Source: Premack, D. (1980). Representational Capacity and Accessibility of Knowledge: The Case of Chimpanzees. In M. Piattelli-Palmarini (Ed.), Language and Learning: The Debate Between Jean Piaget and Noam Chomsky (pp. 205-221). Harvard University Press.
CHAPTER NINE
Interspecies Comparisons of Cognitive Abilities
In his book Intelligence in Ape and Man, published after this symposium, David Premack explains that his motivation in studying the cognitive capacity of chimpanzees was to obtain a description of our own capacities as seen by another species. To some extent Premack is starting a Copernican revolution in psychology: it is no longer a human investigator describing the cognitive structure of another organism, but the reverse. Self-knowledge as imposed by the Socratic motto is pursued through an indirect, painstaking, and unprecedented strategy. An intense flow of communication is first established be- tween the experimenter and the ape, whereby problems are posed, questions asked, and capacities measured and refined. The dominant mode of interaction is clearly the cognitive one. In contrast to more “playful” and “spontaneous” styles of dia- logue established with chimpanzees by other animal psychol- ogists such as Allen and Beatrice Gardner, Premack admits here that “my chimpanzees, poor things, are the victims of a very scholastically inclined experimenter.” This pedagogical bent is congruent with Premack’s interesting assumption that “it is replay, and, of course, problem solving, that language serves uniquely well, and these functions are no less basic than social communication.” Indeed, the latter, Premack specifies, “can get on nicely without language.”
Once the regular “classroom” interaction is established between the experimenter and the chimpanzee, a lot of interesting things happen. Premack describes his methodology D. Premack, Intelligence in Ape and Man (New York: Halsted, 1977). 203 204 / Language and Learning in detail in his book. In this chapter he describes some aston- ishing results (perception of causal links, symbolic replay, hints of structure-dependent transformations, and so on) obtained by careful “interrogation” of his chimpanzees. The conclusions that he draws concerning the cognitive structures available to the chimpanzee and the epistemological presuppositions that have oriented his research deserve a brief commentary here. In the first place, Premack rejects, on the basis of solid experimental evidence, three hypotheses: (1) that cross-modal associations (visual-auditory, visual-tactile, olfactory-visual, and so on) are unique to man (2) that categorial perception is unique to man; and (3) that brain lateralization is unique to man. As he says, “All three are rather good examples of whet one means by innate factors, but they do not appear to be language-specific factors, or in any case factors unique to man..” Second, instead of pursuing a search for a general developmen tal mechanism or a general learning theory, Premack takes the more cautious path of attempting to identify “factors that both participate in human intelligence and play a role in language™ -factors such as representational capacity, accessibility of knowledge, causality, and synonymy. It turns out that all these factors are present, in an unsuspected degree, in the chimpan zee. This finding appears to Premack to be, at least potentially. very important and relevant to the problems raised in this debate between Chomsky and Piaget. For instance, representa tional capacity appears to Premack to be “a more pivotal ca pacity than either syntax or intentionality.” Echoing a previous remark made by Piaget, when referring to the all-important “semiotic function” of which Piaget considers human language to be a subdomain, Premack states: “If we view language as a family of representational systems of which human language is one variant, we can retrieve the study of language from its ethnocentric limitations.” To which he adds a methodological stand, a sophisticated one at that: “It is profitable to take this view even though, on this planet, such a family may have only one natural member, that is, speech.” The careful interspecies comparison of cognitive abilities can, therefore, lead to the perception of human language, unique as it is, as a member of a potentially vast class, even if this class, for contingent reasons (ascribable to the vagaries of phylogenesis), contains only that member. Such a methodological stand is rich in consequences, suggesting further experiments and criteria of interpretation of available data. The second part of Premack’s paper and the ensuing discussion spell out these consequences in detail. Comparisons of Cognitive Abilities / 205
Finally, I would like to attract the reader’s attention to the antibehaviorist argument that Premack develops in the last section of his paper. Originally formulated by Premack’s “old teacher” Herbert Feigl (one of the most authoritative episte- mologists and philosophers of science of our time), the argu- ment attempts to revalue “weaker tests” as a sufficiently reliable basis to infer the existence of a given capacity. The behaviorists’ obsession to obtain only irrefutable evidence by strong negative tests (to prove that such and such an experi- ment is totally compatible with the absence of a given capacity) has paralyzed animal psychologists for a long time. It is indeed possible, according to Feigl and Premack, to “get at what’s what” by avoiding the limitative criteria of “nothing but” and “something more” (that is, behaviorist sterility and pure arm- chair speculation). What is at stake here, as is well known to philosophers of science, is a demarcation criterion between the best available evidence and the best conceivable evidence. Experimental scientists, unlike pure mathematicians, have to be content with the former and avoid being paralyzed by the utopia of the latter. This chapter, like the rest of Premack’s work, shows us where intelligent, rigorous, and creative use of available evidence can lead.
Representational Capacity and Accessibility of Knowledge: The Case of Chimpanzees
By David Premack
Unfortunately, I will have little to say about the question that is of uppermost interest to many of us, namely, does human syntax arise from genetically unique linguistic factors or from general cognitive factors? Ten years ago the former view would have been strongly supported, today it is in decline. The change in view has led to new research, it is now possible to give a richly detailed account of the development of mother-infant preverbal communication. But there is no demonstration that the pre- verbal communication is a necessary condition for adult syntax. 206 / Language and Learning
Moreover, for ethical reasons we cannot find out in a simple, direct way (we cannot withhold preverbal communication from human infants) and thus an answer will depend on finding indirect test models. In the meantime, it is already clear that preverbal communication is not a sufficient condition for adult syntax. Preverbal communication has been shown in the mother-infant dyad in monkeys,¹ let alone apes; and indeed, as regards early social communication, it may be difficult to tell one primate from another. Yet only one primate develops nat- ural language. It seems to me questionable whether the social communication view of language now replacing the unique factor view is any better founded than its predecessor. There is a sense in which the transition we are now witnessing is the re- placement of one vogue by another.
Another quite different approach to resolving the issue would be to obtain grammars for any nonlinguistic domain-play, tool making, art-that is sufficiently well formed so that units could be identified and the sequences of units thus determined. A comparison of the formal properties of the grammars for the nonlanguage cases with those of the grammars for language should help to decide how much the behaviors of language and various kinds of nonlanguage may have in common. Although the proposal to do this has come from several quarters, to my knowledge it remains to be done. Recently Kim Dolgin, Dan Osherson, and I have taken the first steps toward doing this specifically for play, in both chimpanzees and children. There are at least two possible outcomes that would be of interest. If we were to succeed in actually writing a grammar for play, then the desired comparisons with language could be made. Alterna- tively, it may not be possible to write a grammar for play, the problem of units may be insuperable, or there may be other diffi- culties. If so, and if we could show that the difficulty was not trivial but arose from deep factors, there would be a suggestion that play and language are not comparable systems, and thus are not likely to influence one another. This, of course, is only a possibility-a remote one, perhaps-but I mention it to flesh out the possibilities that accompany this comparison, for the range seems to be greater than has been considered. There are a few pieces of evidence of a circumstantial nature that I think bear on the question, not of innate factors but of innate factors that are linguistically unique: I will mention three. First, it has been claimed that cross-modal associations are difficult, if not impossible, in nonhuman species. This claim been falsified not only in apes but more recently even in Comparisons of Cognitive Abilities / 207 monkeys (which, it is my intuitive impression, are farther from the ape than the ape is from man). Cowey and Weiskrantz recently reported a nice experiment in which they showed visual-tactile associations in monkeys. In the ape, not only visual-tactile associations but the most interesting case, visual- auditory, can be demonstrated. We recently found that apes with whom speech had been used informally in the course of their formal training on plastic words could subsequently compre- hend speech alone. The comprehension was very limited, was largely confined to words, and was far less accurate than for the plastic words, but then their “training” on speech was also limited and entirely incidental. In any case, the claim restricting cross-modal association to man can be retired. A second claim that may also be heading for retirement is that categorical perception is unique to speech sounds, on the one hand, and to man, on the other. This two-headed claim is the basis of the view that man is uniquely equipped with speech detectors. It appeared that this claim was corroborated when Eimas and others found categorical discrimination of speech sounds in human neonates. But the relevance of the infant data for the assumption of species-specific speech detectors has been called into question by additional data of two kinds. First, nonhuman species have been found to discriminate certain speech sounds in a categorical manner. Rhesus monkeys and chinchillas, whose inner ear structure resembles man’s, discrim- inate the voice-voiceless distinction in the contrast between /t/ and /d/ in essentially the same categorical manner as the adult human speaker. Second, categorical discrimination is not only not unique to man, but also not unique to speech sounds. Hu- man adults discriminate between a bowed and a plucked instru- ment in a categorical manner. We await additional tests to see if the claim for speech detectors being unique to man can be re- constituted on a more subtle basis. But in the meantime, there is the suggestion that we should think not in terms of speech de- tectors but of auditory hardware that was on the evolutionary shelf for some time before it was picked up by man and incor- porated into language. Thus there exists an innate factor, but not a language-specific innate factor.
Third, quite recently, there have been at least two reports of anatomical lateralization in apes. Lemay and others have re- ported hemispherical differences in the ape’s brain similar to those in man, though far less marked they also reported the absence of such differences in the monkey.10 Now of course, we do not know whether or not the anatomical difference in the 208 / Language and Learning ape is associated with a functional difference, or if it is, whether the difference is comparable to the one in man. From time to time, Gazzaniga and (more recently) Levy have threatened to hunt for functional lateralization in chimpanzees, and on the condition that they leave their drills and knives home, they have been welcomed. Incidently, early studies did not show handedness in the chimp; that is, individual chimps are handed but the distribution of handedness in the population was ran- dom. It could do no harm to look again, however, since this is the result of an early study.
In summary, three factors-intermodal association, speech de- tectors, and anatomical brain lateralization can no longer be unequivocally assigned uniquely to man. Two of the three factors clearly participate in language; the third may not. All three are rather good examples of what one means by innate factors, but they do not appear to be language-specific factors, or in any case factors unique to man.
I will try to talk (very programmatically, I’m afraid) about factors that seem to characterize human intelligence and see whether we can find these factors elsewhere. I think that there are two ways one can make the argument about innate factors and the necessity for them. One is the argument that Chomsky has made in a very elegant presentation: there are some formal properties of language, but there doesn’t appear to be any gen- eral developmental mechanism (GDM) that could generate these factors. This is a legitimate form of argument, but a bit at the mercy of what the next bright fellow might do. Even now, a clever young man, let’s call him Goldschmidt, may be sitting in his attic, figuring out how to build GDMs that will realize Chomsky’s formal properties. I prefer an approach that is less dependent on what Goldschmidt may or may not do-an ap- proach that says: here are the elements that participate in hu- man intelligence this is the time at which we find them in the infant; it is too close to zero for experience to play a pro- found or exclusive role, and moreover, this is the distribution of these elements over the species. So, Goldschmidt, you have a delightful hypothetical GDM, but it is irrelevant to this par- ticular argument.
Let’s look at three or four factors that both participate in human intelligence and play a role in language. Consider repre- sentational capacity, accessibility of knowledge, causality, and (time permitting) synonymy. Memory is obviously critical for both language and representational capacity. So the first ques- tion we might ask is, what is the quality of information a species Comparisons of Cognitive Abilities / 209 can store? If a species cannot store a powerful representation of the world, then even though we might, in some sense, teach it words, the words would be ineffective; they could not be used to retrieve much, since there would not be much information stored in the first place. The power of the word is limited by how much information the species can store.
POWER OF THE WORD
To assess the chimpanzee’s ability to remember the attributes of objects, on the one hand, and the amount of information it could associate with their names, on the other, we used fruits and plastic objects as “names” for the fruits. We divided the fruit into a number of pieces we gave the chimpanzees one piece and then asked them to identify other pieces that be- longed to the same fruit. We assumed that if one knew a great deal about an object, one should need only a small sample to identify it. If given only a stem or a seed or even a taste, an in- formed animal should be able to identify the fruit from which the sample was taken.
What is the difference between what an animal can perceive about an object and what it can reconstruct from memory? When an animal is required to match apple and redness, the object and the color sample match, giving a measure of percep- tion. If, however, the sample remains red but the apple is now painted white, the object and the color sample will no longer match. Instead, they only match on the basis of information reconstructed from memory. Tests that require reconstructing information from memory do not specifically depend on dis- torted alternatives, however. The relation between stem and peel, stem and seed, seed and peel; color and shape, shape and size; and color and stem, shape and seed, and so on, are all un- distorted cases of this kind. Items in these pairs do not share common features but are related simply through being attributes of the same object.
In the present tests, fruits were divided into four canonical components and two features: wedge, stem, peel, and seed; color and shape. Taste was added as the one nonvisual attribute. Eight fruits were divided in this manner: banana, orange, apple, lemon, peach, pear, grape, and cherry. The chimpanzee was given one or another of the features as a sample, along with two other features as alternatives, and was required to select the correct alternative. For instance, the animal was given an apple seed as the sample, along with an apple stem or a pear stem, and was to select the apple stem. 210 / Language and Learning
After completing the series of approximately twenty-4 individual tests with each of the four subjects, we were able to rank-order the components and features according to their in formativeness. Not surprisingly, the whole fruit was the most informative cue. Color and peel were next, followed closely by taste, after which there was essentially a tie between shape, wedge, and stem, and last came seed, the least informative cue of all. Sarah, impressively, was able to use all the cues correctly, but the other three chimpanzees were best able to identify the source of the attribute of a fruit from its color or peel, and least able from its seed.
In the next test series, actual parts of the fruit were given as samples, and the alternatives were plastic words that named the fruits. The results of these tests were unusually clear-cut. Words provided as much information to the animal as did the presence of a whole fruit; furthermore, words were more informative than actual parts of the fruit. In another test series, we used names of colors rather than of objects and obtained identical results. For instance, we could substitute the word red for an actual instance of red in the matching test without loss of ac- curacy. Since in the ape, the word substitutes vigorously for its referent, it seems proper to speak of the “power of the word” for the chimpanzee and not only for man.
The fact that fruit and color names could serve as substitutes for their referents, without any loss of accuracy, shows that a major consequence of giving arbitrary items (such as pieces of plastic) linguistic prerogatives is the transfer to the arbitrary item of some or all of the information that an animal has con- cerning the associated object. Under what circumstances does this transfer of information take place? Perhaps it occurs only after the piece of plastic has been used in a wordlike way, to request or describe the object a certain number of times. That would be the only tenable hypothesis if the only way to produce names was by associating them with their referents in one lin- guistic context or another. However, we already know that this procedure can be short-circuited. Names can be generated more directly by instructions of the form “X is the name of Y,” where X is a previously unused piece of plastic and Y an unnamed ob- ject. Following an instruction of this kind, Sarah used X in all the ways she used names introduced in the more standard fashion. Thus the effect of instructions such as “X is the name of Y” must be to transfer to X some or all of the information the chimpanzee has stored in his memory about Y. This fact clari- fies some of the power of language and at the same time sug- Comparisons of Cognitive Abilities / 211 gests the kind of intelligence a species must have in order to acquire it.
To qualify for language, a species must be capable of storing a rich representation of Y; if not, the information transferred to X would be weak, and the name would be a poor substitute for the referent. In addition, instructions of the form “X is the name of Y” must have the force of transferring to X some- ideally all of the information that the subject has stored about Y. These are not the only capacities a species must have to qualify for language, but they are two that seem to be basic. Sarah was capable of displacement, of comprehending state- ments about “things that are not there.” When given the in- struction “brown color of chocolate” as a means of introducing “brown” and subsequently told “take brown,” she performed correctly, choosing the brown disk from the four offered. The chimpanzee’s ability to comprehend statements about “things that are not there” derives from its demonstrated ability to store adequate representations of items and to use words to retrieve the stored information. In substituting, say, the word apple for an actual apple without loss of accuracy in all the matching tests, it gave direct proof of this ability. Displacement is not a uniquely linguistic phenomenon but the consequence of a cer- tain quality of memory.
REPRESENTATIONAL CAPACITY
Let us turn now more directly to representational capacity. Al- though the language-trained ape has so far shown painfully lit- tle syntactic competence, it has shown a remarkable representa- tional capacity. Beguiled by the linguist on the one hand and the semiotician on the other, we have tended to overlook representa- tional capacity, which in my view is a more pivotal capacity than either syntax or intentionality.
Representational capacity is the ability to judge the relation- ship between actual events and representations of them. For instance, one can place a red card on a green one and ask the chimpanzee “Is red on green?” or “What is on green?” (“? red on green” or “? on green”). In answering these two question forms correctly, the chimpanzee demonstrates that it recognizes the relation between them, a red-card-on-a-green-card, and the representation of the item, “red on green.” The chimpanzee can even answer questions in the absence of the colored cards, demonstrating that it can remember visual situations and recog- nize representations of the absent situations. If we view language as a family of representational systems 212 / Language and Learning of which human language is one variant, we can retrieve the study of language from its ethnocentric limitations. It is profit able to take this view even though, on this planet, such a family may have only one natural member, that is, speech. Neverthe- less, we can imagine other variants, we can find suggestions of other variants (in the developmental stages of the acquisition of speech by children and in pathological human populations); and we can train or synthesize other variants in nonhuman species as we are now doing with apes. While holding this re- laxed view of language, however, it is still questionable to con- sider bee communication as language. Ordinarily, the contrast between bee and human language is made on the grounds that only one of the two systems is learned, but this is a dubious con- trast, since critical aspects of human language, including parts of both syntax and phonology, are probably not learned. More important, even if the bee’s unique system were learned it prob- ably would not qualify as language. The two systems can be better contrasted by asking if the bee shows any suggestion of representational capacity.
Suppose a scout bee were to gather information about the direction and distance of a food source from its hive. The bee encodes this information in its dance, and a second bee decodes the dance; but could the bee, when shown its own dance, judge whether or not this dance accurately represented the direction and distance of the source of food? Could the bee recognize that dance as a representation of its own knowledge? If a bee could judge between the real situation and a representation of that situation, it would be possible to interrogate the bee, just as we can interrogate the ape. A species that can be interrogated, such as the chimpanzee, is well on its way toward being able to make true-false judgments. But I know of no data that even faintly suggest that the bee can recognize the dance as a representation of its knowledge. While we wait for the critical experiments to demonstrate such an ability in the bee, we must adopt an agnostic position with regard to the language ability of bees; we must take the position that the bee has a code, a correlation between items inside and outside its body-not necessarily a language, since language depends on representational capacity. A problem also arises from the tendency to call the elements of the bee’s dance symbols. With the chimpanzee, it is possible to decide in a straightforward manner whether or not a lan- guage element operates as a symbol. If a small blue piece of triangular plastic is consistently associated with an apple in the chimpanzee’s daily experience, whenever the chimpanzee later Comparisons of Cognitive Abilities / 213 wants some apple, he will put the blue piece of plastic on his writing board and be given an apple. If an apple is present and the trainer asks the chimpanzee “What is this?” (in effect), the chimpanzee will answer by placing the blue plastic on the board. How do we decide if the chimpanzee is using the plastic shape as a symbol? We can perform two tests on the chimpanzee, ob- taining what amounts to a features analysis of apple. Like us, the ape sees the apple as red (rather than green), as round (not square), as having a stem (rather than not), and so forth. When we replace the apple with the triangular piece of blue plastic, once more offering the same alternatives and now obtaining a features analysis of the would-be word, we get the same answers as before. The ape tells us that the blue plastic triangle is red, round, and has a stem, and so on. The chimpanzee’s analysis is compatible with the view that the ape is judging not the plastic triangle itself, but what the blue triangle stands for. If the ape could provide no further evidence of linguistic function, we would have to scale down our interpretation of these results; but we accept the interpretation (though only tentatively at first) because the linguistic performance of the ape does not end here but goes on to approximate ever more complex human performances. For example, just as at one stage the ape is able to make judgments about the agreement between an item and a representation of that item, so at a later stage the ape (Sarah) can make judgments about the agreement be- tween two representations. She subsequently made judgments about the relation between sentences such as “apple is red” and “red color of apple.” Thus a capacity for judging between two nonlinguistic items (as in the causality tests, see the section on causal inference) advanced to making judgments between one linguistic and one nonlinguistic item, and culminated finally in synonymy, where the chimpanzee compared two linguistic representations for similarity.
These are only some of the several tests that show the ape to have an impressive representational competence. Now, if you believe as I do that competence means in part an indigenous disposition to exercise the competence, then it is embarrassing to find so little evidence for spontaneous symbolization in the ape. Thankfully, there is a little evidence of this kind from Sarah, not enough to conceal the embarrassment but enough to suggest where we might look to find more. Sarah was offered a photograph of a chimpanzee face that was cut up into eyes, nose, and mouth, to see if she could reassemble it, which she proved able to do and which is interesting in its own right. 214 / Language and Learning
While being given these tests she was also given an opportunity to wear hats and to view herself in a mirror, which she very much enjoys. Incidentally, the ape, unlike the monkey, can recognize the image in the mirror as itself, the monkey stares at the mirror while continuing to handle the mirror, but after about ten hours the chimpanzee ceases to handle the mirror and begins to handle itself while looking in the mirror.11 Now, at a time when Sarah was looking in the mirror while wearing hats, and having a lot of fun, she was subsequently given a disas- sembled face of a chimpanzee. She put in the eyes, the nose, and the mouth, making her usual veridical reconstruction; then she paused, picked up the mouth, turned it over, and put it on top of the head like a hat. When we repeated the procedure in an experimental way, we found that we could evoke the sym- bolic play reliably, provided only that we preceded her puzzle experience with an opportunity for her to view herself wearing hats.
In producing a visual form comparable to the one she had seen earlier in the mirror, Sarah would seem to have been en- gaging in a complex version of a form of behavior in which she had engaged since infancy. For example, when Sarah was about 18 months old, a woman entered the nursery wearing a distinc- tive wool skirt. Sarah palpated the skirt for a moment and then rushed over to her bed and stroked her blanket, which was also wool. Having no language at the time, Sarah had no other way in which to announce or celebrate the discovery of an equiv- alence. Sarah went even further with the chimpanzee face, for there she did not merely detect an equivalence she produced one, or at least helped it along. The face in the mirror and in the picture were only partly comparable, she changed the one, making it more like the other.
All of this behavior would seem to be an example of a basic disposition that takes many forms, and that can be described in several ways. One can speak of a disposition to discover (or produce) equivalences; to imitate oneself; to reproduce previous experience, sometimes with a change in mode. Thus, we see children play and at the same time speak, describing their play, and we see the same thing in the chimpanzees, they too play and then describe their play with the plastic words. Indeed, of all the devices that might be used to reproduce previous experience, none could be more efficient than language. Language is the ideal replay device. People emphasize social communication in canine, yet social communication can get on nicely without replay, and, of course, problem solving, Comparisons of Cognitive Abilities / 215 that language serves uniquely well, and these functions are no less basic than social communication.
Let us turn now to one of the most critical and least under- stood aspects of intelligence, the accessibility of knowledge. Piaget has, of course, dealt with this problem as he has dealt with most problems of intelligence, so I shall do no more than sketch the problem in order to show its relevance to the animal case.
THE ACCESSIBILITY OF KNOWLEDGE
Is access to a cognitive map a sufficient condition for the use of an actual map? Since we have good reason to suspect that it is not, we must explain why a species could have a maplike representation of its home terrain in its head and yet not recog- nize the relation between its cognitive map and an actual map. The question need not be restricted to maps, of course. Why is it that some species can recognize representations, including representations of their own knowledge, and others, apparently, cannot?
From the ape’s success in remembering the location of hidden objects, we infer that it has a cognitive map.12 Yet we can make the same inference for other species as well, since spatial rela- tions is one of the most widely demonstrable forms of knowl- edge. Rats13 and even insects have been shown to have in- formation of this kind.
The differences in the ability of various species to use ex- ternal representations may be due to three factors. The first factor is the form of the information and the level at which it is stored. If the bee’s dance, for instance, did not code for direc- tion specifically but rather coded the distance, the quality, and/ or the quantity of the food source, these factors could be at- tributed to motivational systems, controlled by the gut, taste receptors, or other visceral sites, with little, if any, neural repre- sentation. Second, regardless of how information is stored, species may differ in the extent to which they have access to it. This factor could not only influence the ability of the species to use maps or language, but also affect the degree of self- reflective behavior in which it could engage. Third, even if in- formation were stored in a favorable form and a species had good access to it, we, an alien species, might be unable to design an appropriate map, one that would resemble the form of in- formation stored by the species. At present, we are far from being able to decide which, if any, of these factors-form, ac- cess, or appropriateness-may account for the limitations of 216 / Language and Learning various species. Perhaps the ability of a species to use maps or language depends primarily on the presence of at least two different forms of representation or information storage. This attractive suggestion was offered by my colleague Randy Gallistel during a conversation over lunch. If a species can store information not only in pictorial form (to which in- sects may conceivably be restricted) but also in propositional form (demonstrably the case for man and chimpanzee), then the same information could be stored in both forms. Cases of this kind could provide the opportunity for a species to trans- late one form of internal representation into another. Experience and instruction may help an animal to recognize the relation between an item and a representation of that item, since, in principle, this is similar to the ability to recognize the relation between two equivalent internal representations. Cer- tain kinds of information may naturally lend themselves to an image of maplike form, other kinds to a propositional form (these extremes may never be represented in both ways). Still others may be intermediate with respect to form, and may be repre- sented both as image and as proposition more or less auto- matically, depending on the occasion.
In 1932, Tinklepaugh conducted a series of memory tests with apes and monkeys using a kind of information that could be easily modulated so as to potentiate an image-form of storage on one occasion, a propositional-form on another, and possibly both forms on further occasions.15 Tinklepaugh tested the apes by arranging sixteen pairs of containers in a circle (with a diameter of 7 meters), baiting one container of each pair while the ape observed from the center of the circle, and then, after varying delays, releasing the ape to find the baited containers. The sixteen containers, consisting of a variety of painted and unpainted wooden boxes, tin cans, and cups, were not confined to a special order from trial to trial, and baiting of the containers was essentially random. For an animal to succeed in this experi- ment, it seems that an image-form would be the most practical form of storage. The circular arrangement of the containers would make the coordinate system on which maps are based less suitable, since half the pairs of containers would have the same value on the abscissa and the other half the same value on the ordinate, a factor that would reduce their identifiability. A propositional form of storage also seems unsuited for the Tinklepaugh experiments. It could be used, in theory, but the lack of systematic relations between the type of container and the baiting system greatly reduces the advantages of a proposi- Comparisons of Cognitive Abilities / 217 tional format, which works most efficiently if information is pat- terned or systematic. For example, if for all the tin containers, the left member of each pair were baited, and for all the wooden containers, the right member of each pair were baited, then the information could be stored economically in a propositional format. Moreover, this propositional format does not seem to involve any predicates foreign to the conceptual nature of the chimpanzee. In fact, Sarah was successfully taught quantifiers- the use of various modifiers comparable to “wooden containers” and “tin containers”—and although she was not taught “right- left,” she learned to label distinctions such as “top-bottom,” thus storing information of a kind not unlike that required for the Tinklepaugh studies.
In principle, it would seem possible to change the animal’s form of storage-image, map, propositional-by modulating several parameters: the degree of pattern in the information, the shape of the geometrical arrangement, the reliability of the posi- tion of the containers after baiting is complete. At intermediate values of these parameters, animals capable of more than one form of storage may be inclined to use both forms, setting up an equivalence between two different internal representations. This strategy may help the animal to recognize the relation be- tween a situation and the external representation of the situa- tion. Although this hypothesis seems to elude a direct test at the moment, indirect tests are possible and may lead to more satisfying demonstrations later.
CAUSAL INFERENCE
In acquiring language, one acquires labels for existing concepts; this proposal can be tested with admirable directness in many cases. Can the animal discriminate between conditions that exemplify “same” and “different”? “all” and “none”? “red” and “black”? If so, according to the proposal, it should be possible to teach the animal names for “same-different,” the quantifiers, and the like. Not all concepts are as simple as these, however, and some of them must be approached differently. For instance, we approach the “if-then” or conditional relation by observing that the conditional sentence is a way of expressing a causal relation: “If you drop that, it will break.” “If you smile at Mary, she will smile back.” “If you touch that, you’ll get burned.” These sentences and the infinitely many possible others like them express a causal relation between the antecedent and the consequent. Only a species that made a causal analysis of its experience would use sentences of this form productively. 218 / Language and Learning
Hence, we designed a simple visual test to answer the question: does the subject make a causal analysis of its experience? The subject was given an intact object, a blank space, and the same object in a changed or terminal state, along with vari- ous alternatives, and was encouraged to complete the sequence by placing one of the alternatives in the blank space. For ex- ample, the subject was given items such as an intact apple and a cut apple, a dry sponge and a wet one, and a clear piece of paper and one with writing on it. The three alternatives given the subject consisted of a knife, a bowl of water, and a writing instrument.
Three of the four chimpanzees tested in this way required no more than general adaptation to the test format before respond- ing correctly. Their ability to place the knife between the intact and severed apple, the water between the dry and wet sponge, and the pencil between the unmarked and pencil-marked paper showed that they correctly identified the instrument needed to change each object from its initial to its terminal state. Simple as this outcome is, it can be given a stronger interpretation than may first meet the eye. The visual sequences are infinitely am- biguous: each can be coded in indefinitely many ways, such as red-blank-red, one-blank-two, round-blank-flat, large-blank- small.
Not only the test items but also the three alternatives are subject to indeterminately many codings. “Knife,” for instance, need not be read as knife (instrument that cuts) but can be coded as sharp, metal, long, shiny, and so on, and the same holds for the other alternatives. However, the subjects evidently did not code the sequences or alternatives in these ways, since they con- sistently chose alternatives compatible with only one coding, namely, how do you change the object from the intact to the terminal stage? With what instrument do you produce the change? Because the subjects read the sequences in a specific and consistent way-finding the same question in each of the sequences-I infer that they have a schema, a structure that assigns an interpretation to an otherwise infinitely ambiguous sequence.
The apes’ ability to respond in this fashion was by no means limited to familiar object-implement pairs. They performed equally well not only on pairs they had never experienced, but even on pairs that were anomalous or nonsensical, such as apples that had been written on, sponges that had been cut, and eating paper that had been dunked in water.
It is also noteworthy that the visual sequences were by no Comparisons of Cognitive Abilities / 219 means iconic representations of the actions tested. Cutting, wetting, and marking are analog processes in which an agent brings about a continuous change in an object. In cutting, for example, an agent applies a knife to an apple, exerting pressure until the apple divides. The test items did not portray the gradual division of the apple, but presented only the digital high- lights of the analog process, and did not present the agent at all. Nevertheless, the chimpanzees evidently recognized the test sequences as representations of the actions. If the tests had failed we might then have considered using motion pictures or other iconic forms of representation; but the animals suc- ceeded despite the abstract form of the representation. Thus, chimpanzees not only have a schema for cause-effect relations, but they have one that can be activated by noniconic represen- tations.
Because these tests dealt only with the physical domain, they leave open the question of whether or not the chimpanzee can recognize cause-effect relations in the psychological or social domain. The lack of appropriate stimuli has prevented us from making such tests, yet it is easy enough to describe the form such tests would take. For example, in one test, the three pic- tures would consist of Elizabeth begging food from Peony, a blank frame, and Elizabeth and Peony playing, hugging, and engaging in mutual grooming. The alternatives would include: (1) Peony ignoring Elizabeth’s request, (2) Peony sharing with Elizabeth, (3) Elizabeth stealing Peony’s food, and so on. Of these alternatives, only Peony sharing with Elizabeth would be compatible with the harmonious outcome in frame 3, and the chimpanzee’s appropriate choice in this and comparable tests would indicate that it could recognize representations of social as well as of physical actions. Notice, incidentally, that the test is designed so that selection of the missing frame can not be based simply on knowledge of physical action. If in frame 3 both animals were shown to be eating, one could conclude on physical grounds alone that Peony must have shared with Elizabeth. But since neither animal is shown eating, the content of the second frame can only be inferred from the social char- acter of the behavior in the third frame.
Let us assume for the sake of discussion that the apes can pass the social tests as they have passed the physical ones, which may not be too risky an assumption given the evident social intelligence of the chimpanzee. If the ape can recognize repre- sentations of both physical and social actions, perhaps it can take the next step and recognize higher order structures that 220 / Language and Learning are composed of physical and social actions. Physical and social acts are the building blocks of stories, novels, tales, and the like. Indeed, all narrative prose is formed by appropriately combin- ing physical and social acts. If a species can recognize the basic elements of which stories are formed, perhaps it can also recog- nize stories themselves.
The individual’s ability to recognize a picture story can be revealingly tested with verbal procedures, such as those Walter Kintsch, his students, and I are now using with 3- and 4-year-old children; but it can also be tested in at least some degree with nonverbal procedures. First, the animal can be given story books, of appropriate simplicity, in some of which the pictures are in appropriate order and in others of which they are out of order. A preference for books with appropriate order would be sugges- tive. Second, the animal can be given serial learning tasks re- quiring that he learn to arrange, say, five pictures in a designated order. In some cases the designated order is the normal order in which the pictures would occur in a story book, and in other cases it is not. If serial learning is sensitive to the sequence of the story, so that the animal learns the former faster than the latter, this too can be taken as evidence of the ability to comprehend picture stories. Comprehension of this kind is important, since it would indicate that the animal can not only divide his expe- rience into causal units, but can use still higher order schemata to organize the causal units.
ACCESSIBILITY TO INFORMATION AND CIRCUMSTANTIAL EVIDENCE
One might do a still weaker test, not to prove causal inference, but to produce circumstantial evidence hinting at the merest possibility of causal inference. I propose this approach as an antidote to any approach which, like behaviorism, is unbalanced in its eagerness to avoid false positives, that is, so intent on its quest for an antiseptic position that it commits one false nega- tive after another. Is it not possible to get at what’s what, as my old teacher Herbert Feigl used to call it, without ricocheting wildly from nothing but to something more? Often it does not seem so.
A hint of a capacity for causal inference might be obtained from a test of this kind, which is thus far hypothetical. We know that even nonprimates can learn on an observational basis, al- though exactly what they learn, merely a motivational change or something more cognitive, is not yet resolved. Let us do the experiment with rats, since they are sufficiently low in the Comparisons of Cognitive Abilities / 221 hierarchy for the question to be of interest. We give rats two scenes to observe, and see which of them is more influential. In the first, the rats watch a model rat push a marble off the edge of a table, the marble falls to a dish below, after which food appears in a cup. In the second scene, the rats watch a model rat push the same marble off the edge of the table, but now the marble rises to the ceiling, after which food appears in the cup.
Suppose the rats shown the first scene are more likely to imitate the behavior of the model (and we carry out those con- trols necessary to eliminate competing hypotheses concerned with novelty and the like). For species at this level of intel- ligence, this general kind of circumstantial evidence may be the strongest we can get that the species has any access to the knowledge or information that is stored in it. (Certainly we do not expect rats to produce conditional sentences, or to pass even any version of the visual causality tests described earlier, no matter how simple they may be made.) Even though rats do not leap wildly when faced with a small gap, or jump short for a large one, but behave sensibly in the world, they may have no access to the information underlying their sensible performance. And the failure may not lie in the form in which the informa- tion is stored, but simply in its accessibility-a factor about which we know too little.
I showed in a sentence, simpler than but logically equivalent to Chomsky’s first case (see Chapter 1), that Sarah pluralized not on the basis of physical features but on the basis of the mean- ing of the sentence. The same capacity is employed when the animal makes a judgment about the equivalence of two sen- tences: it is responding not to strings of verbal elements in terms of their physical properties but in terms of what they mean.* These are all examples of the animal’s representational capacity, its ability to process the pieces of plastic not at the level of their physical properties, but at the level of their meaning. *Editor’s note: Such an ability is what Chomsky has previously defined as “structure dependence.” Premack is asserting here that chimpanzees possess the ability to perform structure-dependent operations. The point appears to be controversial, however, as Scott Atran points out in the discussion that follows. (See also the exchange between Putnam and Chomsky in Part II.) Discussion
Bateson: It seems to me that we should pay close attention to
the latter part of Premack’s material. What seems to me out- standing in the whole story is what he referred to at one point as a time lag. He keeps saying that an ape can learn this or that, but obviously, the experiments have a sequential struc- ture to them which is in itself a learning sequence. This is, I think, very important. This is the best data I have ever heard in support of the fact of progress in learning, learning how to learn.
Premack: I’m not a great believer in the word “learning,” and I
am rather more in favor of Fodor’s argument (see Chapter 6). I think that the main thing human training is doing here for the chimpanzee is disclosing capacities that are present. The human intervention is a very modest contribution, we are setting the stage in such a way that the animal’s existing in- telligence can be expressed. I do believe that there is such a thing as learning how to learn, which of course is also found in species a great deal lower than the chimpanzee. But the point I was really trying to make is about the progression from being able to look at a sequence like an apple, a knife, and a cut apple, and judging that sequence relative to one’s previous performance: this already presupposes the very criti- cal psychological capacity of being able to recognize repre- sentations of one’s own behavior. I don’t know how far down the evolutionary ladder that capacity might be found, but I will hazard a guess that it might be strictly a primate phe- nomenon.
Bateson: The thing I’m trying to get across is that to recognize
representations is one thing and these play-repetitions in one sense carry the notion of recognizing representations. On the other hand, to be able to answer a question about represen- tations is much more than that.
Godelier: Let’s return to the first part of Premack’s discussion,
he shows that there is symbolization, and that it is an active process leading to the capacity of representing the form of one’s own behavior. Here is my question: the chimpanzee appears to be capable of symbolization, of organizing his ex- perience, of representing his own scheme of behavior, but it seems that he cannot transform the rules generating his be- 222
Comparisons of Cognitive Abilities / 223 haviors, that is, the rules of his social relationships. Piaget was saying that the function of symbolization-therefore evo- cation-appeared in the course of the second year of child- hood, and he attempts to connect this with sensorimotor practice as a creative practice, and so on. Now, it seems to me that these two examples, the one presented by Premack on the function of symbolization, and the other one on conser- vation presented by Piaget, show us that the symbolic func- tion is, from a certain point of view, already programmed. But what is programmed, then, is the potential capacity to transform-for example, for us, the human species-social relationships, that is, rules of behavior, institutions, which is a much deeper question than that of language. In this case, both the limited formal syntax of the chimpanzee and ours, which is much more complex, would be aspects of the capacity for a species to transform or not to transform the rules of its games, and thus to create any new social relationships, whatever they might be, or to act on anything whatsoever. I would like to ask Changeux a question concerning this problem: can we, with a perspective of the evolution of species and of the forma- tion of the human central nervous system, give an account of the massive capacity to create social relationships, and also to create the languages that are necessary to transform and rep- resent these relationships? At one fell swoop, the question is asked concerning, on the one hand, the capacities of syn- taxes and the possibilities of transformation, and on the other, the nervous system and its evolution, in which the differentia- tion of the species is seen. It is a question that permits one to introduce the function of symbolization, genetically pro- grammed. At the same time, however, it is not merely the function of symbolization, it is the capacity for a species to transform the conditions of its existence. From this point, the problem is to know whether the capacity to transform these conditions is programmed with the differentiation of nervous systems and the central nervous system that Changeux spoke about (see Chapter 8).
Personally, then, I would stand on Chomsky’s side in re- gard to the programming of capacities, but it would not only be a problem of specificity of language for man, it would be the problem dealing with the aptitude for our human species to transform its conditions of existence and the interplay of its behavior.
Premack: The experiment that would maximize the likelihood of that outcome has simply not been done. I was also asked 224 / Language and Learning whether the animal accepts passively the inputs that the world gives it, or whether there are situations in which it oper- ates on those inputs and transforms them. That question is also present in the cases in which the animal, having been given a picture of a chimpanzee’s face with a hat on, takes the puzzle and does not merely reconstruct it so as to produce a chimpanzee’s face, but rather operates on it in a way that produces an equivalence between the input and the given material. I consider this as a weak example of the fact of not merely accepting the input of the world passively, but rather operating on this input. This is admittedly very limited, but nonetheless it shows that the disposition to produce equiv- alence is involved in self-description, in images. Incidentally, this is what is involved when a husband returns from a trip and, while riding home with his wife, tells her what he did, and she tells him what she did both of these conversations appear to be expressions of interest by the husband and wife in each other, but I consider them as examples of replay, for which language is nearly the perfect instrument. I would very much like to understand why we have this deeply rooted disposition to produce equivalences between one experience and another in a somewhat modified form. Now, I think we already see a weak instance of that in the chimpanzee when he transforms the face, operates on it, so as to make it more like the image that he had before.
Papert: Let me mention first an experiment that was carried out recently by Stambak, Sinclair, and others relevant to the idea of schemas in Piaget’s sense.1 It consists of giving human babies a collection of objects, such as a glass, a box, or a small stick, and watching what they do with them. The interesting observation was that during certain periods, perhaps as long as a month, the child would become highly occupied with one activity belonging to a very small set. For example, for a while the child might get very involved with putting things into his mouth. Then after a while the child would turn away from that and get interested in another kind of activity, such as putting sticks into containers. I will use the word “frame” for the data structures that handle each situation (think of it as case grammars if you like): each frame has places for entities, its cases, for example a container, a thing to be put in, and an action, there is the in-frame (the container), the on-frame, and so on.
It does seem to be clear that a small vocabulary of very basic schemata of this sort either develops early in the child Comparisons of Cognitive Abilities / 225 or is innate. Now, it seems to me that people who want to find the innate mechanisms behind language and of conceptual manipulations would do well to look at these things rather than at complex, specially linguistic phenomena such as the specified subject condition.
Now, I would like to look at the situation of apple, knife, cut apple. To describe that as a “frame,” let’s say that it in- cludes three cases: the initial state, the final state, and the trajectory or an instrument. I think this is a better way to look at it than to think of it as knowledge about causality, or rather, let’s say that the knowledge about causality would come later and would be coded in terms of the prior coding of real situations according to schemata of that sort. The first question to ask is whether syntactic properties of the schemata have been pursued. An example comes from the suggestion that chimpanzees aren’t very good at word order; now, are they good at object order in the representation of the sche- mata? For example, does the chimpanzee care whether the cut apple is on the left and the whole apple is on the right? Premack: I regard a schema as a structure that is responsible for the interpretation, the reading that is given to these items, so that they do not end up being interpreted in the many arbitrary ways in which an apple, a blank, an apple can be read. They are read in a specific way which it is necessary to assume to account for the consistency of the results. This is what I mean by schema, and I presume it is very much like what Piaget means by schema: a structure imposing an interpretation on a set of items that would otherwise be susceptible to an infinite number of interpretations. In regard to the question of order, we have excellent evi- dence that the chimpanzee can respond to the order of the items in the visual sequence. To establish this, we first ac- quainted Sarah with actions the reverse of those on which she had been tested. Cut, wet, and mark were the actions on which she had been tested; we therefore acquainted her with join (patch together with tape), dry, and erase. Next we taught her that the visual sequences had to be read left to right. We did this by training her on sequences consisting of the same elements in the opposite order. For instance, we gave her the sequence: blank paper – marked paper; and the opposite: marked paper – blank paper. Her alternatives in both cases were the following: pencil, eraser, and cup of water. You can see that for the first case the answer is pencil (it will con- vert blank paper to marked paper), whereas for the second 226 / Language and Learning case, the answer is eraser (it will convert marked paper into blank paper). We trained Sarah on a limited number of such cases-cases that required her to pay attention to order-and then tested her on sixty new cases. She was correct on forty of the sixty cases with three alternatives; this is highly sig- nificant (P < 0.001).
Papert: It seems that there is a small vocabulary of a few very important discrete schemata that play a crucial role in the structuring of the child’s environment.
Atran: As far as I can see, there is no compelling evidence to indicate that such schemata, which are common to the child and chimpanzee, could constitute the basis for the construc- tion of human language. Even the correlation by analogy is extremely weak, since the examples given simply suggest that a “schema” does little more than present items in succession and restrict the possibility of “reading them off,” or repre- senting them in an infinite number of ways. To look for analogical “equivalents” of grammatical rules in the behavior or the representation of causal behavior that may be exhibited by children or chimpanzees is likely to be a waste of time. I think that these remarks also apply to cases of “formal equivalence” of the kind mentioned here by Premack. For example, as evidence of “structure-dependent” rules in chim- panzee “language,” Premack cites the following experiment: Given markers for the simple declarative, “apple is fruit,” Sarah is taught to pluralize with these markers: “apple”, “ba- nana”; “is”; “plural article of”, and “fruit.” Premack goes on to say: “For from the training we had given her, she could have learned either of two rules, only one of which was cor- rect. She could have learned to use a physical feature com- parable to the physical feature in Chomsky’s example, namely, applying a plural particle whenever there are two words to the left of ‘is.’ Or she could have induced a rule based not on a physical feature but on a grammatical one, namely, applying a plural particle when the subject is plural.” Now, Premack says that this is a formal logical equivalent of Chomsky’s rule for question formation, where the animal is pluralizing not on the basis of physical features but rather on the basis of the meaning of the sentence, that is, the in- ternal representation of the string of words. But this seems to me to be an unwarranted generalization of a principle of syntax for human language.
First, the rule of question formation in human language Comparisons of Cognitive Abilities / 227 does not invariably depend upon any understanding of mean- ing. Second, the notion of structure dependence is not simply meant to reflect the fact that there is some relation between elements other than simple perceptual successions, but also that rules of syntax operate on abstract representations of a certain kind, namely, noun phrases, verb phrases, and so forth. Within the theory of grammar it is widely recognized that the meaning of a sentence is based on the meaning of its parts and the manner of their combination; however, this does not mean that the nature of semantics is the same as that of syntax with respect to the rules operating on represen- tations of phenomena in these two domains. Even the notion of meaning appears to be trivialized when applied to chim- panzees since many, if not all, of the semantic features of language can only be defined with respect to the notion of a sentence, just as rules of structure dependence can only be defined in terms of noun phrases and the like. The use of metaphors for sentences and noun phrases does not help matters at all. Only insofar as some principle is fully and coherently integrated into a well-formed theoretical sys- tem, the whole of which can be tested with respect to a sig- nificant range of phenomena, does it become a principle of any scientific interest. To take any such principle in isola- tion and “demonstrate” its presence in some other cognitive domain or in some other species, whether in spontaneous be- havior or through conditioning, coaxing or evoking, is to demonstrate nothing at all.
Premack: This whole argument confuses theory with evidence.
As theoretical proposals, Atran’s statements are unobjection- able. But he has no proof for any of them, and indeed does not bother with matters of proof. To begin with, his distinc- tion between syntax and semantics avoids the painful fact that it is seldom possible to find actual language performance that realizes the distinction-that gives evidence of syntactic competence while perfectly excluding the role of semantics. So, to my example of Sarah’s pluralization, it is easy enough to say that this had nothing to do with syntax, it is purely semantics. I wonder how many of Chomsky’s examples would escape a similar objection-for example, how many questions do human speakers form that are independent of meaning? Second, the notion of meaning is not trivialized when ap- plied to the ape. On the contrary, at least some ape concepts are reminiscent of our own, specifically with respect to their abstractness. For example, the terms “color of,” “shape of,” 228 / Language and Learning and “wash,” as they appeared in the ape sentences “red color of apple,” “round shape of ball,” and “Mary wash orange,” were used abstractly by Sarah as they are by us. She applied them not only to the training exemplars but to countless new cases. Indeed, she exceeded the usual transfer criterion, for she not only applied these predicates to new cases but used at least one of them to generate a new word. The instruction “brown color of chocolate” was used successfully with Sarah to generate the new word “brown”—showing that she can use the predicate productively, that is, to generate new instances of itself.
This is not to say that there is no difference between human and ape words. But Atran’s account does not point us in the right direction. I suspect that the difference between the two species may be contained in the answer to this question: what defines a word in the chimpanzee and the human mind? Oversimplifying greatly, I suggest the following: in the ape, a word is defined by an image or sensory representation, whereas in man, a word is defined by other words. I can offer circumstantial evidence now, but actually proving this claim is another matter…
Bischof: My question comes back to the problem of teaching.
There is a correspondence between the teaching on the part of the parents and the desire to be taught on the part of the child, and as far as I know, there is a specific age in the de- velopment of the human child (I think around age 2) which is called the “age of what’s that” by developmental psycholo- gists, when children very eagerly try to find out the names of things. Interestingly, a little bit later there is another age called the “why age,” when children inquire about causal re- lations. Of course, it is obvious that there is nothing like a spontaneous “why age” or “age of what’s that” in chim- panzees. Have you ever encountered any instances in which your chimpanzees asked you questions?
Premack: The only thing I can say in that regard is that despite the resolute way in which I followed a checklist of what I suppose to be basic elements in language and forced the ani- mal to carry out a lesson built exclusively around the topic of my interest, so that it had available only those things that I deliberately made available to it, the animal would decide otherwise in the middle of my beautiful lesson and would run off with the plastic words to the middle of the floor, where it was most difficult to reach. It would then ask itself Comparisons of Cognitive Abilities / 229 the questions of the lesson as well as others, and it would answer them, for the most part, correctly. Indeed, on those occasions the animal’s performance level was very close to 100 percent, whereas when the animal was answering my questions the performance level was about 85 percent. Simi- larly, Elizabeth, another trainee who was very active but by no means the equal of some of our other apes, was given a lesson in which she had to describe what the trainer was doing (the trainer carries out simple acts and the ape has to describe them), and there too she was about 85 percent cor- rect. But often, while waiting for the next test during which the trainer carries out an act, the animal performed the act and then described it spontaneously, which we didn’t ask her to do, and in these cases she was 100 percent correct. Finally, one can ask whether the animal has at least the idea of the word, the idea that things can be named. You can do the following experiment with even the least capable of the animals: you include in the set of words a potential word, that is, a piece of plastic which is demonstrably a potential word in the sense that it has all the properties of the class, but it has never been used as a word. You also use an item which is familiar but which has never been named, even the stupidest animal rapidly constructs the sentence, “Give X [the name of the animal] this new piece of plastic.” In other words, the animal requests the unnamed item with the so- far-unused piece of plastic. Thus the chimpanzees recognize that the potential word, which has not yet been so employed, is the appropriate thing to use in requesting the desired item, which is however not yet named. There are more elegant ways of doing that, but people often prefer the nonelegant way.
Wilden: Something that emerged from your data is the question of negation and the various different types or levels of negation. Obviously, what one would expect is a crossing of boundaries between the two domains, between the chim- panzee communication system and the human communica- tion system. My question has to do with how the chimpanzee manages to use negation…
Premack: If the animal is asked what the relationship is between A and B, and if the word “different” is deliberately re- moved from its available lexicon, the animal will answer “no same,” so it can, with proper training, carry out acts of nega- tion. 230 / Language and Learning
Wilden: That is one form of negation, not the only form.
Premack: It is a form of negation, and not a trivial one. There
is also an aversion to the negative; you find the same thing in retarded children. Normally, if you wanted to teach, for instance, the concept of “name of” or “color of” or “shape of,” or any predicate of that kind, by contrasting the positive with the negative case, you would introduce “color of” by contrasting it with “not color of.” But there are occasions when you cannot do that, when the animal rejects negative forms, in which case you are required to train him with “name of” or “color of” in relation to an already known affirmative predicate. Thus you contrast “name of” with “color of” rather than “color of” with “not color of.” As I said, sometimes in retarded or autistic children to whom these procedures are applied, the rejection of the negative form is so pervasive that you can only work on the negative for a short period, and then henceforth you introduce the names of new predicates only by contrasting all positive forms.
Even though Premack, by his exhaustive analysis of the chim- panzee’s cognitive universe, opens the way to a third position between Piaget and Chomsky, one of his sentences betrays his basically innatist attitude. In replying to Bateson (on learning to learn) he says: “I’m not a great believer in the word ‘learn- ing,’ and I am rather more in favor of Fodor’s argument. I think that the main thing that human training is doing here for the chimpanzee is disclosing capacities that are present.” The zone of convergence between Premack and Piaget has to be found elsewhere: in the general (that is, nonlinguistic or pre- linguistic) capacities that are common to higher primates and to man. According to Premack, representational capacity and the complex trait he calls accessibility of knowledge are de- tectable both in man and in the chimpanzee, and language can develop on the basis of such capacities (if suitable brain struc- tures are present). However, Premack considers these cognitive capacities to be innate, in contrast to Piaget. In his “After- thoughts,” Piaget will comment further on Premack’s views concerning the innatist hypothesis and the possible sources of “necessity” in cognitive development.
In the next chapter, the floor is given to Norbert Bischof, an ethologist of the Lorenzian school. In his paper and in the course of his discussion with Piaget, further relevant problems concerning interspecies comparison are brought out.