page: Greenfield on Language, Tools and Brain Philip Swann Faculty of Psychology and Education, University of Geneva, Route de Drize 9, 1227 Carouge Switzerland swann@divsun.unige.ch 1. Introduction Greenfield (1991) is an ambitious attempt to provide a psychobiological account of some developmental patterns identified from observations of young children's object manipulation and early speech. The paper reflects a new interdisciplinary approach to the study of the development and historical origins of human language, an approach which is attracting a wide range of researchers: this is evident in recent important conference transactions, notably Wind (1991) and Gibson & Ingold (1993), as well as several books (Bickerton 1990, Corballis 1992, Lieberman 1991). Greenfield exploits a great variety of previous research and her commentators draw on numerous additional publications: the result is a dense and difficult text in which details of data analysis tend to obscure the radical ideas. Indeed, in the conclusion to her response, Greenfield notes that in the commentaries "data-based challenges [are] especially significant (as compared to theoretical or logical ones)" (p. 588). In this continuing commentary, I shall try to redress the balance by focusing on theory and logic. The rest of this section contains a brief overview of Greenfield's main claims and the evidence she presents in their support, while in subsequent sections I evaluate in detail the key parts of her argument. In earlier work, Greenfield et al. (1972) had investigated the development of complexity in young children's manual combination of various objects, especially sets of nesting cups. This led them to claim that there is a universal developmental sequence for this behavior that is determined by an innate mechanism. They also noted the analogy between combining objects and combining spoken words and speculated that this might in fact be a homology; i.e. the result of a common neural mechanism programming both modalities. A related study by Greenfield and Schneider (1977) introduced the task of copying tree structures with wooden tongue depressors to study older children's developing mastery of hierarchical structure. The same task was used by Grossman (1980) to show that adult Broca's aphasics may also be impaired in their ability to reconstruct or copy tree structure: he suggested that there was a common supramodal hierarchical processor for both speech syntax and manual objection combination tasks. It was Grossman's work that apparently inspired Greenfield to look for more neurological evidence for homology. She found various studies that have located motor sequencing for right manual and speech control close together in or around Broca's area (Brodmann's areas 44 and 45), but also one study (Curtiss et al. 1979) which showed a dissociation between the two skills. Using as a guide Deacon's account of neural circuit development in monkeys (Deacon 1988, 1992), Greenfield proposes that Broca's area starts out in human infants as an undifferentiated hierarchical programmer providing output via two circuits to the left hemisphere orofacial and manual motor regions. The resulting language, limited to 2-3 word combinations with minimal morphosyntax, would correspond to what Bickerton (1990) has called protolanguage. During late in the second year of life neural circuits develop that connect Broca's area forward to two distinct prefrontal cortex areas, one specialized for grammar (Brodmann 46) and the other for manual sequencing (Brodmann 9). In the central part of her target article, Greenfield seeks behavioral and neurobiological evidence for this developmental model. On the one hand, she tries to establish precise structural parallels between the development of children's object manipulation, exemplified by nested cups (Greenfield et al. 1972) and spoons (Connolly & Dalgleish 1989), and their early speech (her own unpublished diary data). On the other hand, she uses two studies of brain development in children (Thatcher 1991 and unpublished data, Simonds & Scheibel 1989) as evidence for the growth of the proposed circuits. The preceeding account of individual development (ontogenesis) is, Greenfield argues, compatible with the following scenario for the evolutionary emergence (phylogenesis) of language in hominids. Tool use and gestural protolanguage originated in the nut-cracking behavior of a common ancestor of humans and other primates. This behavior would be transmitted from mother to infants by explicit pedagogy, as has been observed to happen in wild chimpanzees (Boesch & Boesch 1990). Both the tool use and the gestures would have been supported by a bimodal hierarchal motor control system located in a homologue to Broca's area. In the hominid line, the tool use, pedagogy and gestural language would have then developed futher through "mutually reinforced natural selection" (p. 549) and have pulled human brain evolution in the direction of an expanded prefrontal cortex. And this new cortex would contain the complex grammar module that distinguish human language from that of chimpanzees, who have remained at the pre-hominid evolutionary stage which corresponds to the protolanguage stage in human children. We would thus expect to find cross-species structural equivalences, and we would predict that chimpanzees would not be able to progress beyond protolanguage. While this speculative and nativist theory of language origin occupies a good deal of space in the target article and commentaries, Greenfield has no real evidence by which it could be supported or falsified. As Lewontin (1990) points out, short of time-travel we will probably never know the combination of factors that led to the emergence of higher cognitive functions in humans: the information we would need just isn't there anymore. I will therefore limit my attention to Greenfield's account of ontogeny, since this is at least in principle susceptible to empirical investigation. Like many other contributors to BBS, Greenfield is deeply influenced by mainstream cognitive science in the Fodor-Chomsky tradition, while making extensive use of data and ideas from disciplines that speak a very different scientific language. This leads to persistent problems of interpretation. Taken literally, her claims amount to an absurdly strong form of nativism according which the development of children's speech and object combination is as genetically constrained as is the growth of their teeth. If, on the other hand, her claims are taken as a temporary conceptual framework used to integrate work from many disciplines, or as a sketch of what a general theory might one day look like, then the paper is a more modest but certainly more useful contribution. The problem is that Greenfield adheres strongly to a literal interpretation, while her commentators take her more or less literally according to the point they want to make. For example, Bloom writes that Greenfield "presents no real evidence for her position, there are several studies that refute it, and even if she were right, this sort of 'developmental homology' would have few implications for a theory of the evolution of language". In contrast Gibson believes that "Greenfield's formulation provides a major breakthrough in our methods of approaching the evolution of the human brain." In most of what follows I shall take Greenfield literally and confirm Bloom's negative reaction to the specific proposal she makes. 2. Hierarchically Organized Sequential Behaviour in Children The description of sequential behavior as hierarchically organized has proved a fruitful tool in the analysis of natural and artificial systems. This led Dawkins (1976) to follow Simon (1969) and propose "hierarchical organisation" as a general principle for ethology, and some researchers (e.g. Fentress 1981) have tried to operationalize the suggestion. Many linguists have followed Chomsky (1956) and taken hierarchical organisation as the defining characteristic of natural language syntax. But there are problems with elevating the intuitive concept to the status of general principle and problems with its use in specific empirical research. On the one hand hierarchical organisation can be found in (or perhaps read into) a very wide range of phenomena. On the other hand, there are nearly always other organising principles at work which have to be taken into account. And this has led to most empirical researchers to pay only lip-service to the concept or use it as a temporary place-holder. In computer science, for example, hierarchical structures have played much less of a role than one would have expected after reading Simon's essay. In linguistics, interest has shifted away from the word order and constituent co-occurence phenomena captured by simple trees towards more complex and non- hierarchical relations between words. One can conclude that Greenfield is facing an uphill struggle in proposing such a principle as a source of unification in cognitive neuroscience. Greenfield's goal is to demonstrate that children's language and object combination develop synchronously in a structurally parallel sequence during the period from 9 to 20 months. Her starting point is her own work with nested cups, which she believes established a universal and genetically determined sequence of developmental stages (pair, pot, sub-assembly) for strategies of object manipulation. In the target article she adds data from Connolly & Dalgleish's (1989) study of spoon use in children between 12 and 23 months and, for her reply, fresh data of her own on spoon use. In these cases of simple object manipulation there is a natural description in terms of hierarchical task structures. For example, in the manual syntax for "sub-assembly" two objects are combined and then applied to a third: a spoon is filled with food and then put into the mouth; cup 1 is put in cup 2 and then the combination is put into cup 3 etc. Such manual sequences are presented as having the abstract tree structure ((object1 object2) object3). During the period under consideration, most children's language is limited to associations of one or two words with almost no hierarchical structure, and so object manipulation is in advance of syntax. Greenfield cannot, therefore, use the obvious analogy between verbal syntax and action grammar. Instead, she tries to map the object combinations into the individual phonemes of early words. The actual mapping she proposes (in Section 3 of the target article) is based entirely on her own analysis of her own unpublished diary data for the speech of three children. Her exposition involves five stages of phonological complexity, four of spoon use and three for nested cups and is extremely difficult to follow: I have summarized it in Table 1. There are so many problems with Greenfield's proposal that I shall not try to cover them all here. A number were raised by commentators: the limited data base, the lack of clear metrics for complexity, the poor correspondence in chronology, the apparently arbitrary choice of units ... and so on. One problem that was surprisingly not raised in the original discussion is that Greenfield does not present her child language data in the form of phonetic transcriptions, leading to the suspicion that the original data is in English spelling and thus lacking any precise phonological information ( for example, did her 15 month old subject really articulate the final 'l' in 'ball'?). A far more serious, indeed fatal, weakness is her starting assumption that children build their early words from phonemes in a process of hierarchical construction. Greenfield offers no evidence at all for this surprising claim: for the simple reason that there isn't any. As all parents know, children's early words are largely holistic approximations of the most salient syllable in the target. Indeed, it seems highly unlikely that word formation is ever generative, even in adults, in the strong sense that Greenfield assumes. This last point was made by MacNeilage in his commentary, but Greenfield's opaque response in her Reply indicates that she misunderstood it. MacNeilage also showed that the types of phoneme combination proposed by Greenfield can all be found in pre-speech babbling, where they emerge in no particular order. In addition, he finds no evidence for syllable-internal combinatorial procedures during babbling. Greenfield concedes that this presents something of a problem for her model, but claims that the data is in fact irrelevant because babbling differs from early speech in that babbling is controlled by a different neural system (the SMA as opposed to Broca's area) and is meaningless sound. The function of babbling in the transition to early speech is not yet understood, but what is known suggest a much closer relation than Greenfield admits. But even if the two systems were completely independent, MacNeilage's demonstration that the phoneme strings of babbling are largely non-combinatorial underlines the implausibility of Greenfield's assumption that early words are, in contrast, produced by combining phonemes. In his commentary, Tomasello shows that, by using words as units rather than phonemes, a simple analogy can be constructed between the classic one, two and three word stages and Greenfield's pairing, pot and sub- assembly strategies for object manipulation. Greenfield summarizes his analogy and replies as follows: "He must force the single word into representing the pairing strategy by relating it to a nonverbal element in the communicative situation, however, Then the 'pot' becomes the intonational envelope for the two-word utterances. Finally, two-word phrases become subassemblies in three word sentences. This conceptualization has the problem that its units are apples and oranges - and also that the nature of the combinatorial units changes from stage to stage. The original units presented in the target article accordingly seems preferable." (p. 582) It is clear, I think, that Greenfield's analogy that involves far more forcing and mixing than does Tomasello's! The analogy with words also produces a better temporal mapping between the two domains, eliminating the five month decalage in Greenfield's data between the emergence of sub-assembly in phoneme combination (at 15 months) and in object manipulation (at 20 months). In her reply, Greenfield highlights the weakness of her model by suggesting that the decalage could be a result of individual differences in rate of development or differential environmental stimulation or an actual time lag between the development of the hypothesized circuits for the two domains. Admitting any of these as an explanation for such a large decalage (five months!) is practically equivalent to admitting that there is in fact no significant temporal correlation between developmental stages in the two domains -- other than what would be expected from more general maturational constraints. 3. Greenfield's Model of Brain Development 3.1 Greenfield's Model Greenfield offers a developmental system level account of the brain areas that she claims subserve the hierarchical structuring of object combination and speech. She proposes two main stages. In Stage I (roughly between 9 and 20 months), Broca's area acts as a single unit programming the adjacent manual and speech motor areas. At this stage Broca's area is the "highest level" programming area controlling manual and speech output. It is also an "undifferentiated" neural region. Greenfield idea is that Broca's area would have (or generate) simple syntactic patterns that can be subsequently interpreted as action programs by either the manual or speech motor areas. The undifferentiated character of Broca's area would, among other things, "result in conjoint non- dissociable movements of hands and mouth" (p. 543). Within Stage I there is a progressive increase in the hierarchical complexity of the syntactic patterns produced by Broca's area and the resulting motor outputs: this determines the observed behavioral sequence of pairing, pot and sub-assembly. In Stage 2 (after 20 months) Broca's area participates in two new neural circuits connecting it to two areas in the anterior prefrontal cortex. The first of these areas controls "grammar" (Brodmann's area 46) and the second (Brodmann's area 9) controls "manual object combination". Broca's area would thus lose its independent top-level control function and be subordinated to control by these two anterior areas: "the early circuits constitute sub-processes of the more mature circuits." Greenfield does not offer much detail regarding area 46, but she clearly views it as an amodal grammar module responsible for most of what is considered linguistic syntax. Her view appears, in fact, to correspond closely to Chomsky's idea of language competence: an innate, declarative and linguistically specific grammar module unique to human beings. She is even less specific about area 9, but appears to consider it as providing analogous manual competence. Her commitment to a strong nativist theory is quite explicit: The grammar that a child utilizes from two years on was developed by natural selection, is stored genetically and unpacked into a specific cortical module (area 46) where it determines the development of of complex structure in verbal output. Before considering the evidence Greenfield offers for her model, some general comments are required. To begin with Stage 2, it is self-evident that object manipulation and speech are under functionally distinct neural control in children after 20 months, and there is no reason to doubt that prefrontal areas are involved. This logically requires connections between the various areas mentioned by Greenfield, and we have every reason to expect that, when neurobiologists find a way to trace them in humans, these connections will look something like the circuits that Deacon describes in monkeys. In this general sense, Greenfield's claim that such circuits are in place by about 20 months is no doubt correct. But, of course, the role that such circuits play in the functional organisation of the brain is still unknown: they transmit information, but what that information encodes and how it is transmitted is still a mystery. There is therefore no justification for a rigid mapping of specific high-level computational functions onto specific physical circuits, least of all in the frontal lobes. In addition, the four circuits hypothesized by Greenfield only represent a tiny fraction of the circuits normally implicated in language production and it is hardly plausible that hierarchical structuring should be functionally localized on just one of them. Greenfield's account of Stage 1 is even more problematic. Firstly, because almost nothing is known about the development and functional specialization of Broca's area in the human infant brain. Secondly, because the specific claim that she makes seems wildly improbable: examples of "conjoint non-dissociable movements of hands and mouth" in toddlers are striking precisely because they are so unusual. And Greenfield's sketch of infants as engaged in "action without thought, a state highly typical of the child between one and two years of age, who according to the model being proposed, would lack anterior prefrontal control" appears downright perverse. Indeed, an active, exploratory toddler seems to be the antithesis of the classic patient with frontal lesions. Now, precisely because the behavioral evidence does not support the model, Greenfield would have to find neurobiological evidence for the "undifferentiated" state of Broca's area in toddlers. But, as I shall show in the next section, while the neurobiological evidence she presents can be interpreted as being not in contradiction with the general ideas behind her account of Stage 2, it tells us absolutely nothing about the functional organisation of Broca's area during Stage 1. 3.1 Evidence for the Model in the Target Article As can been seen, Greenfield adopts an information processing style of model for her system level account of the relation between brain and language. She identifies various functional modules that exchange information via pathways. The modules are then localized in specific cortical regions and their interconnection identified with specific neural circuits. At this level of description, Greenfield can cite neuropsychological studies conducted with adults and older children that suggest some kind of association between language and hierarchical construction skills and their common location in the left hemisphere. And she is able to conclude that "the general region in which Broca's area is located has a directive or programming function for simple responses in a variety of modalities" (p. 536). This is hardly new or surprising and offers no support for the specific claims of her model. To strongly support her Stage 1 model, Greenfield would ideally have to demonstrate a sequence of neurophysiological changes in Broca's area between 9 and 20 months that would correlate with the three behavioral sub-stages of pair, pot and sub-assembly. Rather weaker support would be provided by a neurophysiological demonstration that Broca's area is "undifferentiated" with respect to manual or speech programming. Clearly it will be several decades before such detailed evidence becomes available. Instead, Greenfield uses two neurobiological studies (Thatcher 1991, Simonds & Scheibel, 1989) as evidence for the developmental chronology of the four neural circuits she postulates as progressively connecting Broca's area to the motor strip (Stage 1) and to anterior prefrontal cortex (Stage 2). The circuits proposed as the mechanism for the transition to Stage 2 are assumed to act to differentiate Broca's area, and this is implicitly taken as evidence that the area was not functionally differentiated before the circuits developed. In order to make use of the neurobiological data, Greenfield adopts her own speculative account of the process of circuit formation, according to which "multiple short-range connections are 'pruned' to fewer, more specific, and longer-range connections" (p. 544). She is confident that "this is the process by which differential circuits are created" and that "it is this developmental model that allows us to understand why early speech is so closely intertwined with other sorts of action, whereas later grammar is both more independent from action and more abstract" (ibid). She cites no neurobiological literature in support of these assertions (and nor do the commentators). After this build up, the actual data to be found in the Thatcher and Simonds & Scheibel studies is a big disappointment. Thatcher has data on the development of EEG coherence in children that he suggests probably correlates in some way with circuit formation. Simonds & Scheibel conducted a quantitative study of dendritic development in Broca's area and adjacent orofacial motor cortex in human infants from 3 months to 72 months. The data comes from the study of brain tissue collected after autopsy and Simonds & Scheibel are extremely cautious about what, if any, generalizations it supports: they make no mention of circuit formation. Unfortunately, Greenfield presents the data from these two isolated studies in a way that suggests she is using a validated methodology, claiming that "the data converge in providing information about developing neural networks" (p. 542). Even if her methodology was valid, the data itself is quite inadequate to provide a developmental history for the specific circuits and nor does it elucidate or support an "undifferentiated" Broca's area. This emerges clearly in Section 4 of the target article, where Greenfield tries to put together the two sources of data.Averaging over only 3 individuals for an age range of 12-15 months, Simonds & Scheibel found that dendritic growth in the two left hemisphere areas sampled (Broca's area and the adjacent orofacial motor cortex) had increased to catch up with the more precocious corresponding right hemisphere areas. In addition, dendritic growth was significantly higher in the orofacial motor area than in Broca's area. Greenfield believes that this data is symptomatic of the formation of the circuit from Broca's area to the orofacial motor cortex. As for the second circuit of Stage 1, connecting Broca's area to the manual motor areas: "Inspection of Thatcher's cross-sectional data set indicates that this circuit has significant connectivity in this age range, reaching a modest first peak of coherence around 16 months of age" (p. 543).Averaging data from 2 individuals at 24 months and 2 individuals at 36 months, Simonds & Scheibel concluded that in the period 24-36 months dendritic growth in Broca's area had caught up with the adjacent orofacial motor cortex. Greenfield believes that this indicates that Broca's Area is receiving input from elsewhere from about 24 months. She suggests that the input is coming from anterior prefrontal areas and writes: "To test my prediction, Thatcher analyzed his cross-sectional data and found [..] a spurt of increased connectivity between approximately two and four years of age." (p. 543) She thus assumes that dendritic growth at point A plus increased EEG coherence between point A and point B implies a circuit going from B to A. Again this appears to be a purely speculative assumption. It should be clear from the above that Greenfield has no evidence for the specialized circuits described in her model. Even if EEG coherence and dendritic growth were symptomatic of circuit formation in the simple and direct way she assumes, the data would still show nothing more than that connections form between Broca's area and other frontal lobe areas during the second two years of life. But it would be highly surprising if this was not the case. The data tells us nothing about the function of these circuits and nothing about the functional organisation of Broca's area before they formed. It should also be clear that there is no way to map Greenfield's cross-domain behavioral development sequence into the neurobiological data and she does not attempt to do this. Finally, there appears to be a contradiction underlying the neat symmetry of the two pairs of circuits: in Stage 1 circuit formation leaves Broca's area undifferentiated, while in Stage 2 the same mechanism causes differentiation. I conclude that the homology proposed for Stage 1 is not supported by the neurobiological evidence. 3.3 Commentary and Greenfield's Reply Several of the commentators briefly discuss Greenfield's model and her use of the neurobiological evidence. Thatcher, who begins by noting that "the mechanisms of neural circuit differentiation are currently unknown", describes his own general theory of cerebral maturation but, unfortunately, makes no comment on the specific claims of the target article. Jacobs, who has used the same methods as Simonds & Scheibel (and discussed the target article with Scheibel), notes that "what each investigative technique reveals about the brain, and the degree to which these can be complementarily synthesized remains an open question." Deacon, Fuster and Lieberman all make interesting observations about the anatomy and development of various neural circuits, mostly in monkeys, but they offer no new evidence or support for the specific claims of the target article -- and nor does Greenfield's reply to them. 5. Conclusion Greenfield fails almost completely in her ambitious attempt to use thin and patchy data to support an original synthesis of cognitivist and nativist accounts of language development. The proposed synchronous stages for object and phoneme combination in children are not established by the data she presents. Her functional specification of hypothesized circuits is almost entirely speculative, while the neurobiological data she offers does not support the specification. Nor does it seem likely that new data or analysis could save her specific claims, since these are formulated in a simplistic information processing theoretical framework (language production as computation of structure, strong Chomskian nativism, Fodorian modularity ... and so on) that surely now belongs to the past. The general brain development framework that emerges from the target article and the commentaries is, nonetheless, extremely interesting and suggestive. As Fuster formulates it, a caudal to rostral maturational gradient in the frontal lobes reflects a representational gradient from concrete motor output to more abstract and hierarchically structured plans. Greenfield makes a major contribution in associating this maturational process with Bickerton's idea of the transition from protolanguage to complex grammar. But her nativist belief that "brain development drives language development" (p. 550) is untenable. Language acquisition is not something that merely happens to children, it is something they actively do in response to strong environmental stimulation and it is equally likely that language development, in some sense, drives brain development. Elsewhere (Swann, in preparation) I develop a non- nativist model of child language. In conclusion here, I shall briefly indicate how the phenomena Greenfield targets are compatible with such an account. Nativists take as their criteria for "language" the morphosyntactic generative code used by grammarians to describe some aspects of human verbal language. Instead, I shall follow Saussure and Wittgenstein and take the semiotic function as defining language. The semiotic function is a development of the communicative function common among animals and is characterized by arbitrary symbols that require social conventions for their application and reproduction. Like many animals we have an innate ability and drive to communicate using non- arbitrary gestures and signs such as pointing and smiling; and we share with a few higher animals, such as chimpanzees and dolphins, an innate disposition to participate in semiotic systems. It would seem likely, however, that we only become aware of and motivated to use this innate semiotic potential in social development during what Lock (1980) has aptly termed the "guided reinvention of language". From this perspective, acquiring one's mother tongue is analogous to learning to ride a bicycle: we are biologically capable and disposed to acquire the bicycle riding function, but we need a bicycle and the social motivation to do so. This does not mean we must have evolved special purpose neural circuits for cycling: our disposition for cycling is there because the bicycle is a social artifact constructed (with some trial and error) to respect our biological potential limitations - and the same is true, I maintain, for verbal language. During the protolanguage period children acquire their patterns of participation in language by means of predominantly implicit learning in an input - storage - ouput - feedback loop based on many neural circuits, including those connecting Wernicke's area to Broca's area. There is abundant evidence that the hierarchical structures of protolanguage could be assimilated in this way, without any need for innate knowledge (Knowlton et al. 1992).The gradual accumulation of a large number of language games leads to the emergence of complex grammar in a kind of "phase transition" around 24 months. This phase transition is not something that the child generates from within, rather it is an inevitable product of its increasing participation in external dynamic symbol systems. Happily, the problem of dealing with this emergent complexity would coincide with the increasing availability of frontal functionality (Case 1992). In my view, the child would now utilize an increasing proportion of explicit learning and memory, partially subordinating the protolanguage system to frontal control and reorganization and leading to the production of complex morphosyntax. Instead of emerging circuits "driving langauge acquisition", I would imagine a much more flexible relation, along the lines of Bates (1992), in which some part of the capacity of general-purpose circuits would be specialized and directed by the requirements of the child's participation in language. Such a scenario would accomodate a realistic degree of localisation and potential functional dissociation for language, while excluding most of the strong Nativist claims. It would also suggest a great deal of flexibility as to where, when and how language functions get instantiated in the developing infant brain. References Bates, E. (1992) Language development. Current Opinion in Neurobiology, 2, 180 - 85 Bickerton, D. (1990) Language and species. University of Chicago Press Boesch & Boesch (1990) Transmission aspects of tool use in wild chimpanzees. In Gibson, K. R. & Ingold, T. eds (1993) Tools, language, and cognition in human evolution, CUP Case, R. (1992) The role of the frontal lobes in the regulation of cognitive development. Brain and Cognition, 20, 51 - 73 Chomsky (1956) Syntactic Structures. Mouton Connolly & Dalgleish (1989) The emergence of tool-using skill in infancy. Developmental Psychology, 25, 894-912 Corballis, M. C. (1992) On the evolution of language and generativity. Cognition, 44, 197 - 226 Curtiss et al. (1979) How independent is language. UCLA Working Papers in Cognitive Linguistics, 3, 61-91 Dawkins, R. (1976) Hierarchical organisation; A candidate principle for ethology. In: Growing Points in ethology, ed. Bateson, P. P. G. & Hinde, R. A. Cambridge University Press, 7 - 54 Deacon, T. W. (1988) Human brain evolution: I evolution of the language circuits; II embryology and brain allometry. In Jerison, H. & I. eds, Intelligence and evolutionary biology, Springer, 363-416 Deacon, T. W. (1992) The neural circuitry underlying primate calls and human language. in Wind et al. eds, Language origins, Kluwer, 121-62 Fentress, J. C. (1981) Hierarchical Motor Control. In: Psychobiology of language, ed. M. Studdert-Kennedy. MIT Press, 40 - 61 Gibson, K. R. & Ingold, T. eds. (1993) Tools, language, and cognition in human evolution. CUP Greenfield et al. (1972) The development of rulebound strategies for manipulating seriated cups. Cognitive Psychology, 3, 291-310 Greenfield & Schneider (1977) Building a tress structure. Developmental Psychology, 13, 299-313 Greenfield, P. M. (1991) Language, tools and brain: the ontogeny and phylogeny of hierarchically organized sequential behavior. Behavioral and Brain Sciences, 14, 531-95 Grossman, M. (1980) A central processor for hierarchically structured material: evidence from Broca's aphasia. Neuropsychologia, 18, 299-30 Knowlton, B. J., Ramus, S. J. and Squire, L. R. (1992) Intact artificial grammar learning in amnesia: Dissociation of classification learning and explicit memory for specific instances. Psychological Science, 3 (3), 172 - 79 Lock, A. (1980) The guided reinvention of language. Academic Press Lewontin (1990) The evolution of cognition. In Osherson, D. N. & Smith, E. E. eds. An Invitation to Cognitive Science, Volume 3, 229-46 Lieberman, P. (1991) Uniquely human: the evolution of speech, thought and selfless behavior. Harvard Simon, H. A. (1969) The sciences of the artificial. MIT Simonds & Scheibel (1989) The postnatal development of the motor speech area. Brain and Language, 37, 42-58 Swann, P. (in preparation) A non-nativist model of child language. Thatcher (1991) Maturation of the human frontal lobes: physiological evidence for staging. Developmental Neuropsychology, 7, 397-419 Wind, J., Lieberman, P. & Chiarelli, B. eds. (1991) Language origins: a multidisciplinary approach.Kluwer