Kanzi: The Ape at the Brink of the Human Mind

by Sue Savage-Rumbaugh

  Patricia’s analysis produced statistically significant evidence of structure in Kanzi’s utterances. He was developmentally delayed compared to a human child, but parallel syntactic structures were present. I must admit to having been surprised at the regularities that Patricia’s keen perception was able to extract from the body of data that we had gathered on Kanzi.

  Both rules that Kanzi invented (the “action-action” combination and the “gesture follows lexigram” combination) have to do with coordinating actions in some way. While the first reflects bonobo behavior, the second, though arbitrary, is a way of structuring two communicative channels. Why is action-based ordering important? I believe it gives an insight into the way early humans might have imposed order on language as they invented it. Language almost certainly emerged in early humans as their social life and economic subsistence activities were becoming more complex. Therefore, there would have been a premium on the ability to coordinate actions, and a simple syntax of the sort that Kanzi invented would have been an evolutionary toehold onto the more complex syntax that eventually emerged in modern humans. Moreover, the fact that Kanzi is able to invent such rules is strong evidence for the continuity theory—that is, the idea that the mind of man differs in degree from that of the ape, but not in kind. It also gives some indication of the cognitive substrate to language that might have existed in the common ancestor of humans and chimpanzees. The more we learn about what the ape brain cannot do in the realm of language, the more we can pinpoint the specific kinds of skills that must have emerged in early humans.

  One of Chomsky’s arguments for the innateness of language capacity in humans is the existence of common patterns within the grammars of all languages. Underlying this commonality, he suggests, is a universal grammar, the product of a unique language module in the human brain. I agree with Chomsky that language capacity is innate at some level. One would hardly want to attribute even partial linguistic competence to oysters, for example, regardless of their rearing. However, any argument about continuity between humans and apes adduces the fact that humans and chimpanzees share 99 percent of their genetic blueprint as a reason to expect a sharing of some elements of language capacity. Where I disagree with Chomsky is in his assertion that the structure of language is necessarily prewired in the human brain.

  The analysis of Kanzi’s utterances shows that the ordering of action is important in the rules he invents. This ordering is likely to have been the case in human prehistory, leading to a common syntax that is influenced by the environment. Furthermore, just as there are environmental constraints on the way tools can be structured and still function, it is plausible that there are constraints in the way language may be ordered and still be comprehensible. The fundamental similarities of human grammars may therefore be the linguistic equivalent of fundamental similarities found in the cross-cultural design of vessels constructed to carry water. When Chomsky and his followers use the term “innate,” they mean innate and unique to humans; for me, the term “innate” is simply a mark of our biological continuity.

  As I suggested earlier, it is futile to ask whether apes have language, as linguistic orthodoxy demands. The significance of Kanzi’s possession of certain elements of language is, however, enormous. As the ape brain is just one-third the size of the human brain, we should accept the detection of no more than a few elements of language as evidence of continuity. In my view, we had done that.

  With enthusiasm Patricia and I wrote up our results and submitted them to Nature in July 1987. “We demonstrate that an ape, in a communicative environment with humans, develops a productive grammar uncontaminated by imitations, and, most interestingly, invents primitive symbol-ordering rules that he has not been exposed to in his symbolic environment,” we wrote in our letter of submission. In retrospect, it is clear we should not have used the word “grammar,” either in our letter or in our manuscript. To the linguist it means only one thing: human grammar. We were therefore stepping into the linguists’ court, and challenging the cherished notion of human uniqueness.

  We had shown the manuscript to Herbert Terrace before submitting it to Nature, because we valued his views. He liked it, thought it important and worthy of publication, but attacked it with a volley of small criticisms. Overall, he thought we had overstated the grammatical nature of Kanzi’s utterances. When the manuscript reached the reviewing process, some referees thought it too short to do justice to the work, some said it was too long for the amount of new information it conveyed, but all were either skeptical or downright scornful of our claim for grammatical structure. The manuscript was returned to us two months later, with a rejection letter.

  We persuaded the editor to allow us to address some of the issues raised by the referees, and submit a new version. We should not have expended the time and effort, for the response to the new manuscript was precisely the same as previously. “Overenthusiastic overinterpretation” was the general view, with few apparently able to accept the implications of our results: that humans are not unique in possessing a capacity for language, and are therefore not as special as most would like to believe. Our second rejection letter arrived in July 1988.

  A year had passed since our first submission, and we had made little progress in getting linguists to take notice of what Kanzi was able to do. Perhaps we should have used the word “protogrammar” instead of grammar, as indeed it is more appropriate. But if protogrammar is appropriate for the primitive syntactical structures that Kanzi invented and used, then so too is it appropriate for children. “Comparative developmental psycholinguistics has been plagued by a double standard,” Patricia and I wrote later. “Because children ultimately develop language, their early stages are interpreted as having greater linguistic significance than the same stages in primates. When children make up novel words on a one-shot basis, it is called lexical innovation. When chimpanzees do the same thing, it is termed ambiguous.”3 The double standard, I believe, is part of the linguists’ last barricade in defending the notion of human uniqueness. It is not science.

  When our work was finally published, in a conference volume in the fall of 1990, it received wide coverage in newspapers and magazines. Indeed, the volume of publicity and its enthusiasm, while gratifying, was almost embarrassing. It also provided us with some response from our colleagues. “All the evidence suggests that the animals are using sophisticated ways to request things,” said Terrace, damningly, to a writer for U.S. News & World Report.4 “It has nothing to do with language, and nothing to do with words,” scoffed Thomas Sebeok, in a newspaper article. “It has to do with communication.”5 For once Sebeok and I apparently agree: Of course it has to do with communication.

  Chomsky’s comment, cited in Discover magazine, suggested to me a lack of biological sophistication. “If an animal had a capacity as biologically sophisticated as language but somehow hadn’t used it until now, it would be an evolutionary miracle,” he said.6 How do we know that bonobos are not using these abilities in complex ways in the wiid, ways that are not apparent to us yet? The more we learn about them, the more sophisticated their communicative abilities appear. Would Chomsky suggest that an ape’s precise control of a joystick linked to a computer game must be an illusion, because they don’t do this in the wild? No, because it doesn’t threaten the uniqueness of human grammar, the last bastion of the discontinuity theorists and their peculiar creationist position.

  Apart from these and other solicited comments in the popular press, linguists have been oddly silent in the scientific journals. Evidently, Patricia and I had been optimistic in expecting that data would overcome prejudice.

  During the time we were trying to get our syntax paper published by Nature, we initiated a further major test of Kanzi’s language competence: namely, his comprehension of spoken English. The trial, conducted between May 1988 and February 1989, was an important milestone in psycholinguistics, because for the first time a direct comparison was made between an ape and a human child. Kanzi was a litt
le over seven and a half years of age when he began the trial, while Alia (the daughter of Jeannine Murphy, one of my colleagues) was two years old. As the distinguished psychologist Elizabeth Bates puts it: “If we want to understand what an organism knows about language, isn’t comprehension the best place to start?”7 We wanted to find out what Kanzi and Alia knew about language at these different ages, through an exploration of their comprehension of novel sentences.

  Comprehension has been relatively neglected by linguists, for two reasons. First, it is much more difficult to study than word production, particularly when tests require that infants cooperate in test situations where they may be asked to label or identify things that they are not at the moment thinking about or wishing to play with. Second, linguists’ obsession with the putative innateness of language capacity has focused attention on production and away from comprehension. And yet, increasingly, psycholinguists are coming to acknowledge that comprehension is at least as fundamental as production in the acquisition of language, if not more so. It is also being recognized as immensely complex.

  We usually think of language as being composed of discrete words that are assembled into phrases and sentences—this is how it appears when we hear language spoken and, particularly, when we see it written. But the sounds of language when they enter the ear are anything but discrete. Viewed on a spectrograph, a sentence looks like a continuity of sound, with intermittent jumps in amplitude, but with little clue as to where one word ends and another begins. Moreover, the same word in different sentences may look very different. For this reason, it has been unexpectedly difficult to devise a computer program that can decode human speech. And yet, before she reaches the age of one, a human child has already begun to do just that. Beginning with phrases, the child soon moves on to decoding sentences, a process that paves the way for language production.

  Despite progress in breaking language down into its components—such as intentionality, rule learning, imitation, fast associative mapping, and sequencing—there is no widely accepted explanation of how language acquisition occurs. We know that children need early exposure for efficient language acquisition, but it remains a mystery how this puts into place an understanding that words are referential items and that, properly structured, they create information-rich sentences.

  The Chomskian innatist view of language acquisition explains the child’s gradual ability to learn to speak, despite the cacophony of language to which she is exposed, as the result of a developmental switching on of a parsing device, a brain module designed for the purpose. There is no anatomical evidence for the existence of such a structure, and the hypothesis effectively rests on a default premise: namely, that no other hypothesis offers an adequate explanation. It is true that the process of language acquisition through which most of us pass appears to be a near miracle, given the absence of what could be considered effective teaching. But that is an inadequate rationale for assuming the necessity of a unique, undetected brain structure. A plausible alternative hypothesis is that comprehension drives language acquisition.

  Children, as they learn to acquire language, make some linguistic mistakes, but their number is remarkably few, and they are often on the order of incorrectly generalizing correct rules—such as “goed” for went. A minimization of production errors would occur if children came to comprehend much of what others were saying to them before they learned to speak. It has long been recognized that comprehension precedes production, both at the single-word and sentence stages. Comprehension can therefore put in place the fundamental rules for production, with context playing a large role in giving meaning to words that are heard by the child. When they are learning to talk, children do not appear to be building up a complex grammar out of single word units and an innate parsing device. Instead, they seem to be pulling apart the syntactic structure inherent in the speech around them, through the help of speakers who mark these syntactic units by their intonation and use of phrase, and by the context of what is being said. In other words, a child learns language by listening and paying attention, not by talking.

  Although by no means universally accepted as the central aspect of language acquisition, comprehension is being taken more seriously than it used to be, not least in the innateness (uniqueness) debate. With Kanzi, we had an opportunity to explore the elements of comprehension that emerged as a result of his early exposure to spoken language. If his abilities were to extend beyond comprehension of single words, we would have further evidence of evolutionary continuity of language abilities.

  As I described in an earlier chapter, Kanzi showed evidence that he comprehended spoken English words from a very early age. The test we did when he was four years of age showed that he could choose the correct lexigrams with sixty-five spoken words. But it was also clear to me that he understood phrases too, albeit quite simple ones. As the years passed, his comprehension appeared to expand in range and become more sophisticated.

  During our day-to-day activities, we constantly attempted to explore the limits of Kanzi’s comprehension. For instance, one day he and I were down by the river and, as always, Kanzi had his ball with him. Balls had been Kanzi’s favorite toys since he was six months of age. He loves all sorts of balls, large and small, soft and hard, and is never really happy unless he has at least one ball with him, but preferably two or three. His favorite balls are shiny ones that look rather like the genital blossoms or attractive sexual swellings that adolescent female bonobos sport most of the time. When the other bonobos want to tease or hassle Kanzi, they will take his ball if he is not looking at it. Never has Kanzi failed to react when I say, “So and so is about to get your ball.” He immediately whips around and rushes to grab his ball back. When Kanzi has five or six balls and is trying to keep track of all of them with the other bonobos around he has a real job, for one ball or the other is always rolling away where another bonobo can grab it, and as he hurries to retrieve it, others roll away from his pile. We have occasionally made videotapes for Kanzi’s viewing in which an imaginary gorilla steals one of his balls and plays with it. Kanzi is riveted to the screen when such scenes appear and must rush to locations he has seen on the television immediately afterward to search for the ball. Kanzi has an uncanny memory for his balls; he can recall where he has left one days, months, and even years later.

  On this particular day when we were down by the river, I decided to ask Kanzi, “Can you throw your ball in the river?” I knew this was something he had never done before and something no one would have ever asked him to do, as we generally tried to keep everything out of the water except sticks and rocks. However, I decided to violate one rule today, just to see if Kanzi could understand such an unusual request. He promptly tossed the ball in the river. On another occasion, as we were walking through the forest with his half-sister Panbanisha, I said, “Kanzi, would you please give Panbanisha an onion?” He looked around for an onion patch, pulled up a bunch, and handed them to Panbanisha. It might be thought that in such sentences all Kanzi heard was “Kanzi, xxxxx xxx xxxxx xxxx Panbanisha xx onion,” and put two and two together. After all, he could not have given Panbanisha to the onion. But we noticed that, where confusion was possible, such as, “Can you throw a potato at the turtle?” he rarely made mistakes. (In this case, the mistake would be throwing the turde at the potato.)

  Most salient, however, were conditional sentences. One day, for instance, we were visiting Austin, who was busy with a task. As a reward for completing the task, Austin was given cereal, which Kanzi desperately wanted and kept requesting. I knew that Austin would become angry if Kanzi took the cereal, and told Kanzi so. While all this was going on, Kanzi was playing with a monster mask that we had brought in the backpack. Austin was very interested in the mask, so I thought I would offer a deal. I said, “Kanzi, if you give Austin your monster mask, I’ll let you have some of Austin’s cereal.” Kanzi promptly got the mask and gave it to Austin, and then pointed again to the cereal. It had been a linguistic bargain, and Kanz
i had understood.

  We recorded many such examples of Kanzi’s understanding complex sentences—at least, when the subject at hand was of interest to him. When Kanzi wasn’t interested, he either did not understand or simply acted dumb. (I knew he was capable of being contrary when asked to do something he didn’t want to do, often doing the exact opposite.) A compilation of these carefully recorded anecdotes about the extent of Kanzi’s comprehension was regarded as insufficient evidence for most people, especially the skeptics. For this reason we decided we would embark on a strictly controlled study, comparing Kanzi with a human child, Alia. Our goal was not to build a complete picture of their comprehension, but rather to discover what kinds of syntactic markers, if any, they were becoming sensitive to.

  We developed a series of strict criteria for the test, to ensure that Kanzi and Alia were not inadvertendy “trained” during the trial and that we did not cue them. Very simply, in entirely separate but identical experimental settings, we planned to present them with a series of novel sentences (660 in all) and monitor their responses. Initially, the experimental setup was a little unsettling for both subjects, as the person asking the questions had to be out of sight. The questions therefore were delivered by one experimenter as a disembodied voice, from behind a one-way mirror. A second experimenter sat with Kanzi and Alia, sometimes being part of the ensuing action, but always recording what was done. This experimenter wore headphones with loud music, so he or she could not hear the sentences being directed at the subject. The sessions were also videotaped, so that they could be scored by an independent observer.

  We were unsure at the outset how extensive Kanzi’s and Alia’s comprehension competences would be, and so we were prepared to change the complexity of what we asked as the test proceeded. It turned out that we had to increase the complexity of sentences, for both subjects. Whatever the complexity, however, we made sure that many of the sentences were unusual, such as asking them to wash hot dogs. This sometimes caused puzzlement, but mostly comprehension won through. We devised five major types of sentences (with subgroups, giving a total of thirteen) designed to test issues such as word order and complexity of structure.