As a way of exploring categorisation in the chimpanzee Viki, Hayes and Nissen (1971) report the continuation of the generalised discrimination problems presented in Hayes and Hayes (1953). Although not targeted at investigating pictorial compe-tence per se, recognition was a requirement for successful performance.
Testing was done using a Klüver board as in the Hayes and Hayes (1953) study.
Two wells were covered with pictures. One of the wells was baited with food. The same category was always rewarded and the presentation was randomised. The re-sults given (table 2, p.108) are generalisation trials after Viki had discovered the categories to sort by, thus only showing her further application of a given discrimi-nation rule. Pictures were only used once.
Four human subjects of the same age as Viki performed above chance on the same stimuli in all problems but the active versus the inactive people condition.
Hayes and Nissen (1971) ascribe this to the limitation of static pictures to depict activity. Reading activity into pictures might be a late development in both humans (Friedman & Stevenson, 1980) and ape. It remains to be tested if this component of pictorial competence at all can appear in animals. Dynamic content of some sorts, e.g. the recognition of living individuals, seem to pose no problems in even reality mode picture processing.
Verbal reports by the children suggested that it was seldom categorisation as such that posed problems but to remember which option had been correct in the previous trials. This probably held true for Viki as well. In addition she tended to become careless when a task required close attention. “Caution in selection was not one of Viki’s usual working characteristics” (Hayes & Nissen, 1971, p. 85).
There is little reason to believe that a chimpanzee, albeit raised in a human con-text, should form the same categories as an adult language competent experimenter.
This alone is a reason to not draw too far-reaching conclusions regarding Viki’s out-standing results. An additional concern is that Viki in other studies had shown a capacity for single-trial learning, as well as simultaneously retaining in memory up to 10 concurrent discriminations for object pairs (Hayes et al., 1953b). This means that she could remember which one of a stimulus pair that was correct on ten suc-cessive but independent problems.67 In the picture categorisation study this compe-tence of Viki’s made it difficult to control for choice by invariant cues rather than choice based on concepts, even when switching types of pictures.
67 Some chimpanzees in Hayes et al. (1953b) could retain 20 concurrent discriminations.
+ Target - Target n % corr. Type of picture
Animals Objects 41 85 Realistic colour
Men Women 65 67 No details given
Red Green 43 74 Nonsense designs, line drawings, realis-tic with a predominant colour
Children Adults 47 89 Realistic, mostly colour
Circles Crosses 40 80 Drawings
Larger Smaller 25 84 Nonsense designs (matched shape and colour)
Larger Smaller 54 55 (n.s.) 89
Nonsense designs (mismatched shape and colour) With ratio > 2:1
Quiet people Active people 34 61 (n.s.) Realistic pictures (e.g. sleeping, playing baseball)
Complete drawings
Incomplete
drawings 37 79 Black-and-white drawings (e.g. dog without legs, hand missing finger)
Table 2. Generalised discrimination study on Viki (Pan troglodytes, 5 years) in Hayes and Nissen (1971)
Hayes and Nissen (1971) propose that an alternative explanation for Viki’s results with the pictures might be that she continuously changed the cues she worked by, or learned concurrent ones, which could then be generalised to novel pictures. In for example the gender condition she could have started out by basing her categorisa-tion on clothing cues in the full body pictures, then switch to hair length in the head shots, independently from the fact that certain clothes and hair styles have a gender component in common. She might never have thought of applying the gender con-cept to the problem at all, but still be rewarded with food.
Nevertheless, to single out “clothing” or “hairstyle” some sort of apprehension of the pictures’ motifs is necessary, but not necessarily in a pictorial mode. Reality mode would work for the photographs and perhaps some of the line drawings (i.e.
silhouette ones). Even surface mode would suffice if “clothing” and “hairstyle” were substituted with patterns and colourations. The complete versus incomplete
draw-ings, though, seem to again demonstrate Viki’s ability to recognise not only photo-graphs but quite abstracted drawings as being depicting pictures. But unfortunately are descriptions of the stimuli and her performance on individual pictures not given.
A discrimination task was used by Vonk and MacDonald (2002) to test categorisa-tion of natural concepts on “three different levels of abstraccategorisa-tion” in a juvenile gorilla.
The levels of abstractions: concrete, intermediate and abstract, mirror those of Rob-erts and Mazmanian (1988). In this classic experiment humans, pigeons and squirrel monkeys had to discriminate photographs of kingfishers from other birds, birds from other animals and, lastly, animals from non-animals. Human undergraduate students managed to discriminate on all levels of abstraction while pigeons and monkeys performed best on the most concrete, i.e. choosing kingfishers among two pictures of birds. With further training both species could also discriminate novel photographs of animals from photos of inanimate objects. Performance on the in-termediate level on the other hand (birds from other animals) never exceeded chance.
Using the discriminations orangutans/gorillas versus humans, primates versus other animals, and animals versus foods/objects/sceneries, Vonk and MacDonald (2002) mirrored also the results of Roberts and Mazmanian (1998). Zuri the gorilla performed best on the concrete and abstract “levels” and more questionably on the intermediate one. Stimuli used were photographs, mostly in colour, of many differ-ent views of the animals and objects, shown on a computer touchscreen.
At the discrete level (orangutans/gorillas versus humans) discrimination could occur simply by way of recognising invariant visual properties of a single species.
Not even recognition of the species would be necessary. But the authors, making a thorough analysis of properties across pictures as well as reinforcement history, could not find any causes external to the categorisation of the content itself that could ac-count for the successful transfer of Zuri to novel pictures. For example, a test that controlled for the conspicuous colour of orangutans did not reveal an effect of col-our on Zuri’s discriminations. It is therefore safe to assume that Zuri recognised content in most of the photographs and made her judgements accordingly.
The question, however, is which content Zuri ascribed to the pictures. Zuri had limited experience of other animals, and no real life experience of orangutans at all.
The pictorial versions of orangutans can thus count as a class on its own, recognised as a bona fide natural class due to the photographic quality of the stimuli, but not necessarily related to the world outside of computer screens. Zuri’s fastest acquisi-tion was in fact the discriminaacquisi-tion between familiar foods and animals, the former being a category that she did have extensive experience with and that furthermore is more faithfully reproduced in photographs, in terms of size and static nature, than are animals, scenery etc. Worth noting is that food was the positive target stimulus, so in order to discover how to solve the task Zuri had just to continue touching that which she intuitively preferred, and realise that she was differentially rewarded for this.
Here two versions of reality mode processing become evident. One is taking the content at face value, as a novel real entity. This is likely to happen to pictures of objects and animal types that one has no real-life experience of. Since all experience
is from the pictorial version some properties of the real instances is never evoked in the viewing mind, such as the movements of orangutans, their size, or smell. The depicted objects have, more or less, the properties displayed in the static image and nothing more. The second type, recognising well known entities, has the potential to evoke some of the knowledge of these entities which are not directly captured in the picture.68 Expecting a photograph of an apple to smell, expecting a known hu-man to react when you call to him or her on television, etc., are examples of this phenomenon. To a pictorial novice the case of well known objects affords a bigger chance to realise both the referential nature of pictures, and their differences from the real world. However, due to unfulfilled expectations, recognition of this type should also be likely to fail, or break down.
For Zuri, most difficulties were posed by the intermediate level of abstraction:
primates versus other animals. This is because there are no single, or few, visual commonalities among primates that excludes all other animals, especially mammals.
There is rather a mosaic of traits, unevenly distributed across primate species, that constitutes the category. Should one expect a 4 year old gorilla in a zoo nursery to have, or be able to discover and apply, a concept for “primate” (prosimian, monkey and ape) as opposed to “animal” (mammal, bird, fish, insect etc.)? Especially if she is working in a reality mode, where there is little room to infer extra information to the pictures. It is difficult to draw any conclusions about Zuri’s categorisation abili-ties from these results. After all, birds in pictures do not fly, fish do not swim, and insects are not small, fast and crawling. If there is no real-life experience of the in-volved species, Zuri is even more confined to using only visual cues. To give Zuri a fair chance of categorising on an intermediate level, one ought to use real objects instead of photographs of living beings.
That said, from the data it is safe to say that whatever Zuri could retrieve from the photographs was enough to effortlessly solve the “concrete” and “abstract” dis-criminations, and with limited success the “intermediate” ones. Vonk and Mac-Donald (2002) conclude their study with, rightfully, questioning the value of divid-ing abstraction levels in terms of only the “breadth” of the categories that are to be discriminated. Ease of categorisation seems rather to be dictated by the ratio of fea-ture overlap within and between specific categories. In this sense the “intermediate”
level is often by far the most abstract one. (Compare “primates” with “mammals.”) I would hasten to add that the feature overlap within and between categories is also dependent on the interpretation the subject is able to make of the stimuli that is being used.
That categorisation is dynamic and varies with the stimuli and subjects tested is supported by a replication of the Zuri study with six orangutans (Vonk & Mac-Donald, 2004). All but one subjects were adults, touchscreen was used, and except for black-and-white pictures intended to control for the conspicuous colour of orangutans, all photographs were in colour. In this study it was found that the dis-crimination of primates versus other animals was not significantly more difficult
68 Even more properties can of course be inferred when viewing the picture in a pictorial mode, where imagination “fills in the gaps.” It is possible that e.g. orangutans seem even more real in that kind of mode than in reality mode.
than orangutans versus other primates/humans, or animals versus non-animals/foods. In fact, animals versus non-animals was the more difficult discrimi-nation for the orangutans, although, just as in Vonk and MacDonald (2002), food versus animal discrimination seemed to be easily acquired.
On all problems large individual differences were found. This might reflect not only individual differences in learning, but in interpreting the stimulus photographs as well. It is feasible that some pictures made sense so some subjects but not to oth-ers. Interestingly, the adult subjects did not perform unequivocally better than the 2 year old orangutan, which in turn performed better than the 4 year old gorilla.
General exposure to pictorial material during one’s lifetime, which is unavoidable in captive settings, did not seem to have influenced pictorial ability. Experience with the species and objects in the pictures did not seem to have affected performance overall, but is likely to account for some of the variation between individuals.
Using the discrimination task in a touchscreen setup as above, Vonk (2002) did also test concepts for social relationships. Two of the orangutans and the gorilla Zuri were tested on their ability to choose mother-infant pairs as opposed to individuals in other types of configurations, as well as single individuals. Positive and negative choice photographs depicted the same species. In this task orangutans and gorillas were used as stimuli in separate sessions, while unfamiliar species were placed in mixed stimuli sets. In training sessions all subjects performed on par with training in the previous studies, reaching criterion of 80% correct in 20 to 130 trials.
Only one of the animals, an orangutan, showed good transfer to novel photo-graphs. The two remaining subjects were above chance but did not reach criterion in the first transfer sessions. In subsequent sessions all three subjects performed well on novel photographs of unfamiliar species. The orangutans, but not the gorilla, were thrown by a photograph of an adult male gorilla with an infant. It was categorised as a mother-infant pair. Zuri, although reared by humans, had experienced conspecifics of both sexes. This episode highlights the necessity of experience with the depicted stimuli, as well as being able to transfer that experience to its recognised version in a photograph. It was also found that Zuri sometimes preferred to choose gorillas of her own age, regardless of context. This effect did not transfer to other species. This is another example of how the real world can affect the pictorial world. Perhaps es-pecially when viewed in a reality mode. When pictures are truly pictures it makes little sense to prefer to indicate one before another, especially in a non-communicative context.