Aping Language

Language is regarded, at least in most intellectual traditions/as the quintessential human attribute, at once evidence and source of most that is considered transcendent in us, distinguishing ours from the merely mechanical nature of the beast. Even if language did not have the sacrosanct status it does in our conception of human nature, however, the question of its presence in other species would still promote argument, for we lack any universally accepted defining features of language, ones that would allow us to identify it unequivocally in another species. Both the role of language in differentiating ours from other species and contention over the crucial attributes of language are responsible for the stridency of the debate over whether nonhuman animals can learn language. Aping Language is a critical assessment of each of the recent experiments designed to impart a language, either natural or invented, to an ape. The performance of the animals in these experiments is compared with the course of semantic and syntactic development in children, both speaking and signing. The book goes on to examine what is known about the neurological, cognitive, and specifically linguistic attributes of our species that subserve language, and it discusses how they might have come into existence. Finally, the communication of nonhuman primates in nature is assayed to consider whether or not it was reasonable to assume, as the experimenters in these projects did, that apes possess an ability to acquire language.

Aping language

Editors: John Dunn, Jack Goody, Geoffrey Hawthorn Edmund Leach: Culture and communication: the logic by which symbols are connected: an introduction to the use of structuralist analysis in social anthropology Anthony Heath: Rational choice and social exchange: a critique of exchange theory P. Abrams and A. McCulloch: Communes, sociology and society Jack Goody: The domestication of the savage mind Jean-Louis Flandrin: Families in former times: kinship, household and sexuality John Dunn: Theory in the face of the future David Thomas: Naturalism and social science: a post-empiricist philosophy of social science Claude Meillassoux: Maidens, meal and money: capitalism and the domestic community David Lane: Leninism: a sociological interpretation Anthony D. Smith: The ethnic revival Jack Goody: Cooking, cuisine and class: a study in comparative sociology Roy Ellen: Environment, subsistence and system: the ecology of small-scale formations S. N. Eisenstadt and L. Roniger: Patrons, clients and friends: interpersonal relations and the structure of trust in society John Dunn: The politics of socialism: an essay in political theory Martine Segalen: Historical anthropology of the family Tim Ingold: Evolution and social life David Levine: Reproducing families: the political economy of English population history Robert Hinde: Individuals, relationships and culture: links between ethology and the social sciences Paul Connerton: How societies remember G. E. R. Lloyd: Demystifying mentalities

Aping language

JOEL WALLMAN Harry Frank Guggenheim Foundation, New York City

CAMBRIDGE UNIVERSITY PRESS

Published by the Press Syndicate of the University of Cambridge The Pitt Building, Trumpington Street, Cambridge CB2 1RP 40 West 20th Street, New York, NY 10011-4211, USA 10 Stamford Road, Oakleigh, Victoria 3166, Australia © Cambridge University Press 1992 First published 1992 A cataloguing in publication record for this book is available from the British Library Library of Congress cataloguing in publication data

Wallman, Joel. Aping language/Joel Wallman. p. cm. - (Themes in the social sciences) includes bibliographical references and index. ISBN 0-521-40487-8. - ISBN 0-521-40666-8 (pbk.) 1. Apes - Psychology. 2. Human - animal communication I. Title. II. Series. QL737.P96W35 1992 599.88'0451-dc20 91-28564 CIP ISBN 0 521 40487 8 hardback ISBN 0 521 40666 8 paperback Transferred to digital printing 1999

To my parents Ronetta and Louis

A dog cannot relate his autobiography; however eloquently he may bark, he cannot tell you that his parents were honest though poor. BERTRAND RUSSELL

Human Knowledge: Its Scope and Limits

Contents

Acknowledgments

page xii

I. BACKGROUND 1 Introduction 2 History of the ape-language projects

1 3 10

II. THE ARTIFICIAL-LANGUAGE PROJECTS 3 The Lana project 4 The Sarah project

29 31 38

III. APES AND LANGUAGE: ONTOGENY 5 Words 6 Sentences

47 49 79

IV. APES AND LANGUAGE: PHYLOGENY 7 Language, evolution, and anatomy 8 Primate communication in nature

107 109 128

V. CONCLUSION 9 The chimpanzee and the Chinese room

147 149

Notes

154

References

166

Index

184

XI

Acknowledgments

I would like to thank the members of the anthropology and psychology departments at Columbia University who contributed so much to the perspective this book reflects: Alexander Alland, Jr., Don Melnick, and the late, beloved Robert Murphy; and Barbara Landau and Herbert Terrace. Among the many others - roughly one per page - who are responsible, in ways diverse, for whatever is good in this work are Ernie Alleva, Michael Billig, Jeff Bogumil, Brian Burkhalter, Barry Cerf, Marc Edelman, Constantin Fasolt, Brian Ferguson, Sue Herman, Janice Hirota, Robert Hoffmeister, Reamy Jansen, Jessica Kuper, Louise Lennihan, Rosemary Masters, Laura Petitto, Sekhar Ramakrishnan, Alan Richter, David Rosner, Candelario Saenz, Elizabeth Stone, Virginia Valian, and Annie Williams. I know Brian Ferguson will understand when he learns that I declined to pursue his suggestion that the apes were actually speaking "passable French/' I offer a special kind of gratitude to Gary Pasternak, Larry Wallman, Don Zavelo, and, especially, my wife, Karen Kane. I thank the Sigma Xi Research Society for helping me get these monkeys off my back with a grant-in-aid.

xn

Part one

Background

Introduction

This book is about the experiments carried out over the past two decades in which it was attempted to impart a language, either natural or invented, to an ape. The debate engendered by these projects has been of interest consuming for some, passing for others - to all of those whose concerns include the enduring questions of human nature, among them anthropologists, psychologists, linguists, biologists, and philosophers. An adequate treatment of the linguistic capabilities of apes entails consideration of a number of related issues, each of which is an interesting problem in its own right. Continuities in primate mentality, the relationship between language and thought in the individual and in the species, and the origin of language in, again, both the ontogenetic and the phylogenetic senses, are themes that will recur throughout this work. Development of parts of the argument will require moderately technical excursions into American Sign Language grammar, recursive rules in language, the neuropsychology of language, naturalistic primate communication, and language acquisition in children. A grounding in the last topic, in particular, is crucial to the argument, for the apelanguage dispute is essentially a quarrel about how similar the performance of the linguistic apes is to that of young children acquiring language. The method followed in this book is one of detailed (though, hopefully, not tedious) critical analysis of the experimental methods and conclusions of the ape-language projects. The analysis is based on data summaries, published anecdotes, and experimenters' conclusions - in short, on published material rather than on primary data. At the outset of this work, I requested samples of original transcriptions and/or videotaped records of linguistic productions from all but one of the sign-language (as opposed to artificial-language) projects.1 The experimenters uniformly declined to provide such materials. Although this was disappointing, I am confident that enough could be gleaned about the nature of the "raw" data from

Background published material to justify the conclusions about apes and language contained in the following pages. The nature of those conclusions will be evident early on: none of the ape-language projects succeeded, despite employing years of tutelage far more intense than that experienced by most children, in implanting in an ape a capacity for language equal to that of a young child, let alone an adult. This argument is developed in chapters 3 through 6. Chapter 2 provides a description of each of the projects, while chapters 7 and 8 consider alternative explanations for the "language gap." WHY THE APE-LANGUAGE CONTROVERSY IS A CONTROVERSY All scientific arguments have in common at least these elements: (1) a minimum of two positions regarding the subject in dispute, positions generally held to be irreconcilable, and (2) an intensification of the normal emotional investment of the scientist in his or her position, due in some measure to the contending itself but perhaps also related to the ideological significance of the subject within the larger society. If, in addition, the argument includes suggestions of fraudulent or quasi-fraudulent procedures, the disagreement becomes a controversy. To the extent that this is an accurate characterization of scientific controversies, the ape-language debate is an exemplary one. The radical opposition of opinion about the achievements of the various ape-language projects is well conveyed by the following quotations: [Washoe] learned a natural human language and her early utterances were highly similar to, perhaps indistinguishable from the early utterances of human children. (Gardner and Gardner 1978, p. 73) The evidence we have makes it clear that even the brightest ape can acquire not even so much as the weak grammatical system exhibited by very young children. (Premack and Premack 1983, p. 115) On measures of sign performance (form), sign order (structure), semantic relations (meaning), sign acts (function) and sign acquisition (development), apes appear to be very similar to 2 to 3 year old human children learning sign ... Apes also appear to be very similar to 2 to 3 year old human children learning to speak. (Miles 1978, p. 114) [The experimental chimpanzees] show, after years of training and exposure to signing, not the slightest trace of homological development parallel to that of human children. (Leiber 1984, p. 84) After years of gentle teaching Koko has learned to use American Sign Language the very same sign language used by the deaf. With her new-found vocabulary, Koko is now providing us with an astounding wealth of knowledge about the way animals view the world. (Patterson 1985a, p. 1)

Introduction In Koko, A Talking Gorilla, a stirring film documentary that opened last December in Manhattan, Koko does a fine job of acting like a gorilla, but otherwise the film is mostly flimflam. (Martin Gardner 1980, p. 3) It is unlikely that any of us will in our lifetimes see again a scientific breakthrough as profound in its implications as the moment when Washoe, the baby chimpanzee, raised her hand and signed "come-gimme" to a comprehending human. (Hill 1978, p. 109) The systems taught to apes and other species differ from human language at the most primitive and elementary level. (Chomsky 1979, p. 38) What is most likely to be occurring in the ape research is self-deception, in the form of experimental expectancy effects or the "trimming" or "cooking" of data by investigators ... as opposed to outright fraud ... (Umiker-Sebeok and Sebeok 1980, p. 31)

There are several sources of the stridency of the debate, which probably peaked with the threat by Allen and Beatrice Gardner, who conducted the Washoe sign-language project, to sue the principal investigator of the Nim project, Herbert Terrace, for using frames from their films of Washoe in a publication (Terrace, personal communication). Certainly the major catalyst was Terrace's 1979 Science article, in which he became the bete noire of ape-language research by unambiguously concluding that Nim, the subject of his study, had not proved capable of acquiring rudimentary grammatical rules and that, furthermore, this was true as well of the other signing apes.2 Needless to say, the Gardners and Francine Patterson, the teacher of Koko the gorilla, did not agree. That source, however, is merely a historical event - the debate was a debate prior to it, albeit a cooler one. The more interesting causes of both the intellectual differences and the emotionality inhere in the topic of the argument itself. For articulate language is not just one among other capacities thought to be exclusively human abilities. No one would get excited, it has been observed, if it were shown that an ape could mix a dry martini (Atherton and Schwartz 1983). Rather, language, at least in the European intellectual tradition, is the quintessential human attribute, at once evidence and source of most that is transcendent in us, distinguishing ours from the merely mechanical nature of the beast. Language is regarded as the sine qua non of culture, and its presence in our species is the most salient behavioral difference between us and the other hominoids with the relinquishing of tool use and, more recently, tool making (Goodall 1971; Beck 1980) as uniquely human capabilities, the significance of language as a separator has grown. And resistance to losing our quintessential attributes is, arguably, itself one of those uniquely human traits. Hence, some ape partisans (Linden 1974; Gysens-Gosselin 1979)

Background have argued, the prevalent reluctance to accord the talking apes their due. An occasional variant of this interpretation is the accusation that those who refuse to recognize ape language are insufficiently committed to the Darwinian perspective or, worse, are anti-Darwinian. Thus Linden (1987) depicts those who question the likelihood of ape-human linguistic continuities as latter-day Wilberforces, averse to investigating "creatures who threaten to paralyze us by shedding light on the true nature and origins of our abilities" (p. 8). A countervailing vector of our ideology, perhaps peculiar to our culture but possibly pancultural, consists of careless anthropomorphic projection and an irrepressively attractive vision of communication between our own and other species. In fact, it seems correct to observe that, at least until recently in the debate and probably up to the present, the majority opinion, both lay and scientific, regarding the linguistic capabilities of the apes has been positive. People seem not only accepting but positively desirous of the possibility of ape language. Even if language did not have the sacrosanct status it does in our conception of human nature, the question of its presence in other species would still promote argument, for we are lacking any universally accepted, unassailable diagnostic criteria for language. There is no shortage of candidates for the indispensable attribute of language. For Katz (1976) and Limber (1977), the projective capability is crucial, the provision of language for the articulation of any conceivable new proposition through a novel combination of words. Savage-Rumbaugh (1981) holds the referential nature of individual symbols to be the essence of language, while Premack (1984) and Marshall (1971) see the capacity for representation of real-world situations to be paramount, and so on. The property most commonly invoked as definitive of language is its predication on a system of abstract rules for the production and interpretation of utterances - in other words, grammar. Hockett's (1959,1960,1963; Hockett and Altmann 1968) famous list of so-called design features of language - including rapid fading, duality of patterning, and displacement - has provided a useful orientation for those trying to capture the differences between human and nonhuman natural systems of communication. What is wanting, nonetheless, is consensus on what the necessary and sufficient, as distinguished from inessential, property or properties of language are and hence on how we might unequivocally identify language in another species. This problem of defining features is more severe where the language of the young child is concerned, and it is the child's language that is taken by most parties to the debate to be the proper material for comparison with the apes.3 If the young child is not, in fact, capable of linguistically encoding anything she can think of, if her production and understanding

Introduction of utterances do not suggest abstract grammatical constituents and processes, then can it be said that the child has language? Limber (1977) and Lightfoot (1982), at least, would say no. This is a defensible position, its major problem found in the fact that the young child's language, which may not yet be language, will eventually become language. How is this discontinuity in development to be bridged? The difficulty is not the existence of a discontinuity per se - there are a number of others in human development. The physiological transition from prepubescence to pubescence, for example, poses a similar problem - the two developmental phases are identifiably distinct, yet there are no two adjacent points in time about which it could be said that the child was prepubescent in the first but pubescent in the second. What makes the transition from "nonlanguage" (hereafter early language) to language more problematic is that, unlike the case of puberty, in which the first phase is defined largely by the absence of characteristics of the later one, early language has its own, very salient features. Moreover, there are some striking functional and possibly structural similarities between these features and those of adult language. The two-year-old manages the major "speech acts" - the performatives - of the adult speaker, executing declaratives, requests, imperatives, and so forth (Dore 1975; Bates et al. 1979). And, contrary to those who would deny language to the young child, there is extensive evidence for grammatical structure in the earliest word combinations (Bloom 1970; Brown 1973), and, some have suggested (De Laguna 1927; McNeill 1970), in single-word utterances as well. (The proper characterization of this structure, however, is the subject of ongoing debate in developmental psycholinguistics - in fact, this may be the dominant concern of the field. This issue will be discussed in chapter 6.) Language, in summary, is central to our self-definition as a species, even though we have yet to derive an adequate definition of language itself, one that includes the essential but excludes the merely contingent. Behaviorist roots of the ape-language experiments

There is an additional source of the contention surrounding the apelanguage question. The issues in the debate tend to resonate along the longstanding cleavage within the behavioral sciences between those who advocate study of cognition and/or innately determined behavior, on the one hand, and those, on the other, who are behaviorist in method and theory. Behaviorism, or stimulus-response psychology, came into being in the early decades of this century as an avowed antidote to the introspectionist

Background trend in turn-of-the-century psychological investigation. Knowledge, thought, intention, affect, and all other unobservable mental phenomena were banished in favor of overt behavior as the only proper subject of a scientific psychology.4 To explain the behavior of animals, behaviorism, like the eighteenth-century empiricism from which it descends, posits a bare minimum of cognitive apparatus: (1) perception, (2) a capacity to represent in durable format the results of perception, and (3) the ability to form associations among those representations. In the behaviorist paradigm, the acquisition and strengthening of such associations constitute learning. An association may be formed between a perceptual stimulus and an inborn response if that stimulus consistently accompanies another one that is innately connected to the response, as in the celebrated conjunction of the ticking of a metronome, food, and salivation in Pavlov's dogs. Or an animal may form an association between one of its own actions and a subsequent stimulus, as when a pigeon comes reliably to peck a button because its activation results in the dispensing of food. In this process, an association is created between an action and a following stimulus that "reinforces" that action. To qualify as a reinforcing stimulus, a consequence need not be one that we would regard a priori as satisfying or pleasant - in fact, any stimulus that increases the probability of the organism emitting the behavior that preceded it is, by definition, reinforcing. In the behaviorist conception, all behavior is determined either by current stimuli or by past consequences. Language is verbal behavior; words function both as responses to stimuli and as stimuli themselves, eliciting further responses. Thus a sentence can be interpreted as a chain of stimulus-response events, each word a response to the preceding one and also a stimulus evoking the next, with the first word elicited by an environmental stimulus or an internal one, a "private event." Or, in some formulations, the entire sentence is regarded as one complex response to a stimulus.5 The orthodox behaviorist account of learning has little use for traditional distinctions among types of behavior. Nor are species differences in behavioral mechanisms acknowledged. Although sometimes touted as such, the latter attitude is not an appreciation of evolutionary continuity, with the selectively and historically wrought similarities and divergences in behavior that such a theoretical affirmation entails. Rather, it reflects a commitment to cross-species homogeneity, a rejection of the notion that there are important differences across species in the processes that underlie the development and causation of behavior. As Skinner once observed in noting the similarity of learning curves produced in three different

Introduction organisms through reinforcement, "Pigeon, rat, monkey, which is which? It doesn't matter" (1956, p. 230). And thus the Gardners, in contesting the belief that there are differences in kind among language, other forms of human behavior, and behaviors of nonhuman animals, offer their scientific credo: "If a form of behavior such as human language appears to be different in character from other forms of human and animal behavior, then we do not abandon the search for general laws; instead we question the adequacy of existing observations" (1978, p. 37). Like other contemporary adherents of behaviorism, the ape-language experimenters embraced the various concessions to reality that the most primitive versions of behaviorism were forced to make over the years. The Gardners, for example, acknowledge that some parts of the innate behavioral repertoire of a species are more plastic and hence more readily conditioned than others, and also that species differ in their intrinsic propensities for various behaviors. Thus the chimpanzee's inborn motivation to communicate obviates conditioning as laborious as another behavior might require. That language acquisition in the chimpanzee and in the child are similarly dependent on extensive molding, shaping, and imitation, however, is an assumption that is fundamental to their research, and fundamentally erroneous. Indeed, their suggestion that the linguistic performance of the preschool child requires "intensive training" (1971, p. 118) is the opposite of one of the few claims to which virtually all language-acquisition researchers would assent. It would be misleading to suggest that all of the proponents of ape language were behaviorists and all detractors cognitivists or ethologists. It is true that nearly all of the experimenters in this area were trained in the behaviorist tradition. Yet several eventually came to view their projects and the question of language in general in a way quite at variance with the presumptions of radical behaviorism. And, conversely, among the believers in ape language are psychologists of cognitive orientation and linguists. Nevertheless, the aspirations initially underlying the projects derived from behaviorist conceptions of the nature of language, and much of the criticism of those projects has been essentially a critique of these notions along the lines of Chomsky's (1959) vivisection of Skinner's treatment of language in Verbal Behavior (1957). Lastly, it may be worth observing that the potential personal rewards of the ape projects have been substantial and emotional commitment commensurately high - the first person or team to give language to another species would certainly attain scientific immortality.

History of the ape-language projects

EARLY STUDIES

Before the Gardners' innovative attempt, described below, to inculcate a visual human language in a nonhuman primate, at least half a dozen projects had been undertaken with the aim of either actively conferring a spoken language on an ape or observing the possible "natural" acquisition of one within a human home environment. The linguistic results in each case were dismal. Witmer (1909) reported on a chimpanzee that had been trained to approximate crudely the word "mama." Furness (1916) succeeded, after much labor, in getting an orangutan to utter a discernible "papa" and "cup." Hayes and Hayes (1951) attained the greatest success among these early projects; at the end of six and a half years of home-raising a chimpanzee, Viki, the Hayeses had managed to teach her to utter "papa," "mama," "cup," and, less successfully, "up." But Viki's articulation was poor and there was little evidence that these words served a referential, that is, symbolic, function for her. On the other hand, the apes' "comprehension" of spoken words and phrases substantially exceeded their productive abilities. For example, Gua, a chimpanzee home-reared by the Kelloggs (1933), outpaced the Kelloggs' young son in the number of phrases to which she could respond correctly. This was true up to the end of the fourth month of the study, after which the boy surpassed Gua. Rather than positing understanding, Kellogg (1968) speaks of the chimpanzee "reacting correctly," but he does not address the question of how much of this reacting could be attributed to linguistic decoding as distinguished from nonlinguistic cues and contextual information. Several of these investigators noted the ape's proclivity for gesturing. Both Viki and Gua employed numerous gestures that were associated with specific activities in that the chimpanzee made the gesture either 10

History of the ape-language projects prior to engaging in the activity or in an attempt to induce the caretaker to do so. Kellogg (1968) points out that there is no reason to assume these gestures were intentional signals as opposed to mere behavioral correlates of the activities they accompanied. The Hayeses (1954), on the other hand, do not hesitate to interpret Vila's gestures as communicative, and, in fact, Kellogg himself characterizes Gua's in the same way later in the same article. This natural predisposition to gesture, coupled with the ape's wellknown penchant for imitation, prompted Yerkes to speculate: Perhaps the chief reason for the ape's failure to develop speech is the absence of a tendency to imitate sounds. Seeing strongly stimulates to imitation; but hearing seems to have no such effect. I am inclined to conclude from the various evidences that the great apes have plenty to talk about, but no gift for the use of sounds to represent individual, as contrasted with racial, feelings or ideas. Perhaps they can be taught to use their fingers, somewhat as does the deaf and dumb person, and thus helped to acquire a simple, nonvocal, "sign language." (1925, p. 180)

Yerkes' idea was not unprecedented. Ward (1983) provides this extract from the diary of Samuel Pepys, entered August 24,1661: At the office in the morning and did business. By and by we are called to Sir W. Battens to see the strange creature that Captain Holmes hath brought with him from Guiny; it is a great baboone, but so much like a man in most things, that (though they say there is a Species of them) yet I cannot believe but that it is a monster got of a man and she-baboone. I do believe it already understands much english; and I am of the mind it might be taught to speak or make signs. (Latham and Matthews 1970, p. 60 [quoted in Ward 1983, p. 341])

In "Bertran and Bimi," a story about an animal trainer and an orangutan, Kipling created the following scene, in which a sailor describes an encounter with Bimi, the orangutan: Den I felt at der back of my neck der fingers of Bimi. Mein Gott! I tell you dot he talked through dose fingers. It was der deaf-and-dumb alphabet all gomplete. He slide his hairy arm round my neck, und he tilt up my chin und look into my face, shust to see if I understood his talk so well as he understood mine. (1891/1907, pp. 339-10 [quoted in Ward 1983, p. 3411]) Apes and speech

It is not necessary to ascribe the linguistic failure of the animals in the early studies to a general deficit in language-related neurology, for it is now widely held that acquisition of a verbal language by nonhuman primates is precluded by the dominance of the limbic system, the "emotion centers" of the brain, in primate vocalization (see chapter 8). Although human 11

Background vocal production is not wholly devoid of limbic involvement (Robinson 1976), it is clear that the neurological locus of our distinctive language abilities is the neocortex of the cerebrum. Infrahuman primates, by contrast, have shown a modest ability at best to develop volitional, that is cortical, control over vocalization during conditioning experiments (Myers 1976). There is, in addition, evidence that the laryngeal and pharyngeal anatomy of infrahuman primates, especially the comparatively small volume of the region above the larynx, is incapable of supporting what is generally meant by articulate speech (Lieberman 1975). However, this assessment of the primate phonatory apparatus is not universal among experts (see Wind 1976); the soundest assertion one can make would seem to be that attributes of the central nervous system provide more important constraints on vocal language capability than do aspects of peripheral anatomy (Lenneberg 1967). Partly in order to circumvent articulatory limitations, Premack and Schwartz (1966) invented a nonvocal phonemic language. This antecedent of the nonphonemic "chip" language later developed by Premack was designed to incorporate the distinctive features model of speech developed by Jakobson (Jakobson, Fant, and Halle 1952) without requiring the chimpanzees for whom it was designed to do any speaking. The description of the syntactic component above the level of the word was sketchy, but Premack and Schwartz did construct an ingenious articulatory mechanism consisting of a joystick moveable in several dimensions at once. Each dimension controlled one acoustic attribute of the phoneme2 to be produced, including frequency, amplitude, duration, vibrato, and noise. Four of the five dimensions were divided into discrete values, providing the kind of categorical structure that is central to the Jakobsonian analysis of speech production and perception. The sum of the dimensional values of a given movement of the joystick would result in synthesis of a distinctive phoneme. The system even provided for the coarticulation of adjacent phones that is characteristic of speech, "subsequent" vowels influencing "previous" consonants. A word would consist of a CVC sequence, and no sound would be heard until the vowel had been chosen, whereupon a fusion of C and V would be synthesized. Among the goals of this project was determining whether the animals would be able to abstract the phonemes out of their embedment in the speech stream, as we do. Would they "babble" with their joysticks and, on exposure to communication by their trainers, gradually converge on the phonemes used by the trainers? Would this process of analyzing the language input require reinforcement or rather occur inevitably, as it does 12

History of the ape-language projects in children? These questions were never answered, as the system seems to have never been deployed. RECENT STUDIES The studies that are the focus of this work have taken place during the last two decades. There have been six or seven major projects, depending on how one assesses importance - I am using publication volume as a rough measure of significance - and perhaps five minor ones. This classification is crosscut by a second and more important one - the nature of the language medium used. Three of the major projects employed a gestural language, the others using a visual but nongestural language invented for the experiment. What follows is an essentially uncritical presentation of the procedures and claimed achievements of the recent studies. Washoe

Project Washoe, fielded by Gardner and Gardner (Gardner, Beatrice T. and R. Allen Gardner, 1971,1974,1975,1980; Gardner, R. Allen and Beatrice T. Gardner, 1969,1972,1974,1975,1978,1984; Beatrix T. Gardner 1981,1982), was the first of the studies and the most seminal in that the subsequent projects were explicitly influenced by it in conception and method. Washoe, a wild-born female chimpanzee, was acquired by the Gardners in June of 1966. Her age was estimated on the basis of weight and dentition to be between 8 and 14 months. Washoe was housed in a trailer in the Gardners' backyard and also had access to substantial play areas. Her trailer was replete with fixtures, furnishings, and playthings; the Gardners endeavored to provide her with an ambiance as intellectually stimulating as that for a child, including immersion in language. The choice of ASL (American Sign Language), a manual language, allowed the Gardners to exploit the gestural abilities of the chimpanzee.3 And, as the communication standard of the North American deaf, ASL came with a population of human users for comparative purposes. Project personnel were required to "master ASL" (1969, p. 666) and were instructed to communicate with Washoe and with one another, when in her presence, in ASL only. Washoe was accompanied by at least one companion during every waking hour and, in general, she enjoyed all of the activities and materials that are the privilege of the young American child, including trips, games, and other forms of play. Aside from directing signed comments, inquiries, and requests to her and otherwise encouraging her to associate signs with referents, Washoe's 13

Background companions sought to teach her signs through molding and shaping, techniques that were adopted by the subsequent signing projects. In molding, the teacher physically guides the animal's hand into the proper configuration, location, and/or movement for the sign. Shaping is the practice of rewarding successively closer approximations to a target behavior, here execution of a certain sign. The Gardners had hoped that Washoe would actively acquire signs through imitation - observation and copying - of those around her, but, at least during the early phase of the project, this did not occur to any appreciable extent. A controlled study by Fouts (1972) of the relative efficacy of molding and imitation showed the latter to be generally inadequate as a way of teaching Washoe signs. As shaping also proved to be ineffective, molding became the predominant pedagogical tool. The Gardners (1978) assert that the role of imitation in acquisition became proportionately greater as the project progressed, but no quantitative data to this effect are provided. Daily records were kept of Washoe's signing. Eventually the frequency of her signing made the recording of every "utterance" impractical, and a checklist method was adopted in which, for each sign on the list, trainers would note whether Washoe had made that sign each day. As her vocabulary expanded, the likelihood of her uttering any single sign diminished. For this reason, the checklist technique evolved into a more drill-like procedure in which an appropriate context for production of each vocabulary item was arranged. The purpose of these drill sessions was to find out whether Washoe's acquisition of new signs resulted in the loss of old ones. The Gardners concluded that this was not the case. Washoe made her first sign at three months into the project and her first multisign utterances by one year. Her productive vocabulary included four signs by the seventh project month, 30 at 22 months, and 132 at 51 months, when the project ended. The Washoe project emphasized the controlled testing of vocabulary content. Elaborate double-blind controls were employed to test Washoe's ability to utter the correct (noun) sign when presented with slides of objects. Slides were varied so that different exemplars of the target category were presented, ensuring that she was associating her signs not merely with particular items but with classes of items. In addition to vocabulary-test results, the Gardners have included in their publications anecdotal characterizations of Washoe's linguistic performance and, occasionally, statistical information as well, all intended to depict her performance as substantially similar to that of the young child, deaf or hearing. For example, Washoe extended application of signs from their original referents to new ones on the basis of functional or perceptual similarity. Thus "open," initially signed when Washoe was faced with a 14

History of the ape-language projects closed door, was extended to closed containers and eventually to water faucets. Similarly, "flower" was overgeneralized from flowers to other objects or locales with salient odors, including tobacco pouches and kitchens. Washoe coined new modifier-noun combinations to refer to things for which she lacked single signs, the most often-cited example being her signing "water bird" in the presence of a swan. A technique commonly employed in language-acquisition studies is the classification of children's early, two-word combinations according to the elementary semantic relations they embody. Brown (1973) found that, regardless of the language involved, 70 to 75 percent of these constructions fall into just eight to ten semantic-relational categories, such as agent-action, possessor-possession, and demonstrative-entity.4 Using a modified version of Brown's influential taxonomy, the Gardners were able to classify 78 percent of a sample of 294 of Washoe's two-sign strings into seven such categories, suggesting that, like children, Washoe combined signs to encode relationships, rather than merely stringing together signs that are related only in that each is independently relevant to the nonlinguistic context. At the end of 1970, Washoe was moved to the Institute for Primate Studies at the University of Oklahoma. There she was involved in studies by Fouts (1973; Fouts, Chown, and Goodin 1976; Fouts, Shapiro, and O'Neil 1978) of sign acquisition and interanimal signing among several language-trained chimpanzees. Sarah

Premack (1971,1976a, 1976b, 1976c, 1983,1984; Premack and Premack 1983; Ann Premack 1976) took a markedly different approach to investigating an ape's linguistic potential. He made no pretense of teaching a natural language to his subjects. Reasoning from a conception of natural languages as merely one species of language in general, he derived a list of "exemplars" of language, "things an organism must be able to do in order to give evidence of language" (1971, p. 808). His project was designed to instill these "functional prerequisites" of language in his subjects. The exemplars included words, sentences, class concepts (such as color, shape, and size), the copula ("red is/is not color"), quantifiers (all, none, one, several), the hypothetical if-then, and metalinguistics, meaning linguistic reference to linguistic elements themselves, as in learning a new word by means of the word for "name of" (for example [symbol] name of [object]). The language medium was a set of metal-backed plastic chips varying in size, shape, color, and texture, intended to function as words. The appearance of a chip bore no resemblance to its referent. There was no attempt to 15

Background incorporate phonemic or morphological principles - each word was an irreducible whole, the several physical variables providing only for distinguishability. "Sentences"5 in this language were constructed by the subject or trainer by concatenating the chips vertically on a magnetic board. The advantages of this language of physical tokens are essentially two. Using such a "written" system avoids memory-related constraints on production and comprehension of symbol strings, as contrasted with spoken and signed languages, both of which possess Hockett's (1960) "rapid fading" design feature. This eliminates the problem of distinguishing the roles of memory and syntactic ability in the interpretation of experimental results. Secondly, and similarly, using ready-made words circumvents any articulatory shortcomings of the animal.6 The Premack project actually involved four chimpanzees, but the performance of three of the four was so inferior to the fourth that the latter, Sarah, became the focus of Premack's experiments and reports. A wildcaught chimpanzee, Sarah was around five years old at the outset of the study. Like the Gardners, Premack initially envisioned his subject acquiring words through observational learning but, also like the Gardners, he found this approach inadequate. Instead, a shaping procedure was developed in which Sarah was rewarded for incremental increases in the complexity of her performances. For example, teaching her the sequence "Mary give Sarah apple" went as follows. First, Sarah was rewarded with a piece of apple for giving the chip for "apple" (a blue triangle) to the trainer (Mary). (No other chips were available to her.) Next, she was required to write "give apple" to get her reward, followed by "give apple Sarah" and finally "Mary give apple Sarah."7 The general procedure for training new vocabulary elements was to hold constant all items in a mastered sequence except for the new term. In the above paradigm, for example, the chip for "chocolate" might have been taught by providing Sarah with the chips for "Mary," "give," and "Sarah" but exchanging "chocolate" for "apple." Sarah would then be rewarded with the chocolate for writing "Mary give Sarah chocolate." New vocabulary items, whether objects, actors, actions, or relations (such as same/different) were always introduced in this errorless fashion, Sarah being required only to insert a single new chip into a sequence rather than choosing from among alternatives. Testing, on the other hand, designed to assess the generality of her grasp of a syntactic construction, always involved her choosing from among at least two alternatives. Nearly all of the concepts worked on had as their linguistic expression a three-term sequence consisting of a predicate and two arguments. The 16

History of the ape-language projects arguments were not always chips, though; objects were used in several. The same/different paradigm, for example, involved a display of two objects that either were of the same kind or were different. Sarah learned to place either the token for "same" or the "different" token between them. (One of several variants was to have Sarah prefix a "yes" or a "no" chip to a completed three-term sequence depending on whether the "proposition" was true or false.) Sarah learned to associate properties of objects with chips for those properties and could also classify properties into higher-order properties with class names. After learning to request incentives by specifying their color, shape, or size, she acquired the tokens for the class concepts color of, shape of, and size of. Thus she could complete propositions such as "big ball" and "round cracker," where her choices were "color of," "size of," and "shape of." Most impressive in this area was her acquisition of new property names through chip sequences containing class names. For example, she was taught the chip "brown" with the sequence "brown color of chocolate." "Color of" and "chocolate" were already known to her, but "brown" was new. To test her understanding of "brown," she was presented with four disks, each a different color, and instructed to choose the "brown," which she did. This remarkable feat seems to demonstrate not only learning of a new property term ("brown") by way of a class symbol ("color of") but also comprehension of displaced reference (Hockett 1960), for this performance required that, in the absence of chocolate, she retrieve a mental image of it on reading the sequence "brown color of chocolate." A description of each of Sarah's other accomplishments would be a lengthy undertaking, but her apparent mastery of certain chip paradigms involving multiple propositions should be mentioned, as these constructions are arguably equivalent to complex sentences (sentences made up of more than one sentence-like phrase) and are often cited in assessments of her achievements as the most significant relative to human language. Sarah learned to follow instructions of the sort "Sarah insert banana pail apple dish," meaning "insert the banana in the pail and the apple in the dish." In addition, she was able to respond correctly to sequences consisting of two propositions conjoined in a conditional relationship, as in "Sarah take apple if-then Mary give Sarah chocolate" ("if Sarah takes apple then Mary will give Sarah chocolate"). It is not clear when the language training of Sarah and her fellow pupils ended, but they continued to participate in experiments in the Premack laboratory as the focus shifted from language to cognition in general (Gillan 1982; Premack and Woodruff 1978; Premack 1983; Premack and Premack 1983). 17

Background Lana

The LANA (LANguage Analogue) project of Rumbaugh and colleagues (Rumbaugh 1977b) utilized a language created especially for the study. Lana, the chimpanzee subject, was just over two years of age when the experiment began in 1972 at the Yerkes Regional Primate Center in Atlanta, Georgia. The language employed, invented by the psycholinguist Ernst von Glasersfeld, used a grammar of the correlational type, in which rules of sentence formation specify permissible combinations of word types, the word types being based on semantic attributes (1977).8 The language system was thus much more elaborate and explicit in conception than that of the Premack project, which could not be characterized in terms of any general grammatical principles. The physical medium of this language, dubbed "Yerkish" after the project's institution, was a vocabulary of abstract symbols, termed lexigrams, arrayed on a keyboard inside Lana's cage. (Like Sarah, Lana was caged during her language training, in contrast to the relatively unrestricted mobility of the apes in the signing projects.) The surface of each key was embossed with a different lexigram, a lexigram consisting of two to four out of nine possible geometric designs superimposed on each other. These design elements had no systemic features - as in the case of Premack's "Chipese," the physical variables served merely to insure discriminability. There was, however, an additional physical dimension to the key, the color of its surface, on top of which the design elements appeared. Each of the seven possible surface colors was correlated with a broad semantic category. Keys representing autonomous actors were violet, those for ingestibles red, activities blue, and so forth. This correlation of color and meaning was included to provide for eventual testing of semantic organization, but this seems never to have been pursued. Lana's keyboard, connected to a remote computer, was permanently located in her quarters, affording her 24-hour access to it and to the materials and services she could obtain through its use. By composing a lexigram sequence that was grammatically acceptable, as judged by the computer that parsed her constructions as she formed them, Lana could enlist the computer to dispense food and drink, show slides and a movie9, provide a view out the window, and so on. Lana could also communicate with trainers occupying adjacent but invisible areas in order to gain desirables as well as tickling, grooming, and other services beyond the capabilities of the computer and the machinery it controlled. Above the keyboard in Lana's cage was an array of seven small projectors. With each key press the corresponding lexigram would appear on 18

History of the ape-language projects one of these projectors, in a left-to-right progression, providing Lana with a growing image of her composition. (Seven was the maximum allowable "sentence" length, a rule born of the computer's memory limitations.) Above this row was a second bank of seven projectors, which displayed communications, whenever they might occur, from a trainer in an adjacent room, which was also equipped with a Yerkish keyboard and two banks of projectors. The virtues seen by Rumbaugh et al. in the computer-mediated Yerkish system are a superset of those argued by Premack for his language - the ready-made lexigrams pose no articulatory burden on the animal beyond pressing of keys, and the durable image of the lexigram sequence on the projectors eliminates short-term memory problems. In addition, the interposing of a computer between Lana and any humans she might converse with prevents tainting of Lana's linguistic performance with inadvertent cuing by those humans. And, finally, the computer maintains an exhaustive and objective record of all linguistic transactions between Lana and the computer or Lana and a trainer. The course of Lana's language training was as follows. Initially, she was required only to pull down on a bar that activated her keyboard and then press a single key, which would gain her the incentive associated with that key. (A key would brighten when pressed, giving Lana a clear indication that it had been engaged.) Then she was required to prefix a key glossed as "please"10 to such a request and append a "period" key, which was required from this point on as a sentence terminator. Next, all of the keys in a complete request were located in sequence on her keyboard, and Lana was trained to push any one of them, which would activate all of them, after first pulling the activation bar and prefixing a "please." Thus "please" followed by any key in the phrase "machine give M & M" and the period key would gain her an M & M candy. The next phase was identical except that the key for the incentive was not automatically activated on pushing a key from the request phrase; Lana had to activate that key separately. Separate pressing of each key in the sequence was trained next, after which the keys were randomly redistributed around her keyboard to ensure that Lana was learning a sequence of lexigrams rather than a sequence of positions on the keyboard. The practice of regularly rearranging the lexigrams was continued as new items were added to the lexicon. During training, as opposed to testing, human intervention was freely engaged in, Lana's trainers often assisting her in pushing the right keys. Most of Lana's linguistic performance consisted of using stock sentences - the lexigram constructions learned as rote sequences - to get incentives. Strings such as "please machine give chow," "please Tim give apple," and 19

Background "please Tim move into room" were produced voluminously and with little respite - they were, in fact, her only means of acquiring such things (Gill and Rumbaugh 1977). Eventually, Lana came to understand that certain of the lexigrams in her stock sentences had significance apart from the other lexigrams with which they were agglutinated. In other words, she appeared to have learned that lexigrams stood for things and that, conversely, things had names. Thus, she was able to respond correctly to the question "what name of this" on being shown an object. In addition, she spontaneously began to recombine substrings of her stock sentences in appropriate ways, as when, lacking a lexigram for the fruit orange, she requested it with "? Tim give apple which-is orange"11 (Rumbaugh and Gill 1976, p. 573). The majority of such novel constructions were produced by Lana during conversations with a trainer present in her cage rather than in communications to the computer. Lana evinced knowledge of grammatical structure by differential responding to sentence beginnings provided by the trainer: if a grammatically incorrect request beginning were present, Lana would terminate it with the period key, erasing it from the projectors, and proceed to produce her own, correct request. If, on the other hand, the fragment was a correct beginning, she would complete it with the lexigram for some incentive. Moreover, she spontaneously came to treat her own "false starts" in the same way, aborting them with a period and starting again. In general, Lana developed the ability to initiate and sustain a conversation with a trainer. The list of possible topics was very limited, however, and her conversations were virtually always narrowly directed toward the acquisition of some desired item. As occurred in the Premack laboratory, the Yerkes group has reoriented its work in recent years, although they are still very much involved in language-related training. A preoccupation with syntactic ability has been replaced by exploration of the nature of referential communication with single symbols. Koko

In July of 1972, Francine Patterson began instructing a one-year-old female lowland gorilla in American Sign Language, inaugurating what will be referred to here as the Koko project, after the subject (1978a, 1978b, 1978c, 1979a, 1979b, 1980a, 1980b, 1981; Patterson and Linden 1981). The Koko project differed notably from the other major signing experiments - the Washoe project, which preceded it, and the Nim project (described below), which began slightly after it - in three respects. Firstly, it was, and 20

History of the ape-language projects to this date remains, the only linguistic study involving a gorilla. Secondly, unlike the other two signing projects, which mandated ASL as the exclusive language medium, Patterson's employed what she terms "simultaneous communication/' in which an ASL sentence and a spoken English equivalent are articulated to the subject at the same time. The rationale here was that compounding the linguistic input might enhance language acquisition in some way and would also provide for "the possibility of transfer of information between the two modes" (Patterson 1978b, p. 74). Thirdly, the linguistic achievements claimed for Koko substantially surpass those of the other ASL apes - Koko's utterances include not only mundane declaratives and interrogatives but puns, lies, and metaphors (Patterson 1980a). While the Koko project has produced very few publications in scientific journals, it has received quite a bit of attention from the public.12 The attention results from effective dissemination of reports of Koko's extraordinary capabilities through long-term support from the National Geographic Society and Patterson's direct-mail funding solicitations for the Gorilla Foundation, an organization created to support the project. The first eleven months of Koko's tutelage consisted of about five hours of sign training daily. During this period, Koko resided in the nursery of the San Francisco Children's Zoo, where she was born in 1971. At age two, she was moved to a five-room house trailer outfitted, like Washoe's, with toys and furnishings. She had access to the kitchen and living room and limited privileges elsewhere. Koko has been in the company of signing companions for approximately ten hours each day. Molding and imitation (that is, modeling by the trainer) have been the standard instructional techniques, as in the other sign studies. In the early period, Koko's utterances and their contexts were recorded by the trainers at frequent intervals, but when signing frequency and vocabulary expansion precluded this, a daily checklist procedure, similar to that of the Gardners, was instituted. In addition, exhaustive transcriptions are made during eight to ten one-hour time slices each month, and once each month an eight-hour-long sample is taken. These are supplemented with 30 to 60 minutes of videotaping per month. Finally, beginning in the nineteenth project month Patterson designated one of each month's hour-long time slices as "voice only" and one as "sign only," directing only voice or sign communications to Koko during each, respectively, in order to investigate possible differences between the two modalities in Koko's receptive abilities. Patterson has administered at least seven different standardized tests of mental development to Koko over the course of the project. Koko performs comparably to a "low-average" child of her age, with an IQ 21

Background averaging around 80 (Patterson 1979b, p. 222). Tests of language comprehension in which the language medium of test-question presentation was varied indicated that Koko had equal facility in comprehending signed, spoken, and simultaneously signed and spoken sentences (Patterson and Linden 1981). Koko's vocabulary consisted of 100 words by project month 30 and 250 by month 48. Double-blind vocabulary tests have been used to assess the reliability of her sign-referent pairings. Her first sign combinations appeared two to three months into training. The increase in her mean utterance length and upper bound on utterance length during 1974 were reported to be within the range for children of comparable age. According to Patterson (1980b), Koko, like Washoe, has coined new modifier-noun combinations to refer to objects for which she lacked signs. "White tiger" was applied to a toy zebra, "bottle match" to a cigarette lighter, and "quiet chase" to a game of hide and seek. Koko, too, generalizes and overgeneralizes her signs to new referents. And, like Washoe's, the majority (75 percent) of Koko's two-sign combinations can be classified into a small number (12) of semantic-relational categories. In 1976 a male lowland gorilla, Michael, took up residence in Koko's trailer in an area adjacent to Koko's. Michael, who was about three and one-half years old at introduction, is envisioned as a prospective mate for Koko. He is being tutored in sign language with the same methods and intensity applied to Koko, and it is reported that the two animals sign to each other intermittently. In 1979 Koko, Michael, and Patterson moved from their Stanford residence to an empty farm in Woodside, California. It is unclear whether Patterson has continued her research program here, since no scientific publications on the project have appeared since 1981, although funding solicitations for the Gorilla Foundation have continued. Mm The Nim project of Terrace and colleagues (Terrace 1979b, 1981; Terrace et al. 1979, 1980, 1981; Sanders 1985) began in early 1973 with the arrival in New York City of a captive-born two-week-old male chimpanzee.13 Nim Chimpsky, named by Terrace in a whimsical transformation of "Noam Chomsky," lived in an apartment with a human family14 until he was 18 months old, when he was moved to a mansion in Riverdale, New York, owned by Columbia University. Here he lived with three student research assistants for the duration of the study. The Nim project was essentially identical in conception to the Washoe project. The plan was to raise a chimpanzee in a milieu approximating that 22

History of the ape-language projects of a child, including, especially, exposure to a human language, and chart the animal's acquisition, if any, of that language. The language of the project was nominally ASL, although the Terrace group came instead in later publications to describe it (and the language used with Washoe and Koko) as a sort of pidgin sign language. The reason for this conservative characterization of the project language is that while pidgin sign languages do possess grammatical structure, they are in most important respects reduced forms of true sign languages. The grammatical processes governing sentence production in this and the other ape projects were an impoverished subset of those available in ASL. (The structure of ASL will be explored in some detail in chapters 5 and 6.) Nim's exposure to sign language began on his adoption into a New York family, but systematic language training began at nine months. A small (eight feet by eight feet) room at Columbia was prepared as his "classroom," and Nim spent about five hours here each weekday. The advantage seen in conducting the core training in such a physically limited environment was that it would facilitate observation and recording of Nim's language performance. One wall of the room contained a large one-way mirror and below this mirror was a portal for wall-mounted cameras. Detailed analysis of videotaped exchanges between Nim and his trainers, in fact, became a distinctive approach of the Nim project. Trainers whispered the contents of Nim's signing and contextual information into a small tape recorder as the signing occurred and prepared transcriptions of the recordings after each session. Transcriptions and videotapings were periodically made also of Nim's utterances at home. As in the other two signing projects, molding and imitation were the standard means of imparting signs to the subject. Nim's first sign, drink, occurred at four months. By the end of the project, when he was 3 years 8 months old, Nim had acquired some 125 signs. His signing was quite abundant - more than 20,000 multisign utterances (but tokens, not types) were recorded during one two-year period. Nim's use of single signs resembled Washoe's and Koko's, which, in turn, were argued by the Gardners and Patterson to resemble that of the young child. The domain of application of a sign grew from just the original referent to many exemplars, and the sorts of rational but incorrect applications made by children, known as overgeneralizations or overextensions, were evident. While no controlled experiments on Nim's vocabulary were conducted, informal tests and his general pattern of sign use demonstrated that Nim had developed consistent associations between signs and referents. The Terrace group performed an analysis of the distribution of semantic relations encoded in Nim's utterances. Using transcripts from videotaped 23

Background sessions in the Columbia classroom and at the Riverdale mansion, they were able to classify 84 percent of a sample of 967 utterance tokens into eight categories of semantic relationship. This is equivalent to the Gardners' and Patterson's classifications and, again, is arguably parallel to semantic-relational analyses of children's two-word combinations. One marked difference between Nim and at least Koko is in the pattern of development of the mean length of utterance (MLU). Whereas Koko, like the signing or speaking child, showed an increase in this statistic with age, graphing this function for Nim yielded a flat trajectory - Nim's utterances during the last year and a half of the project fluctuated in length between 1.1 and 1.6 morphemes. Terrace's group also carried out an extensive computer study of the distribution of signs in Nim's utterances, looking for nonrandom positioning tendencies of individual signs. Results of this analysis, together with those from the semantic-relational study, provided evidence that Nim's combinations were constrained by grammatical rules. These analyses will be discussed in more detail in chapter 6. It should be mentioned at this point, though, that subsequent fine-grained perusal of videotaped sessions resulted in Terrace's rejecting that interpretation of Nim's productions in favor of a nonsyntactic one. By the middle of 1977, attrition of volunteers and lack of funding had brought the Nim project to an end. Nim was moved back to his natal residence, the Institute for Primate Studies at Norman, Oklahoma, where he joined several other educated subjects participating in studies of chimpanzee language by Roger Fouts. Kanzi

The only ape-language project still generating published results is the most recently undertaken (Savage-Rumbaugh 1984, 1986, 1987, 1988; Savage-Rumbaugh et al. 1986). In 1981, Savage-Rumbaugh began teaching Matata, a wild-caught female pygmy chimpanzee (Pan paniscus), symbols from the Yerkish language used with Lana. Although the language medium was the lexigrams of Yerkish, there was no attempt to elicit combinations from the subject, and she was not constrained to observe the Yerkish grammar in any way. Matata's initial instruction was informal, her teachers simply demonstrating for her the use of the lexigram keyboard, while speaking English, in communicating with each other and with her. This approach proved ineffective, so a practice of systematic pairings of lexigrams and rewards was tried, but Matata still failed to adopt the use of lexigrams as symbols. The beneficiaries of the work undertaken with Matata have turned out 24

History of the ape-language projects to be not her but her offspring, Kanzi, a male, and, to a lesser extent, Mulika, a female. Kanzi was born at the Yerkes Regional Primate Research Center of Emory University in Atlanta in October 1980, Mulika in December 1983. At six months of age, Kanzi accompanied his mother Matata when she was moved to the Language Research Center, a joint operation of Georgia State University and the Yerkes Center, to commence her language training. Kanzi was with his mother constantly for the two years of her language instruction.15 During this time, from age six months to two and one-half years, no attempt was made to impart the lexigram system to Kanzi, although he was allowed to observe his mother's training sessions. The only evidence Kanzi gave of understanding the nature of lexigrams was sporadically pressing randomly chosen keys and then running toward the vending device in anticipation of food, or pressing the lexigram for "chase" to induce others to engage in a bout of chasing. When Kanzi was two and one-half years old, his mother was temporarily removed from the Language Research Center for breeding purposes. It was at this point that Savage-Rumbaugh and her colleagues realized that Kanzi had acquired a remarkable amount of linguistic knowledge merely through observation. He began using the keyboard in a purposeful way, seeking particular lexigrams. After he pressed a key, if Kanzi was presented with an array of several objects including the one whose lexigram he had pressed, he would take only the one he had "requested," indicating that he was able to use symbols to refer to specific entities. He could also "name" something he was shown by pressing its lexigram. Kanzi began to produce lexigram combinations as soon as he was using the keyboard regularly. Although most of his keyboard communications, both single- and multilexigram, are in the nature of requests for desired objects or activities, Kanzi often uses his keyboard to comment, without any sort of prompting, on ongoing events, or simply to state his intended actions. In fact, SavageRumbaugh does not credit a symbol as part of Kanzi's vocabulary until he has exhibited a certain level of concordance between the use of that symbol and his subsequent behavior (Savage-Rumbaugh et al. 1986). If Kanzi presses "treehouse," this usage is accorded a behavioral concordance only if, on being invited to lead, he takes his companion to that location. If he specifies a food, a concordance is granted only if he chooses that particular food when a collection of foods is offered to him. The experimenters do not require him to demonstrate such behavior with every usage, but a lexigram is assigned to his vocabulary only when he has exhibited concordance on nine out of ten occasions when the companion has provided an opportunity to do so. 25

Background As impressive as his production and comprehension of lexigrams has been Kanzi's comprehension of spoken English. Whereas, according to Savage-Rumbaugh (Savage-Rumbaugh et a\. 1986, p. 213), none of the other ape-language subjects has been clearly shown capable of understanding spoken language, Kanzi's facility with words exceeds his knowledge of lexigrams. In fact, Kanzi learns his lexigrams by associating them with word-referent pairings he already knows. To promote Kanzi's understanding of spoken English, his keyboard has been augmented with a speech synthesizer. When a lexigram is pressed, in addition to the button lighting up, the corresponding word is produced. Every lexigram Kanzi "utters" is recorded. When inside, he uses the speech-equipped keyboard, which is connected to a computer that registers the lexigram. When outdoors, Kanzi and his companions conduct their lexigram communication via a portable, non-electronic keyboard. The human companion keeps a record of each lexigram touched. Kanzi has free access to several large areas indoors and to the extensive (55 acre) outdoor area. Though he can spend as much time with his mother as he wishes, Kanzi chooses to be with his human companions most of the time, who continually model keyboard-mediated communication for him, simultaneously speaking in English. Since there are fewer lexigrams than words used with Kanzi, a given utterance in his presence will be mainly spoken English accompanied by intermittent key presses. Kanzi has never been drilled in lexigram use, nor have food or other rewards been withheld contingent on his use of the keyboard. His linguistic abilities, like a child's, are thus derived from observational learning and social interaction rather than operant conditioning. During the warm months, Kanzi's food is located at 16 to 20 named locations in the wooded area surrounding his compound. Kanzi spends most of these days traveling from one such location to another in pursuit of edible desirables and other sources of pleasure. Before setting off on a trip, Kanzi will indicate where he wishes to go with a lexigram for either the food at that site or the site itself. Kanzi's vocabulary has been systematically assessed at age 47 months (Savage-Rumbaugh et al. 1986) and at approximately 65 months (SavageRumbaugh 1988). Several kinds of test have been administered, but Kanzi's task in all of them is essentially to select the correct one of three lexigrams or photographs after seeing a photograph or lexigram or hearing a word. In some tests the word is spoken, in others it is produced by a speech synthesizer to eliminate any prosodic cues such as pitch and intonation, and in others it is prerecorded on tape and played only to Kanzi in order to preclude the possibility that the person administering the test (who does not prepare the tape) might somehow convey the 26

History of the ape-language projects answer to Kanzi. Test items apparently include all of the words assigned to his vocabulary on the basis of the behavioral-concordance criterion. By age five and one-half, Kanzi could respond correctly on 75-100 percent of trials with 149 spoken words, performing at chance on 45 others tested. When these 149 words were articulated by the synthesizer, he performed at the 75-100 percent level on 103 of them and at chance on the others, an achievement roughly equivalent to that of humans, who are able to recognize 65 percent of all words produced by the same synthesizer (Punzi and Kraat 1986, cited in Savage-Rumbaugh 1988). These results indicate that, whether or not the people in Kanzi's environment utter different words with consistently different patterns of prosody, Kanzi uses other acoustical features to identify these words. During a three-month period in the fall of 1986, when Kanzi was six years old, his comprehension of multisymbol utterances was assessed. His response to 310 spoken sentences of two or more words in everyday, nontesting situations was noted. Care was taken to avoid instructions requiring a behavior that would probably have occurred in that situation without the symbolic communication, such as asking Kanzi to put detergent in the washing machine while handing him the detergent and pointing to the washer. Kanzi responded correctly to 298 of the 310 sentences. Savage-Rumbaugh interprets the superior linguistic achievements of Kanzi and Mulika compared to previous ape-language subjects as a reflection of species differences in potential for symbolic communication. Pan paniscus, she believes, is endowed with "a greater propensity for the acquisition of symbols'' than other nonhuman species (Savage-Rumbaugh et al. 1986, p. 231). She suggests that Matata, the mother of Kanzi and Mulika, fell short because her language training began after the passing of a "critical age" for symbol acquisition in her species. OTHER PROJECTS Among the other ape-language projects, a series of studies by SavageRumbaugh and Rumbaugh on referential symbol use, conducted after the Lana project but before the work with Kanzi, would have to be considered a major contribution. A prodigious number of publications has issued from this research, which will be treated in chapter 5. In 1972, the Gardners undertook a sequel to the Washoe project. The goals and procedures of this study (1978,1984) do not differ from those of the previous one except that numerous of the project personnel are native signers of ASL, in contrast to the generally modest ASL competence of those who worked with Washoe. In addition, this second experiment 27

Background involves three chimpanzees (initially four) as subjects, and all were newborns when their sign training began. The first animal was acquired in 1972, the second in 1973, the third in 1975, and the fourth in 1976. Pili, the second acquisition, died just before reaching age 2. The last study that should be mentioned is that undertaken by Miles (1976, 1978, 1983) at the University of Tennessee. This project is notable mainly for the atypical identities of the researcher and subject. The former is an anthropologist, the only nonpsychologist among the ape-language experimenters, while Chantek, her subject, is the only orangutan participating in these studies.16 Chantek, a male, was captive born in December 1977 and was moved to a trailer on the campus of the University in 1978, when he began his language training at nine months of age. Miles employs a language she terms "Pidgin Sign English/' in which sentences are composed of signs combined in English word order but are devoid of articles and lack most of the grammatical morphemes of English. The achievements claimed for Chantek do not differ remarkably from those in the other projects. Miles (1983) observed that, as of 1983, there was no evidence that Chantek's sequences were sentences, as she had yet to perform any analysis toward that determination.

28

Part two

The Artificial-language Projects

The Lana project

THE NATURE OF LANA'S SYMBOL STRINGS The assessments of Lana's performance by the Lana researchers take a strong form of the tendency among ape-language researchers to overinterpret their animals' behavior in the direction of human language, elevating to linguistic status behavior more cogently explained in simpler terms. Lana's "sentences" provide a good example. It is clear that most of Lana's sentences were learned as rote sequences of key presses, yet Rumbaugh et al. gloss each lexigram within them with an English word. The inappropriateness of this practice is most clearly seen in the rendering of the request indicator as "please" (for example "please machine give apple"). Not only is there no evidence to suggest that Lana had any notion of the meaning of "please" or even a child's rudimentary understanding of the sociolinguistic rules governing its usage, but there is also little reason to believe that the other, more concrete terms were meaningful for her either. On the contrary, there is evidence that they were not: long after Lana had learned to "request" these incentives with her sentences, she required 1,600 trials to learn to push the corresponding lexigram when shown an M & M or a banana and asked for its name (Rumbaugh and Gill 1977). The Lana researchers have denied ever suggesting that every element in Lana's strings had linguistic significance (Savage-Rumbaugh and Rumbaugh 1980c), but one must wonder why these elements were nonetheless given English terms. Presumably, this was done to indicate to readers of the publications the concepts the experimenters intended Lana eventually to attain and to convey the pragmatic significance of each term, that is, what would result from its use. Lana did come to respond reliably to inquiries about the names of items presented to her, although we are not told just how general this ability was within the category of things nor whether the lexigrams for actions and prepositions were meaningful for her. 31

The artificial-language projects Thompson and Church (1980) analyzed a small sample of Lana's strings and determined that nearly every one of them was a variant of one of only six sequences. They then wrote a computer program that simulated Lana's performance. If the program was told (1) whether or not an incentive was present at the time of the utterance, (2) if so, whether or not the incentive was food, and (3) if so, whether or not the food had been loaded into one of the computer-controlled dispensers, it would reliably produce one of these stock sentences. Looking at a larger sample of 5,830 of Lana's sequences from a 22-week period, Thompson and Church found that 91 percent of those produced when no experimenter was present (that is, when Lana was directing her communications to the computer) were one of the six stock sentences. Each stock sentence had a fixed part and one or more variable slots, each filled by a lexigram for a food, activity, or addressee. If, for example, a food incentive was present and visible in its dispenser, then, according to the model, the stock sentence "please machine give (variable)" would be produced, with the lexigram for the incentive occupying the variable slot. The significance of the Thompson and Church analysis is its implication for the nature of Lana's string-production mechanism. Her performance could be based largely on conditional discrimination (modeled in the simulation by choice of one of the stock sentences as a function of contextual stimuli) coupled with paired-associate learning (equivalent to filling the variable slot in the stock sentence with the incentive's associated lexigram). Pate and Rumbaugh (1983) performed a similar analysis on a sample of Lana's sequences from a later period in her training, when the sevenlexigram limit on strings in Yerkish had been lifted. Conceding that the Thompson and Church interpretation of Lana's output might be applicable to the early period, they contend that the variability in her later constructions renders the stock-sentence approach inadequate. Their sample covered Lana's responses during question-and-answer drills over a 25-day period. The questions put to her concerned (1) the name of a food or drink, (2) the container that held it, and (3) the latter's location. Although most of the 512 well-formed sequences included highly recurrent phrases such as "... name of this," "(question) you ...," or "Lana want ...," the phrases these units were combined with were numerous, resulting in 69 different "stock sentences," that is, recurrent construction types. This magnitude of diversity, they argue, militates against the notion of the stock sentence and suggests the operation of a system of productive grammatical rules, albeit with phrases rather than lexigrams as the elemental constituents. In order to assess this rejoinder to the Thompson-Church study 32

The Lana project properly, one would need to know certain things about the questions put to Lana during these sessions. Unfortunately, the authors provide no information regarding the phrasing of questions, so that no conclusion can be drawn about possible influences of variability in question structure on variability in the form of Lana's responses. The color coding of Lana's lexigrams according to semantic category undercuts the presumption that each of her lexigrams had a distinct meaning for her. For although this methodological provision was never utilized by the experimenters, it is plausible that Lana used it, and in an unforeseen way - instead of learning to compose sequences of individually meaningful terms, she might have learned merely a finite number of color sequences that reliably resulted in reward. Straub et al. (1979) succeeded in training a pigeon to peck four keys of different color in a consistent sequence regardless of the way in which the keys were arrayed, a performance that may be cognitively equivalent to Lana's producing stock sequences despite frequent rearrangement of her lexigrams. Ristau and Robbins (1979) were the first to point out another flaw in the Lana-project methodology. As mentioned above, the researchers claimed that Lana was sensitive to the ordering of elements in Yerkish as evidenced by her use of the period key to terminate her own ungrammatical sentence beginnings or those created by others. This use of the period key was not part of the training design - Lana developed it spontaneously (Rumbaugh and Gill 1977). The problem with this interpretation of Lana's use of the period key is obvious and severe: How does one "know" that Lana meant "erase," "ignore," or "I didn't mean that" or that Lana meant merely to end her sentence? Surely she had learned that failure to hit the period key led to no food nor drink, and, in general, was not rewarded since all sentences must end with a period for an event to follow. It would appear that when the experimenters determined that the communication was in error, then Lana must have "meant" to erase. Thus, at the most basic level of analysis, data that are inconsistent with some linguistic rule are being eliminated by the experimenters. (Ristau and Robbins 1979, p. 272)

This criticism applies less effectively to Lana's responses to sentence beginnings provided by the experimenters, for she terminated incorrect beginnings much more frequently than correct ones, which she would instead proceed to complete. However, the sequences used in testing Lana were quite short, extremely familiar, and few in number - "please machine give" and "please machine make" were the only correct beginnings provided. The first could be combined with lexigrams for juice, milk, and so forth to make a grammatically correct and hence rewarded expression, while the latter took "music," "movie," and the like. Invalid beginnings included transpositions of elements within a valid one, as in "Please give 33

The artificial-language projects machine/' or parts of a valid one combined with a randomly chosen lexigram. Lana's rejection of the invalid starts and addition of her name and/or an incentive lexigram from the proper class to the valid ones becomes much less impressive when the details are thus presented; it presages the coming of syntax less than it suggests the operation of a chain of well-ingrained stimulus-response sequences. Such response chaining has many precedents in the animal-psychology laboratory, and Lana's performance here can be characterized as "the formal and pragmatic equivalent of that of a rat in a double T-maze" (Terrace and Bever 1976, p. 584). Anecdotes

As is true of the other ape-language projects, the Lana group provides compelling anecdotal evidence of language in lieu of comprehensive statistics on such performances. These accounts, moreover, generally do not detail the linguistic and nonlinguistic context of the episode reported. For example, Lana's production of "novel constructions" is mentioned in a number of articles. Her referring to an orange as "apple which-is orange" (the color orange) before she had learned a lexigram for the fruit is cited several times as a significant instance of this. Standing by itself, this does seem a notable achievement. Seen in context, however, as provided in Rumbaugh 1977b (which contains the most extensive corpus of data from the study available), the construction appears rather more like the end result of a fairly haphazard sequence of attempts to gain an incentive. In the following exchange, Tim, Lana's trainer, was holding an orange outside Lana's cage. The parenthetical material is commentary by the authors of the article from which the exchange is excerpted (Rumbaugh and Gill 1977, pp. 178-79). Tim: ? What color of this. Lana: Color of this orange. Tim: Yes. Lana: Tim give cup which-is red. (This was probably an attempt to request the orange ...) Tim: Yes. Lana: ? Tim give which-is shut. ? Shelly give. Tim: No Shelley. Lana: Eye. (A frank error, probably.) ? Tim give which-is orange. Tim: What which-is orange. 34

The Lana project Lana: ? Tim give apple which-is green. (At this point, Lana frequently confused the keys for the colors orange and green.) Tim: No apple which-is green. Lana: ? Tim give apple which-is orange. Tim: Yes. The "apple which-is orange" locution is thus preceded by several senseless and apparently random variations on it. It is also clear that Lana did not spontaneously decide to indicate a nameless entity by composing a referring expression using one of the item's attributes - the "color" motif of this discussion was created by the trainer, not by Lana. The skeptical description of Lana's output as rote strings of lexigrams with some random combining of strings is corroborated by a statistic mentioned in the Pate and Rumbaugh study (1983) of Lana's syntactic productivity, mentioned above. Of the 881 productions analyzed in the study, fully 369, or 42 percent, were incomplete or otherwise grammatically incorrect by the rules of the Yerkish system. Note that this error rate is from a period in which Lana's linguistic capabilities were alleged by the authors to have exceeded the adequacy of the Thompson-Church model. One of the most striking specimens of overattribution is Gill's (1977) account of Lana's unprecedented use of the past tense. Here again, scrutiny of the context calls into question the generous interpretation forwarded by the researcher. In this exchange, Tim Gill, the trainer, has placed cabbage in the vending device, intentionally refraining from complying with Lana's request for juice in order to elicit conversation with her (Gill 1977, pp. 243-44). The parenthetical comments are Gill's. Lana: Please machine give piece of cabbage (twice). Tim: ? What in machine. Lana: Tim drink. Tim: ? What in machine. Lana: Tim swing. (This was probably a typing error.) Tim: What in machine. Lana: Tim put cabbage in machine. ? Tim put juice in machine. Here is Gill's assessment of Lana's sequences: Lana had not learned the past tense; however, the sentence Tim put cabbage in machine clearly implies that she meant to use it, for that sentence describes what I had indeed just done ... She later used this same form once again to describe a past event... Lana's suggested use of the past tense is important, for it implies that a concept of the past is cognitively present in her. (Gill 1977, p. 244) 35

The artificial-language projects There are at least three reasons to reject this interpretation. Firstly, the epochal statement, like the "apple which-is orange" form, is preceded by two nonsense strings. Secondly, Lana did not indicate in subsequent usages of this or other verbs that she possessed a notion of the past - Gill reports only that she used "put" in the past tense on one later occasion, and no specifics are given regarding this event. Lastly, it seems highly likely that Gill's reading of this usage is an unwitting imposition of the English interpretation of "put" in this context, for it happens that "put" is one of the small class of English verbs having the zero-form past tense indicator, that is, the word has the same phonetic form in both the present and the past tense. I am suggesting, in other words, that Gill would not have made the same interpretation if Lana had said, for example, "Tim load cabbage in machine" or "Tim place cabbage in machine." Such intellectual laxity is fairly pervasive in this project. SavageRumbaugh, Rumbaugh, and Boysen (1980, p. 57), for example, state that "Lana used symbols that were formed by recombining 1, 2, or more of 9 graphic elements, just as English words are written by combining 1, 2, or more of 26 alphabetic elements. Thus Lana, unlike Washoe or Sarah, was forced to attend to each element within the stimulus complex, not just the overall form." The first problem with this assertion is that the elements of the English alphabet or any other alphabet form a code in which, generally speaking, there is a systematic relationship between an element and its pronunciation. The individual design elements of Yerkish lexigrams, by contrast, have no representational properties - they do not stand for anything - and are thus like letters only in the relatively trivial fact that both go to make up other entities. In addition, while Lana may have dealt with her lexigrams by decomposing them into elements, there is no evidence that she did this as opposed to merely processing them as gestalts. Nor would she have any need to; Lana was not composing lexigrams, she was pushing them, ready-made. On the other hand, the signs of American Sign Language are each made up of three or four formational components from a finite set, so that, to the extent that Washoe cognitively exploited the combinatorial make-up of signs (see chapter 5), the intended contrast between Lana's symbols as decomposable and Washoe's as irreducible wholes founders. The Lana experimenters' rather optimistic or, less charitably stated, superficial, assessment of Lana's achievements is not unexpected given the limited, at times crude, understanding of language that underlay it. Rumbaugh (1977a, p. 128) asserted that "the chimpanzee language projects should, in due course, lead to a clarification of a number of problems in the area of learning and of the processes entailed in formulating adaptive behavior patterns such as in the production of sentences." Thus, 36

The Lana project if Lana's sequences were "adaptive" for her - and they clearly were, since the staples of her existence were contingent upon them - then she was engaged in sentence production, provided, of course, that they also gave evidence of grammatical knowledge, such as seen in her completion of valid and rejection of invalid starts. This reduction of language to problem solving is seen in extreme form in Premack's work with Sarah, discussed in the following chapter. SUMMARY The Lana project was an innovative experiment that was flawed in method and misguided in conception. There is a paucity of evidence that the lexigrams in Lana's strings had separable meaning for her and substantial indication that her sequences were rotely acquired wholes, a limited number of stock phrases developed for each of a small number of contexts, with substitution of lexigrams in variable frames, depending upon what incentives and/or trainers were present. Instances of allegedly intentional innovation in construction, when assessed in context, appear at least as likely to be random output.1 The claim that Lana appreciated grammatical principles is undermined by the color coding of keys and indeterminate significance of Lana's use of the period key. The soundness of this deflationary account of Lana's behavior is supported by some authorities one would not expect to be among the project's detractors, namely the Lana experimenters themselves. For in recent years, Rumbaugh and Savage-Rumbaugh have reassessed the Lana project, usually by way of contrasting it with their recent work, which they feel avoids the shortcomings of the Lana project. The Rumbaughs now regard the earlier work as defective because of its untested assumption that Lana's lexigrams were symbols as opposed to mere paired associates for certain stimuli (Savage-Rumbaugh and Rumbaugh 1978,1980a, 1980b; Savage-Rumbaugh, Rumbaugh, and Boysen 1980; Savage-Rumbaugh et al. 1980, 1983a, 1983b). (They have also repeatedly criticized the other language projects on the same grounds.) Thus, one contribution of the ape-language experiments and the debate they engendered has been the reorientation of certain psychologists steeped in behaviorism, which so often substitutes the experimentally tractable for what is scientifically interesting, towards a properly cognitive approach to language.

37

The Sarah project

In Intelligence in Ape and Man, Premack (1976a) characterizes his studies as focused on intelligence rather than language per se, language being of interest only insofar as it illuminates the former. He also maintains that the language imparted to Sarah was not intended to simulate a human one. Despite these disclaimers, it is clearly language and its cognitive bases that are the topic of Premack's writings about Sarah. Furthermore, Premack implicitly and often avowedly regards Sarah's accomplishments as cognitively identical to aspects of human language. In reality, though her intellectual achievements may have exceeded those of Lana, Sarah's are probably no more languagelike than Lana's. THE SEMANTICITY OF SARAH'S CHIPS The major critique of Premack's work with Sarah was made by Terrace (1979a), and the following discussion relies extensively on that work. It is interesting that this most effective critique came not from a linguist anxious to beat back an interloper into the domain of human language but from a fellow behavioral psychologist, whose expertise in experiments on problem-solving by lower animals resulted in a critique "from below" rather than "from above," that is, an argument that Sarah's performance is not different in nature from that demonstrable in rats and pigeons. In a nutshell, the Premack study, like the work with Lana and, less obviously, the signing projects, suffered from rampant overinterpretation. As an example, consider a typical "sentence" that Sarah learned to compose in order to procure an incentive: "Mary give Sarah apple." If one were told that Sarah reliably constructed this sequence in the presence of an apple, substituting the chip for banana if this were the incentive, one might well be impressed. The performance becomes less compelling, however, when one learns that this sequence was composed only during sessions in which this form was the single construction being trained and 38

The Sarah project tested, that Mary was the only trainer involved in giving things to Sarah, and that no actions other than giving incentives to Sarah were involved in the session. Furthermore, no one other than Sarah was receiving incentives. In other words, since the only element taking contrasting forms during the session was the food incentive, there is no basis for attributing semanticity, by assigning them English glosses, to the chips "Mary," "give," and "Sarah." This problem of gratuitous attribution was quite general across the various constructions trained and tested. Sarah learned to construct the appropriate sequence, say "red on green" versus "green on red," when shown one colored card on top of another one, and also to arrange the two cards appropriately in response to such a "statement" assembled by the trainer. But this skill does not support Premack's conclusion that Sarah understood prepositions. Since "on" is not contrasted with other prepositions, such as "under," it would not be necessary for Sarah to attend to that element at all. For example, she could solve the second version, as Terrace points out, merely by noting which item is mentioned first and placing it on top of the remaining one. Considering that almost all of her training and testing sessions involved only a single paradigm (for example, " on ," " same as/different from ," " color of "), and that the alternatives being contrasted within a paradigm were almost always just two in number, it is quite possible that Sarah could attain her characteristic 80 percent level of correctness (Premack 1971) on a paradigm through memorization of that session's constructions and the correct, that is, rewarded, response for each. This approach, of course, would require no linguistic decoding at all - the chips could well be meaningless to Sarah. Farrer (1967, cited in Gardner and Gardner 1978) found that chimpanzees readily memorized the arbitrarily correct (reinforced) response for each of 24 different arrays of visual stimuli. Transfer tests

Obviously, his research would be of little interest if Premack's procedure consisted solely in training Sarah on a problem and then testing her performance on repeated trials with the same items. These results would be subject to the interpretations just mentioned. It is the so-called transfer tests that are of greatest interest. These are intended to determine the extent to which the animal's experience during training has engendered a general capability, one that can handle - transfer to - new instances of the training sort or whether, alternatively, the performance was narrowly adapted to the specific contents of the training tests. 39

The artificial-language projects Sarah's success level on transfer tests was generally equal to the performance she attained by the end of the training phase of a paradigm. For example, if the banana and apple of the "Mary give Sarah" model were replaced by other incentives, Sarah would substitute the appropriate chip. New exemplars did not impair her competence in labeling pairs of objects as "same" or "different," and she could indicate whether red is or isn't the color of banana just as reliably as she had learned to write during training that red is the color of cherry. The problem with Premack's transfer test results, or at least with his manner of presenting them, is the practice of lumping the results of all trials into a single statistic (1971,1976a). If the object of a transfer test is to find out the extent to which an animal has induced an abstract, contentindependent solution for a problem type, then only the results of the first presentation of each new problem should be looked at, since performance on subsequent trials would be confounded with the effect of new learning. Harlow (1949) coined the term "learning set" to describe the discontinuous increase in the rate of learning from one set of experimental trials to the next typically seen in primates.1 Consider an animal that has learned to discriminate the larger within each pair of items in a series of pairs. When introduced to a new series of trials, in which not the larger but, say, the brighter of the two items is the rewarded choice, the animal will typically reach a high level of correct choices much more rapidly than it did on the first series, often attaining 100 percent accuracy after the first trial, which can be said to "inform" the animal what is wanted of it. Learning set thus has to do with learning to learn. The importance of distinguishing Sarah's first-trial performance from subsequent results should be clear in light of the learning-set phenomenon. If she regularly responds correctly on the first trial of her transfer tests, then the hypothesis that she has induced some general rule becomes more likely. If, on the other hand, she performs basically at chance on her first trials, then, regardless of the fact that she might immediately achieve errorless performance, the hypothesis of choice would be that, in the training phase, she acquired not an abstract strategy but rather a way to apprehend quickly what the rewarded behaviors are. Premack, however, generally presents only the ratio of correct to total responses in the series of transfer tests for a paradigm. Occasionally he varies from this practice, specifying the ratio for the first five transfer trials. This indicates his awareness of the importance of the initial transfer trials but, of course, only informs as to the first-trial outcome when the ratio of correct to total responses is unity.

40

The Sarah project Complex constructions

As mentioned earlier, Sarah was trained in two construction types that involved more than one proposition in a single string. Her achievements in this regard are of particular interest, for complex sentences are held by linguists to be a signal feature of human languages. For this reason, a brief digression on the significance of such constructions may be justified. There are several grammatical mechanisms for producing complex sentences. Coordination, the simplest, makes possible constructions such as "Ed Meese eats cheese and cookies/' a conflation derived from the component sentences "Ed Meese eats cheese" and "Ed Meese eats cookies/' The rule for derivation of such a coordination would involve the definition of a sentence as a noun phrase plus a verb phrase connected by "and" to another sentence. (Additional rules would provide for the deletion of redundant constituents.) Complementation refers to structures like "I heard Dan Quayle got a C on his urine test," in which one sentence takes the role of object within another. Relativization provides for "The cheese that Ed eats is green" and "George passed the urine test that Ron administered," where one (syntactically altered) sentence modifies a noun of a sentence in which it is embedded. Complementation is based upon rules in which a sentence is defined as a noun phrase plus a verb phrase and verb phrase is, in turn, defined as a verb followed (optionally) by a sentence. The rules supporting relativization, similarly, define a noun phrase (first example) or a verb phrase (second example) as optionally containing a sentence. What these rules or rule sets have in common is tautologous definitions, certain terms and their equivalences acting as partial definitions of each other. Or, to put it more clearly, these are rules in which the definition of a syntactic entity, such as "sentence" or "noun phrase" or "verb phrase," includes that entity itself as an optional ingredient. Such rules are known as recursive rules. It should be obvious that if, for example, a sentence consists of (is defined as) a noun phrase plus a verb phrase, and a noun phrase can itself consist of or contain a sentence, then there is no theoretical limit to the length of a sentence, since repeated (recursive) application of the rules could produce an endless nesting of sentences (sentence-like phrases, strictly speaking).2 Hence "The woman who lives in the house that sits on the hill that ..." Recursive rules thus confer great power on natural languages and certain other representational systems, since they are a finite means for producing an infinite number of expressions (Chomsky 1957). Such systems can be contrasted with, on the one hand, those natural 41

The artificial-language projects communication systems in which messages can vary only along a single dimension and in a merely quantitative fashion and, on the other, those having qualitatively distinct signals but lacking combinatorial principles. Systems of the first sort provide for an infinitude of messages that are, apart from affective intensity, semantically homogeneous, while those of the latter entail some diversity in message content but a finite and, apparently, small number of messages. (Chapter 8 explores more fully some principles of natural communication systems.) Recursive rules can economically generate expressions to represent an entity that has no simple name or for which the speaker lacks a name (Limber 1977). "Just order the food you most like to have" is an example, or, more complex, "Just order the food you would most like for me to have," which

is, in essence, a sentence inside a sentence inside a sentence. One of the profound lessons in Brown's seminal A First Language (1973) is the demonstration that the young child's evolution from two-word utterances to longer ones entails a process very much parallel to the use of complements and relative clauses at a later stage. The expansion of utterances beyond the two-word threshold involves (1) the addition of formerly unuttered arguments or predicates, as in the elaboration of "Dog bite" into "Dog bite me" and (2) the "unfolding" of single-word terms into multiword terms. It is this second development that is germane here, for it entails the embedding of a formerly autonomous construction, with its own syntactic structure, within another structure. Where before only two-term, two-word constructions, such as "Dog bite" or "My dog" were produced, instances of the agent-action and possessor-possessed semantic relations, respectively, now a two-term construction might consist of one such relationship serving as a single term within another, as in "My dog bite," in which the single-word noun phrase "Dog" in "Dog bite" is replaced by the two-word noun phrase "My dog." The parallel between adults' complex sentences and children's threeword productions is not that both represent members of infinite sets there is, to my knowledge, no evidence that children operate with linguistic rules providing for the embedding of sentences.3 The parallel is found in the fact that, in both cases, a constituent may consist of either a single word or a multiword construction. Against this background, the significance of Sarah's handling of constructions that are at least analogous to complex sentences in that they contain more than a single proposition, propositions being the ideational contents of sentences, becomes apparent. The two major construction types here were the multiple-instruction string and the conditional. Sarah eventually attained her typical 80 percent accuracy level in responding to constructions such as "Sarah insert banana pail apple dish." 42

The Sarah project Here again, however, the significance of the performance is attenuated when its full context is considered, in this case the training that preceded the achievement. First, Sarah was trained to respond correctly to the four "atomic" sentences "Sarah banana pail insert," "Sarah apple pail insert," "Sarah banana dish insert," and "Sarah apple dish insert." Next, she was given two sentences at a time, presented side by side. When she had reached an acceptable level of accuracy in carrying out both commands,4 the two sentences were combined. When, again, she proved competent with the arrangement, a progressive deletion of redundant elements was begun, gradually reducing "Sarah banana pail insert Sarah apple dish insert" to "Sarah banana pail apple dish insert," a process arguably equivalent to grammatical coordination. In transfer tests, Sarah successfully negotiated constructions with different objects, a different verb ("take"), object modifiers (for example, "Sarah cracker red dish candy yellow pail insert"), and multiple objects, as in "Sarah cracker banana dish apple pail insert" (Premack 1976a). Premack ascribes Sarah's success to a grasp of the hierarchical structure of these sequences. She understood that the scope of application of "Sarah" and "insert" was the entire sentence, whereas each receptacle token and the object token immediately above5 it pertained only to each other. This attribution, however, seems unwarranted, given the long sequence of antecedent trials with simple constructions in which Sarah learned to place the food mentioned in the receptacle indicated. It is not remarkable that she was able to "read" a compound sentence, since this terminal performance was methodically shaped in a piecemeal fashion (Brown 1973). In addition, the "Sarah" token need not have had any semantic content, since Sarah would certainly know anyway that it was she who was to solve the problem presented - no one else ever did. Similarly, "insert" was probably superfluous, considering the sequence of training problems, in which she did nothing but insert. Her success when "take" was substituted for "insert" might suggest otherwise, but in these cases she was presented not with empty containers, as before, but with the containers already filled with the items that were to be removed, visual information that would obviate even attending to which food token was adjacent to which container token in the string.6 In short, Sarah could have solved this problem type merely by following such a nonsyntactic strategy as "Operate on all of the objects listed before the name of a container in the manner specified by the verb at the end of the sequence" (Terrace 1979a, p. 167). The other multiproposition form trained was the conditional. Sarah allegedly learned to read strings of the type "if (something is done) then (something will be done)." The simplest case involved constructions like 43

The artificial-language projects "if Sarah take apple then Mary give Sarah chocolate" and "if Sarah take banana then Mary no give Sarah chocolate." Presented with such a pair of compound constructions and apple and banana as alternatives, Sarah would, as indicated in the instruction, be rewarded with chocolate on making the correct choice. When, after numerous trials, Sarah was judged competent with this form, constructions believed to be more challenging were introduced as transfer tests. She was given single instructions such as "if Mary give banana Debby then Sarah insert chocolate dish" or "if Debby give apple Mary then Sarah insert cracker dish." Sarah made only 1 error on 11 such trials, but, as Premack himself points out, the linguistic significance of the performance is diminished by the fact that, in these trials, the antecedent was in fact always carried out, so that Sarah would have had only to attend to the consequent part of the instruction. And, as discussed above, syntactic analysis is a more powerful capability than is required to account for her correctly interpreting simple strings like "Sarah insert cracker dish." (In fact, all of the consequents in this series of trials were either that string or the same one with "chocolate" in place of "cracker.") In the most complicated of her transfer tests, Sarah was presented with pairs of conditional constructions in which the antecedent in each of the two was one of these four: "Mary give green Debby," "Mary give red Debby," "Debby give red Mary," and "Debby give green Mary." The antecedent was combined with either "Sarah eat candy" or "Sarah eat cracker." Thus one test might consist in showing her the string "if Mary give green Debby then Sarah eat candy" together with "if Mary give red Debby then Sarah eat cracker," with candy and cracker as her alternatives. Premack reports that Sarah made 2 errors in 8 such trials. He contends that her performance could not have been based on mere memorization of which consequent went with which antecedent, as opposed to analyzing fully each construction, since each of the four antecedents was sometimes combined with one consequent and sometimes with the other. However, the listing of the sentence pairs (1976a, p. 242) contradicts this description. In fact, there is correlation between antecedent and consequent, for each antecedent mentioning green was combined with the "Sarah eat candy" consequent, each mentioning red with "Sarah eat cracker." Thus Sarah would not have had to analyze the antecedent clause but simply note what color was mentioned in it and choose the correlated consequent. In fairness, though, it is highly unlikely that Sarah - or a person, for that matter - could have observed this correlation quickly enough to exploit it within eight trials. Furthermore, it seems most probable that Premack's description of the transfer-test procedure was correct, the contradictory listing being a publication error. Whatever was the actual case, Sarah's 75 44

The Sarah project percent performance here is at best suggestive, since, with only two items to choose from (candy and cracker), the predicted performance under random choosing would be 50 percent, and the likelihood of 75 percent correct choices over just eight trials is far above the probability threshold below which statistical significance can be claimed. SUMMARY Sarah's evanescent oeuvre entailed, at least, the development of skill in solving a number of problems devoid of linguistic significance and, at most, acquisition of a number of unrelated atoms of linguistic ability. As Brown (1973) observed, Sarah's various constructions evinced little if any systemic connectedness, such as that found in the provision of natural languages for transforming any sentence (for example, "Sarah insert banana pail apple dish") into a different modality ("Did Sarah insert apple pail banana dish?") or a different syntactic structure ("Apple pail banana dish insert by Sarah"). Similarly, there were few instances in which a construction learned in one training and testing set was incorporated into another, such as "if red color-of apple then Sarah take banana." And there were still fewer cases of alternation between single- and multi-term instances of a given sentence constituent, which, as discussed above, is a hallmark of the "nested" structure of language (Fodor, Bever, and Garrett 1974). Let it be said, however, that Sarah cannot properly be faulted for these shortcomings, since, unlike Lana's system, Sarah's did not allow for spontaneous displays of syntactic prowess - Sarah had available only the chips being used during each training paradigm. It thus seems unlikely that Sarah could have come to understand even that her constructions or those of the trainer were communicative in nature. The narrow focus of each training session and small number of alternatives means that the possibility of memorization of correct responses cannot be dismissed. The ascription of semanticity to string elements that had no contrastive function was no more justified here than in the Lana project. And the potential importance of transfer tests in demonstrating the meaningfulness (though not the specific meaning) of Sarah's tokens was vitiated by the statistical obscuring of possible differences between first and subsequent transfer-test results. The nonlinguistic interpretation of Sarah's achievements is corroborated by studies demonstrating that invented "languages" essentially like Sarah's can be learned by persons with profound language deficits, including globally aphasic adults (Glass et al. 1973) and children with developmental aphasias (Hughes 1974/75). (Similarly, some success has been 45

The artificial-language projects achieved in instilling elements of the Yerkish system in severely retarded, alinguistic children [Parkel and Smith, Jr. 1979].) This suggests that whatever faculties underlie mastery of these systems by humans and infrahumans, they are not the same as those responsible for language. Premack, even more than the Rumbaughs, grew increasingly cognizant of the fundamental differences between his chimpanzee's performance and the language of the young child. He now characterizes Sarah's grammatical inductions as involving substitution of tokens in invariant structures, as distinguished from a system of transformational relationships among structures, which is, again, a defining feature of naturallanguage syntax.

46

Part three

Apes and Language: Ontogeny

Words

The emergence of syntax, the combination of meaningful elements into propositions, is developmentally preceded by the solitary appearance of those elements themselves. In the child's development, as in human evolution, the advent of words - culturally forged markers for shared notions - confers the ability to convey the conceptions of one's mind to that of another, albeit imperfectly. In a very immediate way, the word allows us to control other people's minds, which provides a perverse but perhaps useful way of understanding the hackneyed observation that culture is based upon language. While nonverbal forms of communication, in our own and other species, can also be said to consist in this transfer of mental contents, words allow much greater precision in this, sentences an exquisite degree. The first words, which seem to emerge out of nothing around the child's first birthday, do not, of course, appear ex nihilo. There is a developmental history to these earliest utterances, one that has been charted in great detail by Bates (1976,1979), Bloom and Lahey (1978), and Bruner (1983), to name just the Bs.1 PRECURSORS OF THE WORD Utter neonates are communicative to the extent that their vocalizations, gestures, and expressions convey information about their condition to caretakers. Though this communication may be unintended on the infant's part,2 it is effective in eliciting the kinds of nurturance that undoubtedly account for its having been naturally selected. Such controlled utterances are soon supplemented by still unconventional but intentional vocalizations and gestures. These articulations, which tend to accompany the child's taking notice of something, acquire a certain consistency in sound or appearance, although they are quite variable from one child to the next. 49

Apes and language: ontogeny A development of profound significance is the subsequent application of these initially undirected productions to the manipulation of the caretaker's attention, for eventual linguistic communication requires the coordination of attention. In fact, most communicative developments from this point up to the first true words can be seen as refinements of the child's ability to control the attention of her interlocutor. Vocalizing, showing or giving of objects, and pointing, singly or in combination, are used to direct the adult's attention or to procure a desired object by way of the adult. The intentional nature of these actions is seen in the child's monitoring of the adult's attention by repeatedly looking at the latter's face to determine the focus of gaze. This sequence of developments occurs in an essentially uniform manner around the world, and there is reason to regard it as innately ordained. The demonstration by Scaife and Bruner (1975) that infants as young as three and one half months will orient their attention toward the focus of the adult's gaze strongly suggests that the child is attuned to the achievement of joint attention. It is interesting to note that nonhuman primates, though they will follow the line of regard of another group member (Chance 1967), do not appear to direct the attention of others to individuals or things other than themselves. Both Premack (Premack and Premack 1983) and the Rumbaughs (Savage-Rumbaugh and Rumbaugh 1980a) report that over the course of various experiments their subjects developed the practice of pointing, but pointing does not occur in nature.3 LINGUISTIC SYMBOLS

In the following months, the child's vocalizations, including those that already resemble words of the language, are gradually transformed from verbal equivalents of pointing into true linguistic symbols. By linguistic symbol I mean a physical marker for some mental representation, which, in turn, may or may not stand for some thing or class of things in the real, that is, nonmental, world. In addition to functioning as the medium for something like thought transference, the symbols of natural language - words and signs - facilitate the retrieval of information about real-world entities from one's knowledge store and the performance of mental operations on those entities in their absence, in other words, thinking. This definition of the symbol omits mention of two issues that figure importantly in variant definitions. First is the question of arbitrariness versus iconicity. It is true that for nearly all words in spoken languages, there is no isomorphism between the physical properties of the word and 50

Words its referent. But it is not self-evident that this attribute of language derives from any functional imperatives. Words that were to some degree iconic representations of their referents could serve precisely the same functions that words in fact do serve.4 Whether iconic symbols would be easier or harder to learn and retain is uncertain. It might seem that a symbol similar in form to its referent would carry a mnemonic advantage. However, such a system would require as many word-forming elements as there were features in the world of referents (depending upon how finely analyzed that world was to be), whereas the noniconic human languages make do with an economical 20 or so distinctive features for building words. In short, it is not clear whether arbitrariness should be considered essential to the linguistic symbol. The other issue skirted in my definition is whether or not the symbol's user must be consciously aware of the symbol's nature, that is, that it is not the thing it represents, that the association between the two is conventional rather than natural, that it can be used for communication, and so forth. It is safe to say that most adults are aware of these things, inchoate though that knowledge may be in some instances. Here again, however, is an empirical fact about symbol use that may not be indispensable. Insistence on conscious awareness of the symbol as symbol would mean that young children, unintelligent adults, and, most likely, the trained apes were not engaging in symbol use when talking, a proposal that, at least in the first two cases, is untenable. One of the most useful contributions of the recent work on this earliest linguistic stage is the demonstration that even when a word is used by the young child consistently in appropriate contexts - for example, only in the presence of a certain object - it is not the case that she intends by this word what an adult would. Nor is this merely a matter of immature definition or overgeneralization. Rather, the first words may not have any meaning in the conventional sense. Instead, their utterance may be merely a ritualized part of recurrent activity contexts, only nominally more linguistic than the nonverbal behaviors that also define these contexts (Bates 1979), or it may be an attention-directing behavior (Atkinson 1979), or both. This perspective is based on the Piagetian understanding of early words as verbal components of sensorimotor routines. It also involves the application of speech-act theory (Searle 1969) to the prelinguistic era. Thus, these primitive words are termed "pure performatives" by Greenfield and Smith (1976) and "protoimperatives" and "protodeclaratives" by Bates, Camaioni, and Volterra (1979) - utterances conveying no propositions, devoid of referential intent, and serving only to direct the attention of the other. 51

Apes and language: ontogeny REFERENCE The transition to symbolic, referential usage entails a process of decontextualization, a psychological decoupling of word and object. Production and understanding of a word move from initial dependence on a whole activity context to presence of a certain item to, eventually, reference in the absence of the entity mentioned. 5 In referring, we use a word or words to indicate the focus of our attention and, usually, to direct the attention of the interlocutor towards the same focus. This rendering of the notion of reference is intentionally noncommittal regarding the definitional argument as to whether reference should be understood as a relationship among one mentality, a word, and a thing, or rather one mentality, a word, a thing, and another mentality. Whether a recipient of the communication is a defining element for reference seems to me reminiscent of the question about the tree falling in the woods. But a description of this decontextualization is not an explanation of the child's attainment of the "semanticity hypothesis'' in the first place, the notion that things have names and that words are names for things.6 This development has not been captured in any cogent account involving learning, and Macnamara (1982) argues that the act of referring should be regarded as a cognitive primitive.7 The child, in other words, is simply disposed at birth eventually to interpret certain acts of others as referring and, of course, to refer himself. It is well to dwell on the implications of treating [the act of referring] as an undefined primitive. It means that we give up the search for elements out of which it might be constructed. It does not follow, however, that we abandon the search for necessary and sufficient conditions for when the act of referring occurs. These would be extrinsic conditions. For example, referring might be the interpretation that a child automatically imposes on a certain event because of characteristics of the event. For example, he may be so constituted as automatically to impose the category, referring, when his parents use a word when drawing attention to an object for which he has hitherto learned no name. My suggestion is that [children] are endowed with an intentionality space akin to Quine's quality space ... that favors the attribution of one intention rather than another granted certain events. (Macnamara 1982, p. 179) The critical reader might object to the invoking of innate knowledge in explaining developmental phenomena that prove refractory to accounts involving learning. Taken to the limit, this practice would preclude efforts to explain mental development by environmental contingencies, inductive reasoning, or the "epigenetic constructions" of Piaget. In fact, the philosopher Jerry Fodor (1975) makes essentially this extreme argument, 52

Words asserting that the very notion of learning new intellectual structures is infirm, and that mental growth consists instead in the successive revelation of properties that are in the mind from the outset. Whether Fodor is right about the origin of ideas in general or Macnamara is right about reference in particular, it is a matter of fact that young children, certainly by the age of three, are implicitly aware of the relationship between words and things that is denoted by the term symbol. What is important in all of this for the question of ape language is that while children effortlessly come to refer and to interpret referring by others, it is not beyond question that the animals in the projects under discussion demonstrated this ability, notwithstanding their arduous training. Although Savage-Rumbaugh and her colleagues (for example, Savage-Rumbaugh, Rumbaugh, and Boysen 1980) have been raising this question persistently in recent years, the majority viewpoint even among those unwilling to accord grammatical ability to the apes has been that their individual signs were in the important ways identical to the words and signs of children (Bronowski and Bellugi 1970; Brown 1970; Hill 1978; Klima and Bellugi 1972; Limber 1977; McNeill 1974; Slobin 1979). NATURAL GESTURES If they are not linguistic symbols, what are the thousands of gestures, not part of the natural communicative repertoire, that are emitted by these animals? A first answer is that a substantial number of them are, in fact, natural gestures improperly treated by the experimenters as signs (Seidenberg and Petitto 1979). Remarkably, the Gardners (1969) regarded the natural gestures of the chimpanzee not as a hazardous confounding factor, to be carefully distinguished from signs, but rather as an asset, as a welcome addition to their animals' lexicons. The gestures so incorporated include a begging/beckoning gesture (outstretched hand, open palm facing up) and the "hurry" gesture (vigorous shaking of the hand at the wrist).8 These natural gestures made up a significant proportion of Washoe's productions and thus inflated measures of the extent of her signing activity. For example, in their first major description of the Washoe project, the Gardners report that "four signs - 'please', 'come-gimme', 'hurry', and 'more' - used with one or more other signs, account for the largest share of Washoe's early combinations" (1969, p. 671). Thus, two out of four of Washoe's favorite early signs are not, in fact, learned symbols but rather natural behaviors. And there is reason to believe that the "more" sign was, if not a natural gesture, then not a linguistic symbol either. The following quotation, from The Other Side of Silence, an account of the language and 53

Apes and language: ontogeny social life of the North American deaf community (Neisser 1983), contains the impressions of a deaf ex-volunteer to the Gardners' post-Washoe project, described in chapter 2. This man, who spent several months on the project, wished to remain anonymous in his criticism of the experiment. His testimony is worth presenting at some length, not just for the information provided about the "more" sign but because it conveys rather well the general laxity of interpretation endemic to the sign-language projects. They kept all kinds of records. The most important was the logbook of signs. Every time the chimp made a sign, we were supposed to write it down in the log ... They were always complaining because my log didn't show enough signs. All the hearing people turned in logs with long lists of signs. They always saw more signs than I did ... I watched really carefully. The chimp's hands are moving constantly. Maybe I missed something, but I don't think so. I just wasn't seeing any signs. The hearing people were logging every movement the chimp made as a sign. Every time the chimp put his finger in his mouth, they'd say, "Oh, he's making the sign for drink," and they'd give him some milk ... They always held up their arms so you could tickle them under the armpits. Like this ... Sometimes their hands would touch. They'd keep their arms up until they were tickled, then they'd lower them, to stop the tickling. They caught on quick that when they held up their arms, they would be tickled. O.K., a few days later, I look in the log and see the sign more: arms held over head, fingers touching... The more

sign in ASL is not made over the head but in front of the body, and the hands are not just flopping, they have a definite shape. But everybody had seen the chimp do this and it was recorded as a sign ... When the chimp scratched itself, they'd record it as the sign for scratch ... Chimpanzees have their own gestures from the wild. Natural gestures ... The chimps hold out their hands. They do it all the time, without being taught. They want something, they reach. Those things the people in Reno called signs. Sometimes they'd say, "Oh, amazing, look at that, it's exactly like the ASL sign for give\" It wasn't... (Neisser 1983, pp. 214-16) Assuming that the "more" sign was as readily misattributed to Washoe as it was to these later subjects, the tally among these four high-frequency early "signs" becomes one sign, three nonlinguistic gestures. The prevalence of natural gestures in her signing does not seem to have diminished as Washoe matured. In an article on the first three years of the 51-month project (1971), the Gardners report that Washoe produced 294 different two-sign combination types between April 1967 and June 1969. Two hundred and forty or 82 percent of these combinations contained at least one sign from a list of her 12 highest-frequency signs. Among these 12 are "come-gimme" (ranked first), "hurry" (sixth), and "more" (seventh). From this perspective, a photograph of Moja, one of the subjects in the 54

Words post-Washoe project, signing "come" to Pili is much less impressive than the Gardners' caption (1978, p. 45) suggests. In the picture, Moja is employing the begging gesture. The caption explains that Moja is signing "come" and relates that "the chimpanzees use sign language to communicate with each other, especially during play and when changing locations" (my italics). The use of a single natural gesture, undoubtedly often followed by one animal moving nearer the other, is thus promoted to "the use of sign language when changing locations." This aggrandizement through unwarranted generalization pervades the sign studies no less than it does the Lana and Sarah projects, and it will be mentioned again. The failure to differentiate natural gestures from signs was not limited to the Gardners' projects. Roger Fouts, who received his Ph.D. as a student of the Gardners, carried the practice to his own work at the University of Oklahoma and Central Washington University. In a study of possible teaching of signs by Washoe to Loulis, one of the infant chimpanzees she "adopted" on being moved to Oklahoma, Fouts observed that when Loulis was first introduced to Washoe, Washoe would sign come to Loulis and then physically retrieve him. Later she would sign come and approach him but not retrieve him, and finally she would sign come while looking and orienting toward him without approaching him. (1983, p. 72)

Fouts reports that Loulis, within a year or so of his adoption, was "starting to use signs in their correct context. For example, he will now point to an apple, signing that, and sign gimme over and over again" (1983, p. 72). ("That," "this," and other demonstrative pronouns take the form of a simple point with the index finger in ASL.) Thus "come" and "gimme," presumably the same natural beckoning gesture variably translated by Fouts according to context, are an important sign in this project. And, in a study of "signed conversations" among Washoe, Dar, Tatu, Moja, and Loulis in Fouts's lab at Central Washington University, Fouts found that "hurry" was the most commonly used sign (Fouts, Fouts, and Schoenfeld 1984). "Hurry" is, of course, one of the natural gestures on which the Gardners conferred linguistic status. Finally, Patterson reported that "almost as soon as they met, the gorillas [Koko and Michael, her intended mate] were signing COME to each other ..." (1979a, p. 353). This is not surprising, since Patterson, too, has accepted natural gestures as signs, glossing the beckoning gesture as "come-gimme." ASL SIGNS Before considering further the nature of the apes' gestural productions, it will be helpful to have in mind some basic facts about ASL signs, the claimed medium of the signing projects. 55

Apes and language: ontogeny In American Sign Language, each word is a manual gesture made up of one value in each of three or four dimensions. Stated more concretely, each sign is a combination of (1) a handshape, (2) a movement, (3) a place of articulation, and, according to some authorities, (4) a hand orientation.9 These components are visual equivalents of the phonemes of spoken languages; changing a single one in a sign produces a different sign. For example, the signs for "summer" and "ugly" have the same handshape and movement, differing only in place of articulation. "Tape" and "chair" differ only in movement, "apple" and "jealous" only in handshape. The fourth formational parameter, hand orientation, has a much less systematic role in the language, serving to distinguish only a small number of minimal pairs (Siple 1978). According to Stokoe, Casterline, and Croneberg's Dictionary of American Sign Language (1976), the standard

reference on the language, ASL employs 19 handshapes, 12 places of articulation, and 24 movements. In the DASL analysis, however, these 55 values are merely articulatory "primes," each having several allophonic (functionally equivalent, noncontrastive) variants. (Another departure from the idealized characterization above is that a sign may contain more than one basic movement and may be made with two hands, each having a different shape.) The psychological reality of these formational components has been demonstrated in analyses of spontaneous errors of articulation (Klima and Bellugi 1979, pp. 125-46) and in experiments on short-term storage of signs (Bellugi, Klima, and Siple 1975). In the latter, it was shown that, just as hearing subjects recall or misrecall words from a list on the basis of phonetic properties rather than semantic ones, processing of incoming signs by deaf subjects involves decomposing signs into their formative elements for purposes of short-term storage and retrieval. In other words, when a subject, either hearing or deaf, is required to recall lexical items from a recently presented list, a misremembered item (one that was not on the list) is usually one that is structurally rather than semantically similar to the target item. Furthermore, in the same way that each phoneme can be analyzed as a bundle of simultaneously articulated acoustical features distinctive features - there is evidence that the formational components of signs are in turn composed of a small number of recurring visual distinctive features (Lane, Boyes-Braem, and Bellugi 1976; Poizner and Lane 1978). SUBLEXICAL STRUCTURE Leiber (1984) has asserted that this sublexical structure of the spoken word and the ASL sign is not evident in the signing of the apes, that signs for them, in other words, are irreducible, opaque wholes, and that this alone, 56

Words apart from any evidence regarding syntactic competence above the level of the word, means that they have not learned a language. But there is at least putative evidence for the psychological reality of sublexical structure for the ape-language subjects. Savage-Rumbaugh (1988), citing data showing that Kanzi is able to discriminate among the words he hears, suggests that this ability draws on "phonemic processing mechanisms similar to those used by human beings" (p. 230). And Gardner and Gardner (1984) analyzed the errors made by Washoe and her successors in vocabulary tests and found that incorrect signs tended to be related to the target (correct) sign in being either semantically similar or similar in terms of formational elements. On the basis of the latter finding, they argue that their subjects were aware at some level of the componential structure of ASL signs. Obviously, any such evidence that the apes, like humans, unconsciously "unpack" lexical units into molecular constituents, either phoneme-like entities or the lower-level distinctive features, warrants careful assessment. Kami's word discrimination

Savage-Rumbaugh contends that Kanzi's ability to discriminate the words of spoken English is one of several linguistic capacities that separate the pygmy chimpanzee, Pan paniscus, from its common cousin, Pan troglodytes, and that it bespeaks a cognitive affinity with Homo sapiens. There is no reason to doubt the report that Kanzi and his sibling Mulika can tell the difference between various English words while Sherman and Austin, common chimpanzees in earlier work on symbolic communication (see below), could not. However, the evidence regarding Kanzi's word discrimination does not support Savage-Rumbaugh7s inference that he is employing "phonemic processing mechanisms." The relevant data come from formal tests conducted when Kanzi was 47 months old (Savage-Rumbaugh et al. 1986) and again at around 65 months, or five and one-half years (Savage-Rumbaugh 1988). The experimental paradigm, in its several variants, required the subject to listen to an English word and then choose from three lexigrams or object photographs. On both administrations, Kanzi was reliably correct with most of the words presented. Chapter 2 contains a more detailed description of results. In order to determine whether Kanzi was utilizing nonphonetic cues, such as pitch, stress, and inflection, in identifying words, test series were conducted in which the words were uttered not by a person but by a Votrax synthesizer, which produces words lacking these prosodic features. Kanzi's performance declined, but not below the level that would 57

Apes and language: ontogeny be expected for a human given what has been observed of the intelligibility of Votrax productions to humans (Punzi and Kraat 1986, cited in Savage-Rumbaugh 1988). Having established Kanzi7 s ability to discriminate words devoid of prosodic elements, the experimenters reverted to spoken words in a test format designed to assess the resolving power of Kanzi7 s audition for words (Savage-Rumbaugh 1988). As before, Kanzi listened to a spoken word, but now the three lexigram options presented were chosen so that one of the two wrong options corresponded to a word that was phonetically similar to the sample word. This lexigram's word resembled the sample word in sharing either its initial phoneme, its final phoneme, or both. Thus, Kanzi might hear "paper" and be presented with the lexigrams for "paper/7 "peaches/7 and "clover/7 Samples and choices were arranged so that, at the completion of the trials, each word and its phonemically similar alternative had served as both sample word and answer choice. Twenty-four pairs of similar words were tested in this way. Kanzi was correct in identifying both members of these pairs in 21 of the 24 cases. On the basis of these results, as well as those from the other testing formats, Savage-Rumbaugh suggests that Kanzi is perceiving words in the same way that we do. It is not obvious what, precisely, Savage-Rumbaugh has in mind in referring to "phonemic processing mechanisms.77 The most reasonable interpretation is that she means a process by which, on hearing a linguistic utterance, we segment it into phonemes, the clusters of articulatory features that differentiate one word from another, in order to identify the meaning-bearing components of the utterance. Or she may be talking about the categorical perception of certain speech sounds, that is, the perception of a range of acoustically different sounds as identical, a human penchant that may provide a nonarbitrary, neuropsychological basis for the phonemic structure of languages (see chapter 7). In either case, these experiments provide no evidence that Kanzi7 s discrimination of words is governed by phonemic principles. To demonstrate an interesting similarity between Kanzi7s and humans7 decoding of speech sounds, Savage-Rumbaugh would need to show that, like humans, Kanzi makes distinctions between words on the basis of distinctive features within phonemes, such as voicing and place of articulation, and that such features, which actually vary in a continuous way, are perceived in a categorical, discontinuous fashion. But the experiments summarized here do not provide such evidence. Not one of the 21 phonemically similar word pairs Kanzi successfully distinguished constituted what linguists call a minimal pair, that is, words differing only in a single phoneme, such as "shot77 and "shop/7 much less a pair differing only in a 58

Words single distinctive feature, such as "pod," and "pot," which differ only in that the final phoneme of the first is voiced and that of the second unvoiced. English speakers tell such minimally distinct words apart without difficulty, and this is not a function of experience with these particular words - invented words, such as "zod" and "zot," would be as readily discriminated. What Kanzi has shown is an ability to tell the difference between certain often-heard words, including some with similar acoustical profiles. This is not an especially surprising demonstration - though the common chimpanzees in the Savage-Rumbaugh laboratory may not, for some reason, have been able to tell words apart, this feat is not beyond the capabilities of, for example, dogs, as Savage-Rumbaugh herself has noted (SavageRumbaugh et al. 1986, pp. 213-14) (even though dogs may be less gifted in this than pygmy chimpanzees). Yet the dog's ability would not prompt anyone to suggest that dogs are endowed with "phonemic processing mechanisms similar to those used by human beings." They might be, just as Kanzi might, but the ability to discriminate among some English words is certainly not sufficient evidence for it. Sign structure and the Gardners' subjects

The Gardners' vocabulary tests, in both the original (1971) and revised (1984) implementations, involved showing the subject a series of color slide projections of various objects, eliciting a sign in response to each slide, and recording the sign. Careful controls were maintained to prevent experimenter effects. Two observers of the animal's signs were used, one inside the room with the subject and the other outside. Neither observer could see the other or the other's record of signs, and neither saw the slide presented to the subject until after the subject had responded.10 The observers thus provided independent recordings of the animal's responses, and agreement on both correct and incorrect signs averaged around 90 percent (Gardner and Gardner 1984, p. 390). The proportion of subject responses reported as being correct by each of the two observers for Washoe, Dar, and Tatu averaged roughly 80 percent. (Moja, the fourth animal tested, showed markedly poorer performance. This is attributed to inadequate preparation for the tests and lower quality of her test slides, which were prepared by an inexperienced photographer.) The slides presented in these tests included four different examples of each kind of object in order to determine whether the animals were associating signs with object classes or with particular objects. In addition, each slide was presented only once during a test in order to preclude 59

Apes and language: ontogeny memorization of slide-sign pairings, although all four exemplars of each object class were shown over the course of a test. Apparently, initial perusal of the corpus of errors suggested that place of articulation rather than one of the other formational parameters was influencing the error pattern. To represent the relationship between the target, that is, correct, sign and the subject's response sign, a matrix of signs was constructed for each animal. Along the top of the matrix were the objects presented to the subject (the target signs), while vertically spanning the left side were all of the signs that the subject made during the tests. The places of articulation of the signs were collapsed into six major categories that, according to the Gardners, are compatible with similar classifications used in the DASL and by Poizner and Lane (1978) in a study of human sign perception.11 The signs along the top were arranged into these categories, creating six major columns, and those along the left side were likewise grouped, producing six major rows and resulting in a 36-cell matrix. Each cell of this major matrix itself contained numerous "microcells" resulting from the intersection of rows and columns for individual signs. For each microcell, if the object for its column had been erroneously responded to with the sign for its row, then the microcell contained a number indicating hownany times this error had occurred. The crux of the analysis was as follows. Since the sign groupings were arrayed across the top and side of the matrix in the same order, all microcells uniting rows and columns that belonged to the same articulatory category would fall within one of the (six) diagonal major cells of the matrix. In other words, an error in which the subject's sign (a row) and the target sign (a column) were from the same articulatory category would be represented by a number in one of the diagonal cells, and all other errors would fall outside the diagonal. The Gardners sought to demonstrate statistically what is patent from an examination of each animal's matrix: a disproportionate number of erroneous signs shared the same articulatory category as their correct counterparts. This suggested that the animals were somehow utilizing the sublexical structure of the sign in storage and/or retrieval. A chi-square statistic was computed from each animal's matrix and in each case was highly significant, below the .0005 level. This study suffers from several flaws in method, some of which are severe enough independently to invalidate the analysis. The first problem is the reduction of various phonemically distinct places of articulation to a smaller number of more general areas of the body. The Gardners claim that these groupings "are compatible with the DASL [Dictionary of American Sign Language] and with Poizner and Lane's (1978) analysis of adult human signing" (1984, p. 397) (my italics). Although such a nebulous 60

Words assertion is difficult to verify or disprove, it can be said that there is no such classification in the DASL nor anything suggesting that such a reduction is appropriate. The DASL's analysis of ASL signs includes 12 major, phonemically distinct places of articulation and also has numerous signs in which one part of a particular handshape is used as the place of articulation for the other (signing) hand. Neither does the Poizner and Lane article contain anything that would legitimate the Gardners' conflation of places of articulation. The Poizner and Lane study does involve the lumping of places of articulation, but the rationale for their groupings is quite different from that of the Gardners. Poizner and Lane were investigating the effects of visual noise on subjects' ability to discriminate one place of articulation from another in artificial signs presented on a television monitor. They analyzed the pattern of confusions in subjects' discriminations under various levels of noise, using fourteen places of articulation (the DASL's 12 plus 2 of the DASL's handshapes functioning as places of articulation). Places of articulation were then grouped into clusters on the basis of the frequencies of mutual misidentifications of the different places. It would be wholly erroneous to interpret such confusion clusters, which may reveal something about perceptual similarities of ASL places of articulation, as suggesting that the discrete places of articulation they subsume are in some sense equivalent in ASL. This would be as mistaken as asserting that, say, the English phonemes [p]12 and [b] are not really distinct because they are so similar in articulatory and acoustical features. Yet, by describing incorrect signs that fall into the same invented category as their target counterparts as being "correct with respect to place," the Gardners make just this mistake. In short, the only legitimate way to analyze the role of the place-ofarticulation formational component in patterning sign errors would have been to use exact place of articulation in comparing target signs and response signs.13 Putting aside the problems involved in treating as identical places of articulation that are not, in fact, identical, is there a way of accounting for the excess of same-place-of-articulation errors that does not postulate a human-like sublexical analysis of signs by the apes? If the subjects were being influenced in their incorrect responses by the formational similarity of signs, then one would expect that, when a mistake was made and the incorrect sign had the same place of articulation as the target sign, all of the response signs sharing that place of articulation would have equal probability of being the sign substituted for the correct one. If, on the other hand, certain signs were consistently confused with particular other signs, this would suggest simply that the subject's control of these signs as two 61

Apes and language: ontogeny subjectively distinct signs was not stable in the first place. It seems reasonable to predict that, of the three major formational components of a sign handshape, movement, and place of articulation - the latter would be the most readily acquired and reliably executed, being perhaps perceptually more salient than handshape and motorically less demanding than either of the others. This would explain the fact that place of articulation is the formational component most often conserved in the chimpanzees' errors.14 Indeed, there is reason to believe that the animals had difficulty in producing the various distinct handshapes of ASL signs and that the experimenters in these projects were commensurately liberal in the formational standards they imposed on their subjects. The Gardners (1972) relate, for example, that limitations on Washoe's dexterity resulted in the "slurring" together of handshapes that are distinctive in ASL, such as the use of a slightly curved hand for both the flat-hand and the curved-hand shapes, or a slightly open fist for both the fist and the contracted-hand shapes. Likewise, Patterson (1979a) reports that Koko has difficulty assuming certain handshapes and that some are therefore rendered by her in a "simplified" form. Thus, the handshapes for "water" and for "rubber" are reduced from the "w" and "x" (DASL, pp.vii-xxi) configuration, respectively, to "a forefinger extended from a loose hand or a compact fist" (p. 335). While not all or even the majority of errors in the tables are the result of the repeated reciprocal substitution of just a few pairs of signs, in each animal's table there are one or two such recurrent confusions that represent a substantial proportion of the subject's errors.15 One finds in Washoe's table that, of the 331 errors, fully 38 involve the substitution of "pipe" for "drink" (23) or "drink" for "pipe" (15). This single pair of substitutions, a same-place-of-articulation error, accounts for a disproportionate number of such recurrent substitutions; the next most frequent substitution pair occurred 26 times (and was also a same-place error). Eliminating just this first pair of errors from the table on the assumption that Washoe didn't control them as distinct signs reduces her same-place errors from 39 percent to 31 percent of the total errors. Similarly, Tatu made one mutual substitution with disproportionate frequency, signing "sodapop" for "shoe" 12 times and "shoe" for "sodapop" 7 times, a total of 19; the next most common confusion had a frequency of 7. Dropping the "shoe"/"sodapop" confusion, a same-place error, from the table changes Tatu's same-place/total error fraction from 53/262 to 34/243, or 14 percent. Eliminating Dar's most common mutual substitution, on the other hand, raises his same-place proportion from 26 percent to 30 percent, since his most common error (more than twice as frequent as the next most common) was not a same-place error. 62

Words The last shortcoming of this study is perhaps the most serious, for it involves the unjustified omission of data from the analysis. The Gardners explain that there are four possible relationships among the items shown to the subject, the subject's signed response, and the reading of that response by the two observers: (1) agreed correct, where the two observers report the same sign and it is the target sign; (2) agreed errors, where the observers agree on what the response sign was, and that it was incorrect; (3) half-errors, in which only one observer reports the sign as being the target one; and (4) disagreed errors, where each observer reports a different, incorrect sign. The Gardners chose to include in their error analysis only agreed errors. The high level of interobserver agreement is cited in justifying this procedure, and the Gardners argue further that "agreement is itself evidence of reliability" (1984, p. 393). Each subject's corpus of errors was composed of the agreed errors from two tests administered to each.16 In addition, because the animals' errors were so few, errors were included from a number of "pretests" the Gardners had administered before and after the first test in order to accustom the animals to the vocabularytesting procedure. The objection to this method is simply that no justification is provided for discarding disagreed errors and half-errors as irrelevant to the analysis of the distribution of errors. It is certainly not the case that the 91 percent average level of interobserver agreement limits the possible proportion of nonagreed (disagreed or half-agreed) errors to an inconsequential fraction compared with the agreed ones. In fact, told only that each observer reported 20 percent of the signs as incorrect (Gardner and Gardner 1984, p. 390) and that the level of interobserver agreement across all responses, right and wrong, was 91 per cent, one can calculate that the category of nonagreed errors - those discarded - could be fully as numerous as the agreed-errors category. Furthermore, the figures provided to demonstrate high interobserver agreement pertain only to the tests, not the pretests. It can be determined from the number of object presentations over the two tests (207, averaged across Washoe, Tatu, and Dar), the percentage of correct responses according to the observers (81 percent, on average), and the number of errors in the tables for each subject (an average of 276), that the great majority of the errors - 86 percent - come not from the tests but from the pretests. Yet we are given no information about interobserver agreement during these pretests and consequently have no basis for computing even an upper limit on how many data points may have been discarded. There is no obvious reason to assume that inclusion of all errors in the analysis would have produced results less supportive of the Gardners' 63

Apes and language: ontogeny contentions. Nonetheless, this omission of data, apart from any of the other complaints above, would seem to invalidate the analysis. STIMULUS-RESPONSE DYNAMICS, CONCEPTS, AND REFERENCE On leaving the Gardners' vocabulary tests, we should consider a question of more general interest than the nature of the errors produced. What is the import of the animals' 80 percent correct performance and, more specifically, what do these tests have to do with reference, the cognitive event in which some entity in mind is specified through linguistic means? It is clear that the Gardners believe that these vocabulary tests, together with observations of their animals outside the testing context, "demonstrate that the chimpanzees used signs to refer to natural language categories" (1984, p. 383). Yet a skeptic might characterize the animals' performance on these tests as demonstrating only a reliable (80 percent) rate of responding to classes of equivalent stimuli with rotely paired associates. In fact, the Gardners themselves seem to equate the meaning of words with stimulus-equivalence classes and words with responses. As mentioned earlier, most of the detractors to the claims for grammar in apes concede that the apes' application of individual signs is markedly similar to that of children, likening the ape's progressive widening of the scope of application of a new sign to the child's generalization of a new word. An alternative interpretation might dismiss the apes' performance as mere stimulus generalization, but describing it in this manner would not be compelling without stipulating what in the child's use of words cannot be accounted for in terms of stimulus generalization. Furthermore, the evidence from the ape projects strongly suggests that the apes' concepts are not qualitatively different in focus or extent from their counterparts in young children. (Although it is not yet known, for either species, how concepts are organized, that is, whether as clusters of critical attributes [Clark 1973], as prototypes [Rosch and Mervis 1975], or in some other form.) Herrnstein, Loveland, and Cable (1976) demonstrated that pigeons are able to induce abstract representations of "tree," of "bodies of water," and even of particular persons, as evidenced by their ability to discriminate (with rate of pecking) members from nonmembers of these categories across hundreds of diverse photographs.17 Herrnstein et al. argue that stimulus generalization is inadequate to account for this capability, since the photographs classified as category members cannot be summarized in terms of a single sensory attribute or a constellation of them: "We can describe [the pigeons'] behavior pattern better by noting what the pictures are pictures of, rather than by what the pictures them64

Words selves are" (1976, p. 299). If it is necessary to attribute such abstract mental entities to pigeons, it would seem reasonable to interpret similarly the apes' range of sign deployments. It is profitable to make a distinction, however, between concepts and their linguistic expression in reference. When the apes signed, were they referring to things with symbols for concepts or merely producing gestures that had reliably garnered them rewards in that context previously? It is difficult to imagine ways in which one could, in another species, convincingly differentiate linguistic reference from conditioned response. In our own, such a demonstration is unnecessary, as our utterances are, under typical circumstances, simply not controlled by external stimuli. The ability to engage in "displaced reference/' to talk about things removed in time or space, is a compelling refutation of any assertion to the contrary. The young child, of course, is not so unfettered. In fact, as noted earlier, a major theme in linguistic development is the growth of a nondeterministic relationship between situation and utterance, the transformation of verbal gestures wedded to particular contexts into true words. Displaced reference is not apparent in the earliest phase of word development. It does appear, however, within months of the lexical generalizations apparently shared with the apes, whereas this function is, save for the sporadic anecdote, absent from the literature on the signing apes. Telling in this regard is the definition by Fouts and Rigby (1977) of one of the ASL signs attributed to Washoe. The sign for "open/' according to them, "consists of placing the two open palms against the object to be opened and then moving them up and apart, (p. 1041; my italics). This is not a correct description of the ASL sign for "open," nor could it be, since the canonical application of words or signs of language is not specified in terms of physical proximity between a term and its referent.18 On the other hand, there is no reason to doubt that this is an accurate description of Washoe s gesture, since she rarely if ever made reference to absent entities. The virtual absence of displaced reference from the ape studies (with the possible exception of the Kanzi project - see below) is consonant with the prevalent belief (for example Campbell 1971) that this is an exclusive characteristic of human communication, excepting perhaps that of the honey bees. The "waggle dance" of a bee returning to the hive is known to convey information to hive mates about the location of a potential hive site or source of water or nectar, including distance and direction relative to the sun, and also about the "desirability" of the find. But the differences between this sort of referential communication and our own are at least as profound as the similarities: in the bees' code, the domain of referents is 65

Apes and language: ontogeny radically limited - it was exhaustively described above - and both the messages and the potential responses to them are genetically specified in the narrow sense.19 Since displaced reference is absent from the earliest word usage in humans, simple reference would seem to be a more appropriate standard with which to assess the apes. Petitto and Seidenberg (1979), SavageRumbaugh et al. (1983a, 1983b), and Terrace (1985) have made a strong case for the interpretation of ape signing as pragmatic rather than referential:20 The apes appear to have learned not the meaning and linguistic functions of their signs, but rather the consequences of particular acts of signing. (Petitto and Seidenberg 1979, p. 179) Although chimpanzees are presumably being taught the names of objects, actions, states, and so on, they are actually being required to select a particular symbol to bring about a particular state of affairs that has been defined in advance by the context. When a chimpanzee is given what it "names" correctly, the semantic processes of requesting and naming are procedurally confounded. Because of this confounding, it is unclear whether the chimpanzee is learning that a particular symbol is indeed a name for an object or whether it is learning that the execution of that symbol produces that object. (Savage-Rumbaugh et al. 1983a, pp. 463-64) One of the erroneous assumptions of the various recent projects ... was that the symbols the apes learned to use functioned as names of objects, individuals, events or relationships. Once the projections of human meanings were stripped away from the interpretations assigned to those symbols, it became clear that the apes' use of symbols amounted to a means of expressing demands for various incentives. (Terrace 1985, p. 1023)

The interpretation of ape signing from this perspective has been informed by the work of Bates (1976,1979), mentioned earlier. Bates sees the young child's earliest words as similarly context-bound and nonsemantic, as game- or ritual-like behavioral components of recurrent activities. There is less emphasis in her work, though, on the contrast between utilitarian and nonutilitarian utterances. It seems that this distinction, especially Terrace's opposition of demand and name, is not parallel but orthogonal to that between paired associates and symbols. For, technically speaking, every demand or request, at least outside of the period of words as mere performatives, entails the cognitive act of reference, provided the entity so solicited is mentioned. "Give me the book" is a demand with a clear referring expression. Reference is not a property only of declarative sentences; it is independent of sentence mode. Names can be used to declare, to demand, and to inquire. Certainly the instrumental use of their gestures suggests that, for the 66

Words apes, these behaviors were more in the nature of strategies than symbols. But if the main difference between ape and human signing were only a greater proportion or even total predominance of demands versus other sentential functions in the former, the disparity might be dismissed as merely motivational or dispositional in origin. What evidence is there that this difference in the use to which apes7 symbols were put reflects rather a cognitive difference - a failure to appreciate, consciously or not, that their symbols were names for things? Reference studies by Savage-Rumbaugh

A series of symbol-acquisition experiments by Savage-Rumbaugh and her colleagues suggest that Lana, and by extrapolation all of the other apelanguage subjects prior to these experiments, were essentially only simulating referential communication. In addition, these researchers believe they have shown it possible to instill true referential abilities in chimpanzees under appropriate procedures. Two young male chimpanzees, Sherman and Austin, joined the Yerkes Regional Primate Center laboratory of Savage-Rumbaugh and Rumbaugh in 1975. The first of these post-Lana studies involved four subjects (SavageRumbaugh and Rumbaugh 1978). The previous experience of these subjects with language-related training, if any, is unclear from the report. In any case, after four months and over 3,000 trials, none of the animals had learned to depress a corresponding key21 on visual exposure to a particular item held by the trainer. This is, of course, reminiscent of Lana's striking failure to grasp the referential nature of the symbols with which she had supposedly been composing meaningful strings. In further experiments, instead of pressing a key in response to an item the trainer showed them, the subjects were able to press any of several keys on their own initiative, each of which resulted in delivery of a correlated incentive from an automatic dispenser. This condition elicited performances that were more "wordlike." The animals tended to press the different lexigrams according to a food-preference hierarchy that had been independently determined previously, suggesting that they understood that each lexigram stood for one of the incentives present. Additional observations indicated otherwise. When the dispenser's supply of the most desired food was exhausted, the animals would continue to press that food's lexigram, even though the animal could see clearly that the dispenser now contained only the other incentives. The lexigrams, it seems, were functioning as mere conditioned stimuli. In subsequent work, the experimenters used more complicated variants of these training procedures, including the introduction of "verb" 67

Apes and language: ontogeny lexigrams which were combined with the food lexigrams to form simple requests, such as "give banana" or "pour coke." They observed that the most effective methods for eliciting symbolic use of the lexigrams were those in which, like the first variant above, the animal was required to notice the effect of any key choice on the machine's behavior. This approach contrasts with the ineffective one in which the subject is rewarded only for pressing the key associated with an item chosen by the experimenter. Ultimately, they contend, all of the subjects came to use the lexigrams in an effective and productive manner. Once the animals could accurately request a variety of foods and drinks from the machine, they spontaneously transferred the use of these symbols to request food directly from the experimenters. Likewise, with no training, when symbols were left lying about their room, they spontaneously picked them up and brought them to the human companions to indicate desires for foods and drinks the companions were holding; when none were visible, they brought symbols of their favorite foods, placed these symbols in front of the humans, and pointed to them repeatedly. (Savage-Rumbaugh 1979, p. 8) Communicative use of symbols was the focus of experiments in which Sherman and Austin were trained in using the keyboard to inform the trainer or the other animal which tool was needed to obtain an incentive or simply which incentive was desired (Savage-Rumbaugh, Rumbaugh, and Boysen 1978a, 1978b). In the tool-use training, the subject learned to observe the trainer placing food in a container and to request the specific tool needed to open that particular container. Or, in another version of this procedure, one animal would watch the food baiting and then request the necessary tool from the other, who had not witnessed the baiting. If both performed correctly, the trainer would divide the food between them. On the way to attaining what Savage-Rumbaugh et al. regarded as truly symbolic application of their lexigrams, the animals had to divest themselves of associations developed between the tool lexigrams and the locations of the food. Thus, it was only eventually that they could correctly respond to the question "What's this?" when shown one of their tools, even though they had much earlier learned to press the tool lexigram appropriate for a given container. The authors cite this misassociation of tool lexigrams and food sites in criticizing the other ape-language projects for assuming comprehension of a symbol's meaning merely on the basis of its appropriate use in a limited context. In the incentive-specification paradigm, one animal would watch the trainer bait a container with food. The trainer would then ask him, through the keyboard, what was in the container, and the subject would respond by pressing a lexigram. The second animal, who was ignorant as 68

Words to which item had been placed in the container, would note that lexigram and request the incentive by pressing the same lexigram. If both subjects performed correctly, the food would be divided between them. Control tests, in which the same procedure was followed except that the keyboard was turned off, showed that the animals were unable to convey the needed information using only gestural and vocal communication. A variant of this procedure (Savage-Rumbaugh et al. 1983a) placed both animals in a room with an array of foods. The animals took turns "requesting" food items from the other via the keyboard. Roles were reversed frequently, and both subjects became adept at indicating the incentive desired, although the role of complier with the communicated request had to be enforced initially by the trainer. In order to conclude that the chimpanzees were using lexigrams symbolically, one would want to know, among other things, that the requester first had a desired item in mind and then requested it. How can it be determined that the requester wasn't choosing a lexigram in a random way? How, in other words, to demonstrate that the animal was indicating with its key press what it wanted? In a later training regimen, a singlesubject procedure, Savage-Rumbaugh et al. (1983a) attempted to enforce such symbolic indication and seemed to show that the animals were, at least by the end of the study, engaging in it. The subject viewed a small collection of familiar objects and foods, the composition of which was changed each trial. He then went to the keyboard, which was out of sight of those objects, and pushed a lexigram, after which he was allowed to return to the table of objects and select one. If it was the one corresponding to the lexigram pushed, he received praise or, if it was an edible, he was allowed to consume it. Summarizing the trials with edibles, the authors conclude: The results of this test demonstrate that Austin and Sherman can use their symbols to request foods they cannot see; that they know which food it is they intend to eat; and that when allowed to retrieve this food for themselves, their behavior reveals that a close correspondence exists between what they say they want and what, in fact, they actually select. (Savage-Rumbaugh et al. 1983a, p. 472) One interesting aspect of this work is the possibility that the animals were exhibiting not only referential communication but also a modest sort of displaced reference, since the items indicated with lexigrams, though seen moments earlier around the corner, were nonetheless absent when the subject was at the keyboard. On the other hand, one could pose the interpretation that the animals were not referring at all but rather following the rules of yet another opaque but manageable trainer-imposed task. In this one, they had to decide upon a desired item, remember it, press its 69

Apes and language: ontogeny paired associate at the keyboard, and retrieve it. It is true that, under either interpretation, the animal did have a particular incentive in mind before pressing a key, since only rarely did he press a lexigram for an item not among those just viewed at the incentive table. But the demonstration that chimpanzees can have rather specific images or notions is not, in itself, of linguistic interest, though such notions are necessary by definition for referential communication. These experiments, through which Savage-Rumbaugh claims to have engendered referential communication skills in her subjects, are not, in fact, without detractors. In an experimental parody of the work with Sherman and Austin on symbolic communication, Epstein, Lanza, and Skinner (1980) trained two pigeons to emit a choreographed sequence of behaviors that simulated communication. Jack and Jill (both males) were housed in adjoining boxes with a transparent wall between them. Jack pecked a key labeled "WHAT COLOR?" Jill then looked through a curtain in the back of his box to see which of three differently colored lights was illuminated. He would then peck the appropriate one of three buttons in his box, each of which was labeled with the initial of one of the three colors: "R/' "G," or "Y." This would illuminate the key. Jack then pecked a key labeled "THANK YOU," which resulted in dispensing of food to Jill. Finally, observing which key Jill had illuminated, Jack would peck the appropriate one of three colored keys in his box, earning himself a reward. Here is the experimenters' somewhat tongue-in-cheek assessment of their project: We have thus demonstrated that pigeons can learn to engage in a sustained and natural conversation without human intervention, and that one pigeon can transmit information to another entirely through the use of symbols. It has not escaped our notice that an alternative account of this exchange may be given in terms of the prevailing contingencies of reinforcement ... The performances were established through standard fading, shaping, chaining, and discrimination procedures. A similar account may be given of the Rumbaugh procedure ... (Epstein, Lanza, and Skinner 1980, p. 545) Umiker-Sebeok and Sebeok (1981) were similarly unimpressed by the chimpanzee studies, and they invoked this project by Epstein et al. in their critique. In a rejoinder to the Sebeoks, the Rumbaughs rejected the comparability of the pigeon experiment and their own work. In contrast to the rote pecks of the pigeons, Sherman and Austin demonstrated that information had indeed been passed between them and that they knew the topic, nature, and specific message that had been transmitted. Thus the chimpanzee who had not seen the food placed in the container did comprehend that the symbol used by the other animal referred to that container, and to the desire to obtain the contents 70

Words of that container. Because he comprehended these things, the chimp who had not seen the food was able to pick out a photograph of the contents of the container based on the information which he received from the other chimp. Prior to this, these chimps had received no training to select photographs under these or any other conditions ... Sherman and Austin were also able to use symbols to request that specific foods be given and to respond appropriately to one another's requests. Thus if Sherman has an array of foods in front of him, and Austin asks for an orange, Sherman will look over the food until he finds an orange, then give it to Austin. If he has difficulty sighting the orange, Austin may point to it for him. These abilities require an understanding of the nature and purpose of inter-individual communication not shown by pigeons in any study (Savage-Rumbaugh and Rumbaugh 1982, pp. 571-72). The Sebeoks, in responding (Umiker-Sebeok and Sebeok 1982), reiterated their analysis of the symbolic-communication studies as fundamentally nonlinguistic exercises. Austin and Sherman were presented with a number of desirable rewards and painstakingly trained to manipulate human symbols in order to obtain them, first with a human partner and then with each other. Animal A sees a certain object placed in a box and is trained to push a key, which stands for that object. Animal B sees a certain key light up and is trained to pick up the object with which the key has been associated in training and hand it to A. Even overlooking the abundant opportunities for cueing by both the experimenters and the animals themselves, it is difficult to understand in what way such behavior is qualitatively different from simply association of objects and symbols. Lest their readers come to such an impertinent conclusion, the Rumbaughs offer as "proof' that their experiments have demonstrated that chimpanzees have an "understanding of the nature and purpose of inter-individual communication not shown by pigeons," the fact that, on occasion, when animal B was slow in picking up the correct object, animal A would "help" by pointing to or touching the object himself. How, one might ask, does this differ from a pet owner's not uncommon experience of having his dog bring him its leash when its special "Take me for a walk" signal has for some reason been repeatedly disregarded? The Rumbaughs' anecdote does not prove that Austin and Sherman have any greater awareness of the implications of their masters' electronic version of ping-pong than would pigeons. (Umiker-Sebeok and Sebeok 1982, p. 576) The last experiment in the Savage-Rumbaugh symbol-acquisition opus is, ignoring an apparent flaw in the procedure, an impressive demonstration in support of the attainment of referential capabilities by her chimpanzees. In this series (Savage-Rumbaugh, Rumbaugh, Smith, and Lawson 1980), the subjects were first taught to classify objects, then photographs of objects, and eventually lexigrams for objects into two categories, food versus tool (or, as commentators [Ristau and Robbins 1982] suggested was more likely, food versus nonfood). 71

Apes and language: ontogeny Initially, th animals were trained to sort three food items and three tools into each of two bins, one for each category. Then they were trained to push one or the other of two new lexigrams, each representing a bin (and possibly a concept), in response to presentation of the same items. When they became reliably accurate with these six training items, generalization was tested with ten additional ones. Sherman and Austin performed almost flawlessly, but Lana failed just as impressively. It was determined that Lana could readily sort the ten test items correctly into bins but that she could not make the move to the lexigrams representing the bins. In other words, although she, like the others, was able to sort the items on the basis of functional similarity, she was unable to organize these items at a "referential symbolic level/7 The authors attribute this to her early language training, with its emphasis on symbol sequencing and paired associates. Lana was dropped from the later phases of this work. Next, Sherman and Austin were trained to classify photographs of the original training objects. This was followed by testing with what are called "novel" photographs (p. 924). Sherman achieved a perfect score. Austin did not but did so on a retest after being taught to attend carefully to the image in each photograph. In the final phase, which the authors feel truly reveals referential capacity, the subjects were trained to classify the lexigrams for the original six training items and, once they attained criterion on them, were tested with lexigrams for novel items. Sherman got 15 right out of 16, Austin 17 of 17. Thus, it is argued, on viewing a symbol the animals were able to retrieve an image of its referent, decide its status in the food/tool contrast, and indicate that status, using another symbol. The report on this work contains a recurrent misnomer that makes one hesitant to accept the authors' estimation of the study's import. The materials, photographs, and lexigrams used to test the generality of the abilities instilled during training are repeatedly described as "novel." One would assume by this term that the test items were different both from the six (recurrent) training items and from the test items of previous phases (object sorting, object labeling, and photograph labeling) of the experimental series. Yet the illustration (figure 1, p. 923) of the items used in training and testing and its caption show that the same items were used for testing across the four phases of the experiment. In other words, only the testing items in the first phase of the experiment - bin sorting of objects were novel in this experiment. The apes' success in subsequently labeling these objects, photographs of them, and lexigrams for them can be explained at least as cogently, and more parsimoniously, in terms of two facts: (1) their success in sorting the objects in the first phase, a demonstration of conceptual organization that is interesting but far from unprecedented in 72

Words comparative psychology, and (2) associations among each object, its photograph, and its lexigram. This criticism is attenuated somewhat by the (unexplained) fact that in the final phase, labeling of lexigrams, 16 (for Sherman) and 17 (for Austin) test items were used rather than the nine or ten of previous phases. This means, apparently, that in this phase seven or eight of the test lexigrams were novel to the experiment. On balance, then, one would have to accord the results of this experiment a measure of significance greater than that of a fatally flawed study but less than that asserted by the authors. There is another interpretation of these results that is completely at variance with that of the experimenters, one that explains the subjects' performance solely in terms of conditioned and unconditioned responses rather than more complex cognitive functions, such as reference. Epstein (1982) suggests that what Sherman and Austin learned in the early training phase of this work, when they were rewarded for placing the three foods in one place and the three tools in another, was to treat things that induced a physiological response to food, such as salivation, in a certain way, and other things a second way. In subsequent training, they were similarly rewarded for treating photographs and lexigrams of foods, both of which, because of similarity in appearance in the first case and association in the second, would elicit the food-related physiological response, in the same way as they had foods, and other photographs and lexigrams in the other way. In other words, according to Epstein, all Sherman and Austin needed to learn in order to perform as they did was to treat things that made them salivate in one way and all other things in another way. As exemplified by the Gardners' analysis of the vocabulary tests they administered, it is often the case in psychological experimentation that the nature of subjects7 failures is more informative than the pattern of their successes. Under Epstein's reading of this experiment, any misclassified tools should be those that tend to evoke the physiological food response because they have become associated by the subjects with food, while misclassified foods should be ones that fail to elicit the food response, because, for example, the subject does not recognize them as foods, or does not care for them as foods. At the end of the first phase of the experiment, after the animals had been trained and tested on labeling objects as food or tool, SavageRumbaugh et al. tested Sherman and Austin with 14 additional tools and 14 additional foods. Sherman made four errors, Austin three. The authors report that six of these seven errors were of the sort that Epstein predicted would occur: misclassification of tools used in food preparation as foods (Savage-Rumbaugh, Rumbaugh, Smith, and Lawson 1980, p. 923). Epstein's interpretation offers a principled account of these errors; Savage73

Apes and language: ontogeny Rumbaugh et al. do not. However, the hypothesis that the animals were engaging in intentional classification based upon a food/tool (or food/ nonfood) contrast can be defended by assuming that the criterion the subjects were applying was something like tastiness. The report indicates that at least two of the six misclassified food-preparation tools were ones that the animals often licked; the way in which these objects were classified by Sherman and Austin is thus consonant with the claim that their performance reflects reference and conscious classification rather than mindless association. In order to decide between these rival accounts, we would need to know whether the remaining four of the six misclassified tools were also "tasty" ones. However, the fact that, in a subsequent effort, Savage-Rumbaugh (1981) was able to teach Sherman to draw a four-way distinction among foods, drinks, locations, and tools argues in favor of the interpretation of Savage-Rumbaugh et al. Kanzi and reference

According to Savage-Rumbaugh, Sherman and Austin acquired referential abilities only as a result of protracted training, whereas Kanzi and Mulika, her pygmy chimpanzee subjects, adopted referential use of language tokens in the absence of any training whatsoever. Even Kanzi and Mulika, however, who appear to differ so dramatically from previous apelanguage subjects in this regard, use their lexigrams almost exclusively to acquire things or to induce their human companions to do things. SavageRumbaugh (1987) makes much of the fact that, of Kanzi's 25 most frequent two-item and 25 most frequent three-item combinations, not one is a food request (detailed in Savage-Rumbaugh et al. 1986). Yet every one of these fifty combinations seems to be a request for something rather than a noninstrumental comment of the sort that predominates in the early productions of children. It is not clear why Savage-Rumbaugh regards as so noteworthy the fact that Kanzi's most frequent combinations are not food requests, other than that he contrasts with Nim in this respect. It is not even clear, in fact, that he does differ from Nim in this characteristic. Savage-Rumbaugh et al. provide information on the topic of Kanzi's multi-item utterances, yet combinations accounted for only six percent of his productions (1986, p. 222). Since no information is given about the topic of the 94 percent of his utterances that consist of single items, one does not know what Kanzi's most frequent utterances are about. But an indication of the nature of those usages can be gleaned from the utterances presented in the appendix to the same article, a sample of Kanzi's and Mulika's productions from one randomly chosen day. Out of 18 utterances (types as opposed to 74

Words tokens), either 12 or 13 are food requests, depending upon whether or not the request "surprise" is glossed as a food request. 22 And, significantly, all but one of these food requests are single-item utterances. Or, putting it differently, out of 14 single-item utterance types, either 11 or 12 are food requests, depending, again, on how one assigns "surprise." Seidenberg and Petitto (1987) commented on the prevalence of food requests among the utterances listed in the appendix to the 1986 article. In her response, Savage-Rumbaugh (1987) objected that Seidenberg and Petitto had failed to acknowledge that the utterances included in the appendix were "not presented as a random sample, rather, they were purposefully selected to show how Kanzi used such utterances to control his movements about the forest" (p. 290). It is true that the 1986 article stipulates that, due to space limitations, the appendix does not comprise all of Kanzi's and Mulika's utterances for that day. However, nothing in that article suggests a selection criterion that would have resulted in a sample having an unrepresentatively large proportion of food-related utterances. On the contrary, the authors offer the appendix to "illustrate the kinds of utterances that Kanzi and Mulika produce in the context of daily events" (Savage-Rumbaugh et al. 1986, p. 233). It is quite apparent that, food related or not, the great majority of Kanzi's and Mulika's utterances are requests. Savage-Rumbaugh herself agrees with this characterization and also with the observation that such is not the case with children (1987). For her, the explanation for the difference between child and chimpanzee is that Kanzi's language medium makes symbolic utterances much more burdensome than vocal language; she observes that if a child had to go to a keyboard and search through several hundred printed words whenever he wanted to make a comment, he would be much less given to commenting. Futhermore, she adds, analysis of all of Kanzi's communicative behavior reveals that he employs vocal communication in preference to all other modes and that his comments, as distinct from requests, are "almost exclusively vocal." Among the examples presented: 1. Kanzi is working on a video game task ... When he successfully completes this, the experimenter says "You did a good job/' Kanzi looks at the experimenter and comments, "uhh um urn." 3. Kanzi is looking out the door when he sees Rose returning after she had left to get him a surprise. He sees Rose coming ... and comments, "uh ooah" as he goes to the door, looking at Rose. 6. While Kanzi is vocalizing to Matata, the experimenter comments that several days ago Matata bit Kanzi. Kanzi looks back toward the experimenter and responds with the comment, "uh huh/' (Savage-Rumbaugh 1987, p. 291; italics in original.) 75

Apes and language: ontogeny For the ape-language skeptic, the explanation for the predominantly instrumental nature of Kanzi's lexigram usage is not that the lexigram system is so inconvenient that he reserves it only for such uses. It is instead the same interpretation that applies to the signing apes, who were similarly instrumental in the use of their language tokens and yet, unlike Kanzi, no more constrained to be parsimonious in expression than signing children. That interpretation is that their language symbols are not symbols but procedures, not meaningful but effective. It is true that Kanzi's productions do not consist solely of instrumental usages. It is reported, for example, that he "frequently takes the keyboard and goes off by himself to use it" (Savage-Rumbaugh et al. 1986, p. 228), apparently to "talk to himself." Such behavior constitutes perhaps the most compelling evidence that lexigrams are symbols for Kanzi rather than just means to ends (Nelson 1987). However, such noninstrumental utterances were reported from all the other ape-language projects as well (for example, Gardner and Gardner 1969). Yet such anecdotes did not impress critics, such as Savage-Rumbaugh herself, sufficiently to keep them from dismissing these projects for having failed to produce referential use of language, presumably because, as with Kanzi, the overwhelming majority of usages were instrumental in nature. It might also be argued that Kanzi regularly engages in displaced reference, since he uses his lexigrams to indicate the locale to which he wishes to travel. This is a reasonable interpretation, but so is the suggestion that this behavior is no more referential than that of the dog that brings its leash to its master in order to be taken outside. (And, it is worth adding, this practice is typically a spontaneous invention on the part of the dog rather than a trained routine.) Kanzi is much more versatile in this, using different lexigrams for various locales, but it seems plausible that a dog might learn to enact different behaviors according to where it wishes to go. If Kanzi's use of language tokens is different in kind from that of previous subjects, then, the difference does not lie mainly in the pur-poses to which he applies them; Kanzi seems to be nearly as demanding as the next chimpanzee. However, as pointed out earlier, such a bias in usage only suggests rather than proves that an ape is not referring when it uses its language tokens. Moreover, the aspects of Kanzirs linguistic behavior that are unique to this subject (and perhaps Mulika) give Kanzi the best claim thus far for reference. The fact that he acquired his lexigrams without training,23 his untutored ability to press the corresponding lexigram when shown an object, and the reportedly significant frequency of noninstrumental lexigram presses all suggest that Kanzi was functionally aware that lexigrams are names for things. 76

Words SUMMARY This chapter began with highlights in the child's developing mastery of the referential power of words. Words are only gradually transformed from semantically null verbal behaviors into names for things. Reference, the use of a word to indicate the focus of one's attention, is neither present congenitally nor evident within the first year, but it appears soon after in all children with an inevitability suggesting that it is not learned; it is, indeed, hard to imagine how it could be learned. Reference seems to be part of the endowment of the species. Evidence for its natural occurrence elsewhere, on the other hand, is quite limited (see chapter 8). The outcomes of the ape-language projects do not seem to contradict this contention. The constitution and function of the symbols produced by the apes were considered, and both were found to be fundamentally different from those of humans. It was suggested that a consequential proportion of the gestures emitted by the signing apes were part of the natural behavioral repertoire rather than acquired linguistic symbols. Savage-Rumbaugh's claim that Kanzi was disassembling spoken words into phonemes was found to be unsubstantiated, and it was asserted that the Gardners7 efforts to prove their apes cognizant at some level of the formational components of ASL signs were nullified by methodological shortcomings. Furthermore, there are indications in the literature that the apes' limited dexterity precluded the articulation of some contrastive formational elements. This would result in a certain homogenization of the lexicon, a number of "homonyms" distinguishable as different signs only by the experimenter's interpretation of them in context. There is reason to believe that apes construct or, like pigeons, can be stimulated to construct concepts akin to our own. But a distinction should be drawn between concepts and their linguistic expression in reference, and it was suggested that the apes' symbols are better characterized as performative than referential, that the animals had acquired a repertoire of habits that were effective rather than meaningful. The absence of displaced reference from their behavior supports the interpretation of their performance as essentially stimulus bound. The extensive experimental work of Savage-Rumbaugh and colleagues on the development of referential communication was considered. Alternative readings of their results were entertained, leading to a rather timid decision to sit on the fence regarding the possibility of instilling reference in a nonhuman organism. Lastly, Kanzi's use of lexigrams was assessed for its resemblance to 77

Apes and language: ontogeny children's word usage, and it was concluded that Kanzi clusters with the other apes rather than with children on the instrumental/noninstrumental contrast. Nonetheless, Kanzi's use of lexigrams represents the best evidence available for the referential application of learned symbols by an ape.

78

Sentences

A minimal general definition of syntax would be a rule or set of rules dictating the relationships permissible among the elements of a set. These elements, in addition, must be grouped into categories. The syntactic rules are formulated as permissible relations among those categories, rather than as configurations of the elements themselves. Otherwise, the rules would consist merely of an enumeration of all well-formed expressions (whether sentences, propositions, equations, or designs) in the representational system. In that case, one would have specified not the grammar of the system but the system itself, that is, all of its component expressions. Such an approach is not only ungainly in general but impossible as well for natural languages and all other representational systems that have among their rules recursive functions, since, as discussed in chapter 4, such rules provide for an infinite number of expressions. It is beyond dispute that all natural languages are predicated on rule systems - grammars - involving elements grouped into nested categories of increasing abstraction. Thus "bird," an element of English, is a member of the noun category, which in turn is a constituent of the noun phrase, which may be part of a sentence subject or sentence object, components of the category "sentence." The production and understanding of sentences in a natural language entails the speaker's processing of low-level elements, such as "bird," but involves also the mental manipulation of the more abstract grammatical constituents. The rules for forming questions in English are an often-cited and clear example of this. The relationship between "Reagan is sleeping" and "Is Reagan sleeping?" might suggest that the rule for forming the interrogative counterpart of such a declarative sentence specifies simply that one find the first "is" in the sentence and move it to the front. The inadequacy of such a rule, phrased only in terms of low-level elements, is seen in the case of "The man who is sleeping in the White House is the president." The ungrammatical "Is the man who sleeping in the White House is the 79

Apes and language: ontogeny president?" results from application of this rule. The correct rule for deriving this sort of question entails, succinctly stated, the permuting of a sentence's deep-structure subject and main verb, highly abstract constituents having no marking in the manifest sentence itself. More germane to the present work is the observation that children as young as two years evince manipulation of such abstract sentence constituents. For example, Valian (1986) analyzed the productions of children between the ages of two years and two years six months, and found that they observe the English-language constraints on constituent distribution. In other words, nouns, noun phrases, prepositional phrases, and so on occurred in the sentence positions where they would be expected to occur if the children were operating with these grammatical entities in composing their simple utterances. GRAMMAR TYPES Linguists contend over just how abstract the categories of grammars must be, but there is no question of formulating grammars exclusively in terms of words or constituents of words (morphemes, phonemes, or distinctive features). The so-called case grammar of Fillmore (1968) and his adherents employs categories that are relatively concrete and transparently semantic, positing an "agent" instead of a subject, a "patient" in preference to an object. The case-grammar approach has been perhaps more influential among developmental psycholinguists than among students of the "adult grammar," for its concrete entities seem well suited to characterizing the utterances of the child. The early word combinations of children tend to have subjects, for example, with the semantic attributes of an agent, that is, an animate entity, and verbs that stand for physical actions. And, on the whole, the productions of children, unlike those of adults, are devoid of the semantic diversity within syntactic categories that, according to its detractors, renders case grammar inadequate.1 The drawback of interpreting early language in case-grammar terms, and one that has been much discussed, is the difficulty of explaining the developmental transition from such a semantically based grammar to the more abstract, semantically neutral one of the older child and adult (Gleitman and Wanner 1982). Schlesinger (1981), Bates and MacWhinney (1982), and others have offered models for such a transition. On the other hand, some, such as McNeill (1966) and Valian (1977), argue that the child's grammar is highly abstract from the outset, that she is innately equipped with such notions as subject and object. Despite the transition problem, the most influential efforts to capture children's early utterances in a grammar have employed case-like or at least semantically oriented 80

Sentences terms, either exclusively or in combination with more abstract ones. Thus, "agent-action-patient," "agent-action-recipient," and "agent-actionobject" are commonly used sentence paradigms in analyses of children's utterances. A major dimension in the comparing of child and ape must be that of syntax. Syntax, in fact, is the issue that has most occupied the participants in the ape-language debate. Do the apes grammatically encode semantic relations with their signing, or are their sequences merely unstructured strings of signs, some signs being responses to discriminative stimuli in the immediate context and others general-purpose, "wild card" signs, all emitted to obtain rewards? This question provides the focus for this chapter. The function of syntax in natural languages, as distinct from the nature of its rules, is to encode propositions, to translate meanings into sounds (or, in signed languages, gestures) and vice versa. Languages are marvelously diverse in which facets of the cognized universe they mandate be encoded in sentences about that universe. This is why semantics is not identical to cognition. It is safe to say, however, that all languages have grammatical means for unambiguously indicating the actor, action, and thing acted upon in a proposition and for distinguishing objects from attributes. In English, for example, the relationship between actor and acted upon is indicated through word order. Other languages encode this by appending inflections to the nouns or with some combination of word order and inflection. As we will see below, American Sign Language, and presumably other signed languages yet to be studied in detail, have some entirely different mechanisms for encoding grammatical relations, provisions adapted to the visual medium of the language. WORD ORDER Critics of the signing projects observed that the apes7 utterances were devoid of grammatical regularities, noting an apparently random ordering of signs. The ape partisans responded by pointing out that in the (putative) medium of the projects, American Sign Language, the linear ordering of signs is not as important as in English and many other spoken languages. In this section, the significance of word order as a grammatical mechanism in spoken languages will be considered, and ASL alternatives to word order for encoding semantic relations will be discussed. It will be argued that the sign sequences of the apes contain little evidence for grammatical patterning, that neither word order nor the distinctive ASL devices are used to convey semantic relations. The use of word order to mark the actor, action, and recipient of the 81

Apes and language: ontogeny action in sentences appears at such an early point in the acquisition of English and other noninflectional languages that students of early language have engaged in some "overgeneralization" of their own, declaring this to be something of a developmental universal. According to Gleitman and Wanner, in children's earliest combinations "the evidence is massive that the words are sequenced in the sentence" (1982, p. 21). Bever (1971) even claimed that particular orders are universal, that there is an invariant actor-action or action-object order for two-word utterances containing transitive verbs. The promotion of word order to a universal was not based solely on its centrality to the grammar of English. Slobin (1982) observed that, in languages in which noun inflection carries part (Serbo-Croatian) or all (Turkish) of the burden of marking semantic relations, children acquiring the language nonetheless tend to use a consistent order of semantic roles. Bowerman (1973) reported the same for Finnish, an inflecting language. Slobin (1979) has pointed out that pidgin languages characteristically develop word-order constraints. Perhaps the most compelling evidence for ordering as an essential attribute of language is the work of GoldinMeadow (1982; Goldin-Meadow and Mylander 1983) on invented natural languages. She studied deaf children whose parents had not exposed them to sign language. These children had developed gestural communication systems that, though diverse in some respects, evinced some of the defining properties of early child language. Pertinent here is her observation that each child used a consistent sequence for the semantic roles in his or her productions. Commentators on the early reports on Washoe's progress noted that her sequences seemed to show no syntactic patterning of semantic relations2 (Brown 1970; McNeill 1974). Indeed, prior to these comments by observers, grammatical regularities were so unimportant a focus of the project that the Gardners did not record the frequencies of the variant orders of each sign combination (Gardner and Gardner 1971). Subsequently, however, they attended more closely to indices of syntactic patterning, presenting evidence in later articles that Washoe tended to use certain sign orders. Before considering that evidence, it should be pointed out that, despite the reminders by the ape-language advocates that word order is not strictly constrained in ASL, the researchers in these projects did employ English word order in signing to their subjects. This much is stated by the Gardners (1976, p. 176), by Miles (1983, p. 48) regarding her work with Chantek the orangutan, and by Terrace and his colleagues, who worked with the chimpanzee Nim (1979, p. 901). That Patterson, too, used a version of Pidgin Sign English rather than ASL in her work with the gorilla Koko can be inferred from her reported practice of "simultaneous commu82

Sentences nication," the articulation of spoken and signed versions of a proposition at the same time. As Petitto and Seidenberg (1979) have observed, the substantial differences between ASL and English syntax make simultaneous ASL-English articulation a very difficult task, even for persons fluent in both languages. In short, then, the linguistic input to the animals in these studies contained the agent-action-object (subject-verb-object) order of English. What evidence is there in the ape material that the apes extracted this patterning and imposed it on their own strings? By the end of the third year of the Washoe project, the Gardners report, two "regularities of order" had appeared: First, in nearly 90% of the recorded combinations that included you, me, and a sign for action, the sign for the subject preceded the sign for the action. Thus, you me out, you me in, and you me go were typical in "Washoese," but out you me Dennis was exceptional. Second, for just over 90% of the recorded combinations that included both you and me, you preceded me. Thus, a phrase such as please me you go was also exceptional in "Washoese." This preference for you before me remained consistent, even though the preferred position of the verb changed with time. At first, Washoe preferred the format you-me-action, as in you me out. Later, however, the format yow-action-me became more prominent, as in you out me. (1971, p. 177) In assessing these facts, one should bear in mind the distinction between a sign order regularity and a semantic-role order regularity, since only the latter is the sort of patterning that suggests the operation of a grammatical rule. In relating Washoe's practice of putting the sign you first in strings in which it appears with me and an action verb, the Gardners use the term subject in describing the order of signs. While it may be correct to refer to the you sign in these strings as the subject, or at least as the agent, to describe a pattern involving this single sign (you) or the sign pair you me combined with a class of signs (action signs) as one in which "the subject" precedes the action sign implies a degree of abstraction that is not supported by the data. If Washoe does employ a subject-action sequence then that rule should be general across all subjects and should be instantiated in the data base by a number of different signs occupying the role of subject. The same objection applies to Patterson's report that in a sample of Koko's strings from 1974, "in the operations of recurrence, 'more' holds the initial position in 83% of the 94 instances" (1978b, p. 91). Again, a semantic category, in this case "recurrence," which in children is represented by constructions containing, for example, more, 'nother, and again, is equated with a single sign, elevating a regularity of lexical position to a rule of grammar. Terrace, citing similar evidence from Nim, discusses the relationship between these two patterns: 83

Apes and language: ontogeny Only when a semantic role is represented by a large variety of signs is it reasonable to attribute position preferences to semantic rules rather than to lexical position habits ... In combinations presumed to relate an agent and an object, or an object and a beneficiary, one would expect agents and beneficiaries to be expressed by a broad range of agents and beneficiaries: for example, Mm, me, you, and names of other animate beings. However, 99 percent (N = 297) of the beneficiaries in utterances judged to be object-beneficiary combinations were Mm and me, and 76 percent (N = 35) of the agents in utterances judged to be agent-object combinations were you. From these and other examples, it is difficult to decide whether the positional regularities favoring agent-object and object-beneficiary constructions are expressions of semantic relationships or idiosyncratic lexical position habits. Such isolated effects could also be expected to appear from statistically random variation. (1984, pp. 190-91) In all of the evidence presented for regularities of order in the ape productions, there is no convincing indication of a semantically or syntactically based rule as opposed to a lexical position habit. Fouts and Rigby (1977) provide another example of the latter misrepresented as the former: Fouts (personal observation) has also noted the use of sign order to express meaning in signing chimpanzees. One of his chimpanzees, Lucy, has a definite preference for the sign order "tickle Lucy" when asking someone to tickle her; and when the order is reversed to "Lucy tickle," she correspondingly tickles her companion, (p. 1044) Another instance of the aggrandizement through abstract characterization that is found in much of the ape-language literature, "the use of sign order to express meaning in signing chimpanzees" is represented by a "definite preference" of one subject for a certain order in the combination of a single pair of signs. In a later article, the Gardners (Beatrice T. and R. Allen, 1974) state that they have been able to observe the spontaneous emergence of the use of order in combinations. To a large extent, of course, Washoe's preferred orders were the same as ours (for example, agent-action and agent-object). Since an unlimited number of agents, actions, and objects were combined in this way, we can say that it is unlikely that Washoe was copying particular combinations produced by her models ... For example, Washoe's early attributive constructions followed the order of noun-modifier, as in clothes white and baby mine, even though her human companions used the normal English order of modifier-noun, (p. 17) This observation certainly speaks to the objection of insufficient generality, but in the absence of any quantitative data in support of this claim for robust, semantically based ordering rules, the account is not compelling. Patterson (1978b) reports that, in a 40-hour sample of Koko's two-sign 84

Sentences combinations from 1974, of the 98 combinations containing an adjective and a noun, the adjective precedes the noun in 75 percent of the cases. While this does suggest a pattern involving classes of signs, one would want to know how many different signs made up the adjective category. In any case, though, the 75 percent figure reflects a mere statistical tendency, a "preference," rather than a grammatical constraint. By contrast, children, according to Brown, "show something much stronger than a statistical preference for correct order" (1970, p. 227). Bloom (1970) invented a widely used technique for studying children's early word combinations. Sometimes referred to as "rich interpretation," this approach relates children's utterances to their discourse contexts in order to determine the semantic roles of the words within the utterances, which tend to be telegraphic and hence ambiguous versions of their full-sentence, adult equivalents. Thus, a child's "give dolly," uttered when the mother was holding the doll, would be categorized as an instance of the action-object relation, whereas the same utterance made in response to the question "Who should I give this candy to?" would best be designated as the action-indirect-object relation. In rich interpretation, in other words, one utilizes the context of an utterance in deciphering its semanticrelational content. One of Bloom's important discoveries was that children, at least those speaking English, almost invariably use the correct (adult) word order to encode those intentions.3 Patterson (1978b) applied this method to the analysis of some of Koko's verb + noun and noun + verb constructions. Using the situation of utterance to decide on the relationship Koko must have been articulating, Patterson was able to compare the word order employed by Koko with that required (according to the SVO order of English) by the context. In a ten-hour sample from 1974, Koko placed the verb before the pronoun or noun in 82 percent of the 55 cases calling for it and placed it after the pronoun or noun in 76 percent of the 17 cases requiring that order. In a discussion of the use of sign order by Moja, one of their postWashoe subjects, the Gardners (1980) relate that, in the context of preparing to go outdoors, "in almost all recorded phrases containing both me and out, me followed out" (p. 348). On the other hand, in the "toileting" context, Moja more often used me in the subject position, as in me can't and me finish. These tendencies are cited in arguing that Moja used rae-initial andrae-terminalconstructions differentially according to whether or not she was "in control of" the situation. This material, however, indicates only that Moja uses different whole constructions in different activity contexts. This is not the same thing as the use of contrastive word order with the same signs according to who was playing what role in a given activity context. Demonstration of the latter would suggest the operation 85

Apes and language: ontogeny of a grammatical rule, of the former only a correlation between whole strings and situations. This review has covered the bulk of the published evidence bearing on word order, excepting the work of the Nim group, discussed below. Despite the presence of grammatical encoding of semantic relations in their linguistic input, the apes did not learn to sequence their signs in this way. In the words of the deaf man, quoted earlier, who assisted on the Gardners' post-Washoe project, "It was all very messy, and the sign order was completely wrong. They'd learn three or four signs and sign all of them all the time in no order" (Neisser 1983, p. 216). As his comments imply, and as will be explored below, ASL is not, in fact, without any rules of word order. Analysis ofNim's combinations by Terrace

The most thorough analysis of sign order in a corpus of ape productions was performed by the Nim group (Terrace et al. 1979; 1980). They carried out a computer study of over 19,000 of Nim's multisign utterances, representing 5,235 combination types that were recorded in his teachers' reports between June 1975 and February 1977. Among the two-sign combinations, a number of positional regularities involving individual signs or sign classes were apparent. For example, there were many more types and tokens with more in first position than in second. This was true as well for give. In combinations of a transitive verb and Nim or me, the verb was much more often (83 percent) placed first. When Mm or me was combined with a food or drink sign, on the other hand, it was placed second only 65 percent of the time. Terrace et al. explored the possibility that these combination frequencies were simply the outcome of independent position tendencies of signs or sign categories. Each sign was classified into a semantic category (animate human noun, animate nonhuman noun, transitive verb, intransitive verb, and so on) or, if it was the only instance of its category or was inherently ambiguous as to proper category (such as drink), was treated as a onemember category. The relative frequency of each category's occupation of first position and second position was determined, and expected frequencies of various category combinations were computed on the assumption of random combination. It was found that these frequencies were not good predictors of the observed frequencies of the various combinations. This suggested that the observed regularities reflected the operation of ordering rules. The computer analysis treated sign distribution without regard to context, so that any use of contrastive sign order to encode semantic 86

Sentences relations might not have been apparent. Terrace et al. undertook a semantic analysis, following Bloom's method, of 967 two-sign combinations, employing context notes from the teachers' records to assign a semanticrelational interpretation to each utterance. Like the Gardners and Patterson, they were able to assign most (93 percent) of the strings to a limited number (20) of categories of relationship. Some combinations of semantic elements showed consistent sequential patterns. The agent term occupied first position in 80 percent of the combinations in which it occurred. In the relationship of recurrence, the recurrence element was sentence-initial 84 percent of the time. In the location relationship, the location term was initial in 73 percent of its strings. On the other hand, combinations mentioning an action and an object had no consistent ordering, even though some particular sign combinations within this category were sequenced with marked consistency. Because of the idiosyncratic nature of many of the observed regularities and because of the homogeneity of some of the sign classes - the recurrence relationship contained only more combined with other signs, and 90 percent of the location category had the point sign as the locative element Terrace et al. were unwilling to infer the operation of grammatical rules as opposed to lexical position patterning. The major and final source of their skepticism, however, was a meticulous analysis of three and one-half hours of videotape covering nine of Nim's sign-training sessions. The tapes consisted of material spanning the period from February 1976 through July 1977, or the age range of 26 to 44 months. Focusing on the relationship between Nim's productions and those of the teacher in the preceding moments, they found that nearly 40 percent of Nim's utterances were complete or partial imitations of what the teacher had just signed. In addition, some 50 percent of Nim's utterances were begun while his teacher was still signing, suggesting that he was not conversing but rather producing signs in a constant effort to secure desired objects or activities. This stands in striking contrast to the conversational performance of children. Garvey and Berninger (1981) observed children from three to five years old conversing and found that only five percent of their turns overlapped. Terrace et al. assessed available films of Washoe and Koko and concluded that both animals were as imitative in their signing as Nim was. They caution that no matter how compelling the statistical evidence for grammatical patterning adduced in any project, such evidence must be inconclusive without a filmed record that would make it possible to eliminate alternative explanations of that patterning. Sanders (1985) analyzed six videotaped sessions in addition to those used in the Terrace et al. study, elucidating in greater detail the nature of 87

Apes and language: ontogeny Nim's imitative signing. One of the issues he explored was whether NinVs imitation was like that documented in children's language acquisition by Bloom, Hood, and Lightbown (1974). Children vary in the extent to which they use imitation of adult input in acquiring language, but what is common is the imitation of words and grammatical forms that are just coming into the child's language. The elements that are in high frequency in the child's spontaneous speech are not imitated - once an element is firmly controlled, it is no longer subject to imitation. Sanders found Nim's pattern of imitation to be quite different from that of children. The same signs were used spontaneously and imitatively; for a given sign, there was no progression from imitation to spontaneous usage. Of Nim's nonimitative utterances, the great majority were either requests or responses to the trainer's prompting Nim to sign. Sanders offers this explanation of the "semantic-relational look" of the constructions of Nim and the other signing apes: Through the daily use of signs in situations arranged by his trainers in such a way that the trainer had control over something that Nim wanted and could obtain only by using certain signs, Nim learned which signs would be instrumental in obtaining rewards in different situations. Because the trainer would reinforce Nim's signs only when they "made sense" to the trainer, Nim was actually being trained to produce signs that would have a contextually relevant, semantic look. But Nim did not have to learn the semantic relations in order to produce signs that made sense to his trainers. By using signs associated with objects that the trainer indicated nonverbally, and by using the self-referring signs Mm and me, Nim could produce combinations of object names, action names, and self-references that would be readily interpretable as meaningful utterances, for example, the utterance eat banana me signed while attempting to get a banana from the trainer. In addition, Nim could obtain rewards by imitating his trainers' prior signs. This would serve him well when he was uncertain what it was the trainer wanted as well as when the trainer happened to be signing about something of interest to Nim. (pp. 208-209)

The Gardners and Patterson objected to the assertion of Terrace et al. that their own subjects were as agrammatical as Nim. They cited the large number of trainers Nim had and argued that Nim's shortcomings were attributable in large part to the psychological instability engendered by this and other methodological inadequacies (Gardner 1981). Terrace et al. themselves pointed out that the turnover rate among Nim's teachers was far from ideal. The circumstances of Nim's training were not exactly as they have been depicted by critics, however. Although there were some 60 teachers involved for varying periods of time over the four years of the project, Nim spent most of his time with a core group of eight teachers, and many of the 60 volunteers acted as "occasional playmates rather than as regular 88

Sentences teachers" (Terrace 1981, p. 107). It should also be pointed out that the Gardners, in a 1969 article, related that Washoe had several different teachers during the course of each day and stated that, contrary to their initial fear that this routine would affect her adversely, "apparently it is possible to provide an infant chimpanzee with affection on a shift basis" (p. 666). The claimed differences between Washoe's achievements and Nim's, then, cannot properly be attributed to a difference in the number of trainers each experienced. GRAMMATICAL ALTERNATIVES TO WORD ORDER In his 1973 book, A First Language, Brown reconsidered his earlier (1970) skeptical assessment of Washoe's combinations, which had stressed the absence of ordering. Citing recently acquired evidence on language acquisition from around the world, he retracted the earlier emphasis on word order, observing that intonation pattern is really the only universal feature that marks the early combinations as words in grammatical construction rather than strings of independent utterances. Since the Gardners had informed him that Washoe produced the characteristic intonation features demarcating ASL sentences, Brown concluded that Washoe's combinations probably embodied semantic relations. Brown's stature in the field is without equal, but the impeachment of word order as a universal feature of early language has not been his doing alone. Slobin and Bever (1982), for example, have shown that children acquiring Turkish, a highly inflective language with no grammatical constraints on ordering, are as adroit in decoding the semantic relations of the sentences of their language as are English-speaking children and that they compose sentences with various orderings of subject, verb, and object.4 The Gardners (1978) did not fail to note the revised assessment of the centrality of word order, arguing that the chimpanzee evidence and the child evidence were essentially alike with respect to the extent of ordering. However, the implication that children acquiring languages other than English are no more reliable than the apes in indicating the semantic roles within their constructions is not valid. The crucial demonstration of the Slobin and Bever study, in fact, was that children learning languages as diverse as English, Italian, Serbo-Croatian, and Turkish are all attuned to the standard means of their languages - whether word order, inflection, or some combination of these - for encoding relations. The following sections will consider evidence other than word order that the apes, too, were marking semantic relations in their sign sequences. 89

Apes and language: ontogeny ASL syntax and morphology The response of the signing-ape experimenters to the ordering criticism was twofold. Firstly, evidence was marshaled to show that the apes did, in fact, encode semantic relations through sign order. That evidence, I have argued, is quite weak. Secondly, and somewhat in contradiction, they downplayed the importance of sign order in American Sign Language, citing the existence of distinctive ASL mechanisms for encoding semantic relations. Patterson and Linden (1981, p. 117) state that "although Ameslan lacks inflections, by modulating where or how a sign is made, the deaf express grammatical functions simultaneously that are expressed by sequential devices such as word order in spoken language/' Objecting to the Nim group's equation of sign order and syntax, Patterson (1981, p. 87) observes that "by eliminating from consideration such simultaneous grammatical devices as modulation, Terrace et al. have eliminated a key way that apes creatively use the code they have learned." And the Gardners (1980, pp. 332-33) assert that "in many respects the progress of [Washoe's] linguistic development was indistinguishable from that of human children, particularly if one allowed for the fact that the grammatical structure of Ameslan depends more upon inflection and less upon word order than does English." Very little material was presented to substantiate what was implied - that the apes used these mechanisms and what evidence was offered was even less persuasive than that for sign order. Before looking at the published material bearing on the apes' exploitation of these features of ASL grammar, the meaning of "modulation" and "inflection" in ASL should be given. The following description of the syntax and morphology of ASL is based on Baker and Padden (1978), Bellugi (1980), Fischer (1974a, 1974b), Fischer and Gough (1978), Klima and Bellugi (1972,1979), Liddell (1978,1980), Newport (1982), Siple (1978), and Wilbur (1987). The grammar of ASL is complex and fascinating. The most pervasive feature of the language is also the one that best captures the differences between it and spoken languages: ASL thoroughly exploits its spatial medium for the simultaneous expression of numerous grammatical distinctions that, in spoken languages, are encoded by concatenation of morphemes. This simultaneity confers a certain compensatory economy on the language, for it takes twice as long to execute a sign as it does to utter the equivalent English word, yet the rate of production of whole propositions is the same in the two languages. A very productive provision of ASL is a system of spatial "reference points" that allows one to refer to an already mentioned noun entity 90

Sentences (anaphoric reference) simply by indicating a place in space that has been previously established as that item's reference point. Reference points are set up by making the noun's sign at an arbitrarily chosen spot at waist height, preferably to the speaker's right or left, or by pointing to, looking at, or slightly shifting the body toward a spot before, during, or after executing the sign in "neutral" space. Subsequent reference to the noun is achieved with the same sort of indication: pointing, gazing, or body shifting. Each verb sign in ASL has a neutral, "citation"-type articulation, but, when part of an utterance, that neutral form will be embedded within several simultaneously imposed grammatical "modulations," transformations on the sign's basic movement. Verb signs are subject to obligatory modulations for number, manner, and aspect. A verb modulated for the "continuous" aspect, for example, has a slow circular motion superimposed on whatever movement the unmodulated sign possesses. Thus, in ASL, some grammatical morphemes are layered onto the verb stem simultaneously rather than appended in sequence, as in spoken languages. ASL, like every language, has means for indicating the semantic roles of the nominal (noun or nounlike) arguments in a sentence. If none of the distinctive ASL means are used, SVO order is followed. (In simple yes-no questions, SVO is always the correct order.) If, however, the arguments to the verb are semantically irreversible - if there is only one plausible interpretation of the relationship among the arguments and the verb then the sign order is unconstrained, as in "Jane read book" and "Book read Jane." Similarly, sentences with intransitive verbs (verbs taking only one argument) have free word order. Finally, deviations from the canonical SVO form are acceptable if an argument is unmentioned because it is understood or if the verb or one of its arguments is topicalized, that is, moved to the front of the sentence for focus. (An English example of topicalization is provided by a recent magazine advertisement touting the irresistibility of a certain liquor: "A lasting gift it isn't.") The displaced item must be marked, however, with a specific combination of facial expression and head position that indicate a topicalized sentence constituent, and with a slight prolongation in that item's articulation. (In ASL, distinctive head movements and/or facial expressions are used to denote negation and relative clauses in addition to topicalization and are also components of a few lexical items.) ASL has a syntactic process not found in spoken languages that, where applicable, allows for flexible word order while insuring against ambiguity of semantic roles. It is functionally equivalent to case inflection of nouns in spoken languages but entails modification of the verb rather than the nouns. Because of this difference, there is a certain terminological 91

Apes and language: ontogeny inconsistency in the literature on ASL, some scholars contrasting this ASL process with inflection and others calling it inflection. (It is merely this lack of consensus in terminology that accounts for the discrepant assertions above of Patterson and Linden, on the one hand, and the Gardners, on the other, about the presence of inflection in ASL.) In this process, variously termed "verb agreement/7 "referential indexing/' "modulation/' and, as noted, "inflection," the orientation or direction of movement of the verb sign works together with spatial reference points or physically present referents to indicate the semantic roles of the nouns. Some 70 percent of ASL verbs can be adjusted in this way. Signs having no horizontal movement component can be adapted in orientation to indicate subject and object. (Subject agreement is less often observed, since the subject is often understood from prior utterances.) For example, the sign meaning "look at" is made with a single hand, the index and middle fingers extended and spread. To sign "Aaron looks at Anna," the hand would be located at the previously established reference point for Aaron (assuming he had been previously mentioned) or displaced towards him, if he were present, and pointed toward the reference point for Anna or toward Anna herself. The order in which the noun signs were executed prior to the verb would be a matter of narrative choice rather than syntax. For verbs with horizontal movement, the origin and direction of movement indicate the roles of the nouns: The verb MEETS is signed in the open space in front of the body. Both hands [are held with] index fingers pointing upward and palms facing each other. In the citation form of the sign the nondominant hand is held forward with the palm facing the signer. The dominant hand moves forward from just in front of the signer's chest, contacting the nondominant hand at the knuckles. In context, this sign would be glossed "I meet you (or whoever is standing in front of the signer)." To say "You meet me," the locations of the hands would be interchanged, and the direction of movement of the dominant hand would be inward toward the signer. If the signer has previously been talking about two people Bill and Sue, locating them in space, one on the right side and one on the left, the two hands would assume those two positions, and the direction of movement would indicate whether the signer intended "He met her" or "She met him." To sign "They met each other," both hands would move, contacting halfway between the two reference locations. (Siple 1978, p. 12)

Lastly, some ASL verbs, because they are "anchored" to the signer's body at the start of or throughout their execution, limit or preclude modulation through movement or location. The sign for "say" must start at the signer's mouth, regardless of its subject, but it can be adapted in its direction of movement to indicate the recipient of what is said. Other 92

Sentences verbs, such as that for "know/' must contact the signer's body throughout the articulation. In all such cases, a subject other than first person (the signer) is indicated by a point, slight body shift, or gaze shift toward the appropriate reference point. In summary, then, deviations from the canonical subject-verb-object sign order in ASL are common but do not leave the semantic roles of the nominal arguments indeterminate. Sentences with intransitive verbs or those with transitive verbs but semantically irreversible arguments are unconstrained with respect to order. In addition, the order of arguments in sentences with transitive verbs isflexible,but non-SVO orders are mandatorily marked for topicalization by a complex of facial expression, head attitude, and prolonged articulation and/or by inflections of the verb.5 It would be disingenuous to imply that children master these various syntactic devices of ASL early on. They do not, nor would one expect them to, any more than young English-speaking children would be expected to manifest full control of the equivalently complex subsystems of English grammar. However, the few studies of development of the verb agreement system indicate quite clearly that, by the age of three, children are acquiring it. Wilbur (1987, p. 206), reviewing three such studies, notes that "Fischer (1973), Hoffmeister (1977), and Meier (1982) agree remarkably well that the beginning of productive verb inflection emerges before age 3." Newport and Meier (1985), in a review of all published studies of ASL acquisition, conclude that the verb agreement system is in place for physically present referents by age three to three and one-half and that agreement for spatial reference points is controlled by five and one-half. In addition, according to Hoffmeister and Wilbur (1980), children learning ASL consistently impose SV, VO, and SVO order from the very onset of multisign sequences and continue doing so beyond the point at which their use of the verb-agreement system renders such role marking redundant. Collins-Ahlgren (1975) also reports consistent ordering of elements in the signing of deaf children. ASL grammar and the apes

The Gardners, to turn back to the apes, report that "Washoe began to use some of the characteristic inflections of Ameslan before the end of her first year in Reno, and was making considerable use of Ameslan inflections during her last year in Reno" (1978, p. 54). Unfortunately, no evidence is forwarded to substantiate this assertion, nor, to my knowledge, has any been published elsewhere. They report also that their post-Washoe subjects are employing grammatically significant sign modulations. Their description is reproduced here in its entirety: 93

Apes and language: ontogeny We noticed that the infant chimpanzees regularly placed certain signs on the addressee's body as well as on their own. The chimpanzee's lips or the addressee's lips have been used as the place for be-quiet, the back of the chimpanzee's hand, or the back of the addressee's hand as the place for tickle, and so on. By recording the contexts in which this contrast in placement of signs was used, we obtained evidence that Moja, Pili, and Tatu used this device as an inflection for agent of action for the signs be-quiet, and tickle, as well as for such signs as chase and finish. The infants expressed contrasts such as You be quiet and / be quiet in this way within the first 12 months. (Gardner and Gardner 1978, p. 58)

The evidence for variable placement of signs according to agent, though paltry, is mildly suggestive. But, in the absence of minimal documentation to substantiate them, these observations about two signs and the mention of two others must remain at best suggestive. So few specimens of an allegedly systematic practice cannot tip the balance against a counterinterpretation that these observations derive from unbridled subjectivity. Lacking evidence that these usages are representative of a general pattern, one would need to know at least something about "the contexts in which this contrast in placement of signs was used" in order to accept the Gardners' reading of even these signs. It is frankly difficult to imagine, for example, a chimpanzee, less than 12 months old, saying either "I be quiet" or "You be quiet." Patterson (1980a) explains that ASL employs modulations to perform grammatical functions that are encoded in spoken languages with word order and inflection. As an instance of these modulations, she relates that Koko would execute the sign for "pinch" on various parts of her body, "indicating exactly where she wants to be pinched" (p. 528). This variability in location may indicate where Koko wants to be pinched, but it has nothing to do with ASL grammatical modulations, either those using the reference point system for verb agreement or those conveying verb aspect, number, and so forth. A more plausible case of modulation to encode semantic roles is Koko's occasionally turning the sign for "sip" from her own mouth toward that of her companion, which Patterson glosses as "you-sip." Unlike the Gardners' subjects, however, who allegedly modulated four signs in this way, "sip" is the only sign cited by Patterson here or elsewhere (for example, 1980b) to document her misleadingly general statement that "Koko may vary the direction of the motion of a sign to indicate a specific actor or experiencer" (1980b, p. 99).

94

Sentences KANZI'S COMBINATIONS Kanzi began to produce strings almost as soon as he started using his keyboard. As discussed in chapter 5, Savage-Rumbaugh et al. (1986) tabulated Kanzi's 25 most frequent two-item and 25 most frequent threeitem combinations. Apart from the fact that every one of them is a request, the most salient feature of these strings, which Kanzi's experimenters term "multisymbol" and "two- and three-word" combinations, is that only a few of them are, in fact, multisymbol or two- or three-word combinations. Most, instead, consist of a combination of lexigrams and gestures or just gestures. Of the 25 most frequent two-item combinations, only 11 contain two lexigrams. Six consist of a gesture combined with a lexigram, and the remaining 8 are combinations of two gestures. Of the 25 most frequent three-item combinations, exactly 1 is a three-symbol combination. Eleven contain two lexigrams and a gesture, 7 are composed of one lexigram and two gestures, and 6 consist of three gestures. And, although one cannot infer a great deal from the tables presented about the nature of the gestures that figure so prominently in these high-frequency combinations, it appears that the great majority of them are not even iconic representations but rather points toward or manual contacts with persons.6 No evidence is presented for grammatical patterning in these combinations, presumably because none exists. This is apparently what Savage-Rumbaugh means in saying that "in their present form," these combinations "should not be regarded as grammatical" (1986, p. 392). Savage-Rumbaugh et al. (1986) report that many of the two-item combinations seem to be simply requests for two different items that are uttered together. Regarding the three-item sequences, the authors note that, as often as not, the individual specified as the "agent or beneficiary" of action is someone other than Kanzi himself, a pattern that differs from Nim's rather egocentric communications. Kanzi seems to be as desirous of watching someone chase, pat, or grab someone else as he is of participating himself, a penchant that might reflect an interesting psychological divergence between Pan paniscus and Pan troglodytes or might merely indicate that Kanzi is a chimpanzee with unusual tastes. Kanzi's requests suggest that he thinks in terms of action schematas involving an actor, an action, and objects of the action, but he does not appear to encode the relationship among those elements syntactically. There also seems to be an upper limit of two arbitrary symbols per utterance, with most multi-item productions, at least among his frequent ones, containing only one lexigram in combination with one or more deictic gestures. Thus, Kanzi7s most frequent two- and three-item strings 95

Apes and language: ontogeny are "Chase person(gesture)" and "Chase personl(gesture) person2(gesture)." This limitation on the number of elements expressed does not in itself constitute an important difference between Kanzi's utterances and those of young children. Children, too, are constrained in the number of sentence elements they can produce, with the result that their early sentences typically lack one or more sentence constituents (subject, verb, or object). But, in the case of children, there is evidence that underlying this reduced utterance is a complete syntactic structure that, for reasons of linguistic and/or cognitive immaturity, cannot yet be expressed in its entirety (Bloom 1970; Goldin-Meadow 1982). For example, the child who regularly omits one or another of the three major constituents will reliably sequence the other two elements in the order they take in the full, underlying representation. NOVEL COMBINATIONS The ape-language researchers, then, were not ignorant of the grammar of ASL, but their subjects were. A question more fundamental than whether the ape's constructions showed grammatical encoding of semantic relations is whether they were, in fact, words in grammatical construction rather than unarticulated sequences of signs in which the signs related to one another only by their separate relevance to the situation of utterance. The coining of "novel combinations" was seen by the apes' teachers as cogent evidence that their animals were generating phrases when they signed. Washoe's apparent reference to a swan as a "water bird" is the most famous such occurrence. Each of the other signing apes has produced multisign strings of like nature. The problem of validating such attributions to the animals is the same as that involved in assessing reports of their use of ASL modulations: anecdotes are insufficient. One cannot decide their significance without a substantial corpus of such inventions, or at least a statistical "context" indicating what proportion of unprecedented combinations were, like "water bird," sensible and what proportion nonsensical. When these anecdotes, such as the one discussed below, do contain additional information about the linguistic and nonlinguistic context of the coinage, a close look suggests that their appearance as strikingly appropriate is a product of their rendering by their reporters. Note that what is in question below is not whether the animal used a consistent word order to depict an attribute + entity relationship but simply whether the animal was depicting such a relationship at all. In the case of "water bird," for example, was Washoe referring to "a bird of the water" or simply mentioning or responding to water and bird in succession? 96

Sentences Fouts and Rigby (1977) report what they regard as a notable coinage by Lucy, a seven-year-old chimpanzee in language training at the University of Oklahoma at Norman. At the time of this observation, Lucy had undergone two years of sign training and controlled some 75 signs. Fouts had been conducting an experiment investigating Lucy's linguistic classification of various fruits and vegetables. In a very revealing finding, Lucy created novel combinations to describe her perception of the stimulus item. She preferred to call a watermelon "candy drink" or "drink fruit," whereas the experimenters referred to it with entirely different signs, that Lucy did not have in her vocabulary ("water" and "melon"). Another, more striking example occurred with radishes. For the first three days of the experiment Lucy labeled them "fruit food" or "drink." On the 4th day she bit into a radish, spat it out, and called it "cry hurt food." She continued to use "cry" and "hurt" to describe the radish for the next eight days. (1977, p. 1046)

It is the last sentence in this extract that argues against the dramatic conclusion the authors draw. The use of "cry" and "hurt" singly to "describe" the radish in Lucy's subsequent responses suggests that "cry hurt food" was not a noun phrase, a noun modified by two adjectives, but simply a sequence of three parallel responses to the taste of the radish, no more nor less remarkable than the earlier "drink" and "fruit food" or the later "cry" and "hurt." The difference between "cry hurt food" and the other responses is simply that the former sounds like an English noun phrase. Shorn of the surrounding utterances, which, as mentioned above, is the way in which these anecdotes are nearly always presented, these novel combinations convey an impression of linguistic productivity that, to the extent that this case is typical, is unwarranted. Of course, even those anecdotes that do include a relatively full context can demonstrate the creative overattribution affecting the interpretations of those committed to the reality of ape language. Lucy was eventually taken to Africa by Janis Carter, a woman who had befriended Lucy while a graduate student at Oklahoma. Carter had undertaken the heroic task of living in the bush with Lucy and other captive-born or domesticated chimpanzees to assist them in learning to live as chimpanzees in nature. In 1982, some five years after Lucy had last seen anyone from the years of her American upbringing, Eugene Linden, a chronicler and partisan of the ape-language experiments (1974,1987; Patterson and Linden 1981), visited Carter and Lucy in The Gambia. Lucy had met Linden in Norman on several occasions and, on seeing him in Africa, stared at him in a way that suggested to Linden that she recognized him. Although Carter had stopped signing with Lucy a year earlier to expedite Lucy's independence, she was curious enough about Lucy's response to Linden to sign "Who he?" several times. 97

Apes and language: ontogeny At first [Lucy] signed, "Roger," a sign made by grabbing the ear lobe between thumb and forefinger, but then, and much more persistently, she kept signing, "Lucy Janis ... Lucy Janis, Lucy Janis." It appeared that Lucy did associate me with her old life in America. Lucy may have thought that I was there to take her back to America, or to replace Janis, or to come between Janis and her in some way." (Linden 1987, p. 144)

Or it may have meant any of a myriad of other things that one sufficiently convinced of the meaningfulness of these gestures might be willing to ascribe despite having virtually no linguistic evidence for it in the gesturing itself. EXPERIMENTS IN SYNTAX In a report on children's developing ability to formulate Wh questions (Where..., When..., and so on), Brown (1968) observed that, in the period before children begin to manipulate sentence constituents to produce such questions, the best evidence for their knowledge of constituents is in their answers to these questions. Each Wh question requires a particular kind of constituent as a reply. A Where question requires a locative answer, a What a nominal one, a Who a personal name or pronoun. Brown reported that children answer Wh questions correctly about half the time during this period. The Gardners (Beatrice T. and R. Allen, 1975) conducted an experiment in which questions representing ten types of Wh question were posed to Washoe over the course of several weeks. The ten question types included several subtypes of What, Where, and Who questions, such as Who + pronoun (for example, "Who you?"), Who + action ("Who runs?"), and Who + attribute ("Who good?"). Washoe's answers to the numerous exemplars of each type were scored as correct or not according to whether they contained a sign or signs from the response category appropriate for each sort of question. Thus, Who + pronoun required a proper name in response, while Where + object called for a locative sign or phrase. Washoe responded correctly to 84 percent of the questions put to her. Contrasting this performance with the 50 percent rate in children, the Gardners assert that, were Washoe a preschool child, she would be regarded as linguistically advanced by this criterion. The Gardners depict Brown's article as primarily an assessment of children's responses to Wh questions, whereas in fact only a few lines of it - four sentences, to be exact - treat this. The article is about the evolving grammatical rules governing the several phases in the development of children's Wh-question construction. By construing the focus of Brown's paper to be responses, the Gardners are able to present theirs as a parallel 98

Sentences study and to contend that Washoe was linguistically coplanar with the children of Brown's study. They also subtly but importantly misinterpret what Brown does have to say about the significance of children's Wh-question responses. It is not his assertion that these replies are the "best evidence that young children use the items of their vocabulary as sentence constituents" (1975, p. 244) with respect to young children of any age but only for children who are not yet producing their own Wh questions. And, significantly, during the period in which Brown's subjects were responding correctly only 50 percent of the time, they were also beginning to construct Wh questions, albeit with (systematic) grammatical errors. Furthermore, Brown observes that the children were producing "numerous declaratives with noun-phrase subjects, main verbs, nounphrase objects, and locative adverbials," which he regards as "evidence that an underlying grammatical network was in the process of creation" (1968, pp. 283-84). Washoe, on the other hand, did not produce Wh questions in this or any subsequent period.7 Nor is there convincing evidence in her productions for noun phrases or any other multiterm constituents, which, as discussed in chapter 4, appear early on in language acquisition. A possible explanation for her 84 percent correct responses involves the fact that the Gardners scored as correct any response from the correct category of signs, regardless of whether or not it was a factually correct answer. This is justified by pointing out that many Wh questions, such as "Who good?" and "What you want?," call for answers that cannot readily be judged as right or wrong, since they have to do with opinion or preference. At least half of the questions administered to Washoe, though, were not of this nature but rather pertained to matters of fact. In these cases, this practice is not justified. It is conceivable, therefore, that Washoe's performance in this study is based upon a number of well-learned associations between a sign or signs in the questions and classes of suitable responses, as between "Who" and personal names and pronouns. These were, after all, questions whose form and, in many cases, precise contents had been signed to Washoe hundreds of times over the four years of language training that preceded this experiment. In fact, under the Gardners' scoring method, Washoe could have achieved her score merely by answering all questions of a certain form with the same sign, without regard for the factual relationship between that answer and the question. The listing of signs she produced as responses indicates that this extreme possibility was not realized. However, this listing contains sign types rather than tokens, so one cannot determine whether each response category was statistically dominated 99

Apes and language: ontogeny by one or a few signs or contained a number of different signs. Seidenberg (1986) indicates that the former was the case. Analyzing Washoe's responses (unpublished but provided to Seidenberg) to the 500 questions put to her in the study, he found that 71 percent of the responses to the "Who that?" question contained either "Roger," "Washoe," or "you." Seventy-five percent of the responses to the "Who" + action questions contained "you" or "me," as did 58 percent of the answers to "Who" + trait questions. Combining the three types of "Who" question reveals that 83 percent of Washoe's answers contained either "Roger," "Washoe," "you," or "me." Similarly, 100 percent of the responses to "Whose that?" contained either "yours" or "mine." Seidenberg observes that only the "What want?" and "What that?" questions were not associated with a predominating answer. In response to the first, Washoe used a number of different food signs, while the second elicited the correct sign for the object indicated by the questioner, an ability previously demonstrated in the vocabulary tests with slides. Thus, it is not at all obvious that Washoe knew what she was talking about as opposed to knowing simply what signs went with each different cue sign or signs in the various questions. A procedure that would be helpful, though not definitive, in differentiating these two possibilities would be to pose the same questions to Washoe but with all signs except the Wh sign deleted. If her rate of "correct" responses was unchanged with such truncated, nonsense questions, then the interpretation of her earlier performance as mere conditional discrimination would be strengthened. In conclusion, it is uncertain what grammatical significance, if any, should be accorded Washoe's performance in this experiment. It seems safe to say, however, that this study has not, contrary to a remarkably "rich interpretation" of it by Miles (1976, p. 592), demonstrated that "chimpanzees are aware of what constitutes a grammatical syntactic construction." A second experiment that pertains directly to the issue of syntax is described in Fouts 1977 and Fouts, Shapiro, and O'Neil 1978. Fouts regards the results of this study as conclusive enough that they "should put the criticism concerning word order to rest" (1977, p. 122). The subject of the study was Ali, a male chimpanzee raised in a human home for his first four and one-half years and then transferred to the Institute for Primate Studies. At the time of the study he was almost four years old and controlled 88 signs. Ali was trained to produce sign sequences corresponding to a physical arrangement of objects shown to him. These arrangements involved one item, such as a ball, located either on, in, or under a second one, such as a purse. In response to the signed question "Where's X?" X standing for the 100

Sentences first item or "subject," Ali was to respond with a sequence such as "in purse." During the training phase, five items were used as subjects and six as locations,8 while the applicable prepositions were always "on," "in," and "under." In the testing phase, a number (unspecified) of new objects were used as subjects and as locations. In addition, items used exclusively as subjects or locations during training were sometimes reversed in those roles during testing. As a result, 240 different relationships were presented to Ali during testing, some (unspecified) proportion being training relationships and the rest differing from them in one or both positions. The first observation that should be made is that only the so-called novel relationships presented during testing are of interest with respect to the question of whether Ali had acquired a productive grammar, since he had been drilled on the others over hundreds of trials during the training period. In fact, he was allowed to proceed to testing only after attaining a score of 80 percent correct responses twice in a row for three days. His average performance during the training trials was 87 percent correct for preposition (chance performance was 33 percent) and 66 percent correct for location (chance = 16.7 percent). The information on Ali's test performance is broken down according to the degree of novelty of the object arrangement presented. Those repeating arrangements used in training are, as mentioned, irrelevant. In addition, arrangements in which only the subject item was unprecedented are irrelevant, since Ali was not required to sign the subject in his answers. (In fact, he did include the subject in about 18 percent of his responses, and got it right in 42 of those 44 cases.) This leaves arrangements in which new objects were used as locations, on which Ali got the preposition right 71 percent of the time and the location right 39 percent, and those in which training locations and subjects were switched, on which the preposition sign was correct 92 percent of the time and the location 58 percent. The next observation to be made is that, whatever may have been demonstrated in this study, it has nothing to do with word order. Fouts (1977) points out that Ali's errors all involved using the wrong sign rather than reversing the order of, for example, the preposition and the location. This was true as well for those responses (18 percent) in which all three elements were signed. However, only in these 18 percent of responses was there any opportunity for Ali to demonstrate his grasp of contrastive word order, for only in constructions with two "nouns' was there an opportunity to "misplace" a word that could go in either of two slots depending upon the object arrangement presented. Since no information is given on how many different items were the subjects of these 18 percent or what proportion of them were in response to novel as opposed to trained 101

Apes and language: ontogeny arrangements, one cannot determine their significance. Regarding the remaining 82 percent of his responses, on the other hand, which consisted only of a preposition sign followed by a location sign, no linguistic abilities at all are required as an explanation. All of the signing apes as well as Lana, Sarah, and many other primate participants in psychological experimentation have shown themselves capable of substitution in frames, of producing response sequences in which alternative responses are limited to fixed positions in the sequence. Thus, the fact that Ali never put a preposition sign in the location-sign slot nor vice versa is not remarkable in light of his extensive training prior to testing. All's performance on the novel testing items boils down to a reliable (71 percent to 92 percent) signing of the appropriate preposition (chance = 33 percent) and a moderately reliable (39 percent to 58 percent) correct response to the location (chance = 16.7 percent). Savage-Rumbaugh's research with pygmy chimpanzees provides a third experimental study of syntactic capacity. When Kanzi was six years old, his comprehension of sentences was assessed informally by noting his responses to spoken multiword commands (Savage-Rumbaugh 1988). Some of the words were supplemented by lexigram presses. Most of the sentences tested were formulated just for this assessment, but they were given in nontesting, everyday situations over a three-month period. The experimenters were instructed not to test Kanzi on commands that are given so frequently in certain situations that Kanzi might carry out the appropriate behavior anyway, to avoid situations in which the appropriate behavior would be obvious to Kanzi without the linguistic communication, and to refrain from making gestures and glances that might elicit the correct response. A total of 310 sentences were tested. Some twelve sentence types were represented, the most frequent being the action-object paradigm (for example, "Would you please carry the straw"), followed in frequency by action-object-location ("Put the tomato in the refrigerator"), actionobject-recipient ("Kanzi, please carry the cooler to Penny"). SavageRumbaugh notes that there is no evidence that Kanzi attended to words such as "would," "you," "please," and "the," so only those words that she feels would have been needed to formulate the correct response to each command are presented in the tables listing the commands given. Of the 310, Kanzi responded correctly to 298. Savage-Rumbaugh might have been content to summarize the results of this testing as showing that Kanzi can "select the critical words from a wide variety of sentence types and respond appropriately" (p. 239). This observation seems unassailable. However, as in her assessment of Kanzi's ability to discriminate certain words, Savage-Rumbaugh ascribes a linguis102

Sentences tic significance to a performance that does not justify this attribution, suggesting that Kanzi's sentence comprehension "appears to be syntactically based in that he responds differently to the same word depending upon its function in the sentence" (p. 247). It is certainly true that Kanzi did something different with Jeannine depending upon whether he heard "Grab Jeannine" or "Give the trash to Jeannine," but responding correctly to these commands requires no syntactic competence whatsoever. Given that Kanzi knows the meaning of these words, or at least has formed appropriate associations between these words and their corresponding things and actions, and given that he is able to "select the critical words" and "respond appropriately," there is only one way we should expect Kanzi to respond to the coupling of "grab" with "Jeannine" or to the combination of "give," "trash," and "Jeannine." There is, in other words, only one pragmatically sensible way in which to combine the act of giving with the two entities Jeannine and trash, and that is to give the moveable, inanimate one to the one that is capable of being given things, the animate one. No ability to decipher syntactic structure is required in order to respond appropriately; in fact, there is no reason to doubt that Kanzi would have performed as well on these commands if the words of each sentence had been presented as a randomly ordered list. Of the 23 commands containing an action, an object, and a recipient, only one might require that Kanzi attend to word order to discern which word indicated the action, which the direct object, and which the indirect object or, using Savage-Rumbaugh's terms, which was the action, which the object, and which the recipient. Only "Put food (in) straw" contains two objects that are plausibly interchangeable in the act of putting. The 99 commands representing the action-object category, the most numerous in this study, together with 13 object-action commands, are all like "Grab Jeannine" in requiring not only no syntactic analysis but also no reflection about the plausibility of alternative interpretations. If Kanzi knows what the action word requires and what object the object word goes with, and he knows further that he should act in a way that incorporates both, then grabbing Jeannine, getting cheese ("Get cheese"), and slapping the bug ("Slap bug") are the only responses we would expect. Syntax plays no role here. Only the sentences with two objects could require any syntactic decoding. Those in which one of the objects is a recipient were discussed above. The only other major category with two objects consists of sentences composed of an action, an object, and what is called a location, although most of the locations are what we would regard as objects rather than locales. Fifty-three such commands were included in this study. Whereas 103

Apes and language: ontogeny with action-object-recipient constructions there is very little role reversibility between the objects, since in nearly every instance one of the objects was animate and one inanimate, here neither object is animate (with a single exception), so there is at least the possibility of two interpretations of the relationship to be enacted between them. Unfortunately, this possibility was not exploited by the experimenters, for the two objects in almost every sentence were different in type such that, were either Kanzi or a child who was familiar with these objects presented with the action word and the two object words in any order, neither would have difficulty in responding correctly. The following are examples of such commands, with the words presented in reverse order to demonstrate that syntactic analysis is superfluous to determining the appropriate response. (The understood subject - the recipient of the command - is always Kanzi.) me cooler bring backpack it put mouth it put cooler sour-cream put potty this put refrigerator grape take Of the 53 commands containing an action, an object, and a location, only "Put hot water (in) hot dogs/' "Put water (on) it," and "Put water (in) lemonade" seem, in the absence of contextual information, potentially ambiguous as to object roles without syntactic analysis. Before rejecting a pragmatic explanation of Kanzi's handling of these three commands, however, one would need to know the situations in which they were given. For example, in theory one can either put hot water in hot dogs or put hot dogs in hot water. But if the hot dogs were in a pot when Kanzi was given this command, and if Kanzi were accustomed to adding hot water from the faucet to hot dogs in a pot, then the situation would render the "theoretical" role reversibility between hot dogs and hot water irrelevant and syntactic analysis, as a consequence, unnecessary. The testing of Kanzi's sentence comprehension, in summary, demonstrates that he is able to put together the object or objects and the action mentioned in the way that is appropriate given the properties of the objects involved, what he typically does with them, or both. His performance provides no evidence, however, that he was attending to even so simple a syntactic feature as word order. What is new in the Kanzi project is subjects who learn the use - and arguably the meaning - of language tokens without undergoing operant conditioning. What is not new is the penchant of the experimenters for overinterpreting the achievements of their subjects, according them a linguistic competence that is simply not warranted by the data. 104

Sentences SUMMARY This chapter began with a minimal definition of syntax as a rule or set of rules specifying the permissible relationships among the elements of a representational system. The grammars of natural languages are composed of rules stated in terms of categories of symbols rather than the symbols themselves. The question of whether grammatical categories, those of the child in particular, are semantic in character or exclusively syntactic - defined on the basis of distributional privileges and their role in sentence-transformation rules - was discussed. All languages have grammatical means for indicating the semantic roles of the nominal elements in sentences and for distinguishing entities from attributes. English uses word order to encode such relationships, while other languages use inflection, or inflection and order. The first criticisms of the signing apes focused on the absence from their productions of regularities of order. The ape-language experimenters responded by (1) presenting evidence that their animals did order their sentence constituents and (2) asserting that their subjects used ASLspecific alternatives to word order for encoding semantic relations. The grammar of ASL was sketched, and it was shown that ASL does have a canonical sentence order, the subject-verb-object sequence of English. The language also provides, however, syntactic mechanisms that render word order flexible without incurring any ambiguity in semantic roles. It was pointed out that despite the ape-language researchers' referring to the languages in their projects as "ASL," in fact the medium employed in each case was some form of pidgin sign language, which, like English, follows an SVO ordering of constituents. The evidence for systematic adherence to ordering principles by the apes was evaluated, and that evidence was assessed as meager. Allegedly robust grammatical consistencies were shown to be mere lexical position habits, idiosyncratic statistical tendencies, or the product of imitation of chunks of the trainers' prior utterances. The claims that the animals in the signing projects acquired the grammatical modulations of ASL were reviewed. It was concluded that the material presented in support was either far too minimal to be compelling or simply not, in fact, evidence for what was claimed. Kanzi's most frequent two- and three-item combinations were examined. It was found that they not only lacked grammatical structure but usually consisted not of two or three lexigrams in construction but of one lexigram with one or two manual points. There is, in other words, no evidence that Kanzi syntactically encodes propositions. On the other hand, an abundance of evidence from spoken languages 105

Apes and language: ontogeny and signed languages, both conventional and invented, was cited in support of the contention that, in contrast with the experimental apes, children observe rules of role marking from the very beginning of multiword or multisign utterances. Children acquiring ASL, the comparison group of greatest interest, not only mark their utterances for semantic roles but overmark them, compounding sequential ordering of signs with spatial verb agreement by age three to three and one-half. The coining of sign combinations to refer to objects has been reported as a practice of the signing apes (and Lana). The descriptions of these novel combinations were criticized for omitting information necessary to support the interpretation of these strings as words in construction rather than as adjacent but separate utterances. Other studies bearing on the syntactic capabilities of apes were reviewed. The Gardners7 experimental study of Washoe's responses to Wh questions, the work of Fouts et al. on Air's ability to describe object arrangements, and Savage-Rumbaugh's testing of Kanzi's sentence comprehension were discussed. For all three cases it was concluded that the experimental outcome did not warrant the interpretations forwarded by the researchers. Partisans of the apes have protested that detractors employ rubber rulers and moving targets in comparing the apes and children (for example, Fouts 1975; Gardner 1982; Van Cantfort and Rimpau 1982). Criteria are revised "on the run," new criteria are imposed, or different methods of imposing a given criterion are used for apes and for children. Fouts observes that attempts in the past to define language have generally consisted of a list of the characteristics of language ... Usually, however, when evidence is presented that a nonhuman species achieves some of these characteristics, the list grows longer in order to exclude the interloper species. (1975, p. 373) The ape researchers, on the other hand, could be described as having fixated on certain of these criteria, generally the most mechanical and readily quantified, and then having either expressly trained them or searched them out in their data. This, in itself, is not an objectionable practice. One is put in mind, however, of Diogenes, the Athenian cynic who, in response to Plato's definition of man as "a featherless biped/' produced a plucked chicken. In both cases, the claimed identity may be valid on the narrow grounds used, but there is still a profound difference between the alleged equivalent and the genuine article. The difference between the cases - aside from the fact that, as I have argued, the apes are not equivalent to children on the various indices used in the literature - is that Diogenes knew he was holding a chicken. 106

Part four

Apes and Language: Phylogeny

Language, evolution, and anatomy

In case the preceding pages have not made this clear, I should state that I do not believe that any of the ape-language projects succeeded in instilling even a degenerate version of a human language in an ape. There are no persuasive data in support of syntactic patterning. At best, one or more of the animals in these studies acquired an ability, or enhanced an inchoate one, to represent things with symbols. However, the evidence for even this capacity, a prerequisite for syntactic productivity, is equivocal. Is it absurd to consider the linguistic incompetence of apes as something requiring explanation? After all, no one would think to pose the question of, say, why Homo sapiens cannot fly, or why birds cannot swim. These questions are not asked because it is assumed that taxa separated by millions of years of evolution will differ in their adaptations, the extent of the divergence corresponding roughly to the length of that separation, barring parallel or convergent evolution. Given that our lineage diverged from the most closely related hominoid at the very least four million years ago, which separates us by eight million years of independent evolution, why is it thought likely by some that an ape species would possess a human faculty, especially an unused form of it? The molecular geneticists King and Wilson (1975) presented an abundance of evidence that the structural proteins of humans and chimpanzees (Pan troglodytes), as well as the genes coding for those proteins, are 99 percent identical between the two species. The ape-language advocates frequently cite this study, arguing that such a striking genetic affinity makes it plausible that apes would share also our capacity for language (Fouts, Couch, and O'Neil 1979; Patterson 1980b).1 None of the proponents of ape language, however, have clearly stated the basis of this suggestion. If, consonant with the tenets of behaviorism, they view language as a learned behavior resting on nothing more neurologically specific than a superior ability to form associations, then any organism possessing sufficient intelligence could acquire language, and the phylogenetic distance 109

Apes and language: phytogeny between that species and our own would be irrelevant. If, on the other hand, language is thought to involve more specific neurological substrates, then it is a sensible conjecture that a species closely related to Homo sapiens might also possess those neurological attributes. It is important to appreciate, however, that successfully imparting linguistic ability to an ape would not constitute proof that the organic basis of the trait in that species was homologous, that is, phylogenetically connected, to ours rather than merely analogous, as in the relationship of insect wings to bird wings. Establishing an evolutionary homology as opposed to an analogy would require training a range of organisms, at least a number of primates, in order to track the distribution of linguistic competence and construct a phylogenetic tree of its evolution. The fact that the language projects with dolphins (for example, Herman 1984) have attained much the same results as the ape projects demonstrates quite clearly that any putative success in language training of a nonhuman species, including an ape, cannot facilely be ascribed to phylogenetic affinity (Lewontin 1990). Granted that the question of why the apes in these experiments did not develop language is a reasonable one, then, it should be addressed. Were they deprived of adequate pedagogical experiences? Do they lack some organically based capacity for language? If so, is the deficit simply a matter of insufficient intelligence, or are there more specific neural requirements for language, present in humans but absent in apes? If the latter, are these the direct result of selection for language or, alternatively, are they traits selected for other cognitive or perceptual tasks and only incidentally conferring on the species the capacity for language? The ape-language researchers have criticized each other's projects and their own for shortcomings in the educational milieu. Excessive turnover of project personnel, inadequate knowledge of ASL among teachers, and stifling rigidity in teaching procedures have been cited. Yet what variability there is among the projects in these areas seems to have had no effect on the outcomes. The Koko project appears to have had less turnover among the caretakers than the Nim study did. The teachers in the Gardners' postWashoe work (with Moja, Tatu, and Dar) included more native signers than did Washoe's teachers. Miles (1983) contrasts the informality of her orangutan's (Chantek's) language training with the drilling approach taken in the other studies. Despite this diversity, no appreciable differences in the accomplishments of these animals are apparent after a critical assessment of the evidence. The problem, it seems, lies with the students, not their teachers. Are apes unable to acquire language because they lack the requisite level of intelligence? This question assumes either that (1) language is made possible only by general intelligence, or that (2) though language is 110

Language, evolution, and anatomy predicated on more specific mental faculties, it could still be acquired by an organism lacking those traits but possessing sufficient general intelligence. The latter possibility invalidates such assertions as "if we are uniquely endowed with a faculty for language, no other species should be able to learn our language" (de Villiers and de Villiers 1979, p. 118). Horses and dogs are not constitutionally disposed to walk bipedally, as we are, yet they can learn to do a faithful rendition for short distances. Similarly, there is no reason to believe that humans have a specific faculty enabling us to play chess, as does a dedicated chess computer, but this does not prevent us from learning the game, albeit with greater difficulty than we have in learning language. Under the second suggestion, then, an organism of sufficient cognitive potential could construct in its mind a "virtual machine" that was functionally equivalent to the actual language faculty of our species, a virtual architecture duplicating the performance of a neurological one dedicated to language. Apes, according to this interpretation, simply do not possess the brain power necessary for this achievement. The problem with both variants of the inadequate-intelligence account is that, by certain measures of intelligence, chimpanzees and gorillas are the equals of young children, and it is not until after the early stages of language development that children leave the apes behind on these measures (Limber 1977). Redshaw (1978) studied four hand-reared lowland gorillas and two human infants until the age of 18 months, testing them on Piagetian scales of cognitive development. She found that the sequence of cognitive stages and the time between their attainment is the same for the two species but that gorillas are substantially precocious relative to humans on every measure but two. Premack (1983) pursued studies of Sarah's intellectual abilities after her language training ended. These experiments trained and tested her capacity to make causal inferences, draw analogies, conserve quantities (in the Piagetian sense), and so forth. To the extent that these studies were free of the methodological problems of Premack's language work, Sarah demonstrated a capacity for abstract thinking that would be considered remarkable in a young child.2 And Patterson, it will be recalled, has periodically tested Koko on standard intelligence scales, reporting a consistent IQ score of about 85. On measures of cognitive function, then, the asyntactic apes seem to be comparable to children who are putting words into grammatical constructions. Furthermore, children who are categorized as retarded according to the same measures develop language that is essentially the same as that of normal children of a younger age (de Villiers and de Villiers 1978; Fowler 1984). Thus, the apes' syntactic deficits cannot logically be ascribed to lack of intelligence. Ill

Apes and language: phylogeny To be sure, the independence of language from other intellectual capacities can be exaggerated. As Premack (1986) has observed, transplanting the neurological apparatus of language into a chicken would not produce a talking chicken. And the language of the retarded child, while recognizably akin to normal language, is undeniably affected by her deficiencies in the relatively nonlinguistic intellectual abilities. Clearly, in any species, the language faculty must be supplemented by a certain level of functioning in other mental domains for manifestation of normal language. What appears to be the case is that the ape is competent in some or all of the collateral areas but devoid of a language faculty. LANGUAGE AS AN ORGAN The understanding of language as a species capability based upon only general intelligence has been something of an anathema during Chomsky's 30-year dominion of language studies. Language, for him, is made possible by the language "organ" of the brain, a distinct module of cognition discernible in functional characteristics if not yet in anatomy. Piaget and his adherents have been perhaps the most steadfast dissenters, seeing language not as a biologically ordained development but as an inevitable outcome of mental growth during the sensorimotor period.3 Those following Chomsky (Fromkin and Klima 1980; Bickerton 1981; Gleitman 1986) counter by citing structural features of natural language that are logically unnecessary and have no obvious parallels in other cognitive domains, arguing that such traits cannot be accounted for as "constructions of sensorimotor intelligence" (Piaget 1980, p. 31). Others (Ingram 1981; Smolak 1982; Bonvillian, Orlansky, and Lazin 1983) take issue with the Piagetians through research showing that attainment of the cognitive hallmarks of the sensorimotor period is not necessary for the emergence of words or even word combinations. And some (McNeil 1971; Cromer 1981) point out that the existence of any developmental synchronies between such hallmarks and linguistic achievements, such as the parallel development of object-action concepts and subject-predicate expressions, does not necessarily demonstrate that the latter are derived from the former. The nativistsr case is not based simply on the universality of language or even on the existence of universal attributes of language - such things could be seen as the result of cultural diffusion or environmental constants. Rather, it is based on (1) the arbitrariness of certain rules and constraints on language structure, (2) the regularity of the onset of language acquisition, (3) the ease and rapidity of language acquisition, (4) the apparently uniform sequence of language forms in development across 112

Language, evolution, and anatomy languages, and (5) the fact that (4) is unaffected by practice, reinforcement, or intelligence (Cromer 1981). Lightfoot (1982) develops the notion of "the poverty of the stimulus/' the idea that any species-wide trait not completely explicable by environmental determinants requires a biological approach: In general whenever biologists see an intricate system emerging in a more or less uniform way and not simply determined by external forces, they assume a specific genetic structure that guides and directs the growth of that system if certain environmental needs are satisfied. The reasoning is based on arguments from the deficiency of the stimulus, showing that the stimulus, the shaping effect of the environment, is not rich enough to determine the intricacies of the mature system. (1982, p. 12)

The environmental stimulus relevant to language is deficient in several ways. The first is that the speech the child hears consists not only of grammatically correct sentences but also of slips of the tongue and illformed and incomplete sentences (Chomsky 1965). Since these latter inputs are not marked as defective, the child's ability to induce the correct grammar of the language on the basis of the language heard is limited.4 Even if the linguistic input to the child consisted exclusively of wellformed sentences, there would still be the problem of "inductive underdetermination": for any body of evidence, there exists an infinite number of hypotheses capable of accounting for that evidence.5 How does the child generalize to the correct grammar rather than to one of the infinitely many others that would account for the input experienced thus far but make incorrect projections about permissible constructions? The answer must be that the language learner is endowed with knowledge of the possible forms that grammars can take. The process of learning a language, under this conception, is one in which the child homes in on the correct member of the innately specified array of possible grammars, eliminating candidates on the basis of some specific and crucial linguistic inputs. In the remainder of this chapter and in the following one I will discuss some of the evidence that language is a product of evolution, hypotheses about the process of language evolution, and the possibility that living species related to our own possess traits, inherited from our common ancestor, that in humans gave rise to language. PURSUING THE ROOTS OF LANGUAGE Explaining the advent of language on earth is a pursuit longstanding, probably as ancient as the subject itself. The origin of language is at once tantalizingly interesting and unsolvable, a combination that may explain 113

Apes and language: phylogeny both the statutory prohibition of its discussion in the Linguistic Society of Paris in 1866 and the disregard of that prohibition by the membership (Aarsleff 1976). Modeling the evolution of language is problematic for reasons common to all evolutionary accounts of complex and seemingly unprecedented adaptations, and for additional ones as well. Elaborate biological traits, such as the vertebrate eye or winged flight, must appear to be radical discontinuities in evolution in the absence of credible transitional versions, either imagined or demonstrated. Strictly speaking, no characteristics of extant organisms can be phylogenetic links to those of others, since each living species is, to paraphrase Martin Gardner (1980), a terminal node on the tree of life. However, the judicious use of evidence from fossil forms and from relatively primitive living organisms, possibly combined with testimony from embryology, can yield reasonable scenarios for the origin of evolutionary novelties.6 All such accounts must address the standard set of difficulties that attend discussions of the origin of complex traits: How did the intermediate forms enhance fitness?, How could all of the requisite alterations occur in an integrated way?, and so on (Mayr 1960; Frazetta 1975). Evolutionary models of language are beset by these problems and then some, since, unlike the hard-tissue adaptations, and like other behavioral ones, language leaves no fossils. Stories about the evolution of language, as a consequence, are generally some combination of speculation about what the precursors of fully formed language logically must have been, and demonstration of putative functional or neuroanatomical components of language, or precursors of such components, in the animal world. But the quest for neuroanatomical features in other species that are homologous to those responsible for language in humans is problematic because the anatomical correlates of language are not known in great detail (Marshall 1982), although it can be said that we know roughly where language "resides" in the brain. Even a perfect knowledge of the neurology of language, however, would not place the pursuit of protolanguage via neurology on a perfectly logical footing, for the demonstration that a given cognitive or perceptual activity is localized does not prove that that area or areas evolved as an adaptation for that task. Presumably, given adequate techniques, it would be found that every perceptual task or cognitive function takes place in a particular part or parts of the brain, regardless of whether that task was the selective "reason" for those parts developing as they did.7 Thus the presence in a lower primate of some brain feature that in humans subserves a language function does not imply the presence of a rudimentary version of that function. In fact, though, such an argument is not often made. The significance of finding 114

Language, evolution, and anatomy pieces of the neurology of language in infrahuman species would be rather the possibility that such structures evolved initially by virtue of their role in other functions but, under new vectors of natural selection in the hominid lineage, came instead or additionally to subserve language. These structures, in other words, might have been so-called preadaptations for language. Evolutionary accounts of the origin of language, finally, must accommodate the fact that, unlike most morphological, physiological, and behavioral innovations, which are potentially beneficial to the single organism in which they first occur, a capacity for language would be useless unless it developed in concert in at least two individuals. The explanatory challenge posed by this constraint is diminished somewhat when the absurd image of an elaborate present-day language appearing ex nihilo in a cave dweller is replaced by the more reasonable one of a very simple symbolic capacity evolving in an incremental way. Still, the advent of grammatical rules and the replacement of inherited calls by learned words are quite refractory to compelling evolutionary interpretations. Such difficulties confer plausibility on the argument for a conscious invention of language, a proposal that will be considered later. THE NEUROANATOMY OF LANGUAGE The human brain is an exceedingly large one in both absolute and relative terms, even for a primate, an order characterized by large brain-body ratios. The human brain is three times as large as would be predicted from a regression line describing the brain-body weight trend among the primates (Passingham 1979). It has been suggested (Jerison 1973) that the capacity for language is attributable to this gross expansion of the brain rather than to more specific developments. Lenneberg (1967), however, has pointed out that microcephalic dwarfs, with brains no larger than those of chimpanzees, and despite a general mental retardation, nonetheless acquire the rudiments of language, indicating that language depends upon the organization of the brain in addition to sheer size. Passingham and Ettlinger (1974) countered that such dwarfs, because of their commensurately small bodies, show a brain-body ratio substantially greater than the adult chimpanzee's. But, as Holloway (1968) has observed, not all microcephalies are dwarfs, and some of those who are not dwarfs do learn to speak, a fact that, again, argues for the primacy of brain organization over size per se. The embodiment of language has yet to be fully explicated, but some things are known. In the 90 percent of people who are right handed, and a substantial fraction of those who are not, the ability to produce and 115

Apes and language: phytogeny decipher speech is located mainly in the left hemisphere of the brain (Witelson 1977a). Poizner, Klima, and Bellugi (1987) have shown this to be true also of deaf signers of American Sign Language, indicating that the left hemisphere is the seat of language rather than just speech. There is evidence that this functional asymmetry for language is present at birth (Witelson 1977b); infants as young as two months showed greater sensitivity to speech-sound contrasts presented to the right ear (and hence the left hemisphere8) than to the left (Entus 1977), replicating what had been demonstrated by Kimura (1961) in adults. (Musical sound contrasts, though, were better perceived by the right hemisphere, a result consonant with what is known about the cognitive specializations of that hemisphere.) Corresponding to the functional asymmetry is an anatomical one. Portions of the left hemisphere are larger than their right-side counterparts, especially the region containing Wernicke's area, a portion of the temporal lobe critical in speech processing (Geschwind and Levitsky 1968). As with the functional differentiation of the two hemispheres, this anatomical asymmetry appears to be congenital (Wada 1977). The angular gyrus

Geschwind (1965) forwarded another area of the brain as supporting, in fact as being indispensable for, speech. The angular gyrus of the left hemisphere's inferior parietal lobule is located at the juncture of the parietal, temporal, and occipital lobes and lies between the three areas of cortex involved in processing auditory, visual, and tactile sensations. This region is hypertrophied in humans relative to apes, even more so compared to monkeys. It was Geschwind's conjecture that the angular gyrus provides for the formation of associations between sense impressions from two or three of these sensory modalities, especially the visual and auditory. He envisioned this area as central to speech insofar as it mediates the connecting of names (auditory images) to visual impressions. Furthermore, such cross-modal connections do not entail any associations with the limbic system of the brain, which comprises the phylogenetically primitive regions responsible for, among other functions, emotion and the experience of pain and pleasure. Nonhuman animals, by contrast, lacking this brain component, would not be equipped to make intermodal associations free of limbic connections; rather, the argument goes, they form associations between sense images (1) within a single modality or (2) across modalities but only by way of the independent connection of each to the limbic system. Hence their inability to learn and produce words in the dispassionate, unconditioned, and volitional 116

Language, evolution, and anatomy manner that typifies speech - their inability to refer rather than merely to respond. Psychological and anatomical discoveries since Geschwind offered this hypothesis indicate that, although the angular gyrus region may serve the function in language that Geschwind suggested, the absence of words in infrahuman primates cannot be due to an inability to make cross-modal associations. Davenport and Rogers (1970; Davenport 1977) reported that chimpanzees and orangutans in their laboratory were able to associate the visual and tactile impressions of an object. In the basic experiment, the subject was shown an object and was allowed to feel, but not see, two others, one of which was identical to the visually presented specimen. The task was simply to choose the object that felt like the specimen looked, as it were. (A variant involved presenting the sample in the tactile mode and the choices in the visual.) Once the animals came to understand what was wanted, they readily matched the seen and felt images, clearly demonstrating an ability to connect stimuli from different sense modalities. Such cross-modal matching, in which disparate percepts of a single object are connected, is not identical to what Geschwind proposed as underlying the use of words, the association of an object's image in one sense with an auditory image not of it but of its name. However, both processes entail integration of images from different domains, and, though cross-modal matching is obviously not sufficient for naming, it would seem to be necessary. Lacking the ability to collate the diverse sensory images of an object into a "metamodal" concept, we would be burdened with a congeries of independent notions of it, and, among other inconveniences, our dictionaries would be "five times too long," as Desmond (1979) observed. Davenport and Rogers were not able to produce cross-modal matching in monkeys, but Cowey and Weiskrantz (1975) succeeded, eliciting in Rhesus macaques a performance equal to that of human infants. Elliot (1977) succeeded with capuchin monkeys. The anatomical basis of what had previously been regarded as a uniquely human cognitive function may have been elucidated by Pandya and Kuypers (1969), who demonstrated that the Rhesus monkey has in its inferior parietal lobule direct neuronal connections between the cortical regions for the different senses. Histological assays of ape brains in search of equivalent structures have not, to my knowledge, been carried out. At the level of gross anatomy, however, there is evidence that the region of the ape brain corresponding to the angular gyrus in humans is expanded. LeMay and Geschwind (1975) looked for cerebral asymmetries in brains of chimpanzees, gorillas, and orangutans. They found that the sylvian fissure, which defines the upper margin of the temporal lobe, tends to be 117

Apes and language: phytogeny longer and more horizontally oriented on the left side. This is identical to the human picture, in which the left sylvian fissure is extended and deflected horizontally as a result of the evolutionary expansion of Wernicke's area and the adjacent regions, including the angular gyrus. Yeni-Komshian and Benson (1976) reported the same asymmetries in chimpanzees. In summary, it is now known that both apes and monkeys are capable of forming cross-modal associations. There is evidence that both of these primate grades possess the neurological apparatus on which this ability rests, and the angular gyrus, it would seem, should not be regarded as the explanation of the language gap. The markedly greater development of this region in Homo sapiens, however, is probably part, perhaps even the most important part, of an exclusively human evolutionary complex of organic traits undergirding language. Functional lateralization

The experimental literature on lateralization of brain function in infrahumans provides little evidence for its existence; Hamilton (1977) summarizes his review of the literature as offering "little support for the concept of hemispheric specialization in infrahuman mammals" (p. 231). Among the very few studies that do suggest lateralization is that of Dewson (1977). Macaques were taught to push one button or another depending upon the auditory stimulus they received. Dewson then noted the effect of surgical ablation of the superior temporal gyrus, the area homologous to Wernicke's area in humans. He found that right-side lesions had no effect on performance of this task but that left-side lesions clearly impaired it. Since Hamilton's review was published, reports of lateralized perception of vocalization in monkeys have appeared (Petersen et al. 1978; Petersen 1982). This work is best understood in context with some facts about the left-hemisphere dominance of speech perception in humans. Apart from the fact that language as a whole is left lateralized, it is known that the right-ear (left-hemisphere) superiority for discriminating speech sounds is greatest for stop consonants (such as [b], [p], and [k]), intermediate for fricatives (such as [f] and [v]), and generally nonexistent for vowels. On the other hand, such paralinguistic features as pitch and intensity show a left-ear (right-hemisphere) advantage (Petersen 1982). Petersen focused on a particular class of sounds in the repertoire of the Japanese macaque, a type of coo. Coos are emitted in social situations when the animal desires contact, but there are subtypes that differ both acoustically and in subtle functional ways. The subtypes looked at included one in which an ascending-descending frequency modulation 118

Language, evolution, and anatomy occurs early in the call ("early peak") and one in which this transition occurs relatively late ("late peak"). Employing various perceptual tests, Petersen determined that the macaque's ability to discriminate these vocalizations was clearly left localized, while pitch was either not localized or right localized, a pattern that is analogous to the human left dominance for certain speech sounds and right dominance for paralinguistic stimuli. A subsequent experiment by Heffner and Heffner (1984) corroborated Petersen's conclusion. Ablation of the left superior temporal gyrus, but not of the right, impaired the ability of macaques to discriminate the coo subtypes of the Petersen study. Falk (1980) has reported that several genera of Old World monkeys have gross asymmetries in cortical regions homologous to the language areas in humans. Since monkeys do not exhibit handedness, she asserts, these left-side expansions cannot be due to cerebral dominance for righthandedness, and may instead be adaptations related to vocal communication. Petersen's work is cited in support. Her argument thus contradicts both the anatomical findings of LeMay and Geschwind (1975) and YeniKomshian and Benson (1976), who found no such asymmetries in monkeys, and the functional studies of Myers (1976), Robinson (1976), and Maurus and Ploog (1984), who claim that the vocal repertoires of monkeys reside in the limbic system and in those parts of the cortex, bilaterally distributed, that preside over social behavior. Outside of the primate order, in which evidence for functional lateralization related to communication is, save for Homo sapiens, scarce and, as indicated above, somewhat controversial, the only clear-cut cases of such hemispheric specialization are represented, interestingly, by songbirds. Nottebohm (for example, Nottebohm and Nottebohm 1976) discovered that neural control of the production of song in certain songbirds, such as the canary, is essentially a property of the left hemisphere. The fact that this function, like language in humans, is left lateralized should not be accorded undue significance, as there are only two hemispheres for nature to choose from if a function is going to be assigned to one side of the brain. But the lateralization itself is a provocative observation when considered in conjunction with the fact that both songbirds and humans have exceptionally complex vocal productions. Lateralization of song production is thus generally held to be a remarkable and suggestive discovery in the biological study of animal communication, although no one seems to know exactly what it suggests. It would seem to be pertinent to a growing reinterpretation of human left-hemisphere specializations as adaptations not for language per se but more generally for "precise, temporal coordination of hierarchically ordered structures" (Studdert-Kennedy 1981, p. 547), "certain kinds of motor function, both verbal and nonverbal" 119

Apes and language: phytogeny (Kimura 1979, p. 217), or "resolution of complex [acoustical] spectral and temporal information" (Petersen 1982, p. 178). In other words, the competencies of the left hemisphere, according to this recent understanding, evolved for the analysis and/or production of complex motor and sound sequences. Even if this proves to be the best characterization of the specialties of the left hemisphere in both songbirds and humans, the commonality can obviously only be a case of evolutionary convergence of unrelated taxa the neurology of songbird vocalization must remain of analogical rather than homological relevance to work on human communication. It also bears stating that recognizing a common function does not explain, in either case, why a cerebral specialization evolved in a unilateral fashion, in contrast to the bilateral representation of functions that seems to characterize all of nature above the primitive radially symmetric phyla (Levy 1977). Bever (1982a) has speculated that language and other cognitive operations that involve "relational" as opposed to "holistic" processing come to reside in the left side ontogenetically not because of any congenital structural peculiarities in the left hemisphere that better equip it for such tasks but because, for reasons yet unknown, the left hemisphere simply has more computational power than the right. Since relational processing is inherently more demanding than the holistic cognitive tasks, the left hemisphere appropriates the former and the latter end up in the weaker hemisphere. One noteworthy implication of the notion that the left hemisphere excels in complex motor tasks, whether constitutionally or as a developmental outcome of its computational superiority, is the support it lends to arguments for a gestural origin of language in the species. Hewes (1973) cites the verbal shortcomings of nonhuman primates, the manual dexterity evident in the artifacts of early hominids, and the role of gesture in present-day language in developing his belief that the earliest language was gestural.9 A virtue of this model is that cerebral lateralization of language is explained, deriving from an already present left dominance for right-handedness in gesturing and tool use. The recent interpretation of the nature of left-hemisphere competencies, however, rather than taking right-handedness as a given and deriving the left localization of language from the fact that the right hand is controlled by the left hemisphere, would have gestural language, which obviously entails precise control of hand movements, as naturally lodged in the left hemisphere. Subsequent vocal language would develop on the left side either because it too requires complex motor control or simply because the neurological framework for gestural language was already in place there.10 120

Language, evolution, and anatomy Falk's (1980) conception of the evolutionary relationship between left dominance and language is squarely opposed to that of Hewes. Since Old World monkeys, according to her anatomical studies, have left-side adaptations for communication and yet do not show handedness, these cerebral innovations must antedate handedness. In fact, she seems to view human right-handedness as in part an epiphenomenal effect of the expansion of the language areas, which are adjacent to cortical association areas for the hand. Speech-related cortical changes, in short, appear to her to have occurred prior to the divergence of the Old World monkeys on the one hand, and the apes and hominids on the other, and thus preceded both handedness and tool use. The possibility that similar cortical changes occurred in parallel after the cercopithecoid-hominoid split is not considered by Falk. Anatomical study of a diverse sample of Old World species would be necessary to test this possibility; the sparser the distribution of these asymmetries, the less likely her scenario would become. CATEGORICAL PERCEPTION One of the more interesting results of psycholinguistic research is the demonstration that our perceptions of certain speech sounds that grade into one another along acoustic continua do not vary in a commensurate way. In our perception of some classes of speech sounds, we lump certain sounds together as identical and discriminate between others, even though the objective acoustical difference between some of the sounds heard as identical may be greater than that between those heard as different. This is referred to as categorical perception. Remarkably, the points on the acoustic continuum that mark the experimentally determined dividing lines between perceptual classes are precisely the points that differentiate certain phonemes in English and other languages. This will become clearer with an example. Liberman et al. (1967) synthesized an array of syllables, with [ba] at one end of the array and [pa] at the other. The English phonemes [b] and [p] differ essentially only in when vibration of the vocal cords (voicing) occurs relative to the consonant's articulation, [b] is produced simultaneously with voicing (a voice-onset delay of 0), while the onset of voicing with [p] is delayed, typically, for 25 milliseconds. The artificial syllables in Liberman's array differed only in voice-onset time, each one a constant acoustic step away from the adjacent ones in this respect. The range of voice-onset time was from - 40 milliseconds (that is, voicing beginning before consonant production) to + 40 milliseconds (voicing beginning 40 milliseconds after consonant production). Subjects asked to identify each syllable as one or the other did so 121

Apes and language: phylogeny without difficulty. They made a sharp differentiation between those on either side of the 25 millisecond point, which happens to be the voiceonset time that divides a [b] from a [p] in English. In addition, when presented with pairs of syllables drawn from various points along the continuum and asked to say whether they were identical or different, subjects had difficulty in hearing differences between two syllables that they had named as the same sound during the classification task, regardless of the actual difference between them in voice-onset time. Essentially, all of the sounds within a phonetic category were heard as identical, while any two from the different categories were heard as absolutely different. Thus, even the two that were adjacent on the acoustic continuum but fell into the different categories were unambiguously heard as [ba] and [pa]. The same results were obtained when the continuum involved other consonants that differ only in voice-onset time, such as [t] versus [d], as well as with some consonants contrasting in a different articulatory feature. Categorical perception was an interesting finding in itself. It is not what would be expected on the basis of perception in other sense modalities or perception of nonspeech acoustical stimuli. Though all languages partition the continuum of colors into discrete lexical categories, speakers have no difficulty in discriminating slightly different shades within a color category. And pure tones of sound varying continuously are perceived in a continuous manner. The additional observation that the stimulus-equivalence divisions that emerged in the laboratory corresponded to the phonemic categories of the language was suggestive of an innate basis for phonological organization. Further research would be necessary, though, to determine whether these perceptual phenomena were the cause or the consequence of the phonemic distinctions used in English. When Eimas et al. (1971) reported that infants as young as one month showed the same discrimination patterns as adults, with the same voice-onset-time discrimination boundary, and that they also categorically perceived the place-of-articulation continuum (Eimas 1974), which differentiates, for example, a [b] from a [d], categorical perception became a truly exciting discovery for those pursuing the biological bases of the structure of language.11 For example, Marler (1975; 1976) postulated, largely on the basis of the categorical-perception work, the existence of innate feature detectors in humans, "auditory templates" that serve to focus the infant's attention on speech sounds and aid in the acquisition of phonemic contrasts. And, for adults, innately derived categorical discriminators would seem to be a very sensible adaptation for language, allowing us to ignore the inevitable but linguistically functionless minor variations in the articulatory realization of phonemes in normal 122

Language, evolution, and anatomy speech. Lieberman (1984) also has made a strong case for the operation of neural feature detectors evolved specifically for language. Any indications of categorical perception in nonhuman primate species would obviously be relevant to tracing the evolutionary roots of language. In his report on cerebral asymmetry in the perception of species-specific vocalizations (described above), Petersen (1982) reports that although he had not yet done any experiments expressly designed to investigate categorical perception, there was some suggestive evidence for it in his macaque subjects' responses to natural and synthesized calls. The animals, which had been trained during the discrimination experiment to emit a different instrumental response to each of the two coo types (late-peak and early-peak), produced a markedly categorical response pattern to these calls, even though the actual peak position within the coos presented to them varied across a substantial range. The boundary between coo types was around .60; coos in which the frequency modulation occurred earlier than six-tenths of the way into the coo elicited the trained response for early-peak coos, later-occurring peaks causing the response for late-peak coos. Interestingly, this boundary corresponds to that observed to separate statistically these two coo types in nature (Green 1975b, cited in Petersen 1982). Snowdon and Pola (1978) specifically tested for categorical "labeling" of the trills made by the pygmy marmoset, a member of the New World monkey superfamily. The natural "closed-mouth" trill of the pygmy marmoset differs from the "open-mouth" trill in being shorter and in tending to evoke antiphonal calling in hearers. Snowdon and Pola synthesized a range of trills varying in several acoustic features, including duration, rate of frequency modulation, and size of frequency modulation. In playing the synthesized trills to their subjects, they found that duration of the call was related to response but that the response pattern was notably categorical; trills shorter than 250 milliseconds were "labeled" as closed-mouthed trills (for example, they elicited antiphonal vocalization), while those longer than this boundary figure were treated as open-mouth trills (no vocalization in response). And, parallel to the human evidence and that in Petersen's study of macaques, the acoustical dividing line that emerged in the laboratory coincided with the one that statistically divides the natural calls. Reassessing categorical perception

The long list of reasons for concluding that categorical perception is not a phylogenetically continuous trait that eventuated in the phonemic principle of human language might as well begin here, with Snowdon's work. 123

Apes and language: phytogeny Snowdon (1982) relates that, despite the categorical labeling of trills shown in the study above, the fact that, in other work of his, pygmy marmosets have proved capable of distinguishing the calls of different individuals within a trill category indicates that they are not unable to perceive acoustical differences within a category. Thus, these animals are not perceiving categorically in the true sense; depending on the demands of the experimental situation, they can respond to the stimuli categorically or not.12 Next is the rather surprising demonstration by Kuhl and Miller (1975) that chinchillas categorically perceive consonants that differ on the voicing continuum. Moreover, they draw the boundaries between their stimulus-equivalence classes at the same points that English speakers do! Needless to say, this observation militates against the interpretation of categorical perception as an adaptation for language or even for sophisticated nonlinguistic communication in lower primates. That interpretation is further weakened by studies indicating that categorical perception is not unique to the domain of speech sounds. Cutting and Rosner (1974) and Pisoni (1977) have shown categorical perception for certain musical stimuli, and Pastore et al. (1977) have elicited it in vision, using a continuum of flicker frequency. Studies of diverse languages have shown that while all languages utilize voice-onset time to construct some phonemic categories, the boundary points that distinguish the categories do not universally coincide with those demonstrated for English speakers and infants, suggesting a conventional rather than natural determination of the boundaries. The same is true of the place-of-articulation continuum. For example, MacKain, Best, and Strange (1981) tested American native speakers of English and two groups of Japanese immigrants to the United States who differed in previous exposure to English. The native English speakers and those Japanese immigrants who had received intensive training in English perceived the difference between [1] and [r], a contrast within the tongueposition continuum, in a categorical manner, while the immigrants who had not received training in English did not, an outcome almost certainly attributable to the fact that in English but not in Japanese there is a phonemic contrast between these two sounds. Newport and Supalla (Newport 1982) investigated the perception of ASL signs by producing artificial sign continua that covered a range of locations or handshapes. They found that subjects were just as accurate in discriminating small differences between pairs of artificial signs that had earlier been classified as the same sign as they were with pairs in which each sign had been labeled as a different sign. In other words, there appeared to be no categorical perception of the articulatory elements of ASL signs, 124

Language, evolution, and anatomy although ASL is very much a linguistic communication system, possessing the abstract structural properties of spoken languages (Bellugi 1980). Finally, there is the work indicating that categorical versus continuous perception can be manipulated through changes in experimental procedures (Carney, Widin, and Viemeister 1977; Pisoni et al. 1982). The fact that vowels, which are no less essential to language than are consonants, are not perceived categorically should have tempered the conjecturing of those seeking to predicate phonological organization on perceptual constraints. The relationship between the two is well denned by Newport: The presence or absence of categorical perception appears to be irrelevant to the nature of linguistic organization: Both spoken and signed languages are organized phonetically in terms of categories - physically distinct stimuli are treated within the linguistic system as identical - regardless of whether these stimuli are categorically perceived ... (1982, p. 460)

Studdert-Kennedy (1980), who was involved in some of the pioneering research on categorical perception, regards categorical perception as "neither peculiar nor necessary to speech" (p. 51). Categorical perception, in conclusion, seems to be a feature of the auditory system in a number of mammals. In light of this distribution, it is impossible to regard it as an adaptation for language. However, given the recurrence of the experimentally ascertained voice-onset-time division point in phonemic contrasts among the world's languages (Lisker and Abramson 1964), the most sensible interpretation of the role of categorical perception in language is that, rather than the former in some sense dictating the latter, languages exploit certain discontinuities in human auditory sensitivity. It is unfortunate that no experimental evidence for cerebral dominance or language-like perceptual processing exists for apes as opposed to monkeys, since the former, especially the great apes, would be considered on phylogenetic grounds more likely to possess homologous characteristics. The lack of such evidence, however, should not be interpreted as an evolutionary paradox, for the requisite experiments have simply not been carried out. The reason for this is probably a matter of too little tractability and too much physical capability in the potential subjects, as expressed in the answer to the question of where a 500-pound gorilla sits.13 SUMMARY This chapter began by considering alternative explanations for the absence of language in nonhuman primates in nature and their inability to acquire it in captivity. It was concluded that neither shortcomings of the 125

Apes and language: phylogeny pedagogical milieu nor insufficient intelligence are compelling answers. This suggested that language is a unique function of human biology, and arguments from language acquisition in children were presented in support of the nativist perspective on language. Various methodological difficulties attending the pursuit of evolutionary origins of language were discussed. These include the logical impossibility of finding ancestral forms of language in extant species, the challenge of accounting for the gradual development of intricate traits, the absence of fossil evidence of a direct nature, the nondeterministic relationship between structure and function in neuroanatomy, and the fact that language would have no obvious adaptive value until it had "arisen" in at least two individuals. Despite these impediments, a survey of selected anatomical and functional traits thought to be relevant to the capacity for language was undertaken. The focus was on traits in nonhuman primates that might represent preadaptations for language inherited from the common ancestor of monkeys, apes, and humans or of apes and humans. With respect to gross anatomy, it was suggested that overall brain size could not be the major factor in the language gap. General regions of the cerebral cortex implicated in language processing were mentioned. The role of the angular gyrus in the formation of cross-modal associations was considered, and it was concluded that this region could not be the crucial differentiator. Hemispheric asymmetries in human and ape brains were noted, as well as the disagreement about the presence of such asymmetries in monkeys. Humans, it was noted, show left-hemisphere localization for language. Evidence for similar left-right functional asymmetries in nonhuman primates was discussed. Very few studies have succeeded in demonstrating hemispheric specialization outside of Homo sapiens. The major such experiments involving nonhuman primates were summarized, and the work demonstrating left localization of song production in certain birds was mentioned. Linguistic and nonlinguistic characterizations of the left hemisphere's special competence were considered. The implications of this competence for the evolutionary connections among tool use, handedness, and vocal and gestural language were explored. Finally, the role of categorical perception in language was treated. Evidence for its operation in species other than humans in the processing of species-specific signals was presented. Though in several ways suggestive of a biological prerequisite for language, categorical perception was finally rejected as a key to language origins on the basis of its peculiar 126

Language, evolution, and anatomy distribution in nature, its operation in nonlinguistic perceptual domains, and its inessential role in language perception. A survey of the comparative neuropsychology of primate communication, then, neither abounds with nor completely lacks evidence suggesting the presence of structural and functional precursors of language in the relatives of Homo sapiens. It is important to realize, however, that the case for language as a biological attribute and hence as a product of evolution does not rest on the demonstration of such continuities. Rather, the nativist position is supported mainly by the evidence from two disparate sources: (1) the language acquisition process and (2) the localization of the language capacity within the brain.14 Moreover, if the organic hardware underlying language evolved after the hominid-ape split, which occurred sometime between 4 and 15 million years ago, then there would be no reason to expect language-related traits outside of our species. The development of such a complex characteristic in what is, on the evolutionary scale, a very brief period of time is perhaps improbable but not impossible, especially considering the arguably enormous fitness differential between the possessors of language and the inarticulate. The next chapter rounds out the inquiry into the language gap by assaying the evidence from primate communication in nature.

127

8 Primate communication in nature

The literature on naturalistic primate communication has something for everyone - those seeking to document continuity between human language and infrahuman communication as well as those promoting a natural hiatus. Despite this lack of consensus, I think it is accurate to describe the literature as providing much more support for the discontinuity position. The issues in the primate-communication literature bearing on the origin of language lend themselves to a number of contrasts. Some of these contrasts are dichotomous, while others define a continuum of possibilities, with language at one end and primate systems, if not at the other, than at least elsewhere. It will be apparent that several of the oppositions presented in the following pages are conceptually linked; in a more meticulous treatment, some might prove to be essentially synonymous oppositions. VOCAL-AUDITORY VERSUS OTHER CHANNELS Human linguistic exchanges typically employ a visual channel, parallel to the vocal-auditory one, which contains information conveyed by such "paralinguistic" means as hand gestures, body attitude, and facial expression. In addition, the vocal-auditory channel, through such aspects as pitch, stress, and volume, carries information supplementary to the "words" themselves, information about the speaker's current emotional disposition in general, toward the listener, and regarding the objects of reference in the discourse, as well as clues about the speaker's personal identity, age, gender, geographic origin, and so forth. It is the verbal content of the vocal-auditory channel, nonetheless, that is properly considered the central information in language; far more of what is distinctive in human communication would be lost by deleting the words of a conversation than by blocking out the gestures and facial 128

Primate communication in nature expressions. This is why books contain the former but can dispense with the latter. Nonhuman primate communication, on the other hand, is virtually always a multichannel affair, in which vocalization often serves merely to call attention to a complex of visual, olfactory, and tactile signals. Infrahuman primates have not "specialized" in vocal communication as we have. Lancaster (1975) has observed that in primate societies, a blind animal would be much more seriously impaired in communication than a deaf one. Menzel (1973, 1975) conducted a number of studies of captive juvenile chimpanzees in an outdoor enclosure. Chimpanzees, he discovered, are able to extract a remarkable amount of information regarding the location, quantity, and desirability of a hidden item just by observing the gaze orientation, movement, and locomotory posture of a group member that has been shown the hidden object. LEARNED VERSUS INBORN VOCALIZATIONS As discussed in chapter 7, the organic bases of language are far from being a settled matter, but few would quarrel with the claim that the capacity to acquire a language is in some sense inherited. Even fewer, on the other hand, would contest the assertion that the details of the particular language acquired are a function of the individual's milieu rather than her DNA. The bulk of evidence pertaining to the ontogeny of primate vocal signals indicates that an animal's repertoire of calls is basically a congenital endowment. A number of studies (for example, Winter et al. 1973) have shown that infant monkeys, even those deprived of exposure to conspecifics, possess essentially the full complement of species-specific calls. Studies of hybrid offspring of taxa with distinct call structures demonstrate the inheritance of vocalization characteristics (Gautier and Gautier 1977; Newman and Symmes 1982). The observation of Marler and Tenaza (1977) that the vocal repertoires of the gorilla and chimpanzee show a close correspondence despite their ecological differences testifies to the role of phylogeny in primate vocal communication. The literature is not univocal on this score, however. The existence of troop dialects in macaques (Green 1975a) argues for a learned component in vocalization. And Newman and Symmes (1982) cite studies indicating that completely normal development of calls requires acoustic exposure to conspecifics. The fact that the details of language are learned rather than inborn distinguishes ours from nearly all other natural systems of communication. While it is not self-evident that learning as opposed to inheritance 129

Apes and language: phylogeny of the symbols is a prerequisite for the kind of combinatorial syntax that also distinguishes language,1 any adequate theory of the evolution of language must address the transition from a prehominid call repertoire that was, inferring from present-day mammalian systems, innate, to a system based upon vocal learning. Dialects as adaptation

Hill (1974) proposed a rather tenuous explanation for the advent of vocal learning in hominids based upon studies of bird-song ontogeny by Nottebohm (1970). Nottebohm found three distinct patterns of song development in songbirds. In some species, the vocal pattern is innate, and is invulnerable to both isolation from conspecifics and early deafening - the normal song will occur despite one or both deprivations. In others, song development is impervious to isolation but not to deafening - the bird requires auditory feedback for the innate song template to produce a fully normal song. In the third group of species, the bird requires both auditory feedback and exposure to the species song during a critical early period in development. Lacking either, the song does not develop properly. Nottebohm observed that bird populations with the third ontogenetic pattern often also show regional dialects. Populations with regional dialects, in turn, tend to be rapidly expanding into diverse econiches. He speculated that since a bird's dialect was learned prior to maturation and dispersal, dialects might function as at least partial reproductive isolating mechanisms, individuals mating preferentially with members of their own dialect. This would promote a system of partially isolated local gene pools, an evolutionary pattern which, among other things, can facilitate rapid microevolutionary diversification by protecting the adaptive genefrequency changes of each local population against nullification through extensive interbreeding with other demes. Hill suggested that rapid fissioning of small populations into diverse microenvironments was a characteristic of early hominid demography. Given this, vocal learning would have been selectively favored over an immutable call system, she argued, because it would produce dialects, which in turn would promote endogamy, which would result in local demes and consequently an "evolutionarily flexible population structure consisting of small endogamous demes" (p. 146).2 The difficulties with accepting this scenario are several. There is, to my knowledge, no ethnographic evidence that dialect differences operate as determinants, as opposed to mere correlates, of marriage groups. Secondly, one wonders how ecologically distinct the niches of these dialect groups could have been if contact between them was sufficiently 130

Primate communication in nature frequent for preventing mating across groups to be of evolutionary relevance. (Birds, on the other hand, being far less constrained in their dispersal capacity, might be expected routinely to fly far afield of their natal econiches.) Indeed, given this contact, the development of dialects itself seems improbable. There is, thirdly, no reason for believing that early hominid bands living close enough to one another to be able to exchange mates were characterized by selection-caused gene-frequency differences. Such a pattern is certainly not found among contemporary "anthropological" populations, nor would it be expected. The hominid adaptive strategy is, and arguably always has been, one in which culture serves as an adaptive buffer between populations and environmental variation of the relatively minor magnitude under consideration here.3 In fact, Nottebohm presents no evidence that the bird populations occupying distinct habitats and exhibiting local dialects also differ in genetic characteristics; this was only a speculation. Finally, it seems likely that even Rube Goldberg would be unwilling to invoke so tortuous a mechanism for the selective origins of vocal learning in hominids when simpler and more obvious advantages of learning over inheritance, such as the utility of coining new names for things, are available. Hill proposed, in brief, that vocal learning arose because of the adaptive value of one of its more indirect consequences, the formation of dialects. Given that learning has several more proximate advantages, it seems that Hill mistook dialects to be an adaptation rather than merely the inevitable outcome of combining (1) learning, an imperfect system of transgenerational replication of language, with (2) partial communicative isolation.4 Understandably taken with the several parallels between the ontogeny of bird song and of language, including evidence for a critical learning period, left lateralization of productive competence, and dialects, and emboldened by Nottebohm's own speculations along the same lines, Hill offered an imaginative, if untenable, scenario. Her more general argument, that in pursuing the origin of language we should focus more on the relationship between evolutionary ecology and communication system and less on putative similarities between language and primate communication, is worth considering - the songbirds may yet have something important to tell us. VOLITIONAL VERSUS NONVOLITIONAL CONTROL OF VOCALIZATION It is theoretically possible for a set of calls whose acoustical attributes are completely specified genetically and are impervious to learned alteration to be subject to conscious control in their production. It is also conceivable 131

Apes and language: phylogeny that vocal signals acquired through learning would be produced only in an automatic, limbically controlled, involuntary fashion. The contrast between volitional and nonvolitional control, then, is in principle independent of the learned/inborn distinction discussed above. In nature, however, these are not freely varying attributes; there is a correlation between volitional control and acquisition by learning, and between relatively involuntary production and congenital origin. In humans, the occasion of communicating and the content of communication are, ignoring the applicable philosophical cavils, intentionally determined. The cause of vocalizations in lower primates cannot be discovered through introspection (not by us, at least), so the relevant evidence comes mainly from experimental work in which conditionability of vocal output is used as an operational index of voluntary control.5 Various studies of the ability of monkeys to control their vocalizations in accordance with reinforcement contingencies have produced diverse results, but the majority of them corroborate the observation of Myers (1976) that it is unclear "whether or at what level the nonhuman primate has developed even rudimentary mechanisms in its brain that can support any voluntary control of its face or voice" (p. 755). Yamaguchi and Myers (1972), for instance, reported that monkeys that readily learned to press a bar or not, subject to presence or absence of a red light, failed to learn to control their vocalization similarly after two years of training. One of the relatively few studies with contrasting results is the experiment of Sutton, Samson, and Lawson (1978) in which monkeys learned to emit a coo to a red light and a bark to a green one. SEMANTIC VERSUS AFFECTIVE SIGNALS This distinction is intimately connected with the volitional/nonvolitional issue discussed above, and also with the contrast between discrete and continuous signaling, discussed below. A concise way of expressing the semantic/affective contrast would be to say that semantic signals refer6 to things while affective signals only reflect the sender's disposition.71 use the term "disposition," which points ambiguously to both emotional state and behavioral inclination, deliberately, to capture the two most common senses in which animal signals are said to be affective. As in the relationship between the volitional dimension and the learned/inborn one, an independence obtains in theory between the cause of vocalizations and their signification. Semantic signals might be subject to uncontrollable utterance, and affective signals might be intentionally deployed. Again, however, there appears to be an empirical cohesion, here between the voluntary and semantic poles (humans), on the one hand, 132

Primate communication in nature and involuntary and affective signaling (nonhumans), on the other. To the extent that signals are a product of the limbic system, which is also the seat of emotional response, such a relationship would be expected. The received wisdom in primate studies, which has been endorsed at various points in the present work, is that the calls of prosimians, monkeys, and apes are, for the most part or entirely, affective in signification. Speaking of Old World monkeys, Gautier and Gautier observe that "in general, visual, acoustic, and olfactory signals all appear in diverse situations, and they seem to convey no specific indication of the nature of the stimulus ..." (1977, p. 955). In a field study of macaques, Green (1975b) determined that the reliable correlates of the various calls were emotional dispositions rather than contextual elements. At the same time, there is in the literature a certain uncertainty on this issue, and this is reflected in terminology. Students of chimpanzee behavior, for example, use the term "food call" in referring to the signal typically emitted by an animal on discovering food. It would seem, though, that this usage is only justified if it can be shown that this call is not made in other contexts. (However, even this demonstration, which would support a referential reading, could be challenged by an interpretation positing a specific emotion elicited exclusively by the discovery of food.) The semanticity of these calls is often implied but seldom substantiated. It may be a mistaken attribution resulting from the fact that certain calls are made predominantly in certain situations. Marler and Tenaza (1977), citing Fossey (1972), report that the gorilla's "belch," like the chimpanzee's "rough grunt," is typically given in feeding contexts but, unlike the latter, it occurs also in circumstances involving pleasurable group activity. This suggests that the chimpanzee call is referential in nature, the gorilla's affective. On the other hand, specificity is not without problems as a criterion of reference - it is possible that the gorilla's belch is as semantic as the chimpanzee's rough grunt but simply has a more general referent. Deciding whether a signal is referential or affective is not, in short, a cut-and-dried procedure. The best evidence for reference in nature, outside of the honey bees, comes, contra the above observations by Gautier and Gautier and by Green, from field studies of monkey communication. Struhsaker (1967) showed that vervet monkeys (Cercopithecus aethiops) emit acoustically different calls at the sight of each of their main predator classes: leopards and baboons, eagles, and pythons. Seyfarth, Cheney, and Marler (1980) conducted a series of field experiments in which they broadcast recordings of these calls from a hidden tape recorder. They found that each alarm call evoked a different, and appropriate, response, even though no predator was present. Thus, playing the call originally 133

Apes and language: phytogeny given in the presence of a leopard caused the monkeys to climb trees. The eagle alarm resulted in looking up or seeking protection in bushes, and the snake alarm evoked a ground-scanning and mobbing response. To test the possibility that the calls were just variants of the same acoustical pattern, differing only in loudness or duration as a function of the degree of arousal caused by the different predators, the experimenters varied the amplitude and duration of the calls, but this had no effect on response. Their assessment of this work: Because playback of the vervets' alarm calls in the absence of predators evoked the same response as would the predators themselves, and because responses were largely independent of caller arousal, the calls may be regarded as a rudimentary form of representational signalling, in which relatively arbitrary, non-iconic signs are used to designate particular referents external to the signaller. (Seyfarth 1984, pp. 45-46) There are, in fact, reports of distinct alarm calls for different predators or predator classes in other nonhuman species, including ground squirrels, lemurs, and squirrel monkeys (Griffin 1981). All of these observations are relevant to speculation about the origin of referential communication in Homo sapiens. It is plausible that alarm calls such as these were the precursors of words, but it is important to note the differences between these vocalizations and words. An interpretation intended to maximize the contrast between the vervet monkey's alarm calls and human words might posit three different stimulus-response associations, the stimuli being predator classes and the responses the associated calls. It would be pointed out that these calls are mediated by the limbic system of the brain - it is unlikely that these animals could learn to produce these calls in a state of equanimity. Nor could an alarm call be uttered without the immediate presence of the stimulus (except as a mistake). Lastly, the response of the vervets to an alarm call, though different for each type, is invariant and, again, presumably limbically controlled. 8 Words, by contrast, are typically produced without limbic involvement (as far as is known) and are uttered in a dispassionate fashion, are not stimulus bound, and do not evoke a stereotypical behavioral response in the hearer. I am tempted to add that vervets are probably not consciously intending to refer, but this would contradict the earlier decision (chapter 5) not to include such intentionality as a criterion of reference. One implication of the fact that each of these calls occurs only in a certain context and is not syntactically combined with other signals is that there is no basis for determining what, precisely, a call means. The field linguist uses the linguistic and nonlinguistic contexts in which a sound sequence occurs to induce the meaning of the sequence, focusing on those 134

Primate communication in nature elements that are invariant with its usage. While each vervet alarm call does have a consistent environmental correlate - the associated predator class - the behavioral response to each call is an equally consistent concomitant. There is thus no basis for translating a call as "Eagle!" as opposed to, say, "To the bushes!", that is, for deciding whether these vocalizations are designations for animals or exhortations to act in a certain way. In later work, Seyfarth and Cheney have pursued the idea that the incessant grunting (as distinct from alarm calling) of the vervet monkey, though perceptually undifferentiated to humans, takes several forms, each signalling a distinct environmental entity (Seyfarth 1984). They recorded the grunt of a single animal in four different contexts: approaching a dominant animal, approaching a subordinate, watching another animal initiate a group movement into an open area, and spotting another group. They then played each of the sounds back, again from a hidden speaker, to selected individuals. Care was taken to vary the context in which the grunt was heard in order to control for the possibility that different responses by the subject were a function of different contexts rather than different sounds. What they found, in short, was that each type of grunt consistently evoked a different response from the subject, and the response was apt for the message. The subject would look toward the hidden speaker, for example, on hearing the grunt to a subordinate but away from it in response to the grunt for spotting another group. Acoustical analysis revealed several sound components that consistently differentiated the various vocalizations. The signification-signal relationship in vervet grunts would seem less susceptible to a stimulus-response account than in the case of alarm calls. With the latter, the putative referents seem to be classes of species prototypes, while the different types of grunt are associated with fairly abstract situations or at least more complex perceptual gestalts. Gouzoules and Gouzoules carried out equivalent investigations with two other species of monkey, rhesus macaques (Gouzoules, Gouzoules, and Marler 1984) and pigtail macaques (Gouzoules and Gouzoules 1989), focusing on screams emitted in agonistic encounters. Acoustical analyses indicated that such screams took a limited number of discrete acoustical patterns, and distributional analyses revealed that each kind of scream was associated with a particular agonistic situation. For both species the variables defining the agonistic situation were attributes of the opponent, including the opponent's dominance rank relative to the screamer, whether or not there was physical contact involved, and, for the rhesus but not the pigtails, whether or not the opponent was a matrilineal relative. Not surprisingly, the acoustical profile of the scream used in a given 135

Apes and language: phylogeny context was quite different in the two species. For example, when a rhesus monkey suffers contact aggression from a higher-ranking attacker, it issues an atonal scream with a wide bandwidth, whereas a pigtail in the same situation responwith a tonal scream having pronounced frequency modulation. On the assumption that these calls refer to the categories of opponents with which they are associated, Gouzoules, Gouzoules, and Marler (1984) tested the conjecture that the calls functioned to recruit assistance in confrontations by informing the screamer's potential allies about the nature of the opponent. Recordings of four scream classes were made from each of a number of immature rhesus monkeys and then played to its mother from a hidden speaker. It was found that the strength of the mother's response to the scream, measured by probability of looking toward the speaker, how quickly she turned toward it, and how long she looked, varied by scream type. Moreover, the strength of response varied in a manner that might have been predicted on the basis of known patterns of alliance formation in rhesus macaques. Females are more likely to intervene when the victim is a relative, when the victim has been bitten, when the aggressor is dominant to the female, and/or when the aggressor is not related to the female. And, consonant with this pattern, mothers' responses in the playback experiment were strongest to the screams for contact aggression by an opponent of higher rank than the victim (and hence higher rank than the victim's mother) and weakest to those for aggression by a relative of the victim (and hence of the mother). Gouzoules and Gouzoules (1989) found that, as in the case of vervet alarm calls and grunts (Seyfarth and Cheney 1986), the agonistic screams of pigtail macaques are not correctly deployed from their first use; oneand two-year-old infants are responsible for a disproportionate share of calls that are incorrect for the situation, and their correct calls are farther away from the acoustical prototype for the class than those of adults. The developmental improvement in agonistic screaming could result from either learning or endogenous maturational factors, and Gouzoules and Gouzoules, like Seyfarth and Cheney, do not commit themselves to either account. It is undoubtedly significant that the vector of intensities of mothers' responses to screams for various situations maps readily onto the observed gradient of intervention probabilities for the same situations, but Gouzoules, Gouzoules, and Marler's functional explanation of rhesus macaque screams as recruitment signals is problematic. Simply put, it is not at all evident who would benefit from the existence of a set of screams that vary in their likelihood of securing aid in agonistic confrontations. One can easily see the advantage of a signal that increases the probability of 136

Primate communication in nature getting assistance in such encounters, but not of signals that say, in effect, "Come to my assistance with something less than maximal probability/' In other words, it is hard to imagine, under the assumption that recruitment of aid was the adaptive mechanism promoting the natural selection of these signals, how any but the scream with the highest likelihood of recruitment could have been favored by selection. Of course, the validity of Gouzoules, Gouzoules, and Marler's assertion that these signals are referential in nature, or even that they were forged by natural selection, does not stand or fall on the authors' explanation of the adaptive significance of the system. Although the alarm calls of vervets and agonistic screams of macaques should not be equated with words, for reasons presented above, they are an important discovery and suggest the need for a continuum model of signaling, ranging from the more referential to the more affective, rather than a dichotomous contrast between mutually exclusive characteristics. At least for some species, the repertoire of signals should be divided into (1) those, such as vervet alarms and macaque screams, that belong closer to the referential end, (2) those that occur in many contexts, suggesting placement near the affective pole, and (3) those, such as the gorilla belch, that are not readily classified and thus belong, by default, at some intermediate point. DISCRETE VERSUS CONTINUOUS SIGNALS Of the several feature dimensions under discussion, this one is the most multiply connected to the others. The discrete/continuous contrast has to do mainly with the acoustic relationships among the signals in a species7 repertoire, but it is often used to refer also to the signification of signals; the distinction between these two usages is not always clearly drawn in the literature. Discrete signals are acoustically nonadjacent to others, while continuous signals grade into one another along one or more acoustic dimensions, such as volume, frequency, and duration. According to Marler (1977a), primate species are quite variable with respect to the proportion of the signal repertoire made up of discrete as opposed to graded signals. The alarm-call subsystem of the vervet's repertoire consists of calls that are clearly acoustically discontinuous. On the other hand, it is possible that every call in the chimpanzee system fits into an acoustic gradient comprehending all of the signals. Meticulous acoustical analysis is sometimes required to determine whether what appear to be acoustically distinct calls are indeed that or are merely statistical peaks, modal forms, along a continuum of graded signals some of which occur relatively infrequently. 137

Apes and language: phylogeny The signal system of a species could consist of (1) discrete calls, (2) discrete calls with graded variability within each call category, (3) a combination of (1) and (2), or (4) a single continuum with grading both within and between call types. Chimpanzee vocal communication may exemplify the latter possibility, which would mean that the so-called food call, in addition to showing a range of acoustical variability, would be connected to other call categories, all internally variable, through a series of transitional forms. The signals used in language can be said to be continuous in that the combinatory elements out of which they are constructed - the distinctive features - do not, contrary to the idealizing model of contemporary linguistics, actually occur in binary, all-or-none form, but rather in a range of intermediate forms. Thus, an articulated vowel is in reality neither "high" nor "low," "front" nor "back," but instead possesses some degree of each feature; the phonetic realizations of vowels constitute a more or less continuous range of variation (Smith and Wilson 1980). The same is true of the features of consonants, as discussed with regard to categorical perception in chapter 7. At the same time, however, much of this variability in acoustic signal is ignored in language as a system of discrete sound-meaning correspondences. This variability, in other words, is insignificant variation. It can therefore be said that the human communication system combines continuous signals with discrete processing (Marler 1976).9 It is here, in the relationship between the signals and their function, that the discrete/continuous dimension becomes important, for the nature of the signal has implications for the kind of signification it would be expected to bear. It is reasonable to suppose that in a highly graded system the calls would reflect emotional state rather than referring to the environment, since much of nature presents itself as discrete entities rather than continua.10 It is true that there are numerous potential denotata in nature that take the form of continua, such as color, size, temperature, distance, and the like. The honey bee's waggle dance is an example of a continuous signal system that is mapped onto continuous environmental referents (distance and angle). Among monkeys and apes, however, there is no evidence for similar referents. It seems reasonable to suppose also that discrete signals, if they were referential, would refer to discrete entities. It is not that discrete signals cannot be used to refer to continuously distributed properties but that such signals can refer only to nonspecific portions of continua, as in the English terms "red," "heavy," and "early." There are at least two possible interpretations, then, of a graded system. The acoustically intermediate calls connecting the statistically modal 138

Primate communication in nature forms might be functionally equivalent variants of the latter (which may or may not be semantic in nature). This would mean that functionally the system contained afinitenumber of calls. Or the system as a whole might be a complex of acoustical dimensions, each call being composed of some value from each dimension, each value reflecting in analog fashion some component of the caller's disposition. (The values from the various acoustical dimensions might not vary independently; there might be redundancy via correlation among some or all of the dimensions.) Under this interpretation, the system provides for an infinite number of functionally distinct calls. Marler (1977a) has argued that primate species characterized by relatively complex social structure and frequent close-range interaction have predominantly graded systems of vocal signaling, evolved to convey nuances of disposition. In contrast would be those species in which a greater part of the signal repertoire is dedicated to long-distance, interpopulation communications, such as the assertion of territoriality. Such groups would be expected to place a premium on the reduction of signal ambiguity and thus to favor discrete signals. On the basis of the distribution of these characteristics, Marler (1975) considers the phylogenetically primitive primate pattern to consist of discrete signals and discrete processing (that is, ignoring of acoustic variation). The advent of the higher forms saw a shift in emphasis to continuous signals, acoustic variation becoming significant. Humans, by combining continuous signals with discrete processing of them, are something of an aberration, coupling a relatively derived feature with regression to a primitive one. DUALITY OF PATTERNING AND SYNTAX The sentences of language are a product of operations involving at least two levels of syntax. At one level, sentences are composed of morphemes, the smallest meaning-bearing elements. And at the next level down, morphemes are themselves constructed out of phonemes, which are articulatory-feature bundles that have no meaning and serve only to differentiate morphemes. This "duality of patterning" (Hockett 1963) appears to be a property unique to our communication system. An important question is whether other primate species possess even a single level of syntax. For instance, are there systems in which the meaning-bearing units are indivisible but are combined to produce wordor phrase-like structures whose signification is a function, even a simple additive one, of those units? Or are there systems in which elemental units, regardless of whether or not they are independently meaningful, are combined into multiunit constructions whose meanings are not related 139

Apes and language: phylogeny to any that the elemental units may have? A case of the first sort would be roughly equivalent to a language in which sentences were composed of morphemes but the morphemes were not themselves divisible into smaller units. In a language of the second sort, sentences would be composed of either meaningful units (like morphemes) or nonmeaningful ones (like phonemes) but would not be interpretable by reference to the meaning of their component parts. In a trivial sense, all vocal signals are composed of subordinate elements, since any sound can be decomposed into the atomic units that make up the vocabulary of acoustical science. But the atomic units relevant to syntax would be species-specific constellations, equivalent to the phonemes or morphemes of a language, of these most fundamental acoustical elements. Beyond this observation, it can be said that primate signal systems, or at least the discrete signals within them, are overwhelmingly of the sort in which components are variously combined in the formation of structures whose signification is not derivable from those components (Marler 1977b). This is unfortunate in one respect, as an abundance of evidence from contemporary primates of compound signals with meanings systematically related to their components would lend credibility to the interesting hypothesis about the origin of language developed by Hockett and Ascher (1964). The blending hypothesis

Hockett and Ascher took Carpenter's (1940) description of gibbon calls as depicting a typical primate signal system and hence as a plausible account of the vocal-communication system of the ancestor to the hominoids (humans and apes). Accordingly, the aboriginal call system from which language derives would have involved a modest number of inherited calls, each having a consistent environmental correlate. In addition, each call would have been an acoustic gestalt, differing from the others globally rather than componentially. Hockett and Ascher imagined a protohuman confronted with a situation involving both food and danger. Instead of uttering either the food call or the danger call, the animal might have emitted both in sequence or - and this is the more important possibility - might have unwittingly produced a blend of the two calls, a composite made up of acoustic elements from each. If each original call consisted of two such elements, say AB for food and CD for danger, then the blended call might have taken the form AC or BD. If (and this compounding of ifs is plausible only when considered on an evolutionary time scale) such blends were understood 140

Primate communication in nature by the group members as semantic combinations of what the source calls stood for, in this case food plus danger, the practice of inventing new blends might have become established. The process of blending would not only "open up" a closed call system by generating new calls but would also introduce a componential structure to the system. If AC came to mean "food and danger," then the A component would stand for food, and the C for danger. By implication, the original food call, AB, would now mean "food" (A) plus "no danger" (B), while CD would mean "danger" (C) plus "no food" (D). Where before there were only inherited acoustic gestalts, now there were invented words composed of morphemes and requiring learned acquisition. But such a system possesses only a single level of syntax, one in which meaningful but nondecomposable elements combine to form structures whose meanings are a function of those elements. How might the bottom level of language's duality of patterning - phonological structure - have arisen? Hockett and Ascher conjectured that a growing abundance of morphemes, each differing from the others holistically, would have become increasingly unwieldy; many would be so similar that discrimination would be difficult. It would have been fortunate, therefore, if the (by now) hominoids became disposed to analyze the morphemes in terms of the sound elements they comprised. The requisite attentiveness to acoustic detail would have been promoted by the original blending process whereas in a closed call system hearing just the beginning of a call might suffice to infer the rest of it, in an open system, in which calls are composed of variably arranged elements, the listener must attend to all of the several elements of a call in order to identify it. Concomitant with this evolving acoustical analysis of morphemes into their constituent phonemes, articulation would have moved away from production of acoustic gestalts and toward the production of smaller sound features. It is not implausible that the early hominids were capable of analyzing their calls into morphological and phonological regularities. Certainly this capability appeared at some point in our history, for this sort of analysis is central to the child's acquisition of language. What must initially be unanalyzed acoustic bundles are eventually decomposed as the child observes the consistent form-meaning relations that obtain throughout the lexicon. Newport and Supalla (1980) provide dramatic evidence of the child's penchant for analytic decomposition of linguistic input. Ninety percent of the deaf population in the United States are born to hearing parents, which means that, when one of these deaf children learns ASL, he or she is a first-generation signer. These signers tend to acquire the language later in life than do deaf children of deaf parents, and the composition of their lexicon differs in significant respects from that of 141

Apes and language: phytogeny second-generation signers. The signs of first-generation signers show less systematic relatedness within the lexicon, a greater proportion of their signs being unanalyzed wholes. Second-generation signers, who learn their ASL from the former, impose a componential analysis on a linguistic input that is less systematically analytic in structure, regularizing the formation of signs by using recurrent formational components.11 These children thus provide a striking instance of ontogeny recapitulating phylogeny, at least the phylogeny envisioned by Hockett and Ascher. Hockett and Ascher's credible account of the origin of language does not, as mentioned earlier, enjoy corroboration from contemporary studies of primate vocal communication. In particular, their assumption that primate calls are essentially discrete and that each has a referential function is not well founded - the semanticity of certain vervet and macaque vocalizations is notable precisely because these are unusual developments. SUMMARY Surveying the characteristics of naturalistic primate communication yields an impression of systems differing from language in numerous and fundamental ways; few observations suggest continuity with language. Language, in its typical vocal-auditory modality, requires no channel other than the acoustical, whereas nonhuman-primate communication relies on the tactile, visual, and olfactory signal dimensions at least as heavily as on the acoustical. The bulk of a monkey's or ape's repertoire of calls are inborn; their development seems not to require exposure to conspecifics. This contrasts, of course, with the learning required in acquiring the symbols of language. Vocal production in infrahuman species is predominantly a function of the phylogenetically ancient portions of the brain, those presiding over instinctive behavior. Lower primates, in addition, do not readily learn to control their vocalizations; some researchers regard this as essentially beyond the capability of nonhuman primates. In contrast, language clearly involves neuroanatomical regions outside of and more recently evolved than the limbic region and is also a consciously controlled behavior. The signals of language are referential; semanticity in the communication of monkeys or apes seems, with at least two noteworthy exceptions, to be rare, their signals for the most part reflecting affective state. Furthermore, words are in a certain sense psychologically decoupled from their referents, while there is no evidence that lower primates, outside of play contexts, utter a signal in the absence of its referent or the emotional state it reflects. Language signals, though acoustically related to one another in a continuous fashion, are functionally discrete. Those of nonhumans, on the other hand, are largely acoustically 142

Primate communication in nature continuous like ours but appear to be processed in a continuous manner, hearers attending to the affective subtleties encoded by continuous acoustical variation. Finally, language has two levels of syntactic patterning, while monkey and ape calls show at most only one, involving semantically nonproductive compounding of units. CODA: NONBIOLOGICAL ORIGINS OF LANGUAGE Nature is not teeming with obvious precursors of language or elements of language. For this reason, and because it enjoys a certain plausibility in any case, the suggestion that language is both unique to humans and facilitated in its acquisition by neuroanatomical adaptations yet not derived from them should be considered. There are at least two variants of this idea. Language as invention

First, there is the possibility that language was a deliberate invention (Trager 1964; Putnam 1980; Limber 1982; Wells 1987). This might have been the innovation of a particularly thoughtful individual or the outcome of a group effort (although probably not a committee). The tenability of this scenario rests on the degree of intelligence one is willing to attribute to the hominid grade believed responsible for the invention, whether the first stone-tool users three million years ago, more recent species, or even older ones. In Thomas's (1987) unchained rendering, language was the creation of children toying with the primordial call system. I imagine one special early evening, the elders sitting around the fire, grunting monosyllables, pointing at the direction of the next day's hunt or the next field to be slashed, thinking as hard as human beings can think when they are at a permanent loss for words. Then more noise than usual from the children's quarters, interrupting the thought. A rising surf of voices, excited, high-pitched, then louder and louder, exultant, totally incomprehensible to all the adults. Language. It must have been resisted at first, regarded as nonsense. Perhaps resented, even feared, seeing it work so beautifully for communication but only among the children. Magic. Then, later on, accepted as useful magic, parts of it learned by some of the adults from their own children, broken Creole. Words became magical, sentences were miraculous, grammar was sacred, (p. 143)

Thomas's vision is consonant with Jolly's (1985) observation that in nonhuman primate societies it is the young animals who are responsible for most of the important innovations. And, of course, children are irrepressi143

Apes and language: phylogeny bly creative with existing languages, inventing variants with great regularity around the world (Burling 1970). Language as discovery

The second version of the idea that language did not originate in neurological adaptations is a philosophical argument about the ontological status of language. In the so-called Platonist conception of language (Katz 1981, 1984; Bever 1982b), the formal characteristics of language are as they are, not because they are a property of the human brain, as Chomsky contends, but rather because they are "universal abstract objects/' in the same sense that, say, numbers are. The features of language, in other words, are eternal verities. According to this understanding of language, we are the only animal that uses language not because it comes from our brains or because we invented it from scratch, but because we are the only animal intelligent enough to have discovered it, in the same way that we are the only species smart enough to have discovered mathematics. The structure of the essence of language in the child is caused neither by the way it is inherited nor by the way it is learned: It is discovered (like atoms, planets, and America, or logic, geometry, and numbers)... The origin of the essence of language in the species is no longer necessarily ascribed to ... evolutionary causation; rather, it could be the result of the emergence of sufficient complexity (mental or physical) for humans to become susceptible to the relevant forms of language.12 (Bever 1982b, p. 430) Organic "crutches" for the language function

The claim that the appearance of human language was a function of only general intelligence is contradicted but not refuted by the presence in the current species of brain regions such as Broca's area and Wernicke's area, which appear to be dedicated to language. There are at least two accounts of the existence of such features that have language arising in the species independently of them. One possibility is that these organic underpinnings of language evolved after hominids had already come to possess language through either invention or discovery. It is conceivable that a trait conferring substantial fitness value, though acquired initially through individual experience (learning), might come via natural selection to be increasingly a function of the genome and less of the environment. Such a genetic appropriation would be expected to the degree that genetic determination resulted in more rapid or more reliable development of the characteristic. In the abstract, this is a plausible approach to the origin of language144

Primate communication in nature related neuroanatomy, although there does not seem to be a great deal of precedent for the sort of evolutionary dynamic it requires. Limber (1982), in arguing that language was initially an invention but later a genetically determined part of ontogeny, mentions Waddington's notion of genetic assimilation as a possible biological mechanism for this transition. This is an interesting proposition, though not without shortcomings. Waddington (1953) found that exposing fruit-fly (Drosophila) eggs to nonlethal doses of ether resulted in a small number of the adults developing an abnormal morphology known as bithorax. He selected the bithorax specimens as the parents of the next generation. In conjunction with the ether treatment, this resulted in a greater proportion of bithorax types in the second generation. Repeating this process of selective breeding over many generations, he not only progressively increased the proportion of bithorax types but eventually found that a certain number of eggs matured into bithorax adults even without exposure to ether. What had formerly been the result of an environmental determinant had become an obligatory trait. Waddington termed this phenomenon "genetic assimilation of an acquired character/7 a phrase suggesting a Lamarckian mode of evolution. It is clear, however, that what was involved in Waddington's experiment was the artificial selection of individuals already possessing genetic material that was capable, when combined with certain environmental stressors, of producing the distinctive phenotype. Progressively concentrating the genes responsible in each succeeding generation resulted in genotypes for bithorax that were decreasingly dependent on the environmental factor. The fact that genetic variation was being selected rather than created makes this a Darwinian rather than a Lamarckian evolutionary process. The best nonexperimental case of this process is suggested by the presence of calluses on the feet of many newborn mammals. Skin from any part of the body will form a callus when subjected to recurrent friction. It seems plausible that natural selection would have favored those individuals able to callous most rapidly. Apparently, such selection culminated in the capacity, like that in Waddington's Drosophila, to develop the trait even in the absence of the environmental factor, hence the embryonic appearance of calluses in the area certain to undergo friction, the bottoms of the feet. What these two cases have in common is the selective modification of an existing genetically based faculty. It is reasonable to posit the existence of a gene or group of genes that, in response to some extra-somatic stimulus, will deflect development away from a normal thorax or trigger the growth of extra layers of epidermis in stimulated areas. And it is not difficult to imagine that the normal processes of genetic recombination, together with 145

Apes and language: phytogeny the screening effect of selection, could eventually turn out genotypes capable of inducing such outcomes even without the external stimulus. In the case of language, on the other hand, the genetic embodiment of the capacity must be orders of magnitude more complex and thus cannot be taken as pre-existing - rather it is very much part of the matter to be explained. So with respect to language, genetic assimilation could only involve genetic innovations that supplant environmental determinants of the faculty. But the evolutionary process implied in this scenario is no different from what would go on according to the more traditional conception of the evolution of language, in which natural selection promoted both the advent and the subsequent development of language through an evolving neuroanatomical substrate for it. Since it was the implausibility of this gradual evolution of the organic language faculty that made the case for invention worth considering in the first place, the genetic-assimilation version of it, entailing that very process, carries no greater cogency than the traditional account of the origin of language. The second explanation for specific neuroanatomical substrates of language under an invention or discovery account is that these brain areas, though they eventually came to subserve language, originated either before or after the rise of language by virtue of their role in other cognitive or perceptual functions. Language would then have naturally come to reside in these areas because they specialized in the kinds of cognitive processes that make up the language faculty. This scenario, which was presented in more detail in chapter 7, seems the more tenable of the two "nontraditional" models of the origin of the cerebral language areas. A slight variant would have language appearing not through invention or discovery but as a so-called emergent property of the evolution of brain regions selected for nonlinguistic functions.

146

Part five

Conclusion

The chimpanzee and the Chinese room

The argument about language in apes has a striking parallel in the debate among philosophers, psychologists, and computer scientists over the possibility of artificial intelligence. Much of the effort to create intelligent automata is justified by reference to an operational definition of machine intelligence proffered by Alan Turing some 40 years ago (1950). Turing asserted that the question of whether a machine can think is too imprecise and suggested that we assess the intelligence of a machine through an "imitation game." The procedure he proposed was a bit more complex than the current conception of "the Turing test," but the logic is unchanged. In this exercise, humans pose questions, on any subject, to either a person or a computer without knowing which. If the human interrogators are unable consistently to tell from the responses whether they are interacting with a person or the machine, then, according to Turing, we have the best evidence we could have for intelligence in a machine. The most prominent critic of this behavioristic measure of intelligence is the philosopher John Searle (1980), who argues that simply because a machine is able to answer questions as a human would, it is not, ipso facto, engaging in thinking. Thought does not arise in a computer merely because it manipulates symbols; these symbols are devoid of meaning for the computer. Imagine, Searle bids in a now well-known thought experiment, a person ignorant of the Chinese language who is placed inside a room equipped with baskets of Chinese characters. He is also provided with a book of instructions for constructing sequences of characters (identified by shape) in response to sequences presented to him. The instruction book allows him to respond to Chinese sentences in a manner that convinces the questioners, themselves fluent in Chinese, that the occupant of the room also knows the language. But, of course, the occupant has no knowledge of Chinese, since the symbols have no meaning for him. He is, in short, just going through the motions of conversing in 149

Conclusion Chinese, precisely as a computer can only simulate intelligent behavior by operating on symbols according to a program. In describing their aspirations for Washoe, the first of the modern ape-language pupils, the Gardners expressed pessimism about a direct assault on the question "Can an ape talk?" and, as if lifting a page from Turing, adopted instead an unabashedly behavioristic goal: "We wanted to develop behavior that could be called conversation" (1969, p. 665). And critics, anticipating Searle's analogous argument, have maintained that Washoe and her peers, though they may have simulated conversation, acquired neither a human language nor something crucially like one, but rather a system of habits that are crude facsimiles of the features of language. Refuting the claim that apes have the ability to learn a language logically entails proving that they do not have it. This book has not succeeded in doing something that cannot be achieved: proving that something does not exist. The relevant refutable claim, rather, is that one or more of the animals featured in these pages learned a language. Refuting this unequivocally, however, presupposes a set of definitive criteria for language and a demonstration that at least one of them was not met by each of the animals in question. As discussed in chapter 1, such criteria do not yet exist, either for adult forms or for children's forms of language. So it is not possible in principle to show that no ape could learn a language, and it is not possible in practice to show that none has learned a language. One solution, following a recent movement in semantic theory and logic, is to take a "fuzzy" approach to the definition of child language. Instead of conceiving of child language as a discrete and clearly bounded set of exemplars, with a candidate member admitted or not on the basis of some list of necessary attributes, child language can be viewed as a category with fuzzy boundaries - membership in the category is not a yes-or-no matter but one of degree. Entities are members to the degree to which they resemble the prototypical member. Obviously the prototypical member, the type specimen, would have the attributes most common among the languages of normal children. The prototype of the category child language, in other words, is child language. The other members of interest are, of course, the skills attained by the apes in these studies. This has been the approach implicit in the present work. Like the proponents of ape language, I have proceeded undaunted by the lack of a checklist of crucial attributes of language to use in comparing child and ape. Instead, I have simply compared the productions of both in terms of structure and function. Assessing the species with respect to various attributes of prototypical child language distills down to two basic questions about the apes: Do they have words? and Do they have sentences? 150

The chimpanzee and the Chinese room Roughly speaking, a word is a symbol that is arbitrarily linked, through learning, to a concept. A word stands for something the speaker is thinking of and, where there is a listener, elicits in that mind a representation that is, ideally, very similar to what the speaker has in mind. In chapter 5,1 suggested that although the ape's mental representations might be akin to the human's, it is not certain that the ape is capable of using symbols to stand for them. A substantial fraction of the utterances of the signing apes consisted of species-specific gestures rather than learned symbols. As to the remaining gestures in their repertoires, the ones that were undoubtedly learned, it is not clear that for the apes they stood for distinct things or kinds of things. They might instead have been merely instrumental habits deployed on the basis of well-learned contingencies between behaviors and rewards. The rarity in the apes' signing of reference to absent entities makes this a convincing interpretation of their gestures. By their second birthday, typical children have begun to produce sentences, word combinations in which something is predicated of something. Things are assigned to categories ("That dog"), attributes are applied to things ("big dog"), actors and actions are connected ("dog bark"), and so on. These various kinds of predication are said to represent different semantic relations. Each language has rules of sentence formation that specify how a given semantic relation is to be encoded and how to indicate which term plays which role in the relation. Students of language acquisition are not unanimous on this, but certainly the majority would agree that children's earliest combinations observe such constraints, either the conventional ones of the language being acquired or ones of the child's own invention. Perhaps the most salient difference between the utterances of the English-speaking child and those of the signing ape is the presence of a consistent linear ordering of semantic roles in the former and its absence in the latter. In responding to this observation, proponents of ape language asserted that the language of their subjects, American Sign Language, has its own rules for indicating semantic roles and that their subjects observed such rules in their constructions. But, as I argued in chapter 6, (1) American Sign Language does have word-order constraints, (2) the linguistic medium of these projects was not American Sign Language anyway, but a derivative code employing English word order, and (3) the apes used neither word order nor the distinctive syntactic mechanisms of ASL for encoding semantic relations. In light of this, it must be concluded that, whether or not apes can learn to think in terms of semantic relations, they do not linguistically encode them. Assessed in terms of their structure and use, then, the gestural 151

Conclusion compositions of the signing apes reflect an underlying code that has a very low degree of membership in the fuzzy category child language. It might be objected that this description differs little from the claim of the proponents of ape language that their animals attained a linguistic competence cognate with that of children and differing in a merely quantitative way. In fuzzy-set theory, however, anything can be considered a member of any category, albeit in many cases with a vanishingly small degree of membership. So to describe ape language as belonging to the category child language with a low degree of membership is to concede nothing to those claiming that the apes' accomplishments are significantly like those of children. Or the fuzzy-set framework can be discarded, and it can simply be said that what the apes do and what children do are very different. In the last two chapters, I looked to the neuroanatomy of language and to the structure of nonhuman-primate communication for clues to the phylogeny of language. Apes and monkeys do not have obvious nascent versions of the language regions of the human brain, although there are some hints of such developments here and there. Nor do other primates, with a few exceptions, evince language-like principles in their natural systems of communication. Ours, dispensing with qualifications, is cortical, while theirs is limbic; our symbols are learned, their calls are inborn; our language is referential, their communication affective. All of this makes it quite expectable that Lana, Sarah, Washoe, Koko, Nim, and their colleagues would fail to acquire a language despite the formidable efforts of their trainers. What it does not make obvious is how our species came to the point at which every normal human child in every generation succeeds in doing so despite being deprived of such tutelage. The roots of language remain buried, the language gap wide and unexplained. Gallup (1979) exposed a number of chimpanzees to mirrors. Initially, each animal responded to its image in the mirror as if it were another animal, but after three days or so it began to behave in a way suggesting that it took the image to be a reflection of its own body. The animal was then anesthetized, and a spot of odorless and textureless paint was applied to an area of its face outside of its range of vision. On awakening and looking into the mirror, each chimpanzee reached directly to the spot of paint on its body, indicating that it understood the image as a reflection of itself. Here is compelling evidence that our closest relatives share our capacity, if not our propensity, to conceive of a self, to be at once the thinker and the thought-about. No lower primate tested under the same conditions treats the mirror image in a manner suggesting self-consciousness. Interestingly, 152

The chimpanzee and the Chinese room chimpanzees raised alone also did not evince self-consciousness, a striking confirmation of George Herbert Mead's thesis that the self is born of the other's regard. The ape-language experiments confirmed what students dating back at least as far as the gestalt psychologist Wolfgang Kohler have repeatedly demonstrated, which is that apes are highly intelligent creatures, probably second only to us, on measures of human intelligence. We may wonder how the evolutionary process engendered such a powerful mentality in the midst of the African rain forests, asking, like Humphrey (1976), of what use "conditional oddity discrimination" is to an ape in the jungle. But the cognitive prowess of the apes is a fact regardless of our inability to account for it. That the apes, too, are reflexive and capable of impressively abstract mentation suggests that at least modest versions of these faculties arose before the ancestral hominoid lineage diversified into the African apes on the one hand and us on the other. Consider a modest assertion: a capacity for culture requires at least ape-level powers of abstraction. And a case could be made for self-awareness, too, as prerequisite to culture. To the extent that Freud's understanding of humanity's cultural creations as "immortality projects" is sound, an ego is presupposed. If the capacity for language, too, had arisen prior to that last hominoid divergence, then linguistics might have been a branch of comparative psychology, the ape-language experiments would never have been conceived, and this book would have been about something else, say patterns of interspecies marriage. But, for that matter, had language arisen prior to the split that produced them and us - had we all spoken the same language - there might not have been a them and an us.

153

Notes

1 Introduction 1 The exception was Herbert Terrace of the Nim sign-language project, who offered to provide me with such data even without my asking for it. However, since the Terrace group had itself conducted an extensive analysis of its data from a critical perspective similar to the one I intended to employ, I did not pursue Terrace's offer. 2 Considering the stir caused by Terrace's apostasy (he was a student of B. F. Skinner) and the rancor that ensued, it is surprising that anyone familiar enough with the issues to write about them would not know where Terrace stood. Yet a recent item in Scientific American (Horgan 1990) mistakenly reports that Terrace's group claimed to have corroborated the behaviorist account of language learning. 3 The question of whether the linguistic performance of the two-year-old is truly the appropriate comparison for that of the apes, some of whom have been involved in language "training" for well over 10 years, is never raised, at least not by the proponents of ape-language. The reason for this is simple: by any nontrivial criterion, the linguistic abilities of the typical three-year-old child far surpass those of any of the apes. 4 Johnson-Laird (1988, p. 18) relates the story of the behaviorist who, after sex with another behaviorist, says "That was fine for you, but how was it for me?" 5 Lest this appear too preposterous an account of language to be endorsed by any modern researcher, consider Fouts's assertion that Nim, Sarah, and Lana (see chapter 2) did not learn language because their experimenters failed to expose them to the stimuli necessary for language acquisition: "If a behavior does not occur when an organism is deprived of certain stimuli, then those stimuli are obviously very important in the development of that behavior" (1983, p. 64). 2 History of the ape-language projects

1 Ward also cites the philosopher La Mettrie as having suggested in Man a Machine (1912/1748) that an ape might be taught a signed language. A close 154

Notes to pages 12-21 reading of the relevant passage (pp. 100-101), however, indicates that La Mettrie had spoken language in mind. 2 Properly speaking, humans produce phones rather than phonemes. Variation in the acoustical properties of vocalizations is continuous rather than categorical, but in our processing of speech sounds, we treat phones within ranges of variation as equivalent - these functionally equivalent variants are allophones of a single phoneme. In the Premack-Schwartz system, however, the articulatory apparatus itself constrains sound production to a small number of values within each acoustical dimension, which makes the use of the term "phoneme" arguably correct in referring to production. On the other hand, this use of the term ignores the fact that phonemes are not objective elements of speech production but arbitrary impositions on the stream of speech. 3 Whether the medium employed in this and the other signing projects can properly be described as ASL will be taken up in chapter 6. 4 Brown's eight fundamental categories of semantic relationship include agent and action (for example, "you throw"), action and object ("kick ball"), agent and object ("horse hay"), action and locative ("play yard"), entity and locative ("Mommy kitchen"), possessor and possessed ("Daddy hat"), entity and attribute ("cup big"), and demonstrative and entity ("that dog"). 5 In the earlier publications (for example, 1971), Premack refers to the chip sequences as "sentences," but later on (Premack and Premack 1983) uses "sequences" or "constructions," reflecting his increasing conservatism regarding the syntactic abilities of apes. 6 Unfortunately, it also circumvents the entire phonological dimension of human language, which is not merely the physical embodiment of syntax but rather a component of it. And, although Premack contends that his chip language is not intended to incorporate all of the features of a natural language, little justification is given for the omission of phonological (and morphological) structure, that is, syntax at the level of the word. 7 It is worth pointing out something obvious and hence easily lost sight of: Sarah was working not with these English words but rather with plastic tokens that Premack glossed as above (Terrace 1979a). The extent of coincidence of his glosses and Sarah's is, of course, a crucial question in interpreting this work. 8 In a correlational grammar, in other words, so-called selectional constraints on word combinations are the core of the system rather than something applied after the concatenation of semantically neutral syntactic constituents, as in a transformational grammar. 9 Primate Growth and Development - A Gorilla's First Year 10 See note 7 above. 11 In Yerkish the question lexigram is placed at the beginning of a sentence. The noun-"which is"-adjective locution is the only Yerkish provision for modifying a noun with an adjective. 12 Without any systematic effort to compile a list of such materials, I have learned of two major National Geographic articles, one a cover story, on 155

Notes to pages 22-42 Koko. She has been written about in Life, People, and Penthouse magazines. Two books for children, Koko's Kitten (Patterson 1985b) and Koko's Story (Patterson 1987), seem to have enjoyed substantial sales, judging from the prominence accorded them in book stores, and a nontechnical account of the project, The Education of Koko (Patterson and Linden 1981), has been published. At least two films about Koko have appeared, one as part of a National Geographic Society home 'Video/' and she has been profiled on at least two television programs: "60 Minutes/' the CBS program, and "20/20" on ABC. Recently, I received a Chinese publication, in a "Children's Science" series in comic-book format, featuring Koko and Francine Patterson on the cover. 13 The psycholinguist Thomas Bever collaborated with Terrace in preparing the original grant proposal, but his role in the project seems to have ended there save for nominal joint authorship of some publications. An exception is the article coauthored with Terrace in a 1976 New York Academy of Sciences volume (Terrace and Bever 1976). 14 The family was that of Stephanie and William LaFarge. 15 Because Kanzi has become the focus of the Pan paniscus research, I will limit my discussion to his achievements. 16 There is a report of another sign-language project involving an orangutan (Shapiro 1982), but this seems to be the only article to have issued from the work, and the report relates only that the animal acquired some 37 signs after a year and a half of training. 3 The Lana project 1 Trial-and-error is perhaps a more appropriate characterization, as it is unlikely that any behavior in any organism is truly random, that is, literally undetermined. What is intended by random is a contrast with the specifically linguistic constraints on human utterances that make the sorts of lexical and phrasal permutations seen in Lana's utterances nearly unknown in nonpathological language. 4 The Sarah project 1 Warren (1974) presents evidence that learning set is in fact a characteristic of all mammals. Its greater salience in primates disappears, he contends, if experiments are structured in a way that eliminates the advantage of visual acuity, which is a primate forte. 2 A noun phrase is optionally rather than mandatorily expanded into a sentence. Otherwise, obviously, all sentences would be infinitely long. 3 This is not to say that there are no recursive rules in the young child's grammar. The definition of noun phrase in an adequate early grammar, regardless of whether it contained an optional sentence, would have to provide for the appending of an indefinite number of adjectives to a noun, 156

Notes to pages 43-51 as in "big, big, furry, scary, hungry bear/' This is a fairly trivial case of recursion, but recursion nonetheless. 4 Since the propositional content of a sentence is independent of its modality (declarative, interrogative, imperative), the fact that the constructions under discussion here took the imperative form has, in itself, no bearing on the question of whether the strings of Sarah's language truly encode propositions or, alternatively, represent nonsemantic puzzles for her. 5 Recall that in Sarah's chip language, strings were composed on the vertical. 6 Actually, her task may have been somewhat more challenging - the instruction string may have specified that only one of several items was to be removed from its container. It is impossible to determine which was the case from Premack's description: "She was required to remove a specified object from the appropriate container rather than to put them in the container" [my italics] (1976a, p. 329). 5 Words 1 Of the several mysteries in the language acquisition process, none is more puzzling or less explored than the fact that, of the fifteen or so major scholars studying these mysteries, eight are named Elizabeth Bates, Ursula Bellugi, Thomas Bever, Lois Bloom, Melissa Bowerman, Martin Braine, Roger Brown, and Jerome Bruner. 2 This characterization is not contradicted by the studies (for example, Stern 1977) showing that infants are not only capable of but disposed to engage in remarkably synchronized patterns of gazing and vocalizing with the mother. Such behavioral reciprocity need not entail intentionality. 3 The only contrary evidence I am aware of is in a report by SavageRumbaugh et al. (1977) on the use of gestures by male pygmy chimpanzees (Pan paniscus) to direct females into a desired position for mating. SavageRumbaugh also reports (1986) that Kanzi, a pygmy chimpanzee in her studies of symbol usage by apes, spontaneously began to make both indexical (pointing) and iconic gestures at 18 months of age, using an outstretched arm and hand to indicate where he wished to be taken, making twisting motions toward a container to enlist help in unscrewing its top, and so on. 4 Of course, this presumes a solution to the problem of representing with words, whose features (ignoring signed languages) are exclusively acoustical, the properties of their referents, which come in several dimensions. Mapping the features of words to semantic features of referents rather than physical ones seems more plausible, and such correlations do appear sporadically in natural languages, as in the use of reduplication in numerous languages to indicate plurality or intensity, or the cluster of noserelated terms in English beginning with sn-. However, while such a system would contrast with the arbitrary sound-referent relationship characteristic 157

Notes to pages 52-59

5

6

7 8

9

10

of natural languages, it could not properly be described as employing an iconic principle either. American Sign Language (ASL), and possibly other signed languages, incorporates both the arbitrary and the iconic principles in the formation of lexical items. Many signs can be seen as isomorphic representations of the concepts they stand for, yet each ASL sign is made up of several of a small pool of noniconic formational components, equivalent to the phonemes of spoken language. Interestingly, psycholinguistic experiments indicate that the iconic aspect of the whole sign plays no role in the production or recall of signs, and historical studies demonstrate that the language has evolved away from iconicity toward a more thoroughly systemic, nonrepresentational encoding of concepts (Frishberg 1975). Volterra (1983) has shown that the same evolution from sounds lacking stable referential content to verbal symbols occurs also in the gestural realm. In her study, both deaf and hearing children developed referential gestures that supplanted or supplemented the deictic ones. Subsequent development is, of course, rather different in deaf and hearing children. The former go on to combine referential gestures (signs) into propositions, whereas the hearing child's vocabulary of symbolic gestures never attains combinatorial productivity. It may be the case that between the stages of words as verbal gestures and words as referring expressions, in which the uttered word functions to identify some particular object, event, or relationship, there is a period in which words serve only to name, to assert the class to which an entity belongs (Garvey 1984). Bruner (1983) uses the same term, "primitive," in making this same case for the congenital origin of reference, although Macnamara's earlier (1982) work is not cited. Goodall (1968, pp. 350-60) documents the use of the begging gesture among her seminaturalistic study population of chimpanzees at the Gombe Stream Reserve in Tanzania. It might be argued that the inclusion of natural gestures in sign statistics is legitimate to the extent that each such gesture has a consistent signification. Such an interpretation is at least plausible in the case of the begging gesture, but the "hurry" gesture seems to be a typical specimen of the mainly affective significance of animal signals (see chapter 8). The parameter of movement refers to the general trajectory and minor motions that define a sign's execution. Place of articulation has to do with the area of the signing space in which the sign is made. The signing space has a vertical extent from the top of the head down to the waist and ranges horizontally from several inches to the left of the body to the same distance on the right. Hand orientation is the orientation of the palm relative to the signer. In the early tests with Washoe, one of the observers recorded Washoe's signs after looking at the slide screen to determine whether the subject had responded correctly. But the Gardners (1984) note that the high level of 158

Notes to pages 60-65 agreement between the inside and outside observers regarding what Washoe signed indicates that this potential source of experimenter bias was not, in fact, influential. 11 Actually, the Gardners expanded one of the DASL places of articulation, chin/lower face, into two categories, within lips and around lips. After combining the head and side-of-head places into one category and all of the places outside of the head, face, and hand regions into another, however, they had reduced the number of places of articulation to just six: (1) within lips, (2) around lips, (3) midface, (4) upper head/side of head, (5) hand, and (6) periphery. 12 Brackets are used to indicate a phoneme or sequence of phonemes as distinct from a letter or word. 13 By exact place of articulation I mean a class of functionally equivalent variants, in other words a phoneme. 14 In the study of human signers that, I believe, provides the implicit (though uncited) model for the Gardners' error analysis (Bellugi, Klima, and Siple 1975), it was also the case that place of articulation was the most frequently conserved component in errors on short-term sign recall. However, it would be untenable to interpret this as indicating that the human subjects never knew the handshapes and movements of the signs in the first place. 15 In the Bellugi, Klima, and Siple study (1975) of short-term recall of signs, recurrent substitutions did occur. In fact, the experimenters included in their study only those sign substitutions that occurred more than once, in order to minimize any guessing in recall. However, there was no indication in this report of recurrent substitutions having the large relative frequency of those discussed above. Indeed, such results would be very surprising and would cast doubt on the presumption that the study involved welllearned signs. 16 Moja received only one test. Because her performance was adversely affected by what the Gardners describe as her inadequate familiarity with testing procedures and by shortcomings in her testing materials, her results were not included in their analysis of error distribution. 17 It is irrelevant but interesting to note that computer scientists have not succeeded in conferring on the most powerful computers even a fraction of this ability of the pigeon to recognize exemplars of natural kinds. The quest to do so is one of the major research problems in artificial intelligence. 18 An exception to this assertion might be the index-finger point in ASL, which can serve as a demonstrative pronoun, since its articulation entails physical orientation toward the referent. However, this sign can be used to indicate absent entities (see chapter 6), so its status relative to the generalization is not clear. Deictic terms, such as "here," "this," and "those," whose meaning is partly a function of spatial aspects of the context in which they are uttered, are the spoken-language counterparts of the ASL point. And, as with the point, such terms are not limited in meaning or in application to physically present referents. 159

Notes to pages 66-80 19 Actually, the bees' code provides for an infinitude of messages, since there is an infinite number of angles, of distances, and of desirabilities. Strictly speaking, therefore, it is not correct to say that the messages are genetically encoded. Rather, the possible message topics are so constrained. One might say that honey bees can talk about anything they wish to, as long as it's nectar, water, or hive sites. 20 Sugarman (1983) makes the same argument with regard to the claim of Savage-Rumbaugh et al. 1983a (see below) that they implanted referential skills in chimpanzees through lexigram training, suggesting that these animals "might not have the communicative intentions in the first place but only the desire to bring about a given state of affairs" (p. 495). 21 Although these animals were not trained in Lana's Yerkish language, the lexigrams of the Lana project were used in this work. 22 I am counting as requests two utterances that Savage-Rumbaugh et al. categorize otherwise, since, in light of the contextual information provided, and according to the criteria for distinguishing between requests and statements employed with Kanzi and Mulika (Savage-Rumbaugh 1988, p. 291), these should clearly have been classified as requests. The utterances I have so reclassified are Kanzi's "juice" and Mulika's "blackberries" (Savage-Rumbaugh et al. 1986, p. 234). 23 This is to say that Kanzi was not systematically rewarded for connecting lexigrams with things and activities. This is not to say that such connections were acquired by Kanzi without him being exposed to systematic pairings of things and lexigrams beyond what he might have observed in "natural conversation." That Kanzi's likelihood of forming the right associations was intentionally enhanced by the experimenters is suggested by a rather cryptic reference to "lexigram overlays," which Kanzi apparently experienced in abundance: "At night, Kanzi usually asks to watch TV. A number of videotapes, with lexigram overlays, have been prepared that are of interest to the chimpanzees" (Savage-Rumbaugh et al. 1986, p. 216). 6 Sentences

1 Consider these sentences: My dog bit me. I was bitten by my dog. Knowing how to think is useful. The subjects of these three sentences, "my dog," "I," and "knowing how to think," respectively, are classified in a non-case grammar as the subjects of their sentences because, in the terminology of contemporary linguistics, each is the noun phrase directly below the sentence node in the diagram of the sentence's deep structure. And, because they share the same syntactic role, each can be manipulated in the same way by the rules of sentence formation. Yet the semantic roles of these elements are quite disparate. In the 160

Notes to pages 82-109

2

3

4 5

6

7 8

first sentence, it is the subject that executes the action. In the second, the subject is the recipient of the action, and in the third, there is no action. It will be recalled from chapter 2 that Brown (1973) showed the majority of children's two-word combinations from diverse languages to be representatives of a small number (8 to 12) of semantic relations, such as agent-object, agent-action, action-object, possessor-possessed, recurrence, and so forth. Brown's taxonomy of semantic relations has been very influential in child-language studies, and the signing-ape experimenters have used it in analyzing their subjects' utterances. The Gardners, Patterson, and Terrace all succeeded in classifying high percentages of their animals' productions using slight variants of Brown's schema. In arguing against the utility of comparing children and apes by grammatical criteria, Miles (1978) characterizes Bloom's method of rich interpretation as having arisen in response to "the limitations of grammatical analyses" (p. 110) and suggests that it focuses exclusively on semantics rather than syntax. This is a profound misunderstanding of the method, for it is word-order patterning in the child's utterances that practitioners of rich interpretation use to constrain and verify what would otherwise be entirely subjective readings of the semantic structure of those utterances. As mentioned earlier, however, even Turkish children show a disposition, albeit slight, to place the subject first in their sentences. There is a satisfying parallel between ASL and Serbo-Croatian in the encoding of semantic relations. In ASL, SVO order is mandatory for sentences with verbs whose articulation characteristics prevent them from being inflected to indicate subject and object. In Serbo-Croatian, because of historical erosion in the inflectional system, not all inflectional suffixes are phonetically distinct. Thus for some sentences the roles of the nouns are not apparent from morphology, and for these sentences, according to Slobin 1982, an invariant word order (SVO) is the obligatory grammatical solution to the potential ambiguity. The gestures listed as "chase," "pat," and "grab" are possible exceptions. "Come" and "go" might be as well, although they might also be variably glossed manual points. Kanzi certainly seems capable of iconic gestures see chapter 5, note 3. "Who" is the only Wh form included in Gardner and Gardner's 1972 listing of Washoe's expressive vocabulary. These figures are not provided in the articles cited, but have been inferred on the basis of information contained in the articles. 7 Language, evolution, and anatomy

1 In fact, the central point of the King and Wilson study was the striking contrast between, on the one hand, the biochemical similarity of the species and, on the other, the marked anatomical and behavioral differences between them. King and Wilson suggest that these differences are due to changes in the timing or level of activity of regulatory genes, that is, those 161

Notes to pages 111-14 controlling the activity of structural genes, and to the linear rearrangement of strings of genes on the chromosomes. 2 For example, Sarah could correctly respond when shown a bottle threefourths full of water and instructed to match it with either a three-fourths piece or a half piece of apple, placing the three-fourths piece with the bottle. Apparently, she was able to assess the two items as bearing the same relationship to their whole versions. 3 The sensorimotor period is the stage, according to the Piagetian model of development, that lasts from birth through the first year or year and a half of life. This period sees the child's construction of a practical intelligence, including notions of space, time, and causality, as a result of physical interaction with objects and persons. 4 A rejoinder to this well-known claim has been developed in recent years based upon observation of mothers' speech to children. The argument is that when caretakers talk to children, they employ a way of speaking that is attuned to the needs of the language learner. Characteristics of this crosslanguage "Motherese" include careful pronunciation, topics that are clearly related to the situation of discourse, and sentence structures geared in complexity to the child's current stage of linguistic development (Snow and Ferguson 1977). However, other research suggests - convincingly, in my estimation - that the distinctive attributes of child-directed speech could not appreciably assist the child in constructing the grammar of the language she hears (Newport, Gleitman, and Gleitman 1977; Shatz 1982; Wexler and Culicover 1980). For example, Newport et al. found that a substantial proportion of the utterances directed to children are not full sentences but fragments, and that, of the full sentences, most are not the syntactically simplest type, declaratives, but imperatives and questions. 5 Consider, as a concrete example, a collection of three items: a can of soda pop, a fountain pen, and a watch. If asked to guess the rule that had been applied in assembling these items into a set, and thus what additional items would be admissible, one would have no logical basis for choosing from among this Borgesian selection of infinite possibilities: (1) objects found on a desk, (2) objects that are humanly created, (3) objects that either contain liquid or tell time, (4) objects less than 8 inches in height, (5) objects less than 100 feet in height, (6) objects, (7) objects weighing more than .005 ounces, (8) objects that have existed for more than 2 days but less than 307 years. 6 Darwin (1872) presents the logic of the method: "In searching for the gradations through which an organ in any species has been perfected, we ought to look exclusively to its lineal progenitors; but this is scarcely ever possible, and we are forced to look to other species and genera of the same group, that is to the collateral descendants from the same parent-form, in order to see what gradations are possible, and for the chance of some gradations having been transmitted in an unaltered or little altered condition. But the state of the same organ in distinct classes may incidentally throw light on the steps by which it has been perfected" (p. 144). 162

Notes to pages 114-24 7 If, on the other hand, the neurophysiology of language entails little localization but rather complex state changes in the entire brain (Lenneberg 1971), then the prospects for help from other animals are even dimmer. 8 This is due to the contralateral organization of hemispheric authority, the left hemisphere controlling the perceptual input and motor functions of the right side of the body, the right hemisphere responsible for the left side. 9 Hewes envisions the first gestural symbols as mimetic of animal movements or of the characteristic motor patterns of the manufacture or use of various tools. The mechanism by which the vocal mode eventually supplanted the gestural is the weakest link in the theory, involving imitation of hand gestures by the lips, tongue, and mouth. 10 A related hypothesis is that of Lieberman (1984), who sees language arising through recruitment of available left-localized neural mechanisms for complexly structured motor activity. Walking, picking up objects, and chewing food are no less syntactically governed than is talking, he argues, since the nesting of automaticized motor "subroutines" in these activities is crucially similar to the nesting of grammatical constituents in sentences. From this perspective, the emergence of language becomes a less saltatory phenomenon. Lieberman's argument is not without precedent. Fouts (1975), for example, suggested that there are interesting similarities between a chimpanzee's ability to climb any tree and the human ability to create novel sentences. The problem with all such proposals is that they fail to account for the structural properties of language that are not shared with other rulegoverned competencies of our species or of others. Specifically, and to recapitulate a point made in the previous chapter, the categories that figure in grammars of language and, presumably, in the mental processes behind language performance are highly abstract; they can be adequately characterized only in terms of their place in syntactic rules. The constituents of a grammar of walking or object manipulation are quite concrete by comparison. 11 No one could read this sentence without wondering how one-month-old babies indicate how they perceive syllables. The answer involves an experimental design in which the presentation of these sounds is contingent on the rate at which the infant sucks on a pacifier. Infants quickly observe this contingency, and their rate of sucking can thus be used as an index of their interest in hearing a particular stimulus. After a subject becomes habituated to (bored by, roughly) repeated exposure to a certain syllable, say [ba], the experimenter notes any effect on attentiveness (rate of sucking) of presenting the subject with a new syllable, one whose consonant is slightly different in voice-onset time, coming from either (1) the same phonemic category (according to the adult language) as that of the old syllable or (2) across the voice-onset-time division, in the [p] range. 12 Actually, since Snowdon does not indicate whether the marmosets differentiate the calls of different individuals specifically on the basis of the duration component, it is not clear that the animals are not categorically 163

Notes to pages 125-31 perceiving duration. What his study lacks to prove or disprove categorical perception is data on the pattern of his subjects' discriminations of calls, in addition to their "labeling" performance. It is the demonstration that subjects discriminate poorly or not at all among stimuli they have labeled identically but much better or perfectly among those that have been labeled differently that indicates categorical perception. 13 Anywhere he wants to. 14 As pointed out earlier, however, localization of a function does not in itself prove that the brain region involved developed by way of its role in that particular function, or even that that function is a congenital rather than an acquired capacity. 8 Primate communication in nature 1 Jolly (1985) is incorrect in asserting that "the actual forms of language must be learned if it is to be a system of infinite capacity" (p. 425). It is true that a system in which the symbols were inherited would necessarily have a finite number of them, but this in itself would not determine the number of combinations that could be generated. It is the operation of recursive rules in language that provides for an infinite number of combinations (see chapter 4). In principle, a grammar with a single word could generate an infinite number of sentences. 2 Steklis and Raleigh (1979) find merit in Hill's model and attempt to incorporate it into their own account of language origins. Unfortunately, though not inexplicably, given the somewhat bizarre causal mechanism of Hill's scenario, they perfectly misunderstand her contention about the role of dialects, citing her in support of their assertion that dialects would have promoted outbreeding. 3 This is not to say that there are no adaptive gene-frequency differences between contemporary human populations. But the populations showing such differences are generally of a much greater scale, and the distances and thus differences between their environments much more substantial, than those that figure in Hill's model. A possible counterexample is the case of populations adapted to high altitude, manifestly contrasting in physique and physiology with their sea-level counterparts and yet, for purposes of potential "gene flow," effectively adjacent. All evidence, however, indicates that the cluster of high-altitude adaptations represents developmental and physiological responses available to anyone rather than distinctive genetic traits (Frisancho 1979). Malaria-related hemoglobin variation between populations (Livingstone 1967) is a more challenging counterexample, but the environmental distance between these groups is, again, greater than that typifying interacting bands. 4 This error is precisely parallel to the prevalent misunderstanding of genetic mutations as some sort of population-level adaptation to provide new material for natural selection rather than the statistically predictable result 164

Notes to pages 132-44 of, again, an imperfect system of transgenerational replication, in this case of DNA molecules (Williams 1966). 5 This understanding of conditioned behavior as being in opposition to involuntary behavior is an interesting twist on the more conventional conceptual opposition, at least among nonpsychologists, between conditioned behavior and voluntary behavior. 6 On reference, see chapter 5. 7 Altmann (1967) points out that, since the communicating animal's emotional state is defined, at least in part, by the signals it is emitting, it would be fallacious to assert that such signals refer to the corresponding emotional state. This would be to claim that a signal refers to itself. 8 This should not be taken to mean that the call-response complex is fixed from birth. Seyfarth and Cheney (1982) report that young vervets must learn part of the proper evasive behavior for each alarm call and also that they only gradually narrow the category of stimuli for each call down to just the appropriate ones. Initially, for example, an infant will give the flying-predator call to birds in general, including nonraptors, and even to a falling leaf; the snake call is elicited by various long thin items, including inanimate ones. On the other hand, the observation that each call is restricted from the outset to a certain class of perceptually similar items suggests an innate component in the "semantics" of alarm calling. (Seyfarth and Cheney subsequently [1986] became less certain that such developmental changes in vocalizations are the result of learning rather than genetic programming.) 9 Chapter 7 discussed studies suggesting some discrete treatment of continuous signal variability in monkeys. 10 Of course, this observation may simply reflect the limitations of a human worldview rather than the objective state of things - the simian umwelt may be constituted differently. 11 This imposition of grammatical structure by children is reminiscent of the process in which a first-generation trade language, a pidgin, becomes a Creole. Pidgin languages universally lack complex syntax and morphology. A Creole, by contrast, which is the product of acquisition of a pidgin as a native language by children, has inflectional and derivational morphology, relative clauses and other syntactic devices, and the so-called closed-class words, such as prepositions, of mature languages. This structure is apparently institutionalized by the children, who create the Creole as they learn to speak (Sankoff and Laberge 1973; Bickerton 1981). 12 An interesting implication of this position is that a particular human language, or even all human languages, might not be true or correct languages judged from the Platonist standpoint, just as the systems of logic that humans develop may not be complete or consistent though they serve us well enough.

165

References

Aarsleff, Hans. 1976. An Outline of Language-Origins Theory Since the Renaissance. In Origins and Evolution of Language and Speech, 4-13. Annals of the New York Academy of Sciences, Vol. 280, ed. Steven R. Harnad, Horst D. Steklis, and Jane Lancaster. Altmann, Stuart. 1967. The Structure of Primate Social Communication. In Social Communication Among Primates, ed. Stuart A. Altmann, 325-62. Chicago: University of Chicago Press. Atherton, Margaret and Robert Schwartz. 1983. Talk to the Animals. In Language in Primates, 137-45. See Miles 1983. Atkinson, Martin. 1979. Prerequisites for Reference. In Developmental Pragmatics, ed. Elinor Ochs and Bambi Schieffelin, 229-49. New York: Academic. Baker, Charlotte and Carol A. Padden. 1978. Focusing on the Nonmanual Components of American Sign Language. In Understanding Language through Sign Language Research, 27-57. See Siple. Bates, Elizabeth. 1976. Language and Context: The Acquisition of Pragmatics. New York: Academic. 1979. The Emergence of Symbols: Cognition and Communication in Infancy. New York: Academic. Bates, Elizabeth, Luigia Camaioni, and Virginia Volterra. 1979. The Acquisition of Performatives Prior to Speech. In Developmental Pragmatics, ed. Elinor Ochs and Bambi Schieffelin, 111-29. New York: Academic. Bates, Elizabeth and Brian MacWhinney. 1982. Functionalist Approaches to Grammar. In Language Acquisition: The State of the Art, 173-218. See Newport 1982. Beck, Benjamin B. 1980. Animal Tool Behavior. New York: Garland Press. Bellugi, Ursula. 1980. Clues from the Similarities Between Signed and Spoken Language. In Signed and Spoken Language: Biological Constraints on Linguistic Form, ed. Ursula Bellugi and Michael Studdert-Kennedy, 115-40. Weinheim/ Deerfield Beach, Florida/Basel: Verlag Chemie. Bellugi, Ursula, Edward S. Klima, and Patricia Siple. 1975. Remembering in Signs. Cognition 3:93-125. Bever, T. G. 1971. The Integrated Study of Language Behavior. In Biological and Social Factors in Psycholinguistics, ed. John Morton, 158-209. London: Logos Press. 166

References 1982a. Broca and Lashley Were Right: Cerebral Dominance Is an Accident of Growth. In Biological Studies of Mental Processes, ed. David Caplan, 186-230. Cambridge, Mass.: MIT Press. 1982b. Some Implications of the Nonspecific Bases of Language. In Language Acquisition: The State of the Art, 429-49. See Newport 1982. Bickerton, Derek. 1981. Roots of Language. Ann Arbor, Mich.: Karoma. Bloom, Lois. 1970. Language Development: Form and Function in Emerging Grammars. Cambridge, Mass.: MIT Press. Bloom, Lois, Lois Hood, and Patsy Lightbown. 1974. Imitation in Language Development: If, When, and Why. Cognitive Psychology 6:380-420. Bloom, Lois and Margaret Lahey. 1978. Language Development and Language Disorders. New York: John Wiley. Bonvillian, John D., Michael D. Orlansky, and Lesley Lazin Novack. 1983. Early Sign Language Acquisition and Its Relation to Cognitive and Motor Development. In Language in Sign: An International Perspective on Sign Language, ed. J. G. Kyle and B. Woll, 116-25. London: Croom Helm. Bowerman, Melissa. 1973. Early Syntactic Development: A Cross-Linguistic Study with Special Reference to Finnish. Cambridge: Cambridge University Press. Bronowski, Jacob and Ursula Bellugi. 1970. Language, Name, and Concept. Science 168:669-73. Brown, Roger. 1968. The Development of Wh Questions in Child Speech. Journal of Verbal Learning and Behavior 7:279-90. 1970. The First Sentences of Child and Chimpanzee. In Psycholinguistics, ed. R. Brown, 208-31. New York: Free Press. 1973. A First Language: The Early Stages. Cambridge, Mass.: Harvard University Press. Bruner, Jerome. 1983. Child's Talk. New York: Norton. Burling, Robbins. 1970. Man's Many Voices. New York: Holt, Rinehart and Winston. Campbell, Bernard. 1971. The Roots of Language. In Biological and Social Factors in Psycholinguistics, 10-23. See Bever 1971. Carney, A. E., G. P. Widin, and N. F. Viemeister. 1977. Noncategorical Perception of Stop Consonants Differing in VOT. Journal of the Acoustical Society of America 62:961-70. Carpenter, C. R. 1940. A Field Study in Siam of the Behavior and Social Relations of the Gibbon (Hylobates lar). Comparative Psychology Monographs 16:1-212. Chance, Michael R. A. 1967. Attention Structure as the Basis of Primate Rank Orders. Man 2:503-18. Chomsky, Noam. 1957. Syntactic Structures. The Hague: Mouton. 1959. Review of Verbal Behavior. Language 35:26-58. 1965. Aspects of the Theory of Syntax. Cambridge, Mass.: MIT Press. 1979. Human Language and Other Semiotic Systems. Semiotica 25:31-44. Clark, Eve V. 1973. What's in a Word? On the Child's Acquisition of Semantics in His First Language. In Cognitive Development and the Acquisition of Language, ed. Timothy E. Moore, 65-110. New York: Academic. Collins-Ahlgren, Marianne. 1975. Language Development of Two Deaf Children. American Annals of the Deaf 120:524-39.

167

References Cowey, Alan and Lawrence Weiskrantz. 1975. Demonstration of Cross-Modal Matching in Rhesus Monkeys, Macaca mulatta. Neuropsychologia 13:117-20. Cromer, R. 1981. Reconceptualizing Language Acquisition and Cognitive Development. In Early Language: Acquisition and Intervention, ed. Richard L. Schiefelbusch and Diane D. Bricker, 53-102. Baltimore: University Park Press. Cutting, James E. and Burton S. Rosner. 1974. Categories and Boundaries in Speech and Music. Perception and Psychophysics 16:564-70. Darwin, Charles. 1982. The Origin of Species. 6th edn. London: John Murray. Davenport, Richard K. 1977. Cross-Modal Perception: A Basis for Language? In Language Learning by a Chimpanzee, 73-83. See Rumbaugh 1977b. Davenport, Richard K. and C. M. Rogers. 1970. Intermodal Equivalence of Stimuli in Apes. Science 168:279-80. De Laguna, Grace A. 1927. Speech: Its Function and Development. Bloomington: Indiana University Press. Desmond, Adrian J. 1979. The Ape's Reflexion. New York: Dial/James Wade. de Villiers, Jill G. and Peter A. de Villiers. 1978. Language Acquisition. Cambridge, Mass.: Harvard University Press. de Villiers, Peter and Jill de Villiers. 1979. Early Language. Cambridge, Mass.: Harvard University Press. Dewson, J. H. 1977. Preliminary Evidence of Hemispheric Asymmetry of Auditory Function in Monkeys. In Lateralization in the Nervous System, ed. S. Harnad, R. W. Doty, L.Goldstein, J. Jaynes, and G. Krauthamer, 63-74. New York: Academic. Dore, John. 1975. Holophrases, Speech Acts, and Language Universals. Journal of Child Language 2:21^40. Eimas, P. D. 1974. Auditory and Linguistic Processing of Cues for Place of Articulation by Infants. Perception and Psychophysics 16:513-21. Eimas, P. D., E. R. Siqueland, P. Jusczyk, and J. Vigorito. 1971. Speech Perception in Infants. Science 171:303-6. Elliot, R. C. 1977. Cross-Modal Recognition in Three Primates. Neuropsychologia 15:183-86. Entus, Anne K. 1977. Hemispheric Asymmetry in Processing of Dichotically Presented Speech and Nonspeech Stimuli by Infants. In Language Development and Neurological Theory, ed. Sidney J. Segalowitz and Frederic A. Gruber, 64-73. New York: Academic. Epstein, Robert. 1982. "Representation" in the Chimpanzee. Psychological Reports 50:745-46. Epstein, Robert, Robert Lanza, and B. F. Skinner. 1980. Symbolic Communication Between Two Pigeons. Science 207:543-45. Falk, Dean. 1980. Language, Handedness, and Primate Brains: Did the Australopithecines Sign? American Anthropologist 82:72-78. Farrer, D. N. 1967. Picture Memory in the Chimpanzee. Perceptual and Motor Skills 25:305-15. Fillmore, Charles. 1968. The Case for Case. In Universals in Linguistic Theory, ed. Emmon Bach and Robert T. Harms, 1-90. New York: Holt, Rinehart and Winston. 168

References Fischer, Susan D. 1973. The Deaf Child's Acquisition of Verb Inflections in ASL. Paper presented to the Linguistic Society of America Annual Meeting, San Diego, Cal. 1974a. Sign Language and Linguistic Universals. In Actes du Colloque FrancoAllemand de Grammaire Transformationelle, Vol. 2, ed. C. Rohrer and N. Ruwet, 187-204. Tubingen: Niemeyer. 1974b. The Ontogenetic Development of Language. In Language and Language Disturbances: The Fifth Lexington Conference on Pure and Applied Phenomenology, ed. E. W. Strauss, 22-43. Pittsburgh: Duquesne University Press. Fischer, Susan and Bonnie Gough. 1978. Verbs in American Sign Language. Sign Language Studies 18:17-48. Fodor, Jerry. 1975. The Language of Thought. New York: Crowell. Fodor, Jerry A., Thomas G. Bever, and Merrill F. Garrett. 1974. The Psychology of Language: An Introduction to Psycholinguistics and Generative Grammar. New York: McGraw-Hill. Fossey, D. 1972. Vocalizations of the Mountain Gorilla (Gorilla gorilla beringei). Animal Behaviour 20:36-53. Fouts, Roger S. 1972. Use of Guidance in Teaching Sign Language to a Chimpanzee (Pan troglodytes). Journal of Comparative and Physiological Psychology 80:515-22. 1973. Acquisition and Testing of Gestural Signs in Four Young Chimpanzees. Science 180:978-80. 1975. Capacities for Language in Great Apes. In Socioecology and Psychology of Primates, ed. Russell H. Tuttle, 371-90. The Hague: Mouton. 1977. Ameslan in Pan. In Progress in Ape Research, ed. Geoffrey H. Bourne, 117-23. New York: Academic. 1983. Chimpanzee Language and Elephant Tails: A Theoretical Synthesis. In Language in Primates, 63-75. See Miles 1983. Fouts, Roger, Bill Chown, and Larry Goodin. 1976. Transfer of Signed Responses in American Sign Language from Vocal English Stimuli to Physical Object Stimuli by a Chimpanzee (Pan). Learning and Motivation 7:458-75. Fouts, Roger S. and Randall L. Rigby. 1977. Man-Chimpanzee Communication. In How Animals Communicate, 1034-54. See Marler 1977a. Fouts, Roger, Gary Shapiro, and Charity O'Neil. 1978. Studies of Linguistic Behavior in Apes and Children. In Understanding Language through Sign Language Research, 163-85. See Siple. J o u t s , Roger S., Joseph B. Couch, and Charity R. O'Neil. 1979. Strategies for Primate Language Training. In Language Intervention from Ape to Child, 295-323. See Patterson 1979a. Fouts, Roger S., Deborah H. Fouts, and Donna Schoenfeld. 1984. Sign Language Conversational Interaction Between Chimpanzees. Sign Language Studies 42:1-12. Fowler, Anne. 1984. Language Acquisition of Down's Syndrome Children: Production and Comprehension. Unpublished doctoral dissertation, University of Pennsylvania. Frazetta, T. H. 1975. Complex Adaptations in Evolving Populations. Sunderland, Mass.: Sinauer Associates, Inc.

169

References Frisancho, A. Roberto. 1979. Human Adaptation: A Functional Interpretation. St. Louis: C. V. Mosby. Frishberg, Nancy. 1975. Arbitrariness and Iconicity: Historical Changes in American Sign Language. Language 51: 696-719. Fromkin, V. A. and E. S. Klima. 1980. General and Special Properties of Language. In Signed and Spoken Language, 13-28. See Bellugi 1980. Furness, W. H. 1916. Observations on the Mentality of Chimpanzees and OrangUtans. Proceedings of the American Philosophical Society 55:281-90. Gallup, Gordon G., Jr. 1979. Self-Awareness in Primates. American Scientist 67:417-21. Gardner, Beatrix T. 1981. Project Nim: Who Taught Whom? Review of Nim: A Chimpanzee Who Learned Sign Language. Contemporary Psychology 26:425-26. 1982. Comparative Psychology and Language Acquisition. Contemporary Psychology 27:68. Gardner, Beatrice T. and R. Allen Gardner. 1971. Two-Way Communication with an Infant Chimpanzee. In Behavior of Non-Human Primates, Vol. 4, ed. Allan Schrier and Fred Stollnitz, 117-84. New York: Academic. 1974. Comparing the Early Utterances of Child and Chimpanzee. In Minnesota Symposia on Child Psychology, Vol. 8, ed. A. Pick, 3-23. University of Minnesota Press. 1975. Evidence for Sentence Constituents in the Early Utterances of Child and Chimpanzee. Journal of Experimental Psychology: General 104:244-67. 1980. Two Comparative Psychologists Look at Language Acquisition. In Children's Language, 331-69. See Terrace et al. 1980. Gardner, Martin. 1980. Monkey Business. Review of Mm and Speaking of Apes. New York Review of Books March 20:3-5. Gardner, R. Allen and Beatrice T. Gardner. 1969. Teaching Sign Language to a Chimpanzee. Science 165:664-72. 1972. Communication with a Young Chimpanzee: Washoe's Vocabulary. In Modeles Animaux du Comportement Humain, ed. Remy Chauvin, 241-61. Centre National de la Recherche Scientifique, Paris. 1974. Review of A First Language: The Early Stages. American Journal of Psychology 87:729-57. 1975. Early Signs of Language in Child and Chimpanzee. Science 187:752-53. 1978. Comparative Psychology and Language Acquisition. Annals of the New York Academy of Sciences, Vol. 309, 37-76. 1984. A Vocabulary Test for Chimpanzees (Pan troglodytes). Journal of Comparative Psychology 98:381-404. Garvey, Catherine. 1984. Children's Talk. Cambridge, Mass.: Harvard University Press. Garvey, Catherine and Ginger Berninger. 1981. Timing and Turn Taking in Children's Conversations. Discourse Processes 4:27-57. Gautier, J-P. and A. Gautier. 1977. Communication in Old World Monkeys. In How Animals Communicate, 890-964. See Marler 1977a. Geschwind, Norman. 1965. Disconnexion Syndromes in Animals and Man. Brain 88:237-94, 585-644. 170

References Geschwind, N. and W. Levitsky. 1968. Human Brain: Left-Right Asymmetries in Temporal Speech Region. Science 161:186-87. Gill, Timothy V. 1977. Conversations with Lana. In Language Learning by a Chimpanzee, 225-46. See Rumbaugh 1977b. Gill, Timothy V. and Duane M. Rumbaugh. 1977. Training Strategy and Tactics. In Language Learning by a Chimpanzee, 157-62. See Rumbaugh 1977b. Gillan, D. J. 1982. Ascent of Apes. In Animal Mind-Human Mind, ed. D. R. Griffin, 177-200. Berlin, Heidelberg, New York: Springer-Verlag. Glass, S., M. Gazzaniga, and D. Premack. 1973. Artificial Language Training in Global Aphasia. Neuropsychologia 11:95-103. Gleitman, Lila R. 1986. Biological Dispositions to Learn Language. In Language Learning and Concept Acquisition: Foundational Issues, ed. William Demopoulos and Ausonio Marras, 3-28. Norwood, N.J.: Ablex. Gleitman, Lila and Eric Wanner. 1982. Language Acquisition: The State of the State of the Art. In Language Acquisition: The State of the Art, 3—48. See Newport 1982. Goldin-Meadow, Susan. 1982. The Resilience of Recursion: A Study of a Communication System Developed without a Conventional Language Model. In Language Acquisition: The State of the Art, 51-77. See Newport 1982. Goldin-Meadow, Susan and Carolyn Mylander. 1983. Gestural Communication in Deaf Children: Noneffect of Parental Input on Language Development. Science 221:372-74. Goodall, Jane van Lawick. 1968. A Preliminary Report on Expressive Movements and Communication in the Gombe Stream Chimpanzees. In Primates: Studies in Adaptation and Variability, ed. Phyllis C. Jay, 313-74. New York: Holt, Rinehart and Winston. 1971. In the Shadow of Man. Boston: Houghton Mifflin. Gouzoules, Harold and Sarah Gouzoules. 1989. Design Features and Developmental Modification of Pigtail Macaque, Macaca nemestrina, Agonistic Screams. Animal Behaviour 37:383^101. Gouzoules, Sarah, Harold Gouzoules, and Peter Marler. 1984. Rhesus Monkey (Macaca mulatto) Screams: Representational Signalling in the Recruitment of Agonistic Aid. Animal Behaviour 32:182-93. Green, S. 1975a. Dialects in Japanese Monkeys: Vocal Learning and Cultural Transmission of Locale Specific Behavior? Zeitschrift fur Tierpsychologie 38:304-14. 1975b. Variation of Vocal Pattern with Social Situation in the Japanese Monkey (Macaca fuscata): A Field Study. In Primate Behavior, Vol. 4: Developments in Field and Laboratory Research, ed. L. A. Rosenblum, 1-102. New York: Academic. Greenfield, Patricia Marks and Joshua H. Smith. 1976. The Structure of Communication in Early Language Development. New York: Academic. Griffin, Donald R. 1981. The Question of Animal Awareness. Los Altos, Cal: William Kaufmann. Gysens-Gosselin, M. 1979. Concerning Apes, Deaf Children, and Humanists. Communication and Cognition 12:157-69. Hamilton, Charles R. 1977. An Assessment of Hemispheric Specialization in Monkeys. In Evolution and Lateralization of the Brain, 222-32. See Levy. Harlow, H. F. 1949. The Formation of Learning Sets. Psychological Review 56:51-65.

171

References Hayes, Keith J. and Catherine H. Hayes. 1951. The Intellectual Development of a Home-Raised Chimpanzee. Proceedings of the American Philosophical Society 95:105-9. 1954. The Cultural Capacity of Chimpanzees. Human Biology, 26:288-303. Heffner, Henry E. and Rickye S. Heffner. 1984. Temporal Lobe Lesions and Perception of Species-Specific Vocalizations by Macaques. Science 226:75-76. Herman, Louis M., Douglas G. Richards, and James P. Wolz. 1984. Comprehension of Sentences by Bottlenosed Dolphins. Cognition 16:129-219. Herrnstein, R. J., Donald H. Loveland, and Cynthia Cable. 1976. Natural Concepts in Pigeons. Journal of Experimental Psychology: Animal Behavior Processes 2:285-302. Hewes, Gordon W. 1973. Primate Communication and the Gestural Origin of Language. Current Anthropology 14:5-24. Hill, Jane H. 1974. Possible Continuity Theories of Language. Language 50:134-50. 1978. Apes and Language. Annual Review of Anthropology 7:89-112. Hockett, Charles F. 1959. Animal "Languages" and Human Language. In The Evolution of Man's Capacity for Culture, ed. J. N. Spuhler, 32-39. Detroit: Wayne State University Press. 1960. Logical Considerations in the Study of Animal Communication. In Animal Sounds and Communication, ed. W. E. Lanyon and W. N. Tavolga, 392^30. Washington: American Institute of Biological Sciences. 1963. The Problem of Universals in Language. In Universals of Language, ed. Joseph H. Greenberg, 1-22. Cambridge, Mass.: MIT Press. Hockett, Charles F. and Robert Ascher. 1964. The Human Revolution. Current Anthropology 5:135-68. Hockett, Charles F. and Stuart A. Altmann. 1968. A Note on Design Features. In Animal Communication, ed. T. A. Sebeok, 61-72. Bloomington, Ind.: Indiana University Press. Hoffmeister, Robert J. 1977. The Acquisition of American Sign Language by Deaf Children of Deaf Parents: The Development of the Demonstrative Pronouns, Locatives, and Personal Pronouns. Unpublished doctoral dissertation, University of Minnesota, Minneapolis. Hoffmeister, Robert and Ronnie Wilbur. 1980. The Acquisition of Sign Language. In Recent Perspectives on American Sign Language, ed. Harlan Lane and Frangois Grosjean, 61-78. Hillsdale, N.J.: Erlbaum. Holloway, Ralph L. 1968. The Evolution of the Primate Brain: Some Aspects of Quantitative Relationships. Brain Research 7: 121-72. Horgan, John. 1990. Free Radical: A Word (or Two) about Linguist Noam Chomsky. Scientific American 262: 40-44. Hughes, Jennifer. 1974/75. Acquisition of a Non-Vocal 'Language' by Aphasic Children. International Journal of Cognitive Psychology 3: 41-55. Humphrey, N. K. 1976. The Social Function of Intellect. In Growing Points in Ethology, ed. P.P.G. Bateson and R.A. Hinde, 303-17. Cambridge: Cambridge University Press. Ingram, D. 1981. The Transition from Early Symbols to Syntax. In Early Language: Acquisition and Intervention, ed. Richard L. Schiefelbusch and Diane D. Bricker, 261-86. Baltimore: University Park Press. 172

References Jakobson, Roman C, Gunnar M. Fant, and Morris Halle. 1952. Preliminaries to Speech Analysis. 2nd edn. Cambridge, Mass.: MIT Press. Jerison, Harry J. 1973. Evolution of the Brain and Intelligence. New York: Academic. Johnson-Laird, Philip N. 1988. The Computer and the Mind. Cambridge, Mass.: Harvard University Press. Jolly, Alison. 1985. The Evolution of Primate Behavior. 2nd edn. New York: Macmillan. Katz, Jerrold J. 1976. A Hypothesis about the Uniqueness of Natural Language. In Origins and Evolution of Language and Speech, 33-41. See Aarsleff.

1981. Language and Other Abstract Objects. Totowa, N.J.: Rowman and Littlefield. 1984. An Outline of Platonist Grammar. In Talking Minds, ed. Thomas G. Bever, John M. Carroll, and Lance Miller, 17-48. Cambridge, Mass.: MIT Press. Kellogg, Winthrop N. 1968. Communication and Language in the Home-Raised Chimpanzee. Science 162: 423-27. Kellogg, Winthrop N. and Louise A. Kellogg. 1933. The Ape and the Child. New York: McGraw-Hill. Kimura, Doreen. 1961. Cerebral Dominance and the Perception of Verbal Stimuli. Canadian Journal of Psychology 15:166-71.

1979. Neuromotor Mechanisms in the Evolution of Human Communication. In Neurobiology of Social Communication in Primates: An Evolutionary Perspective, ed.

Horst D. Steklis and Michael J. Raleigh, 197-219. New York: Academic. King, Mary-Claire and A. C. Wilson. 1975. Evolution at Two Levels in Humans and Chimpanzees. Science 188: 107-15. Kipling, Rudyard. 1907 (1891). Bertran and Bimi. In Life's Handicap: Being Stories of Mine Own People. Ed. R. Kipling. New York: Doubleday, Page. Klima, Edward S. and Ursula Bellugi. 1972. The Signs of Language in Child and Chimpanzee. In Communication and Affect, ed. T.Alloway, 67-96. New York: Academic. 1979. The Signs of Language. Cambridge, Mass.: Harvard University Press. Kuhl, Patricia K. and James D. Miller. 1975. Speech Perception by the Chinchilla: Voiced-Voiceless Distinction in Alveolar Plosive Consonants. Science 190:69-72. La Mettrie, Julien Offroy de. 1912 (1748). Man a Machine. Ed. G. Bussey. Chicago: Open Court. Lancaster, Jane. 1975. Primate Behavior and the Emergence of Human Culture. N e w

York: Holt, Rinehart and Winston. Lane, Harlan, Penny Boyes-Braem, and Ursula Bellugi. 1976. Preliminaries to a Distinctive Feature Analysis of Handshapes in American Sign Language. Cognitive Psychology 8: 263-89. Latham, R. C. and W. Matthews, eds. 1970 (1661). The Diary of Samuel Pepys. Vol. 2. Berkeley: University of California Press. Leiber, Justin. 1984. The Strange Creature. In The Meaning of Primate Signals, ed. Rom Harre and Vernon Reynolds, 77-88. Cambridge: Cambridge University Press. LeMay, Marjorie and Norman Geschwind. 1975. Hemispheric Differences in the Brains of Great Apes. Brain, Behavior and Evolution 11:48-52. Lenneberg, Eric H. 1967. Biological Foundations of Language. New York: John Wiley. 1971. Of Language Knowledge, Apes, and Brains. Journal of Psycholinguistic Research 1:1-29.

173

References Levy, Jerre. 1977. The Mammalian Brain and the Adaptive Advantage of Cerebral Asymmetry. In Evolution and Lateralization of the Brain, 264-72. Annals of the New York Academy of Sciences, Vol. 299, ed. Stuart J. Dimond and David A. Blizard. Lewontin, R. C. 1990. The Evolution of Cognition. In Thinking: An Invitation to Cognitive Science, Vol. 3, ed. Daniel N. Osherson and Edward E. Smith, 229-46. Cambridge, Mass.: MIT Press. Liberman, A. M., F. S. Cooper, D. P. Shankweiler, and M. Studdert-Kennedy. 1967. Perception of the Speech Code. Psychological Review 74:431-61. Liddell, Scott K. 1978. Nonmanual Signals and Relative Clauses in American Sign Language. In Understanding Language through Sign Language Research, 59-90. See Siple. 1980. American Sign Language Syntax. The Hague: Mouton. Lieberman, Philip. 1975. On the Origins of Language: An Introduction to the Evolution of Human Speech. New York: Macmillan. 1984. The Biology and Evolution of Language. Cambridge, Mass.: Harvard University Press. Lightfoot, David. 1982. The Language Lottery: Toward a Biology of Grammars. Cambridge, Mass.: MIT Press. Limber, John. 1977. Language in Child and Chimp? American Psychologist 32: 280-95. 1982. What Can Chimps Tell Us About the Origin of Language? In Language Development, Vol. 2, ed. Stan Kuczaj, II, 429-69. Hillsdale, N.J.: Erlbaum. Linden, Eugene. 1974. Apes, Men, and Language. New York: Saturday Review Press. 1987. Silent Partners: The Legacy of the Ape Language Experiments. New York: Ballantine. Lisker, L. and A. Abramson. 1964. A Cross-Language Study of Voicing in Initial Stops: Acoustical Measurements. Word 20:384-422. Livingstone, Frank B. 1967. Abnormal Hemoglobins in Human Populations. Chicago: Aldine. MacKain, Kristine S., Catherine T. Best, and Winifred Strange. 1981. Categorical Perception of English /r/ and N by Japanese Bilinguals. Applied Psycholinguistics 2:369-90. Macnamara, John. 1982. Names for Things. Cambridge, Mass.: MIT Press. Marler, Peter. 1975. On the Origin of Speech from Animal Sounds. In The Role of Speech in Language, ed. James F. Kavanagh and James E. Cutting, 11-37. Cambridge, Mass.: MIT Press. 1976. An Ethological Theory of the Origin of Vocal Learning. In Origins and Evolution of Language and Speech, 386-95. See Aarsleff. 1977a. The Evolution of Communication. In How Animals Communicate, ed. Thomas A. Sebeok, 45-70. Bloomington: Indiana University Press. 1977b. The Structure of Animal Communication Sounds. In Recognition of Complex Acoustic Signals, ed. T. H. Bullock, 17-35. Berlin: Dahlem Konferenzem. Marler, Peter and Richard Tenaza. 1977. Signaling Behavior of Apes with Special Reference to Vocalization. In How Animals Communicate, 965-1033. See Marler 1977a.

174

References Marshall, John C. 1971. Can Humans Talk? In Biological and Social Factors in Psycholinguistics, 24-52. See Bever 1971. 1982. On the Biology of Language Acquisition. In Biological Studies of Mental Processes, ed. David Caplan, 106-48. Cambridge, Mass.: MIT Press. Maurus, M. and D. Ploog. 1984. Categorization of Social Signals as Derived from Quantitative Analyses of Communication Processes. In The Meaning of Primate Signals, 226-38. See Leiber. Mayr, Ernst. 1960. The Origin of Evolutionary Novelties. In The Evolution of Life, ed. Sol Tax, 349-80. Chicago: University of Chicago Press. McNeill, David. 1966. The Creation of Language by Children. In Psycholinguistic Papers, ed. J.Lyons and R.Wales, 99-114. Edinburgh: Edinburgh University Press. 1970. The Acquisition of Language: The Study of Developmental Psycholinguistics. New York: Harper and Row. 1971. Explaining Linguistic Universals. In Biological and Social Factors in Psycholinguistics, 53-60. See Bever 1971. 1974. Sentence Structure in Chimpanzee Communication. In The Growth of Competence, ed. Kevin Connolly and Jerome Bruner, 75-94. New York: Academic. Meier, R. 1982. Icons, Analogues, and Morphemes: The Acquisition of Verb Agreement in American Sign Language. Unpublished doctoral dissertation, University of California, San Diego. Menzel, Emil. 1973. Leadership and Communication in Young Chimpanzees. In Precultural Primate Behavior, ed. E. W. Menzel, 192-225. Basel: Karger. 1975. Purposive Behavior as a Basis for Objective Communication Between Chimpanzees. Science 189:652-54. Miles, Lyn W. 1976. The Communicative Competence of Child and Chimpanzee. In Origins and Evolution of Language and Speech, 592-97. See Aarsleff. 1978. Language Acquisition in Apes and Children. In Sign Language and Language Acquisition in Man and Ape: New Dimensions in Comparative Pedolinguistics, ed. Fred C. C. Peng, 103-20. Boulder: Westview Press. 1983. Apes and Language: The Search for Communicative Competence. In Language in Primates: Perspectives and Implications, ed. J. de Luce and H.T. Wilder, 43-61. New York: Springer-Verlag. Myers, Ronald E. 1976. Comparative Neurology of Vocalization and Speech: Proof of a Dichotomy. In Origins and Evolution of Language and Speech, 745-57. See Aarsleff. Neisser, Arden. 1983. The Other Side of Silence. New York: Knopf. Nelson, Katherine. 1987. What's in a Name? Reply to Seidenberg and Petitto. Journal of Experimental Psychology: General 116:293-96. Newman, John D. and David Symmes. 1982. Inheritance and Experience in the Acquisition of Primate Acoustic Behavior. In Primate Communication, 259-78. See Snowdon. Newport, Elissa L. 1982. Task Specificity in Language Learning? Evidence from Speech Perception and American Sign Language. In Language Acquisition: The State of the Art, ed. Eric Wanner and Lila Gleitman, 450-86. Cambridge: Cambridge University Press.

175

References Newport, Elissa L., Henry Gleitman, and Lila R. Gleitman. 1977. Mother, I'd Rather Do It Myself: Some Effects and Non-effects of Maternal Speech Style. In Talking to Children: Language Input and Acquisition, ed. Catherine E. Snow and Charles A. Ferguson, 109-49. Cambridge: Cambridge University Press. Newport, E. L. and T. Supalla. 1980. Clues from the Acquisition of Signed and Spoken Language. In Signed and Spoken Language, 187-211. See Bellugi 1980. Newport, Elissa L. and Richard P. Meier. 1985. The Acquisition of American Sign Language. In The Crosslinguistic Study of Language Acquisition, Vol. 1, ed. Dan Isaac Slobin, 881-938. Hillsdale, N.J.: Erlbaum. Nottebohm, Fernando. 1970. Ontogeny of Bird Song. Science 167:950-56. Nottebohm, F. and M. E. Nottebohm. 1976. Left Hypoglossal Dominance in the Control of Canary and White-Crowned Sparrow Song. Journal of Comparative Physiology: Sensory, Neural, and Behavioral Psychology 108:171-92. Pandya, Deepak N. and Henricus G. J. M. Kuypers. 1969. Cortico-Cortical Connections in the Rhesus Monkey. Brain Research 13:13-36. Parkel, Dorothy A. and S. Tom Smith, Jr. 1979. Application of Computer-Assisted Language Designs. In Language Intervention from Ape to Child, 441-64. See Patterson 1979a. Passingham, Richard E. 1979. Specialization and the Language Areas. In Neurobiology of Social Communication in Primates, 221-56. See Kimura 1979. Passingham R. E. and G. Ettlinger. 1974. A Comparison of Cortical Functions in Man and Other Primates. International Review of Neurobiology 16:233-99. Pastore, R. E., W. A. Ahroon, K. J. Baffuto, C. Friedman, J. S. Puleo, and E. A. Fink. 1977. Common Factor Model of Categorical Perception. Journal of Experimental Psychology: Human Perception and Performance 3:686-96. Pate, James L. and Duane M. Rumbaugh. 1983. The Language-like Behavior of Lana Chimpanzee: Is It Merely Discrimination and Paired-Associate Learning? Animal Learning and Behavior 11:134-38. Patterson, Francine. 1978a. Conversations With a Gorilla. National Geographic 154(4):438-65. 1978b. The Gestures of a Gorilla: Language Acquisition in Another Pongid. Brain and Language 5:72-97. 1978c. Human Communication with Gorillas. In In the Spirit of Enterprise, ed. Gregory B. Stone, 182-86. San Francisco: Freeman. 1979a. Linguistic Capabilities of a Lowland Gorilla. In Language Intervention from Ape to Child, ed. Richard L. Schiefelbusch and John Hollis, 325-56. Baltimore: University Park Press. 1979b. Linguistic Capabilities of a Lowland Gorilla. Unpublished doctoral dissertation, Stanford University. 1980a. Innovative Uses of Language by a Gorilla: A Case Study. In Children's Language, 497-561. See Terrace et al. 1980. 1980b. In Search of Man: Experiments in Primate Communication. Michigan Quarterly Review 19:94-114. 1981. Can an Ape Create a Sentence? Some Affirmative Evidence. Science 211:86-87. 1985a. Direct-mail funding-solicitation letter. Washington: The Gorilla Foundation.

176

References 1985b. Koko's Kitten. New York: Scholastic. 1987. Koko's Story. New York: Scholastic. Patterson, Francine and Eugene Linden. 1981. The Education of Koko. New York: Holt, Rinehart and Winston. Petersen, Michael R. 1982. The Perception of Species-Specific Vocalization by Primates: A Conceptual Framework. In Primate Communication, 171-211. See Snowdon. Petersen, Michael R., M. Beecher, S. Zoloth, D. Moody, and W. Stebbins. 1978. Neural Lateralization of Species-Specific Vocalizations by Japanese Macaques (Macacafuscata). Science 202:324-27. Petitto, Laura A. and Mark S. Seidenberg. 1979. On the Evidence for Linguistic Abilities in Signing Apes. Brain and Language 8:162-83. Piaget, Jean. 1980. The Psychogenesis of Knowledge and Its Epistemological Significance. In Language and Learning: The Debate Between Jean Piaget and Noam Chomsky, ed. Massimo Piattelli-Palmarini, 23-34. Cambridge, Mass.: Harvard University Press. Pisoni, David B. 1977. Identification and Discrimination of the Relevant Onset of Two Component Tones: Implications for the Perception of Voicing in Stops. Journal of the Acoustical Society of America 61:1352-61. Pisoni, David B., Richard N. Aslin, Alan J. Perey, and Beth L. Hennessy. 1982. Some Effects of Laboratory Training on Identification and Discrimination of Voicing Contrasts in Stop Consonants. Journal of Experimental Psychology: Human Perception and Performance 8:297-314. Poizner, Howard and Harlan Lane. 1978. Discrimination of Location in American Sign Language. In Understanding Language through Sign Language Research, 271-87. See Siple. Poizner, Howard, Edward S. Klima, and Ursula Bellugi. 1987. What the Hands Reveal About the Brain. Cambridge, Mass.: MIT Press. Premack, Ann. 1976. Why Chimps Can Read. New York: Harper and Row. Premack, David. 1971. Language in Chimpanzee? Science 172:808-22. 1976a. Intelligence in Ape and Man. Hillsdale, N.J.: Lawrence Erlbaum. 1976b. Language and Intelligence in Ape and Man. American Scientist 64:674-83. 1976c. Mechanisms of Intelligence: Preconditions for Language. In Origins and - Evolution of Language and Speech, 544-61. See Aarsleff. 1983. The Codes of Man and Beasts. Behavioral and Brain Sciences 6:125-67. 1984. Possible General Effects of Language Training on the Chimpanzee. Human Development 27:268-81. 1986. Gavagai! or the Future History of the Animal Language Controversy, Cambridge, Mass.: MIT Press. Premack, David and Arthur Schwartz. 1966. Preparations for Discussing Behaviorism with Chimpanzee. In The Genesis of Language: A Psycholinguistic Approach, ed. Frank Smith and George Miller, 295-335. Cambridge, Mass.: MIT Press. Premack, David and Guy Woodruff. 1978. Does the Chimpanzee Have a Theory of Mind? Behavioral and Brain Sciences 4:515-26. Premack, David and Ann Premack. 1983. The Mind of an Ape. New York: Norton.

177

References Punzi, L. and A. Kratt. 1986. The Effect of Context on Preschool Children's Understanding of Synthetic Speech - A Pilot Study. Unpublished manuscript. Putnam, Hilary. 1980. What is Innate and Why: Comments on the Debate. In Language and Learning: The Debate Between Jean Piaget and Noam Chomsky, ed. Massimo Piattelli-Palmarini, 287-309. Cambridge, Mass.: Harvard University Press. Redshaw, Margaret. 1978. Cognitive Development in Humans and Gorilla Infants. Journal of Human Evolution 7:133-41. Ristau, Carolyn A. and Donald Robbins. 1979. A Threat to Man's Uniqueness?: Language and Communication in the Chimpanzee. Journal of Psycholinguistic Research 8:267-300. 1982. Cognitive Aspects of Ape Language Experiments. In Animal Mind-Human Mind, ed. D. R. Griffin, 299-330. Berlin, Heidelberg, New York: SpringerVerlag. Robinson, Bryan W. 1976. Limbic Influences on Human Speech. In Origins and Evolution of Language and Speech, 761-71. See Aarsleff. Rosch, E. and C. B. Mervis. 1975. Family Resemblances: Studies in the Internal Structure of Categories. Cognitive Psychology 7:573-605. Rumbaugh, Duane M. 1977a. Language Behavior of Apes. In Behavioral Primatology, ed. Allan M. Schrier, 105-38. Hillsdale, N.J.: Lawrence Erlbaum. Rumbaugh, Duane M., ed. 1977b. Language Learning by a Chimpanzee: The LANA Project. New York: Academic. Rumbaugh, Duane M. and Timothy V. Gill. 1976. The Mastery of Language-type Skills by the Chimpanzee (Pan). In Origins and Evolution of Language and Speech, 562-78. See Aarsleff. 1977. Lana's Acquisition of Linguistic Skills. In Language Learning by a Chimpanzee, 165-92. See Rumbaugh 1977b. Sanders, Richard J. 1985. Teaching Apes to Ape Language: Explaining the Imitative and Nonimitative Signing of a Chimpanzee (Pan troglodytes). Journal of Comparative Psychology 99:197-210. Sankoff, G. and S. Laberge. 1973. On the Acquisition of Native Speakers by a Language. Kivung 6:32-47. Savage-Rumbaugh, E. Sue. 1979. Symbolic Communication - Its Origins and Early Development in the Chimpanzee. New Directions for Child Development 3:1-15. 1981. Can Apes Use Symbols to Represent Their World? In The Clever Hans Phenomenon: Communication with Horses, Whales, Apes, and People, 35-59. Annals of the New York Academy of Sciences, Vol. 364, ed. Thomas A. Sebeok and Robert Rosenthal. 1984. Pan paniscus and Pan troglodytes: Contrasts in Preverbal Communicative Competence. In The Pygmy Chimpanzee: Evolutionary Biology and Behavior, ed. R. L. Susman, 395-413. New York: Plenum. 1986. Ape Language: From Conditioned Response to Symbol. New York: Columbia University Press. 1987. Communication, Symbolic Communication, and Language: Reply to Seidenberg and Petitto. Journal of Experimental Psychology: General 116:288-92. 1988. A New Look at Ape Language: Comprehension of Vocal Speech and 178

References Syntax. In Comparative Perspectives in Modern Psychology (Nebraska Symposium on Motivation 1987), ed. Daniel W. Leger, 201-55. Lincoln, Nebraska: University of Nebraska Press. Savage-Rumbaugh, E. Sue, Beverly J. Wilkerson, and Roger Bakeman. 1977. Spontaneous Gestural Communication among Conspecifics in the Pygmy Chimpanzee (Pan paniscus). In Progress in Ape Research, ed. Geoffrey H. Bourne, 97-116. New York: Academic. Savage-Rumbaugh, E. Sue and Duane M. Rumbaugh. 1978. Symbolization, Language, and Chimpanzees: A Theoretical Reevaluation Based on Initial Language Acquisition Processes in Four Young Pan troglodytes. Brain and Language 6:265-300. 1980a. Language Analogue Project, Phase II: Theory and Tactics. In Children's Language, 267-307. See Terrace et al. 1980. 1980b. Requisites of Symbolic Communication - Or, Are Words for Birds? The Psychological Record 30:305-15. 1980c. A Response to Herbert Terrace's Paper, "Linguistic Apes''. The Psychological Record 30:315-18. 1982. Ape Language Research is Alive and Well: A Reply. Anthropos 77:568-73. Savage-Rumbaugh, E. Sue, D. M. Rumbaugh, and Sarah Boysen. 1978a. Linguistically Mediated Tool Use and Exchange by Chimpanzees (Pan troglodytes). Behavioral and Brain Sciences 1:539-54. 1978b. Symbolic Communication Between Two Chimpanzees (Pan troglodytes). Science 201:641-44. 1980. Do Apes Use Language? American Scientist 68:49-61. Savage-Rumbaugh, E. Sue, D. M. Rumbaugh, S. T. Smith, and J. Lawson. 1980. Reference: The Linguistic Essential. Science 210:922-24. Savage-Rumbaugh, E. Sue, James L. Pate, Janet Lawson, S. Tom Smith, and Steven Rosenbaum. 1983a. Can a Chimpanzee Make a Statement? Journal of Experimental Psychology: General 112:457-92. Savage-Rumbaugh, E. Sue, Mary Ann Romski, Rose Sevcik, and James L. Pate. 1983b. Assessing Symbol Usage Versus Symbol Competency. Journal of Experimental Psychology: General 112:508-12. Savage-Rumbaugh, Sue, Kelly McDonald, Rose A. Sevcik, William D. Hopkins, and Elizabeth Rubert. 1986. Spontaneous Symbol Acquisition and Communicative Use by Pygmy Chimpanzees (Pan paniscus). Journal of Experimental Psychology: General 115:211-35. Scaife, Michael and Jerome Bruner. 1975. The Capacity for Joint Visual Attention in the Infant. Nature 253:265-66. Schlesinger, I. M. 1981. Semantic Assimilation in the Development of Relational Categories. In The Child's Construction of Language, ed. W. Deutsch, 223-43. New York: Academic. Searle, John. 1969. Speech Acts. Cambridge: Cambridge University Press. 1980. Minds, Brains, and Programs. Behavioral and Brain Sciences 3:417-24. Seidenberg,Mark S. 1986. Evidence from Great Apes Concerning the Biological Bases of Language. In Language Learning and Concept Acquisition: Foundational Jssues, ed. William Demopoulos and Ausonio Marras, 29-53. Norwood, N.J.: Ablex.

179

References Seidenberg, Mark S. and Laura A. Petitto. 1979. Signing Behavior in Apes: A Critical Review. Cognition 7:177-215. 1987. Communication, Symbolic Communication, and Language: Comment on Savage-Rumbaugh, McDonald, Sevcik, Hopkins, and Rupert (1986). Journal of Experimental Psychology: General 116:279-287. Seyfarth, Robert M. 1984. What the Vocalizations of Monkeys Mean to Humans and What They Mean to the Monkeys Themselves. In The Meaning of Primate Signals, 43-56. See Leiber. Seyfarth, Robert M. and Dorothy L. Cheney. 1982. How Monkeys See the World: A Review of Recent Research on East African Vervet Monkeys. In Primate Communication, 239-52. See Snowdon. 1986. Vocal Development in Vervet Monkeys. Animal Behaviour 34:1640-58. Seyfarth, R. M., D. L. Cheney, and P. Marler. 1980. Monkey Responses to Three Different Alarm Calls: Evidence of Predator Classification and Semantic Communication. Science 210:801-3. Shapiro, Gary L. 1982. Sign Acquisition in a Home-Reared/Free-Ranging Orangutan: Comparisons with Other Signing Apes. American Journal of Primatology 3:121-29. Shatz, Marilyn. 1982. On Mechanisms of Language Acquisition: Can Features of the Communicative Environment Account for Development? In Language Acquisition: The State of the Art, 102-27. See Newport 1982. Siple, Patricia. 1978. Linguistic and Psychological Properties of American Sign Language: An Overview. In Understanding Language through Sign Language Research, ed. Patricia Siple, 3-23. New York: Academic. Skinner, B. F. 1956. A Case History in Scientific Method. American Psychologist 11:221-33. 1957. Verbal Behavior. New York: Appleton-Century-Crofts. Slobin, Dan Isaac. 1979. Psycholinguistics. 2nd edn. Glenview, 111.: Scott, Foresman. 1982. Universal and Particular in the Acquisition of Language. In Language Acquisition: The State of the Art, 128-70. See Newport 1982. Slobin, Dan I. and Thomas G. Bever. 1982. Children Use Canonical Sentence Schemas: A Crosslinguistic Study of Word Order and Inflections. Cognition 12:229-65. Smith, Neil and Deirdre Wilson. 1980. Modern Linguistics: The Results of Chomsky's Revolution. Bloomington, Ind.: Indiana University Press. Smolak, Linda. 1982. Cognitive Precursors of Receptive vs. Expressive Language. Journal of Child Language 9:13-22. Snow, Catherine E. and Charles A. Ferguson. 1977. Talking to Children: Language Input and Acquisition. Cambridge: Cambridge University Press. Snowdon, Charles T. 1982. Linguistic and Psycholinguistic Approaches to Primate Communication. In Primate Communication, ed. Charles T. Snowdon, Charles H. Brown, and Michael R. Petersen, 212-38. Cambridge: Cambridge University Press. Snowdon, C. T. and Y. V. Pola. 1978. Interspecific and Intraspecific Responses to Synthesized Pygmy Marmoset Vocalizations. Animal Behaviour 26:192-206. Steklis, Horst D. and Michael J. Raleigh. 1979. Requisites for Language: Interspecific

180

References and Evolutionary Aspects. In Neurobiology of Social Communication in Primates, 283-314. See Kimura 1979. Stern, Daniel. 1977. The First Relationship: Infant and Mother. Cambridge, Mass.: Harvard University Press. Stokoe, William C, Jr., Dorothy C. Casterline, and Charles G. Croneberg. 1976. A Dictionary of American Sign Language on Linguistic Principles. 2nd edn. Silver Spring, Md.: Linkstok Press. Straub, R. O., M. S. Seidenberg, T. G. Bever, and H. S. Terrace. 1979. Serial Learning in the Pigeon. Journal of Experimental Analysis of Behavior 32:137-48. Struhsaker, T. T. 1967. Auditory Communication among Vervet Monkeys (Cercopithecus aethiops). In Social Communication Among Primates, ed. Stuart A. Altmann, 281-324. Chicago: University of Chicago Press. Studdert-Kennedy, Michael. 1980. Speech Perception. Language and Speech 23:45-65. 1981. The Beginnings of Speech. In Behavioral Development, ed. Klaus Immelmann, George W. Barlow, Lewis Petrinovich, and Mary Main, 533-61. Cambridge: Cambridge University Press. Sugarman, Susan. 1983. Why Talk? Comment on Savage-Rumbaugh et al. Journal of Experimental Psychology: General 112:493-97. Sutton, Dwight, Herman H. Samson, and Charles R. Lawson. 1978. Brain Mechanisms in Learned Phonation of Macaca mulatta. In Recent Advances in Primatology, Vol. 1, ed. D.J. Chivers and J.Herbert, 769-84. New York: Academic. Terrace, Herbert S. 1979a. Is Problem-Solving Language? Journal of Experimental Analysis of Behavior 31:161-75. 1979b. Mm. New York: Knopf. 1981. A Report to an Academy, 1980. In The Clever Hans Phenomenon: Communication with Horses, Whales, Apes, and People, 94-114. Annals of the New York Academy of Sciences, Vol. 364, ed. Thomas A. Sebeok and Robert Rosenthal. 1984. "Language" in Apes. In The Meaning of Primate Signals, 179-203. See Leiber. 1985. In the Beginning Was the Name. American Psychologist 40:1011-28. Terrace, H. S. and T. G. Bever. 1976. What Might Be Learned from Studying Language in the Chimpanzee? The Importance of Symbolizing Oneself. In Origins and Evolution of Language and Speech, 579-88. See Aarsleff. Terrace, H. S., L. A. Petitto, R. J. Sanders, and T. G. Bever. 1979. Can an Ape Create a Sentence? Science 206:891-902. 1980. On the Grammatical Capacity of Apes. In Children's Language, Vol. 2., ed. Keith Nelson, 371-^95. New York: Gardner Press. 1981. Ape Language. Science 211:87-88. Thomas, Lewis. 1987. A Long Line of Cells. In Inventing the Truth: The Art and Craft of Memoir, ed. William Zinsser, 125-48. Boston: Houghton Mifflin. Thompson, Claudia R. and Russell M. Church. 1980. An Explanation of the Language of a Chimpanzee. Science 208:313-14. Trager, George L. 1964. Comment on "The Human Revolution". Current Anthropology 5:163-64. Turing, A. M. 1950. Computing Machinery and Intelligence. Mind 59:433-60. Umiker-Sebeok, Jean and Thomas A. Sebeok. 1980. More on Monkey Talk.

181

References Response to Patterson's rejoinder to Martin Gardner's review of Mm and Speaking of Apes. New York Review of Books December 4:59-60. 1981. Clever Hans and Smart Simians: The Self-Fulfilling Prophecy and Kindred Methodological Pitfalls. Anthropos 76:89-165. 1982. Rejoinder to the Rumbaughs. Anthropos 77:574-78. Valian, Virginia. 1977. Talk, Talk, Talk: A Selective Critical Review of Theories of Speech Production. In Discourse Production and Comprehension, ed. Roy O. Freedle, 107-39. Norwood, N.J.: Ablex. 1986. Syntactic Categories in the Speech of Young Children. Developmental Psychology 22:562-79. Van Cantfort, T. E. and J. B. Rimpau. 1982. Sign Language Studies with Children and Chimpanzees. Sign Language Studies 34:15-72. Volterra, Virginia. 1983. Gestures, Signs and Words at Two Years, Or When Does Communication Become Language? In Language in Sign: An International Perspective on Sign Language, ed. J. G. Kyle and B. Woll, 109-15. London: Croom Helm. von Glasersfeld, Ernst. 1977. The Yerkish Language and its Automatic Parser. In Language Learning by a Chimpanzee, 91-130. See Rumbaugh 1977b. Wada, Juhn A. 1977. Pre-Language and Fundamental Asymmetry of the Infant Brain. In Evolution and Lateralization of the Brain, 370-79. See Levy. Waddington, C. H. 1953. Genetic Assimilation of an Acquired Character. Evolution 7:118-26. Ward, Eric F. 1983. Teaching Sign Language to a Chimpanzee: Some Historical References. Journal of Experimental Analysis of Behavior 40:341^2. Warren, J. M. 1974. Possibly Unique Characteristics of Learning by Primates. Journal of Human Evolution 3:445-54. Wells, G. A. 1987. The Origin of Language: Aspects of the Discussion from Condillac to Wundt. Peru, 111.: Open Court. Wexler, K. and P. Culicover. 1980. Formal Principles of Language Acquisition. Cambridge, Mass.: MIT Press. Wilbur, Ronnie B. 1987. American Sign Language: Linguistic and Applied Dimensions. 2nd edn. Boston: Little, Brown. Williams, George C. 1966. Adaptation and Natural Selection. Princeton: Princeton University Press. Wind, Jan. 1976. Phylogeny of the Human Vocal Tract. In Origins and Evolution of Language and Speech, 612-30. See Aarsleff. Winter, P., P. Handley, D. Ploog, and D. Schott. 1973. Ontogeny of Squirrel Monkey Calls Under Normal Conditions and Under Acoustical Isolation. Behaviour 47:230-39. Witelson, Sandra F. 1977a. Anatomic Asymmetry in the Temporal Lobes: Its Documentation, Phylogenesis, and Relationship to Functional Asymmetry. In Evolution and Lateralization of the Brain, 328-54. See Levy. 1977b. Early Hemisphere Specialization and Interhemisphere Plasticity: An Empirical and Theoretical Review. In Language Development and Neurological Theory, ed. S. Segalowitz and F. Gruber, 213-87. New York: Academic. Witmer, Lightner. 1909. A Monkey with a Mind. The Psychological Clinic 3:179-205.

182

References Yamaguchi, S. and R. E. Myers. 1972. Failure of Discriminative Vocal Conditioning in Rhesus Monkey. Brain Research 37:109-14. Yeni-Komshian, Grace H. and Dennis A. Benson. 1976. Anatomical Study of Cerebral Asymmetry in the Temporal Lobe of Humans, Chimpanzees, and Rhesus Monkeys. Science 192:387-89. Yerkes, Robert M. 1925. Almost Human. New York: Century.

183

Index of Names

AH, 100-2,106 Altmann, S., 165n.7 Ascher, R., 140^2 Austin, 57, 67-74

Elliot, R. C, 117 Epstein, R., 70, 73-74 Ettlinger, G., 115

Bates, E., 49, 51, 66, 80 Bellugi, U., 116,159n.l5 Benson, D. A., 118,119 Berninger, G., 87 Best, C. T., 124 Bever, T. G., 82, 89,120,156n.l3 Bloom, L., 49, 85, 87, 161n.3 Bowerman, M., 82 Boysen, S., 36 Brown, R., 15, 42, 45, 85, 89, 98-99,155n.4, 161n.2 Bruner, J., 49, 50,158n.7 Cable, C, 64-65 Camaioni, L., 51 Carpenter, C. R., 140 Carter, J., 97-98 Casterline, D. C, 56 Chantek, 28, 82,110 Cheney, D. L., 133-35,136,137,165n.8 Chomsky, N., 9, 22,112,144 Church, R. M, 32-33, 35 Collins-Ahlgren, M, 93 Cowey, A., 117 Croneberg, C. G., 56 Cutting, J. E., 125 Dar, 55, 59, 62, 63,110 Darwin, C, 6,145,162n.6 Davenport, R. K., 117 Desmond, A. J., 117 Dewson, J. H., 118 Diogenes, 106 Eimas, P. D., 122

Falk, D., 119,121 Farrer, D. N., 39 Fillmore, C, 80 Fischer, S. D., 93 Fodor, J. A., 52-53 Fossey, D., 133 Fouts, R. S., 14,15, 55, 65, 84, 97,100,101, 106,154n.5,163n.lO Freud, S., 153 Furness, W. H., 10 Gallup, G. G., 152 Gardner, B. T., 5, 9,10,13-15, 16, 21, 23, 24, 27, 53-55, 57, 59-64, 73,11, 82, 83, 84, 85, 86, 87, 88-89, 90, 92, 93-94, 98-99,106,110,150,158n.lO, 159nn.ll, 14,16,161nn.2, 7 Gardner, R. A., See Gardner, B. T. Garvey, C, 87 Gautier, A., 133 Gautier, J-P., 133 Geschwind, N., 116-18,119 Gill, T. V., 34-36 Gleitman, L. R., 82 Goldin-Meadow, S., 82 Goodall, J., 158n.8 Gouzoules, H., 135-37 Gouzoules, S., 135-37 Green, S., 133 Greenfield, P. M, 51 Gua, 10-11 Hamilton, C. R., 118 Harlow, H. F., 40 Hayes, C. H., 10-11 Hayes, K. J., 10-11

184

Index of Names Heffner, H. E., 119 Heffner, R. S., 119 Herrnstein, R.J., 64-65 Hewes, G. W., 120-21,163n.9 Hill, J. H., 130-31,164nn.2, 3 Hockett, C. F., 6,16,140-42 Hoffmeister, R. J., 93 Holloway, R. L., 115 Hood, L., 88 Humphrey, N. K., 153 Jakobson, R. C, 12 Johnson-Laird, P. N., 154n.4 Jolly, A., 143,164n.l Kanzi, 24-27, 57-59, 65, 74-76, 77_78, 95-96,102-4,105,106, 156n.l5,157n.3,160nn.22, 23, 161n.6 Katz, J. J., 6 Kellogg, L. A., 10-11 Kellogg, W. N., 10-11 Kimura, D., 116 King, M-C, 109,161n.l Kipling, R., 11 Klima, E. S., 116,159n.l5 Kohler, W., 153 Koko, 5, 20-22, 24, 55, 62, 82-83, 84-85, 87,94,110, 111, 152,155n.l2 Kuhl, P. K., 125 Kuypers, H., 117 LaFarge, S., 156n.l4 LaFarge, W., 156n.l4 Lahey, M., 49 Lamarck, J. B., 145 La Mettrie, J. O., 154n.l Lana, 18-20, 27, 31-37, 38, 45, 55, 67, 72, 102,106,152,154n.5,156n.l Lancaster, J., 129 Lane, K, 60-61 Lanza, R., 70 Lawson, J., 132 Leiber, J., 56-57 LeMay, M., 117-18,119 Lenneberg, E. H., 115 Liberman, A. M, 121 Lieberman, P., 123,163n.lO Lightbown, P., 88 Lightfoot, D., 7,113 Limber, J., 6, 7,145 Linden, E., 6, 90, 92, 97-98 Linguistic Society of Paris, 114 Loulis, 55 Loveland, D. H., 64-65 Lucy, 84, 97-98

MacKain, K. S., 124 MacNamara, J., 52-53,158n.7 MacWhinney, B., 80 Marler, P., 122,129,133-34,135-37,139 Marshall, J. C, 6 Matata, 24-25, 27 Maurus, M., 119 McNeill, D., 80 Mead, G. H., 153 Meier, R., 93 Menzel, E., 129 Michael, 22, 55 Miles, L. W., 28, 82,100,110,161n.3 Miller, J. D., 125 Moja, 54-55, 59, 85-86,110,159n.l6 Mulika, 25, 27, 57, 74-75, 76,160n.22 Myers, R. E., 119,132 Newman, J. D., 129 Newport, E. L., 93,124-25,141-42,162n.4 Nim, 5, 20, 22-24, 74, 82, 83, 86-89, 95, 110,152,154n.5 Nottebohm, F., 119,130-31 Pandya, D. N., 117 Passingham, R. E., 115 Pastore, R. E., 125 Pate, J. L., 32-33, 35 Patterson, F., 5, 20-22, 23, 24, 55, 62, 82-83, 84-85, 87, 88, 90, 92, 94, 111, 155n.l2,161n.2 Pavlov, I., 8 Pepys, S., 11 Petersen, M. R., 118-19,123 Petitto, L. A., 66, 75, 83 Piaget, J., 51, 52, 111, 112,162n.3 Pili, 28, 55 Pisoni, D. B., 125 Plato, 106 Ploog, D., 119 Poizner, H., 60-61,116 Pola, Y. V., 123 Premack, A., 15 Premack, D., 6,12-13,15-17,18,19, 20, 37, 38^6, 50, 111, 112,155nn.2-7,157n.6 Raleigh, M. J., 164n.2 Redshaw, M., I l l Rigby, R. L., 65, 84, 97 Ristau, C. A., 33 Robbins, D., 33 Robinson, B. W., 119 Rogers, C. M., 117 Rosner, B. S., 125 Rumbaugh, D. M., 18-20, 27, 31-37, 46, 50, 67, 70-71

185

Index of Names Samson, H. H., 132 Sanders, R. J., 87-88 Sarah, 15-17, 36, 37, 38-46, 55, 102, 111, 152,154n.5,155n.7,157nn.4, 5, 162n.2 Savage-Rumbaugh, E. S., 6, 24-27, 36, 37, 46, 50, 53, 57-59, 66, 67-76, 77-78, 95, 102-3,106,157n.3,160nn.20, 22 Scaife, M., 50 Schlesinger, I. M., 80 Schwartz, A., 12-13,155n.2 Searle, J., 149-50 Sebeok, T. A., 70-71 Seidenberg, M. S., 66, 75, 83,100 Seyfarth, R. ML, 133-35,136,137,165n.8 Sherman, 57, 67-74 Siple, P., 159n.l5 Skinner, B. F., 8-9, 70,154n.2 Slobin, D. I., 82, 89,161n.5 Smith, J. H., 51 Snowdon, C. T., 123-124,163n.l2 Steklis, H. D., 164n.2 Stokoe, W. C, 56 Strange, W., 124 Straub, R. O., 33 Struhsaker, T. T., 133 Studdert-Kennedy, M., 125 Sugarman, S., 160n.20 Supalla, T., 124-25,141-42 Sutton, D., 132 Symmes, D., 129

Tatu, 55, 59, 62, 63,110 Tenaza, R., 129 Terrace, H. S., 5, 22-24, 38, 39, 66, 82, 83, 86-88, 89,154nn.l, 2,156n.l3,161n.2 Thomas, L., 143 Thompson, C. R., 32-33, 35 Turing, A. M, 149-50 Umiker-Sebeok, J., 70-71 Valian, V., 80 Viki, 10-11 Volterra, V., 51, 158n.5 von Glasersfeld, E., 18 Waddington, C. H., 145 Wanner, E., 82 Ward, E. F., 11,154n.l Warren, J. M, 156n.l Washoe, 5,13-15, 20, 21, 22, 23, 27, 36, 53-54, 55, 57, 59, 62, 63, 65, 82, 83, 84, 87, 89, 90, 93, 96, 98-100,106,110,150, 152,158n.lO Weiskrantz, L., 117 Wilbur, R. B., 93 Wilson, A. C, 109, 161n.l Witmer, L., 10 Yamaguchi, S., 132 Yeni-Komshian, G. H., 118,119 Yerkes, R. M, 11

186

Index of Subjects

adaptation. See evolution affect, in communication, 11, 42,128, 132-39,142-43,152,158n.8,165n.7

A First Language, 42, 89

alarm calls, of monkeys, 133-37 American Sign Language. See ASL Ameslan. See ASL anecdotes, 34-36, 96 angular gyrus, 116-18,126 aphasia, and Sarah's communication system, 45-46 arbitrariness, of linguistic symbols, 50-51 artificial intelligence, 149-50,159n.l7 ASL acquisition of, by deaf, 93,106,141-^2 categorical perception in, 124-25 and functional lateralization, 116 grammar of, 81, 83, 90-93,105,151, 161n.5 Koko's training in, 20-21 Nim's training in, 23 pointing in, 55 in post-Washoe project of Gardners, 27 structure of signs, 36, 54, 55-56,157n.4 psychological reality of, for apes, 36, 56-57, 60-64, 77 Washoe's training in, 13-14, 27 word order in, 81-82, 86, 91, 93,105, 151,161n.5 associations in behaviorism, 8 cross-modal, 116-18,126 experiments on, 117 in Kanzi project, 160n.23 in language, 109 in Sherman and Austin's training, 71, 73 in Washoe's responses to Wh questions, 99-100 asymmetries, cerebral. See functional lateralization

attention, place of, in language, 50, 51, 52, 77 behaviorism, 7-9, 37,109,154nn.2, 4 and artificial intelligence, 149-50 and Washoe project, 150 bird song, 119-20,126,130-31 blending hypothesis, 140-42 brain. See also neurology size and language, 115,126 Broca's area, 144 calls, primate. See vocalization case grammar, 80 categorical perception, 58,121-25,126, 163n.l2. See also speech, perception of cerebral dominance. See functional lateralization channels of communication, 128-29,142 Chinese-room debate, 149-50 cognition, 81,110-12, 146 color coding, of Lana's lexigrams, 33, 37 combinations of children, 49, 80-81, 85,151,161n.2 comprehension of, by Kanzi, 27,102-4, 106 of Kanzi, 74, 95-96,105 of Nim, 86-88 novel (see novel combinations) of Sherman and Austin, 68 of Washoe, 89 communication, natural systems of nonlinguistic, 6, 41^2,128-43, 152 complementation, grammatical, 41 complex sentences, 17, 41-45 computer analysis of Nim's utterances, 86-87 chess-playing, 111 (see also artificial intelligence) in Lana's training, 18-19, 20

187

Index of Subjects concepts, 49, 50, 64-65, 77,151 conditional discrimination in ape-language projects, 81 in Lana project, 32 in Washoe's responses to Wh questions, 100 conditional relationship, in Sarah's training, 17, 42, 43-45 constituents, grammatical, 79, 80 in child language, 80, 96, 98-99 in Washoe's utterances, 98-100 coordination, grammatical, 41 Creoles, 143,165n.ll cross-modal association. See associations cueing, 10,19, 26-27, 59, 71,102,158n.lO culture, 5, 49,153 Darwinian perspective, 6,145-46 DASL, 56,60-61,159n.ll deaf acquisition of ASL by, 93, 106, 141-42 invention of language by, 82 North American community, 13, 54, 141-42 demands, 66-67. See also instrumentality; requests design features of language, 6 dialects, 130-31,164n.2 Dictionary of American Sign Language. See DASL

discontinuities between child language and adult language, 7, 80 between language and natural primate communication, 128-29,142-43 discovery of language, 144,146,165n.l2 discrete vs. continuous signals, 41-42,132, 137-39,142-43 displaced reference by apes, 65, 69, 76, 77,151 in child language, 52, 65, 66 by honey bees, 65-66,133,138 in natural primate communication, 134, 142 in Sarah's training, 17 distinctive features, 58 in ASL signs, 56, 57 and discrete vs. continuous signals, 138 and Jakobson's model of phonology, 12 dolphins, 110 dominance, cerebral. See functional lateralization duality of patterning, evolution of, 139-42 ecology, and communication systems, 131,139

emotion. See affect endogamy, and dialects, 130-31,164n.2 errors on Gardners' vocabulary tests, analysis of, 57, 60-64, 73,159nn.l4,16 Lana's, frequency of, 35 in Savage-Rumbaugh's labeling task, analysis of, 73-74 evolution and hominid adaptive niche, 131, 164n.3 and homology versus analogy, 110,120 of language, 113-15,126-27,152,162n.6, 163n.lO and categorical perception, 123,125 Hewes's model of, 120-21,126,163n.9 Hill's model of, 130-31,164nn.2-4 Hockett and Ascher's model of, 140-42 invention models, 144-46 of new traits, 114,126 and species differences, 8,109-10 formational components of ASL signs, 36, 55-56, 57, 60-62,124-25, 157n.4 functional lateralization, 115-16,118-21, 125,126,131,163n.8 fuzzy sets, 150,151-52 generalization of sign application by apes, 65 by children, 64, 65 by Koko, 22 by Nim, 23 by Washoe, 14-15 genetic assimilation, 145—16 genetics and ape-human relationship, 109, and evolution of language, 144-46 and hominid evolution, 131,164n.3 and microevolution of songbirds, 130-31 gestures of children, 158n.5 and evolution of language, 120-21,126, 163n.9 in Kanzi's productions, 95, 96, 105, 161n.6 natural, of apes, 10-11, 53-55, 77,151, 157n.3,158n.8 Gorilla Foundation, 21, 22 graded signals, 41-42,132,137-39,142-43 grammar. See also syntax; semantic relations as component of language, 6, 79

188

Index of Subjects invention of language, 143-46 IQ testing, of Koko, 21-22, 111

types of, 18, 80-81,155n.8,160n.l grammatical constituents. See constituents, grammatical grammatical structure in apes' utterances, 5, 87,109 in child language, 7, 93, 96,105-6, 111, 161n.3,165n.ll in Kanzi's productions, 95 in Lana's combinations, 20, 32-33 in Nim's utterances, 5, 86-88

joystick articulation device, 12-13,155n.2

hand orientation, in ASL signs, 56 handedness, 119,120-21,126 handshape, in ASL signs, 56, 62,124 hemispheres, cerebral. See functional lateralization hemoglobin adaptations, 164n.3 high-altitude adaptations, 164n.3 hominids, 115,121,127,130-31,141,143,144 hominoids, 5,109,121,140,153 homology, in evolution, 110,114,120,125 honey bees, 65-66,133,138,160n.l9 iconicity, of linguistic symbols, 50-51, 157n.4 imitation ape talent for, 11 by apes, 87-88,105 in children's language acquisition, 9, 88 as training technique, 9,14,16, 21, 23 induction, in language acquisition, 113, 162n.5 inflection in apes' utterances, 93-94,105 in ASL, 90, 91-93 in children's ASL utterances, 93 in spoken languages, 81, 82, 89, 91, 94, 105 inheritance of acquired characters, 145 innate capacity, 7, 9, 50, 52,122,142. See also learning language as, 112-13,129-30 reference as, 52-53 vocalization as, 129-30,132,136,142, 152,165n.8 Institute for Primate Studies, 15, 24,100 instrumentality, of apes' language-like behavior, 66-67, 77, 81,151,160n.20. See also requests and Kanzi's productions, 74-76 and Nim's utterances, 88 intelligence, 109-12,153 artificial, 149-50 Intelligence in Ape and Man, 38

intention, and communication, 49, 50, 134,157n.2. See also volition

Lamarckian evolution vs. Darwinian evolution, 145^6 language criteria of, 6,150 evolution of (see evolution, of language) as innate capacity, 112-13,129-30 as uniquely human attribute, 5,152 Language Research Center, 25 lateralization. See functional lateralization learning and cognitive development, 52-53 and dialects, 130-31 and language acquisition, 109-13, 129-31,142,144,152,164n.l and primate vocalization, 129,131-32, 136,165n.8 and reference, 52-53, 77 learning set, 40,156n.l lexigrams in Kanzi's training, 160n.23 Lana's representation of, 36 in Lana's training, 18-20 in Sherman and Austin's training, 160n.21 structure of, 18, 36 limbic system and language, 11-12,142 and primate vocalization, 11-12,116, 119,133,134,142,152 meaning. See semanticity; reference mean length of utterance, 22, 24 modulation. See inflection molding, 9,14, 21, 23 monkeys and affective signalling, 133-37 agonistic screams of, 135-37 alarm calls of, 133-37 and angular gyms, 116 cross-modal associations in, 117-18 functional lateralization in, 118-19,121, 126 vocalization of, 129,132-37 and volitional control over vocalization, 132 morphemes, 139-40 in ASL, 91 in protolanguage of hominids, 141-42 in spoken language, 90 Motherese, 162n.4

189

Index of Subjects grammar of, 82,165n.ll place of articulation, in ASL signs, 56, 60-62,158n.9,159nn.ll, 14 Platonist view of language, 144, 165n.l2 playback experiments on primate vocalization, 133-37 pointing by apes, 50,157n.3 in ASL, 159n.l8 poverty-of-the-stimulus argument, 113. See also learning, and language acquisition; nativist view of language; innate capacity, language as preadaptations, 115,126

movement, in ASL signs, 56, 62,158n.9 nativist view of language, 112-13,126, 127. See also innate capacity, language as; learning, and language acquisition natural selection and dialects, 130-31 and genetic vs. acquired traits, 145-46 and human infant vocalization, 49 and language, 110,127,144-46 and mutations, 164n.4 and neurology, 114,144-46 and vocalization, 49,137 neurology, and language, 11-12,109-12, 114-21,126-27,142^6,152,163n.7, 164n.l4 novel combinations, 96-97, 106 by Koko, 22 by Lana, 34-36, 37 by Lucy, 97 by Washoe, 15, 96 number of trainers, 88-89, 110 orangutans, in language training, 10, 28, 156n.l6 order, word. See word order Other Side of Silence, The, 53-54

overinterpretation by ape-language researchers, 31, 54, 55, 84 in Kanzi project, 102-3,104 in Lana project, 31, 35-36 in Sarah project, 38-39 in Washoe project, 94,100 paired-associate relations, 66 in Lana project, 32, 37 in Sherman and Austin's training, 70 in Washoe project, 64 Pan paniscus, contrasted with other species, 27, 57, 95 performatives, in child language, 7, 51 phonemes, 139-40,155n.2 ASL equivalents of, 56 and categorical perception, 121-25 and joystick articulation device, 12-13, 155n.2 in Kanzi's word discrimination, 58-59, 77 in protolanguage of hominids, 141 Piagetian psychology, 51, 52, 111, 112, 162n.3 pidgin languages in ape-language projects, 23, 28, 82-83, 105,151

recursive rules, 41-42, 79,156n.3, 164n.l reference, 52, 64, 66, 70, 77,134,152, 157n.4. See also semanticity; symbols; words in child language, 51, 52-53, 77,158n.5 in honey bees, 65-66 as innate capacity, 52-53, 77,158n.7 in Kanzi project, 25, 65, 74-76, 77-78 and language-trained apes, 10,17, 65, 66-67, 76,109,151 in natural primate communication, 77, 117,132-39,140,142 in Savage-Rumbaugh's projects, 20, 27, 67-78,160n.20 reference points, in ASL, 90-91, 92, 94 reinforcement, 8,12, 70 in Nim project, 88 relativization, grammatical, 41, 91 requests. See also instrumentality in apes' utterances, 66-67 in children's utterances, 74, 75 in Kanzi's productions, 25, 74-76, 95, 160n.22 in Nim's utterances, 88 in Sherman and Austin's training, 68 retardation, and language, 46,111-12,115 rich interpretation, 85, 87,161n.3 screams, agonistic, of monkeys, 135-37 selection, natural, and language. See natural selection, and language self-consciousness, 152-53 semantic features, in storage and retrieval, 56, 57,157n.4 semanticity. See also reference; symbols; words of Lana's lexigrams, 31, 33, 37, 45 of primate signals, 132-40,142 of Sarah's chips, 39, 43, 45 understanding of, by children, 52-53

190

Index of Subjects semantic relations in apes' utterances, 81, 90,151,161n.2 Brown's taxonomy of, 15, 155n.4 in children's utterances, 42, 82, 85, 106, 151,161n.2 grammatical encoding of, 81, 89,105 in Kanzi's productions, 95-96 in Koko's utterances, 22 in Nim's utterances, 23-24, 86-88 in Washoe's utterances, 15, 82 semantics, and cognition, 81 sensorimotor period, 112,162n.3 Serbo-Croatian, 161n.5 shaping, 9,14,16, 43 sign order. See word order simultaneous communication, in Koko project, 21, 82-83 songbirds, 119-20,126, 130-31 speech anatomical basis of, 12 comprehension of, by Kanzi, 26-27, 57-59, 77

comprehension of, by Koko, 21 perception of, 58,118,121-25,127 (see also categorical perception) speech acts, in child language, 7, 51 stimulus generalization, 64 stimulus-response psychology. See behaviorism stimulus-response relations in alarm calling, 134-35 in behaviorism, 8 in Gardners' projects, 64 in Lana project, 34 stock sentences, in Lana's combinations, 19-20, 32, 35, 37 symbols, linguistic, 50-51. See also reference; semanticity; words syntax. See also grammar defined, 79,105 development of, in children, 49 function of, 81 in natural primate communication, 139-43 and phonology, 56-57,140-42,155n.6 and transformational relationships, 45, 46 tools and evolution of language, 120-21, 126 as not unique to humans, 5 topicalization, 91 transfer tests, in Sarah's training, 39-40, 43, 44, 45 Turing test, 149

utterance length, 22, 24 Verbal Behavior, 9

videotaping, 21, 23-24, 87 visual channel, 128-29 vocabulary testing in Kanzi project, 26-27 in Koko project, 22 in Washoe project, 14, 57, 59-64, 73,100 vocal-auditory channel, 128-29,142 vocalization. See also reference; speech; volition as innate capacity, 129-30,132,136,142, 152,165n.8 of Kanzi, 75 of monkeys, 129,132-37 and natural selection, 49,137 perception of, 118-19,123-24,126 (see also speech perception) of preverbal children, 49-50 volition, and control over vocalization, 116-17,131-33,165n.5 Wernicke's area, 116,118,144 Wh questions, 98-100,106,161n.7 word order in Air's utterances, 100-2,106 in apes' utterances, 81-82, 84, 90,100, 105,151 in ASL (see ASL, word order in) in children's utterances, 81-82, 89, 93, 106,151,161nn.3, 4 contrasted with semantic-role order, 83-84,105 in invented languages, 82 in Koko's utterances, 83, 84-85 in Lana's constructions, 33-34 in Lucy's utterances, 84 in Moja's utterances, 85-86 in Nim's utterances, 83-84 in spoken languages, 81, 89, 90, 94,105 in tests of Kanzi's comprehension, 103-4 in utterances of apes' trainers, 82-83 in Washoe's utterances, 82, 83, 84, 89 words. See also reference; semanticity; symbols contrasted with primate signals, 134, 137 development of, in children, 49-53, 65, 77,112,151,158n.6 as media for concepts, 49, 50, 65,151 Yerkes Regional Primate Center, 18, 25, 67 Yerkish, 18, 32, 35, 36,46,155n.ll

191