal that subjects not only notice B, they also notice that it is a less reliable predictor of reinforcement. Retarded learning in the experimental Ab group, compared to the b alone control group, indicates subjects do not simply fail to notice or attend to the less salient stimulus, but that they do notice it and specifically learn to ignore the less salient stimulus. Otherwise, learning should not be retarded compared to the control group when the experimental group is subsequently conditioned to the less salient stimulus alone. In order to learn to ignore the less salient stimulus, subjects must initially attend to both stimuli and then select the more salient stimulus in favor of the less salient stimulus. This contradicts conditioning-extinction theory, which assumes learning is an automatic process of acquiring excitatory and inhibitory potential through reinforcement and the omission of reinforcement and that all stimuli present when a response is reinforced acquires excitatory potential and all stimuli present when a response is not reinforced acquires inhibitory potential. Instead, these experiments reveal that the associability of a stimulus with a reinforcer declines as a function of experience with that stimulus in which it is less informative than some other event at predicting reinforcement. Not only have comparative studies shown that overshadowing is a very robust phenomenon, they have also shown that animals selectively learn to attend to some stimuli and ignore others by virtue of their relative informative value. That is, animals actively process stimulus information.
In standard overshadowing experiments to determine what stimulus information comes to control response during conditioning, there are two groups of subjects. One group is conditioned with a compound of more and less salient stimuli presented simultaneously. The other group is conditioned with the less salient stimulus presented in isolation of the more salient stimulus. After conditioning, both groups are tested with the less salient stimulus alone and strength of response in the two groups is compared. The results typically show that response to the less salient stimulus is significantly stronger in the less salient stimulus alone group compared to the compound group. Although a more salient stimulus will overshadow conditioning to an otherwise conditionable weaker stimulus when they are conditioned in compound, response to the weaker stimulus can be conditioned when the weaker stimulus is conditioned in isolation of the stronger stimulus.
Alternatively, early tracking experiments to determine what odor information controlled response during tracking involved only one group of subjects; dogs trained to track on a compound of more and less salient stimuli. Most and others thought training should proceed in stages from the easiest to the more difficult. From working experiments with track-sure dogs, Most found the easiest tracks to follow were on ground overgrown with grass and vegetation when the air was moist, winds were moderate, and there was shade or no sunshine; whereas, no tracking could be undertaken on dry stone, asphalt, or sandy ground with or without undergrowth. Thus, Most advised dry flat sandy places without undergrowth, as well as stone or asphalt pavements, were unsuitable for tracking. Böttger (1936) also reported that highly skilled Berlin police tracking dogs would not track on stone, asphalt, or hot dry sand free of vegetation. Based on the evidence available, all test subjects were initially trained to track on a compound of stimuli, such as crushed vegetation, disturbed soil, and human scents. Early tracking experiments did not involve another group of test subjects that received human scent discrimination training in isolation of other track elements. Therefore, although it may appear dogs cannot detect and discriminate between human scents while tracking, we cannot know from early experiments alone because there was only one group of subjects tested; dogs who’s training involved compound conditioning with more and less salient stimuli.
Most conducted a number of interesting experiments with trained tracking dogs. In some of his experiments a cable was strung above the ground from one point to another on which an aerial transport apparatus was hung. During the experiments, a track layer walked a certain distance from point A to point B where he next sat on the floating apparatus and was transported along the length of the cable to point C, all the while with his feet above the ground about thirty centimeters. In all test trials conducted, trained tracking dogs tracked only from point A to point B, and did not follow a human scent trail to point C. Even when track layers waved their arms and legs while riding the aerial transport apparatus, dogs did not respond to a human scent trail between points B and C.
Apparently without knowing about Most’s experiments in Germany, Budgett (1933), in England, conducted similar aerial transport experiments, first with two of his bloodhounds and later with Colonel Roberts’s field trial winner, St Christopher of Trethill. After walking about a quarter mile from point A to point B, the track layer sat on a floating apparatus and was transported to point C with his feet about a foot off the ground, which took about ten minutes. Upon arriving at point C, the track layer got off the apparatus and walked to the far corner of the field in which the experiments were conducted. During testing, the two bloodhounds followed the track from point A to point B, arriving at point B fifteen minutes after the track layer got off the apparatus at point C. Although cast and recast several times between points B and C, the bloodhounds did not follow the line or indicate the direction in which the track layer moved along the cable. However, when the hounds were put on the track at point C, they tracked the line to the corner of the field. Budgett repeated the experiment with the bloodhounds on several occasions, with varying conditions of wind, temperature, and start times from fifteen minutes to five hours, but the hounds never succeeded following the track layer’s scent trail after he left the ground at point B.
During an extension of the aerial transport experiments, Most found trained tracking dogs would follow tracks laid with porcelain feet on a rolling wheel without any human scent. In these experiments, first wooden and later porcelain, feet were attached to a wooden wheel that hung from another cable strung from point B to a location away from points A, B and C. When a track layer walked from point A to point B and was then transported on the suspension apparatus to point C, while at the same time the wheel laid a different track away from point B, dogs subsequently tracked from point A to point B and then followed the human scent free track made by the wheel rather than following a human scent trail in the absence of a track to point C. On the basis of the finding from the two types of aerial transport experiments, Most supposed human odor originating above ground from an individual does not leave behind enough odor that dogs can detect and follow.
Similarly, Budgett (1933) conducted an extension of his aerial transport experiments, which involved elaborate precautions to ensure human scent could not be used to solve the second part of the tracking problem. In these experiments, rather than using rolling wheel with porcelain feet, Budgett arranged to have a ten pound glass bottle, filled with lead, pulled from point B some distance over grass in the absence of any human laid tracks or scent. During the experiments a track layer walked from point A to B and was aerial transported to point C where he then walked to the corner of the field. Meanwhile, the ten pound glass bottle was pulled from point B over grass away from points A, B, and C. Budgett found that upon arriving at point B from point A, his old hound trained for police tracking would follow the line of the clean ten pound weight over grass rather than follow a fresh human scent trail from point B to C. He reported that the hound appeared to hunt the weighted drag, in the absence of human scent, with greater readiness than displayed when following the track layer’s track. Prior to the aerial transport tests, Budgett’s wife suggested that the weight of the track layer crushed the grass he walked on, causing it to release an odor that was followed by hounds. Because the old hound followed the weighted drag with greater readiness than when following foot tracks, Budgett thought it clear evidence that a continuous line of scent was released from the grass as the weight was dragged over it, to which the hound could follow with greater ease than footsteps that left a gap of about a yard between them.
Supposing the weight of the track layer crushed the grass a track was laid on, causing it to give out an odor that was followed by the hounds, Budgett tested his assumption by drawing the weighted bottle across a ploughed field in the absence of vegetation. Here again elaborate precautions were taken to insure human scent could not be used to solve the weighted bottle problem. In this experiment a track layer walked from point A across a grass field to point B at the edge of a ploughed field where he climbed a tree. From the base of the tree, a weighted bottle was dragged across the ploughed field. While being careful that the wind was exactly right so the bloodhound would not detect the track layer in the tree, the hound was started at point A and tracked to point B where after some searching, found the weighted line, followed it with reduced enthusiasm into the ploughed field for a short distance, lost the line, and finally went back to the tree to search some more. Alternatively, when the experiment was repeated in another ploughed field that had been sown the previous month with a crop of winter oats, the hound followed the weighted line over the crop that was beginning to show, but with less assurance than on grass.
Body scent and skin raft theory
Body scent theory, popular when large scale police tracking was in its infancy, supposes a plume of volatile organic compounds emanates from individuals as they walk, leaving behind a scent trail that dogs can detect and discriminate between. On a still day [when the barometric pressure is high], the plume is presumed to remain suspended in the air along the path a person walks. However, the plume of volatile organic compounds is dispersed [when the barometric pressure is low or] when it is carried downwind on a breezy day. Furthermore, when an individual is stationary the human scent plume can be blown downwind to be detected by dogs. In conjunction with volatile organic compounds that float in the air, it was also supposed during the time of Most and others that heavier odorous particles fell to the ground as someone walked. In a later theory, Clifford (1958) reported that human skin is composed of about two billion cells. In a twenty four hour period, the human body sheds about 14.35 grams of dead skin cells. And, on average, forty thousand skin cells are shed every minute. He argued dogs can detect and discriminate between the dead skin cells, termed rafts, which shed and fall to the ground as we walk (see also Syrotuck, 1972).
Alternatively, Budgett (1933) supposed bloodhounds discriminate between individual human scent, but that individual human scent along a track came solely from physical touch or items that had been touched by a particular individual, such as minute odorous particles left behind from a track layers boots that had become impregnated with his scent. To test his assumption, Budgett conducted yet another set of experiments to see if bloodhounds could discriminate between human scent trails left behind without the trail layers coming in contact with the ground. In these experiments, it was first established that both Budgett’s experienced bloodhound and Colonel Roberts’ bloodhound, St. Christopher, would discriminate between human laid foot tracks and follow a horseback track. Following the preliminaries, a track layer walked some distance to a group of horseback riders and mounted an extra horse that one member of the group was holding. The mounted group then proceeded together for about 100 yards before they departed, riding off in different directions to a location where the hounds could not respond by sight or air-scent. In all test trials, the bloodhounds followed the wrong trail. On the basis of all his experimental results, Budgett argued they left no doubt “body scent” theory is a myth.
Both body scent and skin raft theories predict dogs/bloodhounds can detect and discriminate between human scent trails left behind when track layers are transported above ground from one location to another. Yet, Most and Budgett’s experimental results seem to indicate the theories are wrong.In their experiments, dogs did not respond to and discriminate between human scent trails left behind when track layers were transported above ground. However, absence of response does not prove absence of human scent or absence of an ability to detect and discriminate between aerial scent trails, as Budgett and others thought. A possible explanation as to why the dogs tested failed to respond to an aerial human scent trail is that during tracking training, conditioning to human scent was overshadowed by some other element present in its compound. Based on the evidence available, the possibility of a more salient element of track scent overshadowing learning about human scent was not controlled against during the training of any of the dogs tested. Nor was the possibility of overshadowing investigated.
Although Budgett used horses in some of his experiments, he challenged the same results could be obtained if bicycles were used instead of horses. Therefore, I tested one of my hounds on several occasions during hard surface mantrailing training on bicycle trails laid on concrete and asphalt surfaces in populated city environments.*9 Much of the hound’s training involved all stranger scent matching mantrailing in city environments in order to control against the hound learning undesired alternative solutions. All test trails were run blind as to where the trail layer went. In all tests, the hound readily followed the target bicycle trails and reliably discriminated between the target trail and other bicycle, car, and human laid foot trails of the same age, older, or fresher.
Although Budgett did not describe how the bloodhounds in his experiments had been trained, trainers commonly subscribe to the idea of training in stages from the easiest to the more difficult and consequently train on soft vegetative surfaces that can overshadow learning about human scent. Thus, the discrepancy between test results can be explained by supposing during soft surface tracking training of the bloodhounds used in Budgett’s experiments, conditioning to individual human scent was overshadowed by some other element present in compound with human scent.
*9. Trailing and mantrailing are more specific terms than tracking, used under the assumption dogs are following and discriminating between human scents. Some measures of human scent trailing are following the fringes of the actual line someone walked, following and discriminating between lines of human travel on hard surfaces, bicycles, or sometimes in cars.
Compound conditioning group compared to human scent alone group
Both Most and Böttger reported trained police tracking dogs would not follow a track on dry stone or asphalt. However, in their experiments, as well as Budgett’s, there were no reports of dogs that received human scent discrimination training in isolation of other track elements prior to testing. In overshadowing experiments there are two (or more) groups of subjects. One group is conditioned to respond to a compound of more and less salient stimuli presented simultaneously, while the other group is conditioned to respond to the less salient stimulus presented in isolation of the more salient stimulus. After conditioning, both groups are tested with the less salient stimulus alone and strength of response in the two groups is compared. The results typically show response to the less silent stimulus is significantly stronger in the group that received conditioning to the less salient stimulus in isolation of the more salient stimulus. Overshadowing experiments reveal successful conditioning to a less salient stimulus requires either there be no stronger stimulus present to overshadow conditioning to the less salient stimulus or that the less salient stimulus be better correlated with reinforcement.
On the basis of Mackintosh’s attentional theory, four mixed breed puppies were trained to discriminate between two human scented articles; one article handled by one person and the other handled by someone else. All puppies were four months old at the start of human scent discrimination training. Using a simple simultaneous classical discrimination procedure, the human scented articles, each placed in an apothecary jar, were simultaneously presented in random order over trials, with a constraint that S+ should not be presented in the same position relative to S- for more than three consecutive trials. Scent from one individual, designated the positive stimulus (S+), was always paired with reinforcement, while scent from the other individual, designated the negative stimulus (S-), was always paired with the omission of reinforcement. After simultaneous presentation of S+ and S-, a nose-poke to S+ was always follow with a click from a clicker and a food reward, a small piece of liver. Alternatively, if a puppy nose-poked S-, reinforcement was omitted until the puppy switched sides and nose-poked S+. Since in classical conditioning there is a relationship between a CS+ and reinforcement regardless of the subject’s behavior, the procedure was an errorless discrimination procedure, which was thought to be better for puppies. Each click and food reward marked the end of a trial. There were on average, twenty trials in every daily training session. Once the puppies were reliably responding to the positive stimulus, which was measured by persistent nose-poking to S+ when reinforcement was momentarily withheld, they were overtrained an additional 300 trials to increase attention and associability to the individual unique discriminative component of human scent and prepare them for transfer of the discrimination to a new task. Thus, simple human scent discrimination training took about one month.
After simple simultaneous human scent discrimination training to increase attention and associability to the individual unique features of human scent, the puppies were required to discriminate between trails laid on concrete and asphalt surfaces to see if the discrimination would transfer to a new task. Thus, the people who supplied human scent during stage one also laid trails during stage two without changing their S+ and S- designation. It was supposed if the discrimination would transfer to hard surface trailing, it would provide evidence that during soft surface tracking training a more salient track element overshadows learning about human scent. However, it should be noted that the response rate while tracking on soft surfaces is typically faster than when working on hard surfaces, which is a measure of the presence of a more salient element than human scent contributing to response during soft surface tracking. The puppies required three or four puppy runaways to learn the task requirement to follow their nose. During puppy runaways, S+ (the same S+ as in human scent discrimination training) teased a pup with liver treats and then ran way to be found shortly after by the puppy. After a few puppy runaways, all puppies were tested on dry concrete and asphalt surfaces. All test trails were fresh and short, three to four blocks or less. Most discrimination involved cross tracks of the same age, but the tests were conducted in a small town so there were also fresher and older trails laid by members of the general public. If the puppies responded correctly, they were rewarded by the S+ trail layer, but if they went to S- the trail layer ignored them and the test trial was terminated without reward. The locations the trail layers walked was unknown to the handler and onlookers. During the first five test trials, only one puppy showed transfer of the discrimination out of the four tested. He responded correctly on the first, third, fourth, and fifth test trial. Nevertheless, all mixed breed puppies were able to follow a scent trail on hard surfaces starting with the first puppy runaway and all learned to discriminate between S+ and S- hard surface trails as training continued. Conversely, Most and Böttger reported adult police dogs trained on soft surfaces would not track on hard surfaces. Thus, the results indicate during soft surface tracking training a more salient element of the track overshadows learning about the human scent element.
It is not clear however, that transfer of the discrimination did not occur in the three puppies that responded at chance during the first five trials. Although first trial accuracy is certainly a very good measure of transfer, acquisition rate is more realistic. Transfer tests typically involve two or more groups of subjects. For example, during stage one an experimental group may be trained to discriminate between human scents and a control group may be trained to discriminate between hay and dirt. In stage two, both groups may be trained human scent discrimination involving new response requirements. If the experimental group reaches an arbitrary criterion of say 18 out of 20 correct responses consecutively faster than the control group, it can be supposed discrimination learning in stage one transferred to the new task during stage two. Since all puppies did learn to discriminate between trails laid on hard surfaces, it is reasonable to suppose acquisition rate was higher in puppies initially trained human scent discrimination than in dogs initially trained to track on soft surfaces without any explicit human scent discrimination training.
Learning to ignore human scent
Some overshadowing experiments have shown that when one group of animals initially receives compound conditioning in which more and less salient discriminative stimuli are presented simultaneously and are subsequently conditioned with the less salient discriminative stimuli presented in isolation of the more salient discriminative stimuli, they not only continue to perform less accurately compared to another group initially conditioned with the less salient stimulus alone, acquisition rate of the less salient discrimination is significantly slower in the compound group. Retarded learning indicates animals do notice the less salient discriminative stimuli during compound conditioning and selectively choose to ignore the less salient stimuli.
In order to test whether dogs initially trained to track on soft vegetative surfaces do notice the individual unique human scent component present in compound with other track scent during tracking training but selectively choose ignore it, both a compound group and a human scent alone group of adult hounds were trained human scent discrimination in isolation of tracks. There were two hounds in each group. The compound group was initially trained to track on soft surfaces, whereas the human scent alone group received no discrimination or tracking training prior to experimental conditioning. After tracking on soft surfaces was well established in the compound group, both groups of adult hounds were trained simple human scent discrimination using the same simultaneous classical conditioning procedure already described in the previous experiment involving mixed breed puppies. During conditioning, the human scent alone group readily learned the discrimination within the first few trials of the first twenty trial session. However, because the measure of response was persistence to S+, the exact rate of acquisition could not be determined. Nevertheless, it was clear the novice hounds readily learned the discrimination, which was significantly different from the puppies who took between two to three sessions to reliably learn the discrimination.*10 Alternatively, the compound group that had initially been trained to track on soft surfaces was still responding at chance after fifteen sessions or three hundred human scent alone discrimination training trials. The compound group not only continued to perform less accurately during subsequent human scent alone discrimination training, the rate of human scent discrimination acquisition was significantly slower than the human scent alone groups involving both the adult hounds in this experiment and the puppies in the previous experiment. The results indicate dogs initially trained to track on soft surfaces not only notice the human scent component, but also learn to ignore human scent during compound conditioning, which retards subsequent human scent discrimination training.
*10. That the mixed breed puppies took longer to learn human scent discrimination than adult bloodhounds was not because they were incapable of discriminating between human scents. Prior to human scent discrimination training, the puppies were trained to target the mouth of the jars human scent articles were place in. During that initial stage, the positive human scent that was to be used in human scent discrimination training was paired with reinforcement over trials. Thus, they were familiar with the S+ scent. During subsequent human scent discrimination training, on the first occasion the puppies smelled the novel negative human scent some of them responded fearful or apprehensively. In order to respond apprehensively to the novel human scent on the first occasion it was smelled, the puppies must have noticed it was different from S+. Therefore, the slower learning rate cannot be due to an inability to discriminate between human scents.
Equal saliency/equal reinforcement of discriminative components during compound discrimination training
When compound discrimination involves two or more dimensions of a complex stimulus in which the discriminative stimuli are equally salient and equally correlated with reinforcement, stimuli may differ in the degree to which they come to control behavior during conditioning. A common finding is that many subjects come to solve such compound discriminations mainly in terms of one of the dimensional components. For example, Reynolds (1961) trained two pigeons on a successive go/no-go discrimination in which a white triangle on a red background projected onto a pecking key was always paired with reinforcement and a white circle on a green background was always paired with the omission of reinforcement. When triangle/red randomly appeared, pecks to the pecking key were reinforced. Alternatively, reinforcement was omitted following pecks to the pecking key when circle/green appeared. Thus, the arrangement of training involved two dimensions in which subjects could learn the discrimination; both shape and color. The pigeons readily learned the discrimination. If all discriminative stimuli were equally salient and equally correlated with reinforcement, it seems reasonable to suppose attention and associability should increase to both shape and color. However, that is not what is commonly found. Subsequent test trials with the component stimuli presented in isolation of one another — triangle alone, circle alone, red alone, green alone — revealed that the birds only learned to discriminate between stimuli of one dimension. During test trials, in which reinforcement was necessarily omitted, one pigeon pecked the triangle (S+) but did not peck red (S+) or circle and green (S-); whereas the other pigeon pecked red (S+) but did not peck triangle or circle and green. Thus, one bird learned a shape discrimination and the other bird learned a color discrimination, but none learned both discriminations. Additionally, because one bird learned a shape discrimination and the other learned a color discrimination, all of the discriminative stimuli must have been equally salient. Reynold’s experiment shows that equal saliency and equal correlation of discriminative components during compound multi-dimension discrimination training does not ensure subjects will learn all discriminations possible.
Furthermore, Cohen et al. (1969) and Telegdy and Cohen (1971) have shown that the tendency to solve a compound discrimination in terms of one of the dimensional components may be increased by an increase in the level of drive to obtain the particular reinforcer used during conditioning.
Following the discovery of overshadowing, Kamin (1968) found not only can conditioning occur to a less salient stimulus when it is conditioned in isolation of a more salient stimulus, but also that prior conditioning to one component of a compound can prevent or block conditioning to the other component when they are subsequently presented in compound. In a set of experiments that had an important impact on how associative learning and conditioning is now viewed, Kamin showed that prior conditioning to one component of a compound would block conditioning to the second component when it was subsequently presented in compound with the first component.
Thus, if dogs receive human scent discrimination training in the absence of crushed vegetation and any other track elements, not only can they learn about the informative significance of the individual unique component of human scent predicting reinforcement, knowledge of the informative significance of individual unique human scent can also block learning about other stimuli when they are subsequently presented in compound, provided the particular reinforcer used during conditioning is not changed (unblocking will be reviewed shortly).
The basic design of blocking experiments involves three stages. In the first stage, an experimental group learns a relationship between stimulus A and a particular reinforcer by pairing stimulus A with the reinforcer until conditioning reaches asymptote. In the second stage, both the experimental group and a control group receive compound conditioning to A and B involving the same reinforcer used in stage one. Thus, the experimental group receives A alone training followed by compound AB training, whereas the control group only receives compound AB training. In the third stage, subjects from both groups are given test trials with B alone to see how much they learned about stimulus B. The tests reveal that prior training with one component of a compound (A) prevents the second component (B) from acquiring control over behavior when it is subsequently presented in compound with the first.
In other experiments, Kamin (1969) was able to show that the blocking effect is a function of pretraining with one of the components, A. However, even if blocking is a function of pretraining with A, this does not explain why blocking to B occurs. There are three ways in which attention has been used to explain blocking. The first is that during A trials, subjects in the experimental group (A alone followed by compound AB training) learn to attend to A. Since A is consistently paired with reinforcement during the A alone trials, subjects have no reason to disregard A and attend to B during subsequent AB trials in which the same reinforcer was used. Therefore, subjects in the experimental group continue to attend to A and learn nothing about B. The second is similar to the first in that it assumes a limited attentional capacity in animals. During A alone trials, conditioning to A causes the subjects in the experimental group to allocate all of their attention to A so that when A and B are subsequently presented in compound they have no attentional capacity left to attend to B. As a result, experimental subjects neither notice B nor learn anything about it. Finally, the third possible explanation is that during initial A alone training subjects in the experimental group learn that A predicts the particular reinforcer used, resulting in the subjects learning to attend to A. When B is added on subsequent AB trials, B is redundant and predicts nothing new about the outcome or reinforcer used during training. Consequently, subjects learn specifically to ignore the redundant stimulus B. In other words, in the third explanation subjects do notice and learn something about B, whereas in the first two explanations, subjects do not notice or learn anything about B.
In the first two explanations, subjects are assumed to have a limited capacity to attend to stimuli and do not notice the added stimulus B during subsequent AB trials. In the third explanation, subjects in the experimental group do notice B during subsequent AB trials but blocking occurs because they also notice that B is redundant in that it predicts nothing new about the outcome or reinforcer, and thus subjects learn (choose) to ignore B.
In order to determine which attentional explanation was more accurate, Kamin compared the response behavior on the first compound AB trial with the previous A alone trial and found that the experimental subjects responded differently on the first compound trial, which was evidence against the first two explanations. In order to respond differently on the first compound trial, the experimental subjects must have noticed the added stimulus. Therefore, blocking cannot be due to a failure to notice the added or blocked stimulus B on subsequent AB trials.
Additional evidence against the first two explanations was found in unblocking experiments. If blocking is the result of a learned predictive relationship between a specific stimulus and specific reinforcer conditioned in stage one then it is reasonable to assume that if the specific reinforcer is changed on the first compound trial, conditioning to the added stimulus should occur. This is exactly what has been found. Unblocking, or conditioning to the added stimulus B, occurs not only when the specific reinforcer used in prior A alone conditioning is changed, but also when the intensity of the reinforcer is changed, when another reinforcer is added, and when the predicted reinforcer is omitted on the first AB trial. Thus blocking is stimulus-reinforcer specific. After A alone conditioning, the specific reinforcer that follows the compound presentation of A and B controls blocking. When the addition of B on subsequent AB compound trials does not signal a change in reinforcement, blocking occurs. When the addition of B does signal a change in reinforcement, subjects learn the added stimulus-reinforcement relationship. Additionally, during unblocking experiments, often AB has disappeared by the time the reinforcer is presented. Thus, subjects can neither attend to nor ignore B. Therefore, blocking cannot be due to a failure to attend to B when it is first presented on AB trials.
Based on the experimental data, it is reasonable to conclude that animals selectively attend to certain stimuli or stimulus dimensions and not others, but that it is not due to a limited attentional or associative capacity. Instead, blocking and unblocking experiments reveal that selective attention is voluntary, resulting from the learned predictive relationship between a particular stimulus (or event) and particular reinforcer compared to other events that predict nothing new. In other words, during conditioning, animals learn not only about which stimuli or stimulus dimensions are informative in predicting a given outcome but also about which stimuli are redundant and do not predict anything new. Subjects come to ignore the added stimuli, not because of a limited attentional capacity, but because they learn that the stimuli are uninformative, in that they predict nothing new about the outcome.
Kamin’s blocking experiments had an important impact on how associative learning and conditioning is now viewed because they led theorists to question the importance of contiguity for conditioning. Contiguity refers to the nearness of events in time and space. Behaviorists thought that stimulus-response connections were automatically established either by simple contiguity between a stimulus and a response, in classical conditioning, or by a combination of contiguity and the strengthening of a response habit through reinforcement and repetition, in instrumental conditioning. However, relative validity, overshadowing, and blocking experiments show that the contiguous presentation of a stimulus with reinforcement does not guarantee conditioning to that stimulus. Instead, it is the predictive or informative value of a particular stimulus that is most important. If the addition of B in blocking experiments offers no information beyond the original A about the occurrence of the reinforcer, then the added B will fail to become conditioned.
Contiguity is still important, but not in the way it was initially envisioned. Researchers now know that the contiguity between a to-be-conditioned event and reinforcer is a clue for the subject that there is a possible causal or predictive relationship between a particular event and outcome. Therefore, the subject pays attention to that event until its informative value is determined relative to other possible cues predicting the same reinforcer.
Mackintosh (1973) reported that prior uncorrelated presentations of a stimulus and reinforcer retarded subsequent conditioning and that the interference was greater than that caused by exposure to either the stimulus or reinforcer alone. He argued that during initial exposure to uncorrelated stimulus-reinforcer presentations, animals come to learn that the stimulus is “irrelevant” as a predictor of the reinforcer, which retards subsequent conditioning when there is a contingency or correlation between the stimulus and reinforce.
Mackintosh’s learned irrelevance experiment involved four groups of subjects; a learned irrelevance appetitive group, a learned irrelevance aversive group, a latent inhibition group, and a control group. During the initial phase of the experiment, the learned irrelevance appetitive group was exposed to uncorrelated presentations of a tone and an appetitive reinforcer, the learned irrelevance aversive group was exposed to uncorrelated presentations of the tone and an aversive reinforcer, the latent inhibition group was exposed to the tone alone (without reinforcement), and the control group was simply place in the conditioning apparatus with no events scheduled. Uncorrelated presentation in the learned irrelevance groups involved a random control procedure, in which presentation of the tone in the absence reinforcement on some trials was randomly interspersed with the presentation of the reinforcer in the absence of the tone on other trials, so the probability of the tone occurring on any trial was entirely independent of the occurrence of the reinforce.
In second phase, retarded learning was used as a test measure of learned irrelevance. In this phase, half of the each group received paired presentations of the tone and the appetitive reinforcer to establish an appetitive approach response. While the other half received paired presentations of the tone and the aversive reinforcer to establish suppression of response. The results of the test phase showed that prior experience with the tone in the first three groups retarded subsequent conditioning, whereas the control group, for whom the tone was novel, associated the tone with reinforcement, either appetitive or aversive, more rapidly than the first three groups. Furthermore, both the learned irrelevance groups, who were initially exposed to a zero correlation between the tone and reinforcement during the initial phase, were the slowest to learn the tone-reinforcement relationships during the second phase of the experiment. Of the groups given tone paired with appetitive reinforcement conditioning in the second phase, the slowest to learn were those from the learned irrelevance appetitive group, who had initially been given the opportunity to learn that the tone and appetitive reinforcer occurred independently. Similarly, of the groups given tone paired with the aversive reinforcer in the second phase, the slowest to learn were those from the learned irrelevance aversive group, who had initially been given the opportunity to learn that the tone and aversive reinforcer occurred independently. Learning that there was a contingency between events suffered significantly from prior learning that there was no contingency in the learned irrelevance groups. A substantial body of evidence indicates exposure to a set of stimuli uncorrelated with reinforcement can specifically reduce attention and associability of those stimuli, which is apart from any increase in attention to a stimulus correlated with reinforcement.
Taken together with other effects found in discrimination learning, such as latent inhibition, overshadowing, and blocking, learned irrelevance experiments reveal the course of conditioning is not only affected by the detection of a contingency or correlation between stimuli or responses and reinforcers, it is also affected by prior learning that events are uncorrelated, or are less informative relative to other events that are better correlated with reinforcement. All show that the associability of a stimulus with a reinforcer declines as a consequence of earlier experiences with that stimulus in which it predicts nothing (latent inhibition), is less salient (overshadowing), is redundant (blocking), or uncorrelated with reinforcement (learned irrelevance).
Romanes reported that his setter trained to follow his track at the exclusion of others, had difficulty following his track when he walked barefoot or wore new or previously worn socks without his boots. Yet, subsequently, when he wore his boots on the first leg of a track, then removed them and walked first in his socks and then barefoot, she learned to follow his track. Böttger (1936) also reported that tracks laid with bare feet were less readily followed than those laid with shod feet (see also Most p. 158, 1954).
Assuming human scent from sweat permeated the leather soles of track layers shoes and boots and that no such scent permeated the soles of rubber boots, Budgett also conducted barefoot experiments to determine the importance of individual human scent compared to that caused by a track layers weight crushing vegetation as he walked over it. After walking some distance on a soft vegetative surface in rubber boots that had been cleaned to remove residual human scent, the track layer took off his boots and socks and proceeded for an equal distance on bare feet. Subsequently, when his old bloodhound was tested on the track after six hours had passed, Budgett was surprised to find that the hound more readily followed the first leg of the track in which rubber boots had been worn.
In a second experiment, after five hours had passed since a soft surface track was laid, the old hound was brought to the place where the track layer discarded his rubber boots and continued walking barefoot. When given a choice upon testing, the old hound selected the rubber boot backtrack in preference to the forward direction barefoot track. Upon repeating the experiment after a one and a half hour time lapse, Budgett’s old hound chose the forward direction barefoot track laid on a vegetative surface.
In another experiment conducted by Romanes, after it was well established his setter bitch would reliably discriminate between Romanes track always positive and other tracks of any age always negative, he found she was unable to follow his track when a single layer of brown paper was glued to the sides and soles of his boots. Later, he found she could readily do follow his track after “a small portion of the brown paper first became worn away at one of the heels … although the surface of the shoe leather which touched the ground was not more than a few square millimeters” (Dec. 16, 1886, Journal of the Linnean Society).
In yet another experiment Romanes found his setter would refuse follow his track after he changed boots at a choice point with a stranger, but would readily track the stranger with whom he had changed boots. Budgett (1933) repeated Romanes’ boot changing experiment with Colonel Robert’s bloodhound, St. Christopher of Trethill, which took place across a mowed polo field. After providing two of his staff with rubber gloves to control against scent from their hands contaminating the boots, Budgett had them walk from opposite directions towards one another across the polo field, then at the point in which they met the men changed boots and parted in opposite directions [assuming at right angles away from their original tracks]. St Christopher followed the target track to the choice point where the track layers switched boots, hesitated for some moments, and finally followed the track of the man wearing the boots that had previously been worn by the target track layer.
Taken together, the experiments indicate the setter and bloodhounds did not learn to follow or discriminate between individual human scents. Although Budgett thought bloodhounds discriminate between individual human scent particles left behind from a track layers boots that had become impregnated with the track layer’s scent, the boot changing and barefoot tracking experiments did not support his assumption. If the setter and bloodhounds learned to follow and discriminate between human scents, St Christopher should not have switched tracks at the choice point, Romanes’ setter should have followed his track after he changed boots, and the setter and old hound should have readily followed the barefoot track on the first test trials.
Mackintosh argued that during initial exposure to uncorrelated stimulus-reinforcer presentations, animals come to learn that the stimulus is “irrelevant” as a predictor of the reinforcer. While Budgett did not report how the bloodhounds had been trained, it was popular at that time, and for some still today, to apply Romanes method of laying and working his own track during the initial stages of tracking training. Assuming that was the case, it is not unreasonable to consider that during initial training in which handlers both lay the track to be followed and walk behind their dogs when they are subsequently tasked to follow the track, dogs learn that human scent is irrelevant.
The discovery that the relative predictive value of a stimulus is more important for successful conditioning than contiguity led researchers, starting in the late 1960s, to change their focus from investigating how responding to a single stimulus paired with a biologically important outcome affected numerous controlling variables (the laws of conditioning) to investigating how responding is affected by the interaction between multiple probable signals of reinforcement. This was the start of a radical change from behaviorists’ views to cognitive views. Unfortunately, although many dog trainers know about the old views and arrangements of training in keeping with old outmoded views, still today it seems many do not know about the significant changes that have taken place since the late 1960s and early 1970s. All of those who train dogs to detect and discriminate between stimuli can greatly benefit by learning about stimulus selection.
Not only can more salient elements from tracks laid on soft surfaces overshadow learning about the individual unique discriminative element of human scent along human laid tracks during initial training, but learned irrelevance experiments indicate the likelihood of dogs learning to ignore the discriminative elements of human scent during initial soft surface tracking is increased when training involves trainer laid tracks that are subsequently worked with the trainer following behind the dog.
Belleville, R. (1938). Neue Versuche auf dem Gebiet der Fährtenreinheit und des Identifizierens von Gegenständen auf der Fährte. Z. Hundeforsch, N. F., 13, 17-28.
Bennett, C. H., Wills, S. J., Wells, J. O., and Mackintosh, N. J. (1994). Reduced generalization following preexposure: Latent inhibition of common elements or a difference in familiarity? Journal of Experimental Psychology: Animal Behavior Processes, 20, 232-239.
Blunk, R. (1926). Die Ausbildung des Hundes zur Spurenreinheit. Rostock: Hinstorffs Hofbuchdruckerei.
Böttger, P. (1926). Anweisung zum Abrichten von Suchhunden auf Grund praktischer Erfahrungen. IV. Aufl. Augdburg: Verein für deutsche Schäferhunde (SV), Sitz München (EV).
Böttger, P. (1932). Von Werdegang des Polizeifährtenhundes. Z. Hundeforsch, 2, 196-222.
Böttger, P. (1936). Hund im Dienste der Kriminalpolizei. Krim. Monatsh, 10, 121-127.
Böttger, P. (1937). Hunde im Dienste der Kriminalpolizei: unter besonderer Berücksichtigung des Mordfalles Seefeldt. Z. Hundeforsch. N. F. 5, 32.
Brough, E. (1902). The Bloodhound and its Use in Tracking Criminals. London: Illustrated “Kennel News” Company.
Budgett, H. M. (1933). Hunting by Scent. London: Eyre & Spottiswoode.
Clifford, R. J. (1958). Some notes and theories on ‘scent’, its formation, properties, and usage as derived from observations on and experience with tracker and patrol dogs in Malaya. JR Army Veteran Corps, 29, 145.
Cohen, J. S., Stettner, L. J., and Micheal. D. J. (1969). Effect of deprivation level on span of attention in a multi-dimension discrimination task. Psychonomic Science, 15, 31-32.
Dominik, D. (2000). Attempt toward determination of influence of scent background upon indications by dogs. Paper presented at conference: Osmology: overestimated or neglected area of forensic science? Problemy Kryminalistyki 230: 56-58.
Ehmann, E., Boedeker, E., Friedrich, U., Sagert, J., Dippon, J., Friedel, G., and Walles, T. (2012). Canine scent detection in the diagnosis of lung cancer: revisiting a puzzling phenomenon. European Respiratory Journal, 39(3), 669- 676.
Hansmann, J. (1929). Die Physiologie des Riechens in ihrer Bedeutung für die Nasenarbeit des Hundes. Ztg. Ver. Dtsch. Schäferhunde, 28, 736-741.
Hansmann, J. (1931). Unter Welchen Gesichtspunkten erfolgt die praktische Verwendung des Polizeifährtenhundes? Z. Hundeforscg, 1, 14-30.
Hansmann, J. (1949). Der Gebrauchshund auf der Menschenfahrte. Unsere Hunde, 3, 275-278, 280-281; (1950) Unser Rassehund, 2, 334-336, 385-386, 430-431; (1951) 3, 84, 114, 129-130.
Hansmann, J. (1955). Hunde- und andere Nasen, Geruch und „Roter Faden“. Terrier, 48, 206-207, 210-212.
Honhon, J. (1967). L’Olfaction chez le Chien: Son rôle dans le pistage et la localization d’une source odorante. Dissertation, Paris.
Hull, C. L. (1952). A Behavior System. Yale University Press, New Haven.
Jenkins, H. M. and Harrison, R. H. (1960). Effects of discrimination training on auditory generalization. Journal of Experimental Psychology, 59, 246-253.
Kamin, L. J. (1968). ‘Attention-like’ processes in classical conditioning. In M. R. Jones (Eds.), Miami Symposium on the Prediction of Behavior: Aversive Stimulation. University of Miami Press, 9-33.
Kamin, L. J. (1969). Predictability, surprise, attention and conditioning. In B. A. Campbell and R. M. Church (Eds.), Punishment and Aversive Behavior. New York: Appleton-Century-Crofts, 279-296.
Konorski, J. (1948). Conditioned Reflexes and Neuron Organization. Cambridge: Cambridge University Press.
Lashley, K. S. (1940). Studies of simian intelligence from the University of Liége. Psychological Bulletin, 37(4), 237-248.
Löhner, L. (1924). Über menschliche Individual- und Regionalgerüche. Pflügers Archiv European Journal of Physiology, 202, 25-45.
Löhner, L. (1926a). Untersuchungen über die geruchsphysiologische Leistungsfähigkeit von Polizeihunden. Pflügers Archiv European Journal of Physiology, 212, 84-94.
Löhner, L. (1926b). Von entscheidenden Versuchen über das Anzeigen von Witterungsübereinstimmung. Dtsch. Schäferhund, 5, 497-499.
Löhner, L. (1926c). Zur Methodik geruchsphysiologischer Untersunchungen von Polizeihunden. Hund, 10, 330-332.
Löhner, L. (1927). Über Fremdidentifizierung und Anzeigen von Witterungsübereinstimmung. Ztg. Vereins dtsch. Schäferhunde (SV) Sitz München, 26, 566-572.
Löhner, L. (1931). Grundsätzliches aus der Geruchsphysiologie. Ztg. Vereins dtsch. Schäferhund, 30, 422-436.
Mackintosh, N. J. (1973). Stimulus selection: learning to ignore stimuli that predict no change in reinforcement. In In R. A. Hinde and J. S. Hinde (Eds.) Constraints on Learning. London: Academic Press. 75-96.
Mackintosh, N. J. (1974). The Psychology of Animal Learning. London, New York, San Francisco: Academic Press.
Mackintosh, N. J. (1975). A theory of attention: variations in the associability of stimuli with reinforcement. Psychological Review, 82, 276-298.
Mackintosh, N. J. (1983). Conditioning and Associative Learning. Oxford: Clarendon Press.
Mackintosh, N. J., Kaye, H., and Bennett, C. H. (1991). Perceptual learning in flavour aversion conditioning. Quarterly Journal of Experimental Psychology, 43B, 297-322.
Marciniak, R. (1999). Attempts towards determining the influence of beauty products on olfactory performance in dogs. Problemy Kryminalistyki 226: 45-46.
McCartney, W. (1968). Olfaction and Odours: An Osphresiological Essay. Springer-Verlag, Berlin, Heidelberg, New York.
McCulloch, M., Jezierski, T., Broffman, M., Hubbard, A., Turner, K., and Janecki, T. (2006). Diagnostic accuracy of canine scent detection in early- and late-stage lung and breast cancers. Integrative Cancer Therapies, 5(1), 30-39.
Menzel, R. (1933). Die Arbeit mit Witterungskonserven. Z. Hundeforsch. Heft ½, 101-113.
Menzel, R. and Menzel R. (1928a). Das Problem der Verwendung des Hundes als „Feinriecher“. Dtsch. Polizeihund, 28, 387-395.
Menzel, R. and Menzel R. (1928b). Das „Mikrolfak“ Hund und seine Leistungsfähigkeit. Ztg. Ver. Dtsch. Schäferhunde, 27, 155-161.
Menzel, R. and Menzel R. (1929a). Beobachtungen über Abstraktionsvermögen des Hundes. Arch. Ges. Psychology, 71, 339-350.
Menzel, R. and Menzel R. (1929b). Die Bedeutung der Gesetze über Schwellenwert und Reizsummation bei der Spürarbeit des Hundes. J. Psychol. Neurol., 38 258-264.
Menzel, R. and Menzel R. (1929c). Der Eigengeruch des Hundes als Fehlerquelle bei Witterungsreinheits-Arbeiten. Ztg. Ver. Dtsch. Schäferhunde.
Menzel, R. and Menzel R. (1930). Die Verwertung der Riechfähigkeit des Hundes im Dienste der Menschheit. Berlin: Kameradschaft Verlagsgesellschaft.
Menzel, R. and Menzel R. (1932). Tierpsychologische Voraussetzungen für die Nasenarbeit des Hundes. Z. Hundeforsch ,2,17-28.
Menzel, R. and Menzel R. (1935). Die Vererbung der Nasenleistungen und sich daraus ergebenden Zuchtziele. Ber. Dritter kynologischer Weltkongreß, 88-90.
Misiewicz, K, (2000). Survey of nicotine scent on correctness of special dog’s identification in human scent traces identification. Problemy Kryminalistyki 229: 38-40.
Morgan, C. L. (1894). An Introduction to Comparative Psychology. London: W. Scott. Most, K. (1926). Neue Versuche über Spürfähigkeit. Hund Nos, 18, 20, 22.
Most, K. (1928). Das Problem der Spurenreinheit auf der menschlichen Spur im Lichte der zumalmit der Fährtenbahn erzielten Versuchsergebnisse. Hund No. 2, 31-35.
Most, K. (1951). Die Abrichtung des Hundes. Braunschweig: Gersbach. Most, K. (1952). Über Fährtenarbeit des Hundes. Pirsch 4, 120-124, 158-160.
Most, K. (1954). Training dogs: a manual. London: Popular Dogs Publishing Company. Translated in English in 1954 from the German version, Die Abrichtung des Hundes, originally written in 1911, last revised German edition published in 1951.
Most, K. u. Brückner, G. H. (1935). Die Vererbung der Nasenleistungen und die sich daraus ergebenden Zuchtziele. General-Ber. Dritter kynologischer Weltkongress, 48-58.
Most, K. u. Brückner, G. H. (1936). Über Voraussetzungen und derzeitigen Stand der Nasenleistungen von Hunden. Z. Hundeforsch. N. F. 1, Heft, 5, 9-30.
Pavlov, I. P. (1927). Conditioned Reflexes. Oxford University Press.
Premack, D. (1978). On the abstractness of human concepts: why it would be difficult to talk to a pigeon. In S. H. Hulse, H. Fowler, and W. K. Honig (Eds.), Cognitive Processes in Animal Behavior, (423-451), Hillsdale, NJ: Erlbuam.
Rescorla, R. S. (1988). Pavlovian conditioning: it’s not what you think it is. American Psychologist, 43(3), 151-160.
Reynolds, G. S. (1961). Attention in pigeons. Journal of the Experimental Analysis of Behavior, 4, 179-184.
Romanes, G. J. (1982). Animal Intelligence. London: Kegan Paul.
Romanes, G. J. (1983/1977). Animal Intelligence. Washington, D. C.: University Publications of America.
Romanes, G. J. (1984/1969). Animal Intelligence. New York: Appleton. Romanes, G. J. (1887). Experiments on the sense of smell in dogs. Nature, 36, 273-274. Also (Dec. 16, 1886). Journal of the Linnean Society.
Schmid, B. (1933). Wie weit sieht der Hund, und auf welche Entfernung erkennt er seinen Herrn? Z. Hundeforsch, 3, 188-208.
Schmidt, F. (1910). Verbrecherspur und Polizeihund. Praktische Versuche, Beobachtungen und Winke für Polizeihundmänner u. Behörden. Augsburg: Selbstverlag des Vereins für deutsche Schäferhunde.
Schmidt, F. (1911). Die menschliche Geruchspur und die Fähigkeiten des Polizeihundes. Jber. Naturwiss. Vereins Osnabrück (1907/1910), 17, 31-55.
Schoon, G. A. A. (1997). The Performance of Dogs in Identifying Humans by Scent. Thesis, University of Leiden, the Netherlands.
Seraganian, P. (1979). Extradimensional transfer in the easy-to-hard effect. Learning and Motivation, 10, 39-57.
Skinner, B. F. (1938). The Behavior of Organisms. New York: Appleton-Century-Crofts.
Spence, K. W. (1956). Behavior Theory and Conditioning. Yale University Press, New Haven.
Sutherland, N. S. and Mackintosh, N. J. (1971). Mechanisms of Animal Discrimination Learning. New York: Academic Press.
Syrotuck, W. G. (1972). Scent and the Scenting Dog. Arner Publications, Westmoreland, NY.
Telegdy, G. A. and Cohen, J. S. (1971). Cue utilization and drive level in albino rats. Journal of Comparative and Physiological Psychology, 75, 248-253.
Thorndike, E. L. (1898). Animal intelligence: An experimental study of the associative processes in animals. Psychological Review Monograph Supplements, 2(8), 1-109.
Thorndike, E. L. (1911). Animal Intelligence: Experimental Studies. New York: Macmillan.
Wagner, A. R., Logan, F. A., Haberlandt, K., and Price, T. (1968). Stimulus selection in animal discrimination learning. Journal of Experimental Psychology, 76, 171-180.
Watson, J. B. (1913). Psychology as the behaviorist views it. Psychological Review, 20, 158-177.
Watson, J. B. (1914). Behavior: An Introduction to Comparative Psychology. New York: Holt.
Watson, J. B. (1916). The place of the conditioned reflex in psychology. Psychological Review, 23, 89-116.
Zell, T. (1909). Der Polizeihund als Gehilfe der Strafrechtsorgane. Ein Hilfsbuch für Staatsanwälte, Untersuchungsrichter, Verteidiger, Polizeibehörden usw. Berlin: Guttentag.
Copyright © E Hale 2015.
I am not real great with English but I find this real easygoing to interpret.