Kanzi: The Ape at the Brink of the Human Mind

by Sue Savage-Rumbaugh

  “It required just a few seconds’ thought to realize that if we could learn how language develops in a chimpanzee, we would surely have a leg up in learning how to cultivate language in mentally retarded children,” Duane now says, modestly. “I knew that the technology we developed would be vital, both for enabling ease of use by chimps and humans, and for making economical use of manpower.”1 Duane wanted a partly automated system, which would therefore be less demanding of instructors’ time and effort than the systems used by the Gardners and Premack. Knowing he had the unique resources of Yerkes to draw upon, he decided to aim for an electronic system that would bring efficiency and objectivity to the project.

  Duane recruited the help of his friend and colleague Harold Warner, chief of the Biomedical Engineering Laboratory at Yerkes, and the two of them quickly focused on the idea of a computerized system, probably with a keyboard input. A linguist was needed, and so they asked Ernst von Glaserfield, of the University of Georgia, to join them. Von Glaserfield recommended that his longtime colleague Pier Pisani, a computer expert, also be part of the team. During the winter of 1970, the four men met many times to create the outlines of a system and assemble a research grant proposal for submission to the National Institutes of Health. “We were confident that ultimately the project would lead to a better understanding of human language and some of its cognitive prerequisites,” Duane, Warner, and von Glaserfield later wrote.2 A four-year grant was awarded in the spring of 1971, and the LANA Project was born.

  LANA stood for LANguage Analogue, which relates in part to the symbol system devised for the project. Lana was also the name of the first chimpanzee to work with the system. Von Glaserfield was principally responsible for developing the syntax that formed Lana’s symbol strings, which he called Yerkish, in honor of Robert Yerkes, who had founded the Primate Research Center in 1924. Each symbol, or lexigram, was arbitrary and stood for a single word, including verbs, nouns, and adjectives. Each lexigram was built from combinations of one, two, three, or four geometric forms, of which there were nine in all; there was a total of several hundred potential Yerkish words. Different classes of words were denoted by the use of one of three different primary colors.

  The Lana keyboard.

  A subset of the lexigrams was displayed on a five-by-five matrix keyboard. When a key was pressed and released, the lexigram went from being dimly lit to brightly illuminated. In addition, the lexigram was displayed on a projector above the keyboard, and a sequence of keystrokes produced a sequence of lexigrams—a “sentence”—on the projector, such as “Please Machine Give Juice.” Being computerized, the system could operate day and night and did not require the presence of an instructor. Lana could interact with the machine and control aspects of her environment at will. In addition, all use of the keyboard, by chimp and instructors, was stored in the computer, bringing a measure of objectivity to the recording of communicative interactions.

  Lana gradually acquired a large productive vocabulary and learned to generate a series of stock sentences, by which she typically gained food, turned music on, or obtained some other object she wanted. In addition to the stock sentences, Lana occasionally generated novel sentences, such as asking for an overripe banana by saying, You give banana which is black. But these were the minority of her utterances. Although Lana’s behavior was impressively languagelike, we eventually realized that her abilities were limited in important ways, as I described earlier. The focus on engendering productive behavior meant that Lana’s comprehension was not well developed. As I described earlier, only when our work with Sherman and Austin was well advanced did we come to appreciate fully this key distinction. Nevertheless, Lana’s success with the computerized keyboard system was sufficient to encourage Duane to embark on the second aspect of the project: to discover whether nonspeech communication could be taught to severely mentally retarded children who are unable to speak.

  There are at least 1.25 million children in the United States who are either without speech, or whose speech is severely impaired, as a result of neurological, physical, or psychological disability. For a long time, clinicians favored persisting with speech therapy with such children, fearing that the teaching of nonspeech communication might inhibit whatever latent capacity for spoken language still existed. As a result, the development of nonspeech communication systems—namely, signs of some kind—has only recendy become fully established. For instance, when the first volume of Language Perspectives, which is now a leading journal in the field, was published in 1974, its inclusion of a section on such systems was considered a major advance on earlier publications concerning language acquisition and intervention.

  Nonspeech communication systems, or augmentative language systems, can be divided into two types: aided and unaided. The terms refer to whether or not some piece of equipment is required for communication. Gestures, mime, and manual sign languages, for instance, are unaided systems, whereas anything that uses physical symbols, such as pictographs or arbitrary lexigrams, are aided systems. The LANA system is an aided system. An important advantage that aided systems offer, particularly those like the LANA system that use visual-graphic symbols, is that the signal does not fade. In spoken language, words as symbols have to be brought to mind, and then produced. And once produced, they reside only in the listener’s memory. Manual gestures rely on recall of the symbol by the producer, and are similarly transitory. A visual-graphic symbol, on the other hand, is available for recognition, not recall, by the producer, and once indicated for use may remain visible for as long as is required. Visual-graphic systems therefore place fewer cognitive demands on mentally retarded individuals, thus giving them an advantage among nonspeech systems.

  The key issue facing those who wish to teach symbolic communication to speech-impaired children is the absence of a clear understanding of the nature of language acquisition and how it relates to other cognitive development. The different domains of cognitive development are by no means independent, and must interact synergistically. As Elizabeth Bates and her colleagues have shown, the best predictor of a child’s skill on nonverbal tasks is the level of language comprehension obtained by the child. The mentally retarded child is therefore doubly disabled, often failing to acquire language, which would promote further mental development.

  Since nonimpaired children acquire language spontaneously and with ease and speed, they offer no real clue as to how impaired children might learn. The very reasonable assumption was therefore made that children with extreme learning difficulties would have to be taught a communicative system in a very structured way, rather than be encouraged to learn as other children do.

  Traditionally, language interventionists have pursued two different routes to structured teaching: One is remedial, the other developmental. The remedial model supposes that children “being taught language relatively late in their lives, because they have failed to acquire it adequately in their earlier experience, no longer possess the same collection of abilities and deficits that normal children have when they begin to acquire language.”3 This very pragmatic approach attempts to teach useful communication skills as quickly as possible, without consideration of the prelinguistic or cognitive skills the child might already possess.

  The developmental model, on the other hand, is driven very much by theory, specifically, theory based on the observations of nonimpaired children. According to this approach, the impaired child, in learning communication skills, must pass through the same steps that a normal child spontaneously does. But as these steps are at best incompletely understood due to the largely hidden nature of normal language acquisition, only limited guidance can be developed for teaching strategies.

  Neither the remedial model nor the developmental model is of much help to those children who completely fail to speak. Such children have often been given up as hopeless cases, but it is here that the experience of teaching language skills to chimpanzees has been especially fruitful. Earlier chapters of this book have shown how o
ur language-teaching strategies with apes evolved through time, as our experience increased and we understood more about the components of language and how they were learned. We progressed from the insights of the LANA project, through Sherman and Austin, and on to Kanzi and the other bonobos. Advances in the work with humans tracked those with chimps every step of the way, often surprisingly so.

  Duane initiated the first of his studies with humans in 1975, in conjunction with the Georgia Retardation Center. Nine institutionalized individuals took part in the study. They were aged between eleven years, eleven months and eighteen years, three months, and were all profoundly retarded. Only one of them spoke in any way, this individual producing some ten word-approximations that were unintelligible to the naive listener. Traditional speech-language treatment had failed with these individuals, and they had been judged to have insufficient prelinguistic skills to warrant participation in a language intervention program. “We stacked the cards against us by working with severely handicapped individuals,” Duane now recalls. “The firm expectation by staff at the Georgia Retardation Center was that they would learn nothing at all in our program either. So, if we did make progress, no matter how small, that would have to be recognized as important.”4

  Five of the original nine people in the study continued in it for more than three years, and all of them did learn how to use symbols in a simple, communicative way. The size of the acquired vocabulary varied considerably, with a range of twenty to seventy-five symbols. With LANA-based teaching, the program achieved LANA-like abilities in the subjects, namely, the apparently competent symbol production but rudimentary comprehension. Like Lana, Washoe, Sarah, and Nim, these children learned to make their wishes known and to answer simple questions by stringing symbols together. Yet they also suffered similar deficiencies. Although they could answer questions with words, they often could not respond to commands with appropriate behavior. For example, Sam might say Yes want cookie if asked whether he was hungry, but if asked to go to the table and get a cookie, he often failed to respond. Nevertheless, the program had achieved its aim: to test the feasibility of using the computer-based keyboard system to teach symbol association and use to nonspeaking, mentally retarded children.

  Duane believes that some of the physical characteristics of the communicative system—such as the striking shapes of lexigrams, their colors, and the fact that they lit up when selected—were important in its unexpected success. This helped gain the children’s attention, which is vital but difficult when teaching mentally retarded children. Duane’s insight of teaching with a computer-based keyboard lexigram system had been vindicated, and it soon became possible to pursue the concept further by applying what I had been learning with Sherman and Austin.

  An important conclusion from my work with Sherman and Austin was that chimpanzees are capable of using symbols referentially, as humans do, but that the skill emerges only through the systematic teaching of the various components of language. These components include requesting, naming, and comprehension. The individuals in the original collaborative study with the Georgia Retardation Center had acquired languagelike skills, but they lacked a strong referential quality. The LANA project had provided the model for teaching and, inevitably, LANA-like performance was the result. The obvious step to take was to implement a teaching paradigm based on what I had learned with Sherman and Austin, with the hope of developing symbolic communication among the human students. Mary Ann and Rose initiated such a program at the Language Research Center with James Pate, beginning in 1981. The program was run in collaboration with the Developmental Learning Center of Georgia Regional Hospital.

  The computer-based communication system would remain essentially the same as in the earlier study, but in this case there would be far more emphasis on comprehension. As Mary Ann and her colleagues wrote in a description of the project, “Symbolic communication is a complex phenomenon requiring an individual to comprehend the meanings of symbols, to produce symbols in appropriate situations, and to use arbitrary symbols for interindividual communication.”5 Important in the new study was a set of operationally defined components of symbolic communication. They are:

  An arbitrary symbol that stands for, and takes the place of, a real object, event, person, action, or relationship.

  Stored knowledge regarding the actions, objects, and relationships relating to that symbol.

  The intentional use of symbols to convey this stored knowledge about an object, event, person, action, or relationship to another individual who has similar real-world experiences and has related them to the same symbol system.

  The appropriate decoding of, and response to, symbols by the recipients.

  The most fundamental lesson in acquiring language is learning that symbols have a function in communication. As Mary Ann and her colleagues point out, for severely retarded individuals, “this first step toward learning to communicate symbolically is perhaps the most difficult to achieve, regardless of the modality in which the instruction occurs.”6 A teaching paradigm based on my experience with Sherman and Austin was designed to surmount this difficulty, moving step by step. The first step was to teach the request function.

  In unimpaired children, requesting visible objects is one of the earliest communicative actions to occur in development, with other skills emerging rapidly thereafter. Language comprehension begins as early as three to six months after birth, and by one to one and a half years, children spontaneously begin to make their needs known. In children who suffer brain impairment, both comprehension and production may fail to occur. Previous attempts to train such children to communicate have typically focused on symbol production, as there seemed to be no efficient means of measuring progress in language understanding. Additionally, most teachers have assumed that if such a child learns to use words, he or she will automatically understand them. However, a major problem with approaches that fostered symbol production was that while children learned words, they often failed to use them readily outside the training situation. Thus, though the children acquired words such as “potty” or “thank you,” they nonetheless failed to tell their teachers in all circumstances when they needed to go to the potty or even to ask where it was located when placed in a new setting.

  Mary Ann wondered if such generalization failures resulted from a lack of comprehension skills. Perhaps these children had learned what to say to gain the teacher’s approval, much as Washoe and Lana learned what to say to win the approval of their caretakers. Could it be that these children, like apes, did not really understand that words did more than get them things? Did they realize that words were being used in all manner of ways set by others, and that to “catch on” to this they had to listen and decipher the meaning or intent behind each utterance that others made? The meaning or intent of others was often very different from one’s own and had to be “figured out” anew for each occasion.

  Mary Ann decided to try to determine whether it was possible to give such children an introduction to language that provided them with a thorough grounding in the myriad of different ways that symbols function—as a means of asking for things, naming things, making statements about intended actions, and understanding and cooperating with others.

  Four individuals took part in the study: Bev and Connie, whom I mentioned earlier, and Ruth (aged eighteen years, eleven months) and Max (aged fourteen years, one month). Like Bev and Connie, Ruth and Max had failed to acquire speech or augmentative communication, despite extensive teaching efforts. Because of their ages (relatively old) and absence of progress in traditional language-teaching regimes, the four individuals were reckoned by their caregivers to have little chance with the Sherman and Austin regime, either. Mary Ann and Rose, however, were confident of making progress.

  In order to establish as strong a motivation as possible in the teaching regime, Mary Ann and her colleagues used favorite food items as objects to be requested. This would provide a strong connection between the use of a symbol and the con
sequences of that use. The subjects were to be introduced to one food item and its lexigram at the beginning of the program; when the association between lexigram and food item had been learned, a second would be introduced; and so on. Mary Ann and her colleagues established a series of stages of learning each lexigram, until the subject could accurately pick it out among a field of other lexigrams. Bev learned quickly, requiring only 125 trials to reach the required accuracy for the first lexigram. Connie and Ruth were much slower, requiring 469 and 903 trials, while Max never really succeeded, even after more than 2500 trials. He was excluded from the study at that point.

  The difficulty that the subjects faced with the first lexigram paled against the experience with the addition of a second. The number of trials Connie and Ruth required for the establishment of accurate requesting with this second lexigram were 2217 and 1319, respectively. Even Bev found the task more difficult, requiring 298. Thereafter, however, the subjects found the learning task much easier. For instance, Connie required 102 and 455 trials for learning the third and fourth lexigrams, respectively; for Ruth, the figures were 262 and 337; and for Bev, 93 and 42. “They learn how to learn,” observed Mary Ann, “and the quality of what they learn improves.”7 One mark of this improvement was the subjects’ eventual production of multilexigram utterances. As Elizabeth Bates has pointed out, in mentally normal individuals, the transition from one-word to multiword utterances is the third milestone in the acquisition of language. (The first is the emergence of communicative intentionality and the second is the appearance of reference.) Bev, Connie, and Ruth can therefore be considered to have reached that milestone.