Categorisation in animals

Experimental work in the field of animal categorisation is rather aimed at studying to what level animals are able to follow “human” rules in grouping objects than investigating how far in the animal kingdom does this mode of conceptualisation extend. A useful general framework for the investigation of categorisation in animals was provided by Herrnstein (1990) who described categorisation abilities in five levels of increasing abstractness, including (1) discrimination; (2) “categorisation” by rote; (3) open-ended categorisation (namely, category formation resting on a perceptual similarity between objects that belong to a given class; (4) conceptual categorisation and (5) abstract relations. Herrnstein (1990) uses two criteria to define conceptual categorisation (level 4). The first criterion is met when a rapid generalisation about members of a class of items is observed. The second criterion, which is related to conceptual processing, implies categorisation abilities that go beyond perceiving a similarity between exemplars or a class. Thus, level 4 is more complex than open-ended classification, the latter being related to the use of perceptual dimensions of stimuli. To perform this, a subject has to discriminate polymorphous stimuli, i.e. stimuli for which no single feature is either necessary or sufficient to determine category membership. Level 5 is attained when a subject is able to use abstract relations not only between objects but also between concepts, such as in conceptual matching or in conceptual identity (for example, the mastery of “sameness” relations).

In their attempts to investigate animal mentality, experimenters test ideas about what in called prototype effect, which was initially reported in the human studies (Rosch, 1973). This effect is expressed by a better categorising performance with prototypical stimuli representing the central tendency of the category than with other, less typical exemplars. For example, humans think that a sparrow is a better exemplar for the “bird” category than an ostrich. In fact, prototype effect is a matter of one of three theories of categorisation which have been elaborated in human psychology. These are the exemplar, the feature, and the prototype view as three types of representation.

The evidence for capacities to perform the first three levels of categorisation is abundant for several animal species. It is, however, much less clear concerning levels 4 and 5 (for review, see: Thomson and Oden, 2000; Huber, 2000; Vauclair, 2002).

Several experiments were conducted with baboons in order to assess the abilities of these monkeys to discriminate objects on the basis of their membership in a category and to study the nature of the representations of categories the baboon formed (Vauclair and Fagot, 1996). A video task required the baboons to manipulate a joystick that controlled the movements of a cursor on a screen. The subject was required to manipulate the joystick so as to touch with the cursor a response stimulus that matched the sample stimulus on an arbitrary (experimenter-defined) basis. In one set of experiments the baboons were tested to categorise characters displayed in various type faces. For this purpose, the baboons were first trained in a symbolic matching-to-sample task with 21 different fonts of the characters “B” and “3” as sample forms, and colour squares as comparison forms. After training, novel fonts were displayed. This task can be traced back to the basic concept of gestalt psychologists who investigated human’s ability to grasp whole figures (such as letters) by their fragments. Baboons showed positive transfer of categorising performance to the novel stimuli of the characters used in the original training. These results demonstrate that the original learning was not achieved by rote, because in that case the animals would have demonstrated no transfer to the novel typefaces. Thus the baboons’ performance indicates that these monkeys were able to exhibit level 3, i.e. open-ended categories in Herrnstein (1990) sense.

Identical polymorphous stimuli were presented to humans and baboons in a symbolic matching-to-sample task (DJpy et al., 1997). Subjects were trained to classify two out of three feature stimuli (colour, shape, position), and then to assess transfer of performance with the prototypes of each category. Whereas human participants solved the task in a propositional way, the baboons did not extract the prototypes; instead they used a mixed procedure that consisted in memorising salient cues between stimuli or specific associations between exemplars and response associations.

In series of experiments of Bovet and Vauclair (1998, 2000, 2001), olive baboons living in small social groups in an outdoor enclosure were individually trained and tested on the natural category on food versus non-food with real objects with the use of an adapted version of Wisconsin General Test Apparatus. The apparatus was made of a vertical wooden board comprising a one-way screen, a horizontal board to present the stimuli behind a Plexiglas window and two openings for two ropes. When the experimenter placed one or two objects on the board, the subject had to respond by pulling one of the two ropes, according to the categories or to the relations presented. A food reward was provided when the baboons’ response was correct. In each task, the baboons were trained with two stimuli (objects or pairs of objects) and when they succeeded, new objects were presented in order to assess transfer abilities. Four baboons were first trained to categorise two objects, one food and one non-food; then 80 other objects (40 foods and 40 non-foods) were presented and the response to each object was recorded. The baboons showed a high and rapid transfer of categorising abilities to the novel items. A similar performance for vervet monkeys was described by Zuberbhhler et al. (1999). The set of data in which various modes of picture presentations were used further demonstrated the abilities of the baboons to relate real objects to their pictorial representations. Using the procedure of successive simple discriminations, the experimenters demonstrated monkey’s abilities of categorisation corresponding to the level 5 of Herrnstein’s classification scheme. In a first experiment, the monkeys had to judge two physical objects as “same” or “different” (perceptual identity). For example, they were required to judge two apples as being the same, or an apple and a padlock as being different. In a crucial test of conceptual identity (corresponding to Herrnstein’s level 5), the baboon had to combine their previously acquired skills in order to classify as “same” two (different) objects that belonged to the same functional category (food or non food) and apply that learning to new exemplars. For example, they had to classify as “same” an apple and a banana, or a padlock and a cup, and as “different” an apple and a padlock. The monkeys attained a high level of performance at the end of the experiment with totally novel objects (i.e. objects novel in the task but left in the monkey’s enclosure before the experiment). Such results demonstrate the mastery of the “same-different” concepts and the ability to conceptually judge as same or different objects in the previously learned categories (Vauclair, 2002).

Outstanding capacity of pigeons for categorisation still remains a mystery. In fact, these were Herrnstain’s works on pigeons that initiated investigations on categorisation and concept formation in animals. Herrnstein and Loveland (1964) trained pigeons to peck for food at any of a number of photographic pictures that contained a person or people somewhere in the picture. There was a wide variation in the appearance of the people shown in these slides - e.g., their number, orientation, size, colour of their clothing, etc. Other slides that did not contain a person in them were also shown to the pigeons; when these appeared, pecking at them did not produce food. With training, pigeons soon came to peck quickly and rapidly at the " people" slides and not to peck at the " non-people" slides. Similarly, when shown a new slide without a person in it, the pigeons either did not peck at it or did so much less frequently than to a " people" slide, again despite never having experienced non-reinforcement with that particular slide. These latter results represent crucial data for inferring categorisation: the pigeon's explicitly taught behaviour generalised to new instances of the " people" and " non-people" groups (or " tree" and " non-tree" in other experiments).

This type of experiment has received considerable attention not only because the results suggest that pigeons possess an ability that transcends the discrimination of simple stimulus dimensions such as wavelength, intensity, or frequency, but also because they imply that the pigeons' classification behaviour is mediated by abstract, or conceptual, rules, and therefore resembles the cognitive solution accomplished by humans.

An even closer analogue to human categorisation by appearance was developed by Wasserman and his colleagues (Bhatt et al., 1988). They applied the pigeon version of the " name game", which parents often use to teach their young children the names of different objects from a number of perceptually distinct categories. On each training trial in the pigeon version of the name game, hungry pigeons were shown a photographic slide on a centre viewing screen that depicted one of a variety of examples drawn from four different categories: cats, chairs, cars, and flowers. After pecking at the slide appearing on the centre screen, pigeons could then obtain food by " naming/labelling" it appropriately. This meant pecking one of four different keys located at the vertices of the viewing screen. Thus, if a picture of a flower appeared on the centre screen, pecking the top left key produced food; if a picture of a car appeared, pecking the top right key produced food, and so on. Like the pigeons in Herrnstein and Loveland (1964) study, these pigeons readily learned to correctly categorise the various pictures shown to them in each training session. In this case, they quickly learned to confine their pecks to one key when shown any instance of a cat, to another key when shown any instance of a flower, and so on. The results suggest that the pigeons had grouped the various slides together by appearance. Things that looked alike were treated alike. The four different, explicitly taught pigeon " names" generalised immediately and appropriately to novel instances of each group that were later shown to them during a subsequent generalisation test. To distinguish between " accident" and true categorisation, researchers have compared pigeons' performances on the task with the performances of other pigeons that have been shown exactly the same set of photographic slides but have been taught a " pseudo-category" task (Wasserman et al., 1988). In this task each picture is again associated with one of the four possible responses but different pictures within a particular group (e.g., different cats or different flowers) are associated with different responses. For example, the pigeon might receive food for pecking the top left key after seeing one of the flower pictures, but the bottom right key after seeing a picture of a different flower, and so on. Pigeons learn this task, too, but they do so much more slowly and they are not as accurate in their choices as the " true category" pigeons.

Some experiments demonstrated pigeons as being able to use multiple facial features to discriminate human faces (Jitsumori and Yoshihara (1997). Researches have also examined prototype effects in pigeons by using human faces, i.e., naturalistic visual stimuli, created by the morphing technique. A physical prototype created by averaging training exemplars is assumed to correspond to a prototype abstracted by categorisation training. With the categories constructed to mimic family resemblance of natural categories, pigeons showed clear evidence of prototype effects (Makino and Jitsumori, 2001). Pigeons trained to discriminate typical and atypical exemplars of a category, showed multiple prototypes of the sets of atypical exemplars, each of which were perceptually disparate to one another (Jitsumori, 2004).

Huber (2002) suggests that pigeons use characteristics of objects that differ from those of humans’, such as tiny details of texture and shape.

Texture is mostly related to the surface properties of aerial photographs, landscapes or industrial materials. Huber and colleagues (Huber, 1995; Huber et al., 2000) used human faces as stimuli and the concept " sex" to define class membership. Experimenters compared the classification performance of pigeons presented with different versions of the same set of stimuli (Fig. V-2). The stimuli could be distinguished according to their texture and shape information, and were derived from laser scanned models of the faces of 100 men and 100 women. The faces were free from any kind of accessories such as glasses or earrings. The men were carefully shaven and the hair on the head was digitally removed from the 3D-models. The 200 faces were randomly divided into two sets (A and B), each containing 50 male and 50 female faces. Group O were shown the original images, while Groups T and S were shown images only after they had been subjected to a technique described in Vetter and Troje (1997) which involved separating the texture and shape components of each image. Group T was shown images generated by combining the original texture of each face with an average shape. This yielded an image set that varied with respect to texture but not shape. Group S was shown images generated by combining the original shape of each face with an average texture, which yielded an image set that varied with respect to shape but not texture. The results of this experiment indicated that Groups O and T learned very quickly and accurately to discriminate faces, whereas Group S failed to do so. These results suggest that pigeons are extraordinarily sensitive to texture differences, but that they find it very difficult to discriminate shapes.

The stimulus parameters that controlled the performance of successful subjects remain to be determined. Male and female faces differ both in average size and in average intensity. Female faces are generally smaller and brighter than male faces. Therefore, the experimenters computed the rank correlation between pecking rate to individual faces and either the average size or the average intensity of these images. The five parameters that describe the texture of images (energy, contrast, entropy, homogeneity) as well as the three components that describe their colour (red, green, and blue) were also quantified. The correlation was computed separately for male and female faces. The pecking rates of almost all Group O and Group T subjects correlated significantly with intensity, but not with any of the other texture parameters or size. Pigeons assigned to these groups appeared to use the intensity of faces as a cue to discriminate between male and female faces. Computer analysis of images revealed a subtle difference in colour between male and female faces; male faces are redder than female faces, while female faces are more blue and green than male faces. It is likely that pigeons use their extraordinary physiological capacity for the exploitation of colour. These findings raise question of whether animals might sort the complex objects of the natural environment, even the so-called higher-order concepts like " persons" and " fish", by fixing on some specific, single feature. Huber (2002) concludes that although pigeons have strong resources for learning specific exemplars, and display surprising cognitive capacities, neither categorisation in terms of exemplar memorisation nor in terms of abstract concept formation is plausible; it seems reasonable that representations of stimuli at the level of relationships between two or more arrays are at the limit of their cognitive capacity.

We are induced to leave an open question of whether categorisation research with pigeons have already revealed their extraordinary cognitive skills or “only” their capacity to exploit very subtle variation in texture of natural objects. Nevertheless, it is indicative that advanced forms of categorisation are currently intensively investigated in many species including small brained but specifically gifted creatures such as pigeons.