Beyond Modularity: Innate Constraints and Developmental Change |
MRC Cognitive Development Unit, London
Some years ago, Gleitman, Gleitman, and Shipley (1972) made a simple yet very thought-provoking statement, which I used as a colophon to an article published in the same journal some 14 years later (Karmiloff-Smith, 1986). The statement was: “Young children know something about language that the spider does not know about web weaving” (Gleitman et al., 1972, p. 160). My chapter is not, of course, about spiders. Rather, my intention is to explore a number of speculations about what it is for a mind to “know” (about language, the physical environment, etc.) and what makes the human mind special in contrast to the innately specified procedures by which the spider produces its seemingly complex web. How can we account for human flexibility and creativity?
Let me begin by suggesting the following: for as long as Piaget’s constructivist description of the human infant held, (i.e., an assimilation/accommodation organism with no constraints from built-in knowledge), then it followed that the human mind, acquiring basic knowledge via interaction with the environment, might turn out to be cognitively flexible and creative. However, in the last decade or so, exciting new paradigms for infancy research have radically changed our view of the architecture of the human mind, which is now considered to be richly endowed from the outset.
For many psychologists, accepting a nativist viewpoint precludes constructivism completely. Yet nativism and constructivism are not necessarily incompatible. Together with a now growing number of developmentalists, I have been grappling for some time with a paradox. On the one hand, I was dissatisfied with Piaget’s account of the human infant as a purely sensorimotor organism with nothing more to start life than a few sensory reflexes and three ill-defined processes: assimilation, accommodation, and equilibration. There had to be more to the initial human structure than that. Yet I felt that a purely static, radical nativist/maturational position had to be wrong, too. I continued to be attracted by the more general aspects of Piaget’s epistemology—his epigenetic constructivist view of biology and knowledge and his vision of the cognizer as a very active participant in his or her own cognitive development. An attempt at reconciliation between nativism and constructivism is thus an important thread throughout this essay (see, also, Feldman & Gelman’s [1986] arguments in favor of a rational-constructionist view of development).
Why are some developmentalists still so reticent about attributing innate knowledge to the human mind? A connectionist learning theory, for instance, does not necessarily preclude innately specified underpinnings, yet most developmentalists of the connectionist persuasion are adamantly antinativist (e.g., Bates & MacWhinney, 1987; Bates, Thal, & Marchman, 1989). No one would hesitate to accept that the spider, the ant, the beaver, and the like use innately specified knowledge structures. So why not the human? But if the human has innately specified knowledge too, then what is special about human cognition? My argument throughout this chapter is that although all species have knowledge in their cognitive systems, knowledge in the human mind subsequently becomes knowledge to other parts of the mind. In other words, a major part of human development is transforming special-purpose, procedurally encoded knowledge into data structures which then become available to other parts of the mind.
My conviction that a nativist stance is required for an account of the initial architecture of the mind stems from three directions. First, I fail to see how it could be biologically tenable that nature would have endowed every species except the human with a considerable amount of biologically specified knowledge. Second, the bulk of infancy research over the past decade points to a certain amount of innately specified perceptual, conceptual, and linguistic structures (Anderson, 1988; Baillargeon, 1987; Butterworth, 1981; Cohen, 1988; Diamond & Gilbert, 1989; Gelman, 1986; Johnson, Dziurawiac, Ellis & Morton, 1989; Jusczyk, 1986; Kellman, 1988; Leslie, 1984; Mandler, 1988; Mehler & Fox, 1985; Rutkowska, 1987; Slater, Morison, & Rose, 1983; Spelke, 1982, 1990; Starkey & Cooper, 1980, and many others too numerous to mention). The human infant is biologically set to process constrained classes of inputs that are numerically relevant, linguistically relevant, relevant to physical properties of objects, of cause-effect relations, and so forth. Third, taking language as an example, I fail to see how complex syntax could get off the ground without invoking a linguistically constrained input analyzer and linguistically specified principles and parameters (Borer, 1984; Chomsky, 1981; Gleitman, Gleitman, Landau, & Wanner, 1988; Valian, 1990). However, accepting the nativist position with respect to initial human development does not necessarily preclude a constructivist approach with respect to subsequent development. Nor does a constructivist approach necessarily imply domain-neutral, stage-like changes à la Piaget. One can invoke cyclical phases that reoccur at different ages in different domains (Karmiloff-Smith, 1986; Keil, 1986; Mounoud, 1986) and fundamental, domain-specific knowledge reorganizations (Carey, 1985; Brown, 1990).
The infancy literature abounds with examples of the wealth of complex, domain-specific knowledge with which the human infant enters the world. Young infants’ knowledge of principles of physics provides an excellent example of the innately specified principles and those that are subsequently learned during infancy (see Spelke, 1988a, 1988b, and this volume). Spelke has shown that very young infants are surprised when viewing an impossible event in which one solid object passes through another. However, it is not until considerably later in infancy that they show surprise if objects behave as if they were not subject to gravity and in need of stable support. Spelke’s work offers many other examples suggesting that some physics principles are available to the baby very early on and probably are innately specified, whereas others involve learning.
A particularly interesting theoretical discussion of innately specified knowledge and subsequent learning can be found in the work of Johnson, Morton, and their colleagues on infant face recognition (Johnson, 1988, 1990, in press; Johnson & Morton, in press; Johnson et al., 1989). Extending Johnson’s theory of species recognition and imprinting in the domestic chick, the existence of two mechanisms was also postulated for human species recognition. The first mechanism operates from birth and is predominantly mediated by subcortical structures. Newborn infants (within the first hour of life) will track certain types of face-like patterns further than other patterns. The exact stimulus characteristics that give rise to this preferential orientating (which may be as minimal as three high-contrast blobs in the appropriate locations for eyes and mouth or as detailed as the actual features of a human face in their correct locations) are still under investigation (Johnson et al., in preparation). The second mechanism is controlled by cortical structures but is constrained by the functioning of the first and gains control over behavior at around 2 months of age. Two months is the age when many authors have suggested that there is the onset of cortical control over visually guided behavior (Johnson, 1990). Thus the first system serves to constrain the range of inputs processed by the second system and in some sense “tutors” it before it gains subsequent control over behavior.
All the new neonate and infancy data that are accumulating serve to suggest that the nativists have won the battle in accounting for the initial structure of the human mind. So, does that put the constructivists completely out of business? Not necessarily, and for two reasons. First, the examples from Spelke’s and Johnson’s work previously mentioned involve both a detailed innately specified aspect and subsequent learning. The actual mechanisms involved in the process of learning must be innately specified. However, from the point of view of the content of the subsequent learning, although it is constrained by innate specifications, it feeds on and is crucially affected by its interaction with properties of the input. Second, we know that human cognition manifests flexibility with development. It is true that the greater the amount of primitively fixed formal properties of the infant mind, the more constrained its computational system will be (Chomsky, 1988). In other words, there is a trade-off between the efficiency and automaticity of the infant’s innately specified systems, on the one hand, and the rigidity of such systems, on the other. But if systems were to remain rigid, there would be little if any room for cognitive flexibility and creativity. This is where a constructivist stance becomes essential. As we draw up a much more complex picture of the rich innate structure of the infant mind and its complex interaction with environmental constraints, cognitive flexibility and creativity require specific theoretical focus within a constructivist epistemology. Thus, both nativist and constructivist approaches are essential for a comprehensive account of human development.
Until recently, I had remained relatively agnostic about the structural constraints pertaining to the neonate, because my research has concentrated on constraints on developmental change and flexibility in older children (above 3 and, more frequently, above 5 years of age). However, partly because I found the new infancy research very convincing and uninterpretable within previously accepted theoretical paradigms such as Piaget’s, and partly because of some misinterpretations by the so-called neoPiagetians (e.g., Johnson, Fabian, & Pascual-Leone, 1989) about the relationship I wish to draw between language and other cognitive processes (Karmiloff-Smith, 1989), I found myself focusing more specifically on the relationship between the constraints on the neonate system and the constraints on the subsequent flexibility that the system displays in the normally developing child.
Let me begin then, by making four assumptions about the innate structure of the human mind. They are as follows:
1. The human mind has an appreciable amount of innately specified knowledge about persons, objects, space, cause-effect relations, number, language, and so forth. This does not, of course, preclude the necessity for subsequent learning, but it constrains the way in which such learning takes place.
2. Something like Fodor’s distinction between transducers, input analyzers and central processes holds for the biological specification of the basic organization of the human mind at birth. However, I disagree with Fodor’s unitary conception of a single central processor. I accept the existence of central processing of a domain-neutral kind, but I also argue for dedicated central-like processing within the specific codes of the various input systems themselves. Moreover, I reject Fodor’s pessimism about the possibility of studying central processing.
3. The human mind has at least five innately specified processes: it is a self-replicating and self-describing system, on the one hand, and a self-organizing/self-restructuring system, on the other. Further, it specifies mechanisms for inferential processes, for deductive reasoning, and for hypothesis testing.
4. The human mind not only tries to appropriate the external environment that it is set to begin exploring and representing from birth, but it also tries to appropriate its own internal representations. I argue for the representational redescription hypothesis, that is, that the human mind re-represents recursively its own internal representations. For a number of years I have argued that it is the pervasiveness of this recursive capacity to represent one’s own representations, that is, to construct metarepresentations, that sets us apart from other species (Karmiloff-Smith, 1979a, 1979b and, in much greater detail, in Karmiloff-Smith, in press).
What, then, are the implications of the preceding assumptions for the question of innate constraints and developmental change, the title of my chapter? Following are two speculations that I would like to explore here.
The first speculation runs as follows: None of the initial, special-purpose knowledge that is built into the human system is available to the system as data structures. Initial, special-purpose knowledge is, in my view, represented as procedures that are activated as a response to external stimuli but to which other parts of the cognitive system do not have access. This procedurally encoded knowledge has a similar status to knowledge in nonhuman species. Subsequently, in humans, a process of representational redescription (which I outline later) enables certain aspects of knowledge to become accessible to other parts of the mind. Thus, human development crucially involves the passage from representations that constitute knowledge in the mind to representations that acquire the status of knowledge to other parts of the mind. I therefore submit that there are no innately specified “theories” in the mind. Despite the fact that innately specified principles and knowledge may form a coherent, interrelated set that supports inductions (Spelke, 1990), they do not, in my view, have the status of a “theory.” They are simply embedded in procedures for responding to the environment. For knowledge to have “theoretical” status for the cognizer, it must be explicitly represented via a process of redescription that extracts the knowledge from the procedure and encodes it in a new form (see Karmiloff-Smith, 1979a, 1979b, 1984, 1986, and discussion later in the chapter).
Fodor’s dichotomy between input systems and a central processor, although correct in essence, is too narrow. Fodor’s view is that at birth, basic input analysis is modular. With development, the outputs of modules are sent to central processing, where the human belief system is built up. Whereas modules are cognitively encapsulated, central processing is not. The second speculation that I explore here endorses Fodor’s basic distinction but argues that, with development, central processes end up mimicking the basic characteristics of the modular/central organization of the initial state. This leads to different degrees to which processing is ultimately modular or central. I am not suggesting that the modules themselves gradually open up their processes to the rest of the cognitive system (see subsequent discussion on Rozin, 1976). What I am suggesting is that, with development and in interaction with the constraints of the environment, the organism recreates its basic organization to form modular-like processes within central processing and central-like processors within specific input systems. The notion of central-like processes within input systems leads to the hypothesis that some rudimentary domain-specific metacognitive processes will be available to otherwise severely retarded children. In other words, in their particular domain of expertise (e.g., William Syndrome children’s language), retarded children may have dedicated aspects of central-like processing, although lacking domain-neutral central processing. The notion of modular-like processes within central processing leads to the hypothesis that some late-acquired knowledge may become encapsulated and to all intents and purposes have the basic characteristics of a Fodorian module (Fodor, 1983), although not innately specified nor unassembled. We take up these points shortly hereafter.
As I mentioned before, the abundance of infancy research has shown that the neonate is biologically set to compute—and, according to some, to actively select—constrained classes of inputs that are numerically relevant, linguistically relevant, relevant to species recognition, to the identification of features of objects, of cause-effect relations, and so forth. That being so, however, we are still left with the question of the form of the neonate’s knowledge representations. Piaget granted no initial innately specified knowledge representations and no capacity for symbolic representations. For Piaget, only sensorimotor encodings of a domain-general type existed during infancy. Others argue for domain specificity via a process of modularization that results from complex interaction between innately specified knowledge structures and the constraints of the environment (Johnson, 1990). This is a form of epigenetic constructivism but on the basis of innately specified constraints that guide subsequent selection in the developing infant’s mind (Johnson & Karmiloff-Smith, 1989). Mandler (1983, 1988) has made a very convincing case for the existence of symbolic representations in young infants (but perhaps not neonates) who show the capacity for both procedural and declarative (symbolic) representations of new knowledge. I agree that symbolic representation is a capacity available to the young infant, and not solely at 18 months at the end of the sensorimotor period as Piaget would have it. I question, however, one of Mandler’s other assumptions. She has argued that the infant’s procedural and declarative stores are independent of one another and that knowledge is often stored simultaneously in each. By contrast, I hypothesize that, except perhaps where new knowledge is directly encoded linguistically (Mandler, 1988), the procedural representations actually constrain the content, form, and timing of what is ultimately in declarative form, as a result of representational redescription.
Irrespective of one’s position on the issue of symbolic representations in early infancy, an important question arises with respect to the neonate’s domain-specific principles and knowledge that are innately specified, that guide its initial responses to the external environment and constrain its subsequent learning. As I mentioned earlier, my hypothesis is that the neonate’s representations are initially procedurally encoded. What form might they take? Two solutions are possible: One is to hypothesize that the innate principles and knowledge are represented in the form of a mere sketch, a skeletal outline to be filled in by experience. A second, somewhat different solution is to hypothesize that the innate principles and knowledge are very detailed and complex, rather than mere skeletal outlines, but that they undergo a process of selection. Both may be true, depending on different domains of cognition. As I mentioned previously, there is some controversy in the literature as to the exact stimulus characteristics for face recognition that give rise to preferential orientating in the neonate (see discussion in Johnson, 1990) but it could turn out to be a mere sketch in the form of three high-contrast blobs. For other domains of cognition (perhaps language and certain principles of physics), it may be biologically untenable to think in terms of a skeletal outline, because, as Gould and others have argued (Gould, 1982), the more complex a behavior is, the more likely it is to be innately specified in detail.
However detailed or skeletal the innate specifications are, I submit that they are inaccessible as data structures to other parts of the cognitive system. They are knowledge in the mind but not yet data to other parts of the mind. They merely run as procedures within specific input systems, given appropriate external stimuli. Yet these initial representations ought to become available for subsequent cognitive change, without which the system would always be starting again from scratch. Carey (1985) has argued that knowledge built up in the infant mind is modified by processes underlying subsequent theory change. However, these later developments must be preceded by a developmental process that enables the child to appropriate, via representational redescription, the knowledge that is innately specified (Karmiloff-Smith, 1979a, 1979b, 1984, 1986). It is such redescriptions, and not the procedurally embedded knowledge, that are then open to cognitive change of various sorts (reorganization, further explicitation, theory building, etc.).
To some extent, I share the views that Rozin spelled out in some detail in a now classic article published in 1976 on access to the cognitive unconscious. Rozin described evolution in terms of adaptive specializations that originate as specific solutions to specific problems in survival, as opposed to what he called calibration learning. He argued for a similar account of the evolution of human intelligence. For Rozin, progress involves gaining access to the cognitive unconscious, that is, “bringing special-purpose programs to a level of consciousness, thereby making them applicable to the full realm of behaviour or mental function” (Rozin, 1976, p. 246). Rozin argued that accessibility can be gained via two processes: “… establishment of a physical connection of one system to another or by duplication of one system’s circuitry in another part of the brain by use of the appropriate genetic blueprint” (p. 246).
There are ways in which my views concord with Rozin’s, but they differ as follows: First, it is hard to envisage how use of the same physical circuitry and genetic blueprint could lead to increased cognitive flexibility. Second, I believe that to account adequately for human development, we need to invoke more than a simple dichotomy between the cognitive unconscious and the cognitive conscious. Third, it is highly unlikely that gaining access to the cognitive unconscious results in making special-purpose programs available to the full range of behaviors and problems. This is where I also disagree with what is implicit in Fodor’s single central processor position. It is implausible that input analyzers deliver their products immediately in a single language of thought such that the central processor has available to it all the information represented centrally by the organism. There must be constraints on central encoding and access, and on interrelations between different representational codes (spatial, linguistic, kinaesthetic, etc.). I return to this later. The fourth difference with Rozin’s (1976) view of phylogeny and ontogeny (and with Leslie’s copying mechanism [Leslie, 1987] which is really about a maturationally guided structure’s on-line use rather than an ontogenetic process in interaction with environmental constraints) is that I doubt that human development involves the mere “duplication”/“copy” of programs and representations. My position regarding metarepresentation has always been in terms of “redescription” of representations, rather than simple duplication (Karmiloff-Smith, 1979a, 1979b, 1984, 1986). The lower levels are left intact; copies of these are redescribed. Redescription involves a loss, at the higher level, of information that continues to be represented at the lower level. Our multiple levels of representation are not, I submit, simple duplicates of lower levels; rather, they involve increasing explicitation and accessibility at the cost of detail of information.
I thus argue that there are three ways in which knowledge gets into the child’s mind:
1. It is innately specified, for some aspects of knowledge in precise detail, for other aspects as a skeletal outline to be filled in by experience.
2. It is acquired via interaction with the physical and sociocultural environments.
3. It involves an endogenous process whereby the mind exploits the knowledge that it has already stored (both innate and acquired), by re-representing recursively its own internal representations. This process may, of course, be triggered by external constraints but it can also be self-generating.
It is this third way of gaining knowledge, endogenous exploitation via representational redescription and restructuring, that has been the focus of almost all of my work in linguistic and cognitive development. In the rest of this chapter, I examine the acquisition of language within this theoretical framework: linguistic knowledge acquired at the procedural level on the basis of innate linguistic specifications, as compared to the subsequent process of redescription of linguistic representations. We then go on to look at an example of representational redescription of an efficiently functioning procedure learned in the sociocultural environment, using a study of children’s drawing, in both normal and retarded children.
Let me now briefly outline my current thinking about constraints on the process of representational change. There are at least three levels at which knowledge is represented and re-represented (Karmiloff-Smith, 1986).
The first level I have termed implicit (I-level). By this I mean that the knowledge is embedded in an efficiently functioning procedure, activated as a response to external stimuli, but not available as knowledge, i.e. as a data structure, to other parts of the system. As hypothesized earlier, the neonate’s knowledge is of this type, and so is all innately specified knowledge under subsequent maturational constraints, such as language. Once the procedures are automatized, some form of pattern recognition mechanism scans the internal representations for stable states, and then a metaprocess is set in motion that redescribes the knowledge embedded in procedures. This process does not depend on information currently coming into the system for processing. It can be self-generating. At this next level of explicitation (E-1 level), knowledge embedded in procedures becomes available as data to other parts of the system, but not yet to conscious access. The latter requires yet further levels of redescription (levels E-2 and E-3) (see Karmiloff-Smith, 1986 and in press, for fuller details). Let us consider these levels as they pertain to the acquisition of language.
I begin with an outline of the infancy work on constraints on language and some discussion of the difference between normal language and the language of fluent-speaking yet very retarded children. Chomsky has argued that what is built into the system is some form of universal grammar, that is, linguistic principles that are innately specified and constrain the child’s acquisition of his or her mother tongue. Also built in are a series of parameters either with a default setting (Hyams, 1986) or with more than one setting available (Valian, 1990), to be fixed one way or the other in the light of the characteristics of the child’s particular linguistic environment (the mother tongue). Note that the principles and parameters are not mere skeletal outlines but are specified very precisely and then selected via interaction with the constraints of the environment. However, as Gleitman and her colleagues have demonstrated (Gleitman, Gleitman, Landau, & Wanner, 1988), the Chomskyan model fails to address a prior problem. Between birth and the onset of language maturationally, how does the infant build up linguistically relevant representations from the mother tongue model on which to base the mapping with the innate linguistic constraints of universal grammar? Do the child’s input systems constrain the interpretation of sound waves to distinguish between linguistically relevant and other, nonlinguistic acoustic input? Gleitman and her colleagues argue that the child is preset biologically to represent the linguistically relevant aspects of sound waves. Recent research suggests that the infant’s mind computes a constrained class of specifically linguistic inputs such that their interpretation of sound waves makes a distinction between linguistically relevant and other, nonlinguistic sounds (Mehler & Bertoncini, 1988). The normal infant attends preferentially to human language. The development of speech perception of the particularities of the infant’s mother tongue has been argued to be an innately guided learning process (Jusczyk & Bertoncini, 1988). Results of studies have suggested that, well before they can talk, young infants are already sensitive to word boundaries (Gleitman et al., 1988) as well as to clause boundaries within which grammatical rules apply (Hirsh-Pasek, Kemler-Nelson, Jusczyk, Wright, & Druss, 1987). Babies show distinct preference for a recording into which pauses are inserted at natural clause boundaries as opposed to a recording in which the pauses violate such linguistic boundaries. Experiments have also been devised to demonstrate that infants are sensitive to relative pitch, which is linguistically relevant, versus absolute pitch (e.g., male versus female voice), which is socially relevant, that they are sensitive to rhythmic aspects of linguistic input, to vowel duration, to linguistic stress, to the contour of rising and falling intonation, and to subtle phonemic distinctions (Eilers, Bull, Oller, & Lewis, 1984; Eimas, Siqueland, Jusczyk, & Vigorito, 1971; Fernald, in press; Fernald & Kuhl, 1981; Fowler, Smith, & Tassinary, 1986; Kuhl, 1985; DeMany, McKenzie, & Vurpillot, 1977; Spring & Dale, 1977; Sullivan & Horowitz, 1983). Work has also shown that the stabilization of phonologically relevant perceptual categories does not require the prior establishment of sensorimotor programs (Mehler & Bertoncini, 1988). Moreover, according to Mehler and his colleagues in Paris, 4-day-old infants are already sensitive to certain characteristics of their mother tongue (Mehler et al., 1988). Indeed, Mehler found that 12 hours after birth, babies distinguished between linguistically relevant input and other, nonlinguistic acoustic input. They preferentially attend to human language. However, they do not yet react to differences between particular languages. Thus, the 9 months in utero do not provide the necessary input for the child to attend preferentially at birth to its mother tongue. However, already by 4 days of age, that is, after exceedingly little experience, the infants studied by the Parisian team showed preference for listening to the prosody of French over Russian. Given the chauvinism of the French, and despite glasnost, one might start to wonder whether the results would have been published had the French babies turned out to prefer listening to Russian!
What this and other infancy work shows is that by the time the child is maturationally ready, there is already a large bulk of linguistically relevant representations in the child’s mind to interact with the innate universal grammar. These are a result of multiple external and internal constraints that give rise to a discrete number of possible emergent functions or parameter settings (see discussion in Johnson, 1990; Johnson & Karmiloff-Smith, 1989).
So, is that all there is to language acquisition? A set of constraining biases for attending to linguistically relevant input and subsequently, with maturation, a universal grammar and a number of parameters to be set via some sort of inductive mechanism? Does language acquisition involve nothing more? Little more is needed to provide an adequate account of language acquisition as far as the initial mapping operations are concerned to generate efficient output. A Connectionist model of learning equipped with an “innate” base could probably account for this. This may be all there is to the language acquisition of certain fluent-speaking retarded children. The semantically and syntactically adequate language of certain very retarded children with William’s syndrome (Bellugi, Marks, Bihrle, & Sabo, 1988; Thal, Bates, & Bellugi, 1989) and of hydrocephalic children with spina bifida (Cromer, in press) suggests that the need to invoke domain-neutral cognitive prerequisites to linguistic development (Sinclair, 1971, 1987) is highly challengeable. Whereas fluent language in a very retarded child is difficult to accommodate theoretically within a traditional Piagetian perspective, it is unsurprising within modularity theory. Indeed, the preservation of a module with innately specified constraints on attending to a class of linguistically relevant acoustic inputs, together with innately specified linguistic principles and parameters and some form of induction and mapping mechanism, is all that the retarded child would need to carry out automatically, without cognitive effort, the mapping operations between the input model and the prespecified internal constraints. With the linguistic module intact, this would allow the fluent-speaking retarded child to display similar behavior to that of the normal child. Thus, only I-level representations would be needed to support fluent output.
However, this is not all there is to the language acquisition of the normally developing child. Although the fluent-speaking retarded child probably has an intact linguistic module, the normal child also has the potential to become a grammarian. By contrast, not only do other species not have a linguistic module, but the constraints on spiders, ants, beavers, and the like are such that they could never become potential describers of the knowledge embedded in their procedures for interacting with the environment. The human system’s capacity to re-represent recursively its internal representations allows us eventually to become grammarians, poets, philosophers, physicists, and so forth (Karmiloff-Smith, in press).
Normal children tend to reach a form of behavioral mastery in their linguistic output and go beyond that mastery to reorganize their linguistic representations and to generate theories about how language functions as a system (Karmiloff-Smith, 1986). Severely retarded, yet fluent language users such as those identified by Cromer (in press) and by Bellugi and her colleagues (Bellugi et al., 1988) seem in some cases to have practically nonexistent metalinguistic knowledge (Karmiloff-Smith & Grant, in preparation), whereas in others some metalinguistic reflection on language is manifest (Bellugi & Karmiloff-Smith, in preparation). Much more in-depth probing is required to understand these differences, but they suggest that input systems can function in a totally rigid fashion or that an input system can have dedicated aspects of central-like processing.
Linguistic representations of the fluent-speaking retarded child, although analogous at first to those of the normal child, may subsequently be very different because of representational reorganization in normal development. Initially, to get linguistic representations into the mind, the child focuses on the external input model, involving complex interactions between the input model and universal grammar. During this first phase, children’s representations are procedurally encoded and isolated one from the other. Subsequently, in normal development and in some, but by no means all, retarded development, the child’s mind ignores the external stimuli and focuses on its internal representations. In other words, the normal child goes beyond successful output to exploit the knowledge that he or she has already stored. Thus external reality serves as input to form representations, but it is the internal representations that serve to form theories about the linguistic system. This necessitates the redescription of procedurally embedded knowledge and is typical of the normal child’s linguistic development.
Having reached behavioral mastery in output, the normal child’s mind then proceeds to redescribe, at a higher level of abstraction, the knowledge embedded in linguistic procedures, which then becomes available to the system as a data structure. It is now explicitly defined in the internal representations at E-1 level. But being explicitly defined does not mean that the knowledge is available to conscious access nor, of course, to verbal report. This requires further redescriptions (levels E-2 and E-3). Many authors have merely used a dichotomy between an undefined notion of “implicit” and the verbally stateable knowledge. They have failed to distinguish between implicit knowledge and a level at which knowledge is explicitly represented but is not consciously accessible. The human representational system is far more complex than accounts using dichotomous models can offer.
Let us now take a simple example from language showing the passage from language procedurally encoded to the normal child’s subsequent spontaneous formation of theories about how language functions as a system. In other words, knowledge of language is first embedded in an efficiently functioning procedure, then it is redescribed and available as data to the system but not yet to conscious access, and finally to conscious access and verbal report. But this example of progressive conscious access will be followed by one from some new research that illustrates how the constraints on allocation of competing resources result in a process of modularization, whereby the linguistic system closes off parts of itself from conscious access. In some sense, then, this involves the encapsulation of the newly developed system. In other words, in human development, a process of modularization can arise as a result of development and not merely as part of the mind’s innate specification (Karmiloff-Smith, 1986; see, also, Johnson & Karmiloff-Smith, 1989; Jusczyk & Cohen, 1985).
The first example is from the acquisition of simple words like the articles “a” and “the” (Karmiloff-Smith, 1979a). Children learn to use and interpret articles correctly in a number of their different appropriate contexts between the ages of 2 and 5. But at 5, they cannot explain the articles’ conditions of use. Yet the knowledge of conditions of use must be represented (at I-level, as knowledge still embedded in a procedure) to make it possible for young children to produce the subtle linguistic distinctions conveyed by the articles. However, the knowledge represented at that level is not available to them as consciously reportable data. By around 9 or 10, for this particular aspect of language, children have elaborate linguistic theories about the conditions of use of the articles and how they are related to other members of the nominal determiner system (Karmiloff-Smith, 1979a, 1986).
But between the correct usage for several functions at around 5 and the capacity for conscious verbal report about these different functions at around 10, children make self-repairs in their use of articles that bear witness to the fact that they have linked explicitly in their internal representations some of the knowledge that was previously embedded in independently stored procedures. I have explored in great detail the phenomenon of late-occurring repairs that are not in the child’s language earlier (Karmiloff-Smith, 1979a; see also Bowerman, 1982; Newport, 1981). Numerous examples have been gathered of the progression from efficiently functioning procedures to repairs denoting that new representations have been formed about the underlying linguistic subsystem and, finally, to children’s capacity for verbal report (Karmiloff-Smith, 1979a, 1986).
To have knowledge available to conscious access requires yet another level of redescription. My explanation of why this takes time developmentally is that conscious access is at first constrained by the fact that the knowledge is still in the code in which it was represented at the previous level. Thus, it may not always be available for verbal report. It is here that my disagreement with Fodor is highlighted, for in my view, the system does not transform all input into a common, propositional representational system—that is, a common language of thought. Rather, the translation process is constrained by the multiplicity of representational codes (spatial, linguistic, kinaesthetic, etc.) available to the human mind.
I submit that each input system, as opposed to input analyzer (Fodor uses the terms as synonymous), has its own representational code and its own dedicated aspects of central-like processing. A simple dichotomy between modular and central processing is too rigid to account adequately for the architecture of the human mind, either in its developing or adult states. New work in progress on William’s syndrome children (Bellugi & Karmiloff-Smith, in preparation) has shown that when an exceedingly high level of linguistic output is achieved, these children go on to develop central-like processing in their linguistic input system and thus manifest some metalinguistic knowledge, despite their very low IQs.
It is only at the final level of redescription that all products are sent to a common, centralized processor. Then, represented knowledge becomes verbally stateable and thereby communicable to others. It is at this level of representational redescription that the different representational codes compete to bring representations into a common format. The linguistic one frequently wins out because it can be used to actually violate the temporal, spatial, and causal constraints inherent in the other representational codes (e.g., one can express in language time and space relations that are in the wrong sequence).
It is possible that knowledge that the child receives from the sociocultural environment, already in the form of verbally encoded information, is processed directly at the top level and is immediately available to conscious access (as suggested in Mandler, 1988). How such directly linguistically encoded representations are ultimately brought into relationship with existing procedurally encoded representations remains an open question, inasmuch as Hennessy, for example, has shown that children can successfully learn elaborate mathematical principles in linguistically stateable form, but that for several years such principles do not constrain the procedures that children use to actually solve mathematical problems (Hennessy, 1986).
Finally, it should, of course, be recalled that natural language is also constrained in its expressibility, so there is again a loss of information that continues to be represented in other codes. As humans, however, we constantly transcend the constraints of our current powers of reasoning, and, in the case of natural language, we have invented formal languages to surmount the constraints on expressibility of natural languages.
Does a similar pattern of progressive conscious access hold for the entire linguistic system? A recent series of experiments has focused on metalinguistic awareness of what might be called the “processing instructions” aspect of language (Karmiloff-Smith, Johnson, Grant, Jones, Karmiloff, Bartrip, & Cuckle, 1989). This is in contrast to the propositional content of utterances. Both aspects of language are simultaneously encoded by speakers and decoded by addressees. By the processing instructions aspect of language, I have in mind discourse organizational markers (terms like pronouns and full noun phrases in their discourse functions of denoting the thematic structure of a discourse, and aspectual marking on verbs in its function of marking foregrounding and backgrounding in a span of discourse).
Take the pronoun “he” as a brief example of how a linguistic marker functions both at the level of propositional content and at the discourse organizational level. From the point of view of the semantics of propositional content, the pronoun provides information at the sentential level about a referent being human, male, singular, and so forth. But in the on-line flow of discourse, the use of the pronoun also conveys information about the speaker’s mental model of the overall discourse structure, about whether the referent can be taken by default to be the main protagonist, and so forth.
The results of the experiment show that in the allocation of shared computational resources, the discourse organization aspect of language turns out to be unavailable to conscious access, whereas the semantics of the propositional content can be reflected on metalinguistically. Indeed, subjects can give elaborate explanations about, say, the pronoun “he” and how it relates to “she,” “they,” and “it” but almost nothing about the function of pronouns at the discourse level. This new research suggests that the aspect of language involving implicit instructions to the addressee on how to process an utterance within the global discourse structure, rather than the local semantics of its propositional content, is not available to conscious access in children or adults (Karmiloff-Smith et al., 1989).
Discourse organizational structure is acquired rather late in development and is unlikely to be part of the innate endowment in the same way as sentential syntax. Moreover, unlike the deterministic rules of sentential syntax, discourse rules are probabilistic. The feedback from each decision regarding the choice of a discourse marker becomes the input for the generation of the next choice, that is, a closed loop control. It is also important to note that the marking of discourse structure shows several signs of having undergone the first level of representational redescription that can be gleaned from children’s late-occurring self-repairs (Karmiloff-Smith, 1985). Moreover, when discourse cohesion markers first appear consistently in children’s output, the propositional content of their output decreases compared to younger subjects (Karmiloff-Smith, 1985; Karmiloff-Smith et al., 1989). So, it looks as if there are definite constraints on subsequent conscious access due to competition in the allocation of processing resources for encoding and decoding a fast-fading message in real time.
Development thus seems to involve two opposite processes: on the one hand, the progressive access to knowledge previously embedded in procedures and, on the other hand, the progressive modularization of parts of the system. But if discourse organizational processes are modular-like, then how does the psycholinguist or linguist have access to that part of language inaccessible to normal adults and children? A necessary prerequisite to such access is to re-represent the language and change the spoken code to a written form with a trace, that is, freeze the fast-fading message of spoken language into the static form of written language. It is impossible for the psycholinguist to analyze, on-line, the discourse organizational markers of spoken language output. Both speaker and listener have no conscious access to it. In a modular-like fashion, it runs off automatically and is cognitively inpenetrable.
Earlier work within both language and cognitive development has rendered plausible the model of representational redescription and how there is progressive conscious access in some cases and a process of modularization in other cases (Karmiloff-Smith, 1979a, 1979b, 1984, 1986). But what has not been addressed thus far is the nature of the constraints on the first level of representational redescription. What constraints obtain when the child moves from knowledge embedded in a procedure to redescribing that knowledge, thereby making it accessible as a data structure to the system? Recent research on children’s drawing has focused directly on this issue (Karmiloff-Smith, 1990).
The same general research strategy was used, by selecting an age group of children who already have efficiently functioning procedures in the particular domain of interest. Children between 4½ and 10 years of age were asked to make drawings of a house and then to draw a house that doesn’t exist. The same procedure was used for man and for animal. The rationale for this design was as follows: By roughly 4½ years of age, children have efficiently functioning procedures for rapidly drawing houses, for example. Asking them to draw a house that doesn’t exist should force them into operating on the knowledge embedded in their procedures.
My hypothesis was that there would be an initial period during which the child could merely run a procedure in its entirety. This I-level knowledge, I hypothesized, would be followed by a period during which the child would be able to access, via E-1 representational redescription, some of the knowledge embedded previously in the procedure but that there would be constraints on how the flexibility was realized, that is, a movement from a sequence of instructions in which relatively fixed order is part of the redescribed specification (E-1 level) to a much more flexibly ordered list of features in which the sequential constraint is relaxed (E-2 level).
Let me very briefly run through some of the data. First, the hypothesis that, initially, children may only be able to run through a successful procedure but be unable to access the knowledge embedded in it (because represented at I-level) was borne out by the results of a few of the youngest subjects. Thus, although announcing that they were going to draw a silly/pretend house, they proceeded to run through their normal house-drawing procedure. In this chapter, we focus on those children who did achieve some flexibility (levels E-1 and E-2). What is of particular interest in this study is the analysis of the constraints that obtained on the types of change that children made.
Figures 6.1–6.3 illustrate the types of change that children of all age groups introduced. They involved the following:
• shape and size of elements changed;
• shape of whole changed;
• elements deleted.
Note that, in most cases, the changes do not involve interruption or re-ordering of the sequential constraints on the procedure. Although these three types of change were found in children of all ages, important differences emerged with respect to younger and older subjects who used deletion as a solution. We shall look at this once we have illustrated the other changes introduced.
Figures 6.4–6.6 illustrate far more flexible and creative solutions to the task. Changes here involved:
• insertion of elements from same conceptual category;
• position and orientation changed;
• insertion of elements from other conceptual categories.
These changes were found almost totally in older children only. They involve reordering of sequence, interruption and insertion of subroutine, using representations from other conceptual categories, and so on.
If we now reconsider Fig. 6.3 illustrating deletions, it can be noted that although children of all ages used deletions, this category showed particularly interesting differences between younger and older subjects. It is not possible to make a formal statistical analysis of the differences in the sequence of deletions between the two age groups, because video recordings were not taken. However, written notes were made on the protocols wherever they were deemed necessary (e.g. “added wings at the end of the drawing,” “left leg was last thing drawn,” etc.). These notes, together with a systematic analysis of what can be inferred from the product of many of the drawings, made it possible to assert with some assurance that the children from the older age group frequently made their deletions in the middle of their drawing procedure. By contrast, the younger age group tended to make deletions of elements that are drawn towards the end of a procedure, and they did not continue drawing after deletions. The subjects in a follow-up study, for which sequence was carefully recorded on a copy drawn on-line during their productions by a second experimenter, confirmed this tendency (see Karmiloff-Smith, 1990 for details).
FIG. 6.1. Shape and/or size of elements changed.
FIG. 6.2. Shape of whole changed.
FIG. 6.3. Deletion of elements.
FIG. 6.4. Insertion of new elements.
FIG. 6.5. Position/orientation changed.
FIG. 6.6. Insertion of cross-category elements.
The very few 4–6-year-olds who made changes classifiable in the last three categories (insertions of elements, position/orientation changes, cross-category insertions) all added elements after finishing a normal representation of the subject (e.g., by adding a chimney emerging horizontally from the side wall of a house, by adding a smile on a house, etc.). They did not make insertions into the middle of their drawing procedure, as did older children who, for instance, drew a man with two heads, which involves an insertion towards the beginning of the procedure.
FIG. 6.7.
The histogram in Fig. 6.7 shows the differences between the two age groups with respect to those changes constrained by sequence and those in which the sequential specification has been relaxed. The figures speak for themselves. This very striking developmental difference suggests that, initially, when children are able to work on a redescription of the knowledge embedded in a procedure, the new representation is not yet very flexible and sometimes a sequential constraint obtains. When children build up a new procedure, like learning to draw a house, they do so laboriously, working out the different elements, watching the way others perform, and so forth. The process can be long. Next, they start to consolidate the procedure so that it runs off more or less automatically, much in the manner of skill acquisition. Ask the child to draw a house, and now he or she can draw it easily and quickly. The child is calling on a relatively compiled procedure, which is specified sequentially in the internal representations. The results of this study suggest that although at the first level of redescription, the knowledge becomes accessible, it is also to some extent constrained by a sequential specification. By contrast, subsequent representations, having undergone further redescriptions, are far more flexible. By then, representations involve a flexibly ordered list of features that makes it possible for children to be creative about insertion of new elements, change of position, orientation, and so forth.
We have seen evidence in this new study and in previous work that suggests several characteristics of procedures (I-level representations) in normal children’s development: (a) they become automatized and run in a fixed sequence; (b) once initiated, a procedure must run in its entirety or, if interrupted, stops; (c) procedural representations are cognitively impenetrable; (d) knowledge embedded in procedures only become data structures to the system after redescription; and (e) initially, redescribed representations lack flexibility. Thus, in the drawing study, once 5-year-olds can reproduce drawings of houses effortlessly, their outputs show the hallmarks of a procedure as characterized previously. By contrast, an in-depth case study of a Down syndrome 9-year-old (Karmiloff-Smith & Grant, in preparation) showed that although he could produce efficient output (e.g., draw a house like a 5-year-old), his behavior was not consistent. A week later, his drawings of a house were closer to those of a 2-year-old. Thus it is questionable, even when his drawing product looked like that of a 5-year-old, that the processes used to produce it were controlled by a compiled procedure. In other words, not only could he not do the draw-an-X-that-does-not-exist task, but it is even doubtful that he had reached behavioral mastery, that is, that compilation of a normal house-drawing procedure had occurred. Further, unlike William’s syndrome children, many aspects of the Down syndrome child’s linguistic output showed a similar pattern of inconsistency and lack of representational redescription. Behavioral mastery thus seems to be a prerequisite for representational redescription.
Interestingly, the interpretation in terms of some sequential constraints on initial redescriptions helps to make sense of what turns out to be very similar phenomena from phonological awareness tasks done with both children and illiterate versus newly literate adults (Morais, Alegria, & Content, 1987). It is easier for subjects to delete final phonemes than to delete initial phonemes. Here, too, then, initially a sequential constraint obtains for children and for the newly literate, which is surmounted by the fully literate adult. This is not, therefore, just a developmental phenomenon. The movement from the sequential constraints of procedural encoding to the flexibility of subsequent knowledge seems to hold for new learning in adults, too.
It is important to note that the sequential specification does not only act as a curtailing constraint on flexibility. The fact that initial levels of representation are sequentially specified can, in fact, potentiate development in just those areas where sequence is an essential component—language, counting, and so forth. Whereas in each case, the child ultimately needs to surpass the sequential constraint, initially such a constraint actually potentiates learning by getting the system off the ground, getting representations into the mind.
Let me conclude by summarizing the main speculations explored in this chapter on innate constraints and developmental change:
1. Nativism and constructivism are not necessarily incompatible.
2. Piaget’s view of the initial state of the neonate mind was wrong. It is clear that, at the outset, some aspects of the human mind are innately specified, and often in some detail. Knowledge is initially domain-specific and constrains subsequent learning in complex interaction with the environment. It is not based solely on the outcome of domain-neutral sensorimotor representations. Subsequent development can be viewed within a constructivist framework.
3. Some innate knowledge is specified in precise detail, whereas other knowledge may be specified as a skeletal outline to be filled in by experience.
4. Innately specified knowledge is initially procedurally encoded, activated as a response to external stimuli, and is not available as data structures to other parts of the system. Apart from knowledge directly encoded linguistically, newly acquired knowledge is also initially represented procedurally.
5. The coherent sets of principles that underlie the infant’s procedures for interacting with the external environment and support inductions cannot be considered as innately specified “theories.” To have the status of a “theory,” these principles must be explicitly defined in the internal representations. This takes place via a process of representational redescription.
6. The child is a spontaneous theory builder (about language, physics, etc.) and exploits the innate and acquired knowledge that he or she has already stored via a process of representational redescription, in other words, by representing recursively his or her own representations. Once redescribed, knowledge becomes explicitly represented and thus available as data structures to the system for theory building and ultimately, in some cases, to conscious access.
7. Dichotomies such as implicit/explicit, unconscious/conscious, automatic/controlled, and the like are insufficient to capture the complexities of representational change. One needs to invoke several different levels, namely, a level of implicit knowledge represented but embedded in procedures, a level of explicitly defined knowledge but not available to conscious access, a level of consciously accessible representations but not available to verbal report, and a level available to verbal report. A developmental perspective is crucial to identifying these different levels of representation in the human mind.
8. Those aspects of development that are open to representational redescription seem initially to remain partially under a sequential constraint. Subsequently, knowledge is re-represented in the form of a flexibly ordered list, allowing for inter-representational relations and cognitive creativity.
9. Some aspects of development are clearly domain-specific and modular and have their own dedicated aspects of central-like processing. Others have more general implications for the system in domain-neutral central processing. But only some aspects of cognitive and linguistic development are available to consciousness. Others, despite developing rather late, undergo a process of modularization and are not available to conscious access, even for adults.
Although representational change can be accounted for in terms of cycles of domain-specific recurrent phases, it remains an open question as to whether a change in a belief built up from, say, linguistically encoded information and represented in central processing may be extended across the whole central system in a sort of domain-neutral stage-like change. A question that permeates the developmental literature is whether the mind is a general purpose, domain-neutral learning mechanism or whether knowledge is acquired and processed in a domain-specific fashion. I have italicized the exclusive “or” purposely, because I feel it is misplaced. It is doubtful that development will turn out to be entirely domain-specific or entirely domain-general. The more interesting developmental question for the future will hinge on the relative weights of constraints imposed by innate specifications, constraints that are domain-specific and those that are domain-general.
ACKNOWLEDGMENTS
This chapter is a slightly adapted version (alas, all the jokes have had to go!) of my keynote address of the same title to the Symposium of the Jean Piaget Society, Philadelphia, April 1988, and of my opening address to the Developmental Section Annual Meeting of the British Psychological Society, Harlech, September 1988. I thank Mike Anderson, Ellen Bialystok, Susan Carey, Rochel Gelman, Julia Grant, Mark Johnson, Jean Mandler, and Liliana Tolchinsky-Landsmann for comments.
Anderson, M. (1988). Inspection time, information processing and the development of intelligence. British Journal of Developmental Psychology, 6, 43–57.
Baillargeon, R. (1987). Young infants’ reasoning about the physical and spatial properties of a hidden object. Cognitive Development, 2, 179–200.
Bates, E., Thal, D., & Marchman, V. (1989). Symbols and syntax: a Darwinian approach to language development (Tech. Rep. No. 89–04). University of California at San Diego, Neurodevelopmental Research Center.
Bates, E. & MacWhinney, B. (1987). Competition, variation and language learning. In B. MacWhinney (Ed.), Mechanisms of language acquisition (pp. 157–193). Hillsdale, NJ: Lawrence Erlbaum Associates.
Bellugi, U., Marks, S., Bihrle, A.M., & Sabo, H. (1988). Dissociation between language and cognitive functions in William’s syndrome. In D. Bishop & K. Mogford (Eds.), Language development in exceptional circumstances (pp. 177–189). London: Churchill Livingstone
Borer, H. (1984). Parametric syntax. Dordrecht: Foris.
Bowerman, M. (1982). Reorganizational processes in lexical and syntactic development. In E. Wanner & L. R. Gleitman (Eds.), Language acquisition: The state of the art. Cambridge: Cambridge University Press.
Brown, A. C. (1990). Domain-specific principles affect learning. Cognitive Science, 14, (107–133).
Butterworth, G. (Ed.). (1981). Infancy and epistemology: An evaluation of Piaget’s theory. Brighton: Harvester Press.
Carey, S. (1985). Conceptual change in childhood. Cambridge, MA: MIT Press.
Chomsky, N. (1981). Lectures on government and binding. Dordrecht: Foris.
Chomsky, N. (1988). Language and problems of knowledge. The Managua lectures. Cambridge, MA: MIT Press.
Cohen, L. B. (1988). An information-processing approach to infant cognitive development. In L. Weiskrantz (Ed.), Thought without language (pp. 211–228). Oxford: Clarendon Press.
Cromer, R. F. (in press). A case study of dissociations between language and cognition. In H. Tager-Flusberg (Ed.), Constraints on language acquisition: studies ofatypical children. Hillsdale NJ: Lawrence Erlbaum Associates.
Diamond, A., & Gilbert, J. (1989). Development as progressive inhibitory control of action: Retrieval of a contiguous object. Cognitive Development, 4, 223–249.
Eilers, R. E., Bull, D. H., Oller, K., & Lewis, D. C. (1984). The discrimination of vowel duration by infants. Journal of the Acoustical Society of America, 75, 1213–1218.
Eimas, P. D., Siqueland, E. R., Jusczyk, P., & Vigorito, J. (1971). Speech perception in infants. Science, 171, 303–306.
Feldman, H., & Gelman, R. (1986). Otitis media and cognitive development. In J. F. Kavanagh (Ed.), Otitis media and child development (pp. 27–41). Parkton MD: York Press.
Fernald, A. (in press). Four-month-old infants prefer to listen to Motherese. Infant Behaviour and Development.
Fernald, A., & Kuhl, P. (1981, March). Fundamental frequency as an acoustic determinant of infant preference for Motherese. Paper presented at the meeting of the Society for Research in Child Development, Boston, MA.
Fodor, J. A. (1983). The modularity of mind. Cambridge, MA: MIT Press.
Fowler, C. A., Smith, M. R., & Tassinary, L. G. (1986). Perception of syllable timing by prebabbling infants. Journal of the Acoustical Society of America, 79, 814–825.
Gelman, R. (1986). First principles for structuring acquisition. Paper presented at the meeting of the American Psychological Association, New York.
Gleitman, L. R., Gleitman, H., & Shipley, E. F. (1972). The emergence of the child as grammarian. Cognition, 1, 137–164.
Gleitman, L. R., Gleitman, H., Landau, B., & Wanner, E. (1988). Where learning begins: Initial representations for language learning. In F. Newmeyer (Ed.), The Cambridge linguistic survey (pp. 150–193). New York: Cambridge University Press.
Gould, J. L. (1982). Ethology. New York: Norton.
Hennessy, S. (1986). The role of conceptual knowledge in the acquisition of arithmetic algorithms. Unpublished doctoral dissertation, MRC-CDU, London.
Hirsh-Pasek, K., Kemler-Nelson, D. G., Jusczyk, P. W., Wright, K., & Druss, B. (1987). Clauses are perceptual units for young children. Cognition, 26, 269–286.
Hyams, N. (1986). Language acquisition and the theory of parameters. Dardrecht: Reidel.
Johnson, J., Fabian, V., & Pascual-Leone, J. (1989). Quantitative hardware stages that constrain language development. Human development, 32, 245–271.
Johnson, M. H. (1988). Memories of mother. New Scientist, 18, 60–62.
Johnson, M. H. (1990). Cortical maturation and the development of visual attention in early infancy. Journal of Cognitive Neuroscience, 2, 81–95.
Johnson, M. H. (in press). Cortical maturation and perceptual development. In H. Bloch & B. Bertenthal (Eds.), Sensory motor organisation and development in infancy and early childhood. Dordrecht: Kluwer Academic Press.
Johnson, M. H., Dziurawiec, S., Ellis, H. D., & Morton, J. (in press). Newborns preferential tracking of face-like stimuli and its subsequent decline. Cognition.
Johnson, M. H. & Karmiloff-Smith, A. (1989). Neural development and language acquisition: Parcellation and selective stabilization. Manuscript submitted for publication.
Johnson, M. H., & Morton, J. (in press). The development of face recognition: A biological approach to cognitive development. Oxford: Blackwells Ltd.
Jusczyk, P. W. (1986). Speech perception. In K. R. Boff, L. Kaufman & J. P. Thomas (Eds.), Handbook of perception and human performance: Vol. 2. Cognitive processes and performance (pp. 27:1–57). New York: Wiley.
Jusczyk, P. W., & Bertoncini, J. (1988). Viewing the development of speech perception as an innately guided learning process. Language and Speech, 31, 217–238.
Jusczyk, P. W., & Cohen, A. (1985). What constitutes a module? The Behavioral and Brain Sciences, 8, 20–21.
Karmiloff-Smith, A. (1979a). A functional approach to child language. Cambridge, UK: Cambridge University Press.
Karmiloff-Smith, A. (1979b). Micro- and macro-developmental changes in language acquisition and other representation systems. Cognitive Science, 3 (2), 91–118.
Karmiloff-Smith, A. (1984). Children’s problem solving. In M. E. Lamb, A. L. Brown, & B. Rogoff (Eds.), Advances in developmental psychology (Vol. 3, pp. 39–90). Hillsdale, NJ: Lawrence Erlbaum Associates.
Karmiloff-Smith, A. (1985). Language and cognitive processes from a developmental perspective. Language and Cognitive Processes, 1, 61–85.
Karmiloff-Smith, A. (1986). From meta-processes to conscious access: evidence from children’s metalinguistic and repair data. Cognition, 23, 95–147.
Karmiloff-Smith, A. (1989). Commentary. Human Development, 32, 272–275.
Karmiloff-Smith, A. (1990). Constraints on representational change: evidence from children’s drawing. Cognition, 34, 57–83.
Karmiloff-Smith, A. (in press). Beyond modularity: A developmental perspective on cognitive science. Cambridge, MA: MIT Press.
Karmiloff-Smith, A., Johnson, H., Grant, J., Jones, M. C., Karmiloff, Y. N., Bartrip, J., & Cuckle, P. (1989). Detection and explanation of speech repairs by children and adults. Manuscript submitted for publication.
Keil, F. C. (1986). On the structure-dependent nature of stages of cognitive development. In I. Levin (Ed.), State and structure: Reopening the debate (pp. 144–163). Norwood NJ: Ablex.
Kellman, P. J. (1988). Theories of perception and research in perceptual development. In A. Yonas (Ed.), Perceptual development in infancy: The Minnesota symposium on child psychology (Vol. 20, pp. 267–281). Hillsdale, NJ: Lawrence Erlbaum Associates.
Kuhl, P. K. (1985). Categorization of speech by infants. In J. Mehler & R. Fox (Eds.), Neonate cognition: Beyond the blooming, buzzing confusion (pp. 231–262). Hillsdale, NJ: Lawrence Erlbaum Associates.
Leslie, A. M. (1984). Infant perception of a manual pick-up event. British Journal of Developmental Psychology, 2, 19–32.
Leslie, A. M. (1987). Pretense and representation: The origins of “Theory of Mind.” Psychological Review, 94, 412–426.
Mandler, J. M. (1983). Representation. In J. Flavell & E. Markman (Eds.), Handbook of child psychology (Vol. 3, pp. 420–495). New York: Wiley.
Mandler, J. M. (1988). How to build a baby: On the development of an accessible representational system. Cognitive Development, 3, 113–136.
DeMany, L., McKenzie, B., & Vurpillot, E. (1977). Rhythm perception in early infancy. Nature, 266, 718–719.
Mehler, J. & Bertoncini, J. (1988). Development: a question of properties, not change? International Social Science Journal, Cognitive Science, 115, 121–133.
Mehler, J., & Fox, R. (1985). Neonate cognition: Beyond the blooming buzzing confusion. Hillsdale, NJ: Lawrence Erlbaum Associates.
Mehler, J., Jusczyk, P., Lambertz, G., Halsted, N., Bertoncini, J., & Amiel-Tison, C. (1988). A precursor of language acquisition in young infants. Cognition, 29, 143–178.
Morais, J., Alegria, J., & Content, A. (1987). The relationships between segmental analysis and alphabetic literacy: An interactive view. Cahiers de Psychologie Cognitive, 7, 5, 415–438.
Mounoud, P. (1986). Similarities between developmental sequences at different age periods. In I. Levin (Ed.), State and structure: Reopening the debate (pp. 40–58). Norwood NJ: Ablex.
Newport, E. L. (1981). Constraints on structure: Evidence from American Sign Language and language learning. In W. A. Collins (Ed.), Aspects of the development of competence: Minnesota symposia on child psychology (Vol. 14, pp. 93–124). Hillsdale, NJ: Lawrence Erlbaum Associates.
Rozin, P. (1976). The evolution of intelligence and access to the cognitive unconscious. In J. M. Sprague & A. A. Epstein (Eds.), Progress in psychobiology and physiological psychology (Vol. 6, pp. 245–280). New York: Academic Press.
Rutkowska, J. C. (1987). Computational models and developmental psychology. In J. C. Rutkowska & C. Crook (Eds.), Computers, cognition and development, Chichester: Wiley.
Sinclair, H. (1971). Sensori-motor action patterns as the condition for the acquisition of syntax. In R. Huxley & E. Ingrams (Eds.), Language acquisition: Models and methods (pp. 121–135). New York: Academic Press.
Sinclair, H. (1987). Language: A gift of nature or a home-made tool? In S. Modgil & C. Modgil (Eds.), Noam Chomsky: Consensus and controversy (pp. 173–180). London: The Falmer Press, 173–180.
Slater, A., Morison, V., & Rose, D. (1983). Perception of shape by the new-born baby. British Journal of Developmental Psychology, 1, 135–142.
Spelke, E. S. (1982). Perceptual knowledge of objects in infancy. In J. Mehler, E. Walker & M. M. Garrett (Eds.), Perspectives on mental representation (pp. 409–430). Hillsdale, NJ: Lawrence Erlbaum Associates.
Spelke, E. S. (1990). Principles of object perception. Cognitive Science, 14, 29–56.
Spring, D. R., & Dale, P. S. (1977). Discrimination of linguistic stress in early infancy. Journal of Speech and Hearing Research, 20, 224–232.
Starkey, P., & Cooper, R. G. (1980). Perception of number by human infants. Science, 200, 1033–1035.
Sullivan, J. W. & Horowitz, F. D. (1983). The effects of intonation on infant attention: The role of the rising intonation contour. Journal of Child Language, 10, 521–534.
Thal, D., Bates, E. & Bellugi, U. (1989). Language and cognition in two children with William’s syndrome. Journal of Speech and Hearing Research, 32, 489–500.
Valian, V. (1990). Positive evidence, indirect negative evidence, parameter-setting and language learning. Cognition, 35, 105–122.
