LANGUAGE AND THE INFANT BRAIN ELIZABETH BATES University of California, San Diego

Three logically and empirically independent issues are often conflated in theory and research on brain and language: localization, innateness, and domain specificity. Research on adults and infants with focal brain injury support the following conclusions: (a) linguistic knowledge is not innate, and it is not localized in a clear and compact form in either the infant or adult brain; (b) the infant brain is not, however, a tabala rasa—it is already highly differentiated at birth, and certain regions are biased from the beginning toward modes of information processing that are particularly useful for language, leading (in the absence of local injury) to the standard form of brain organization for language; (c) the processing biases that lead to the “standard brain plan” are innate and localized, in both infants and adults, but they are not specific to language; and (d) the infant brain is highly plastic, permitting alternative “brain plans” for language to emerge if the standard situation does not hold. © 1999 by Elsevier Science Inc. Educational Objectives: The reader will understand how insights from neurobiology illuminate the process of language acquisition. KEY WORDS: Language acquisition, Brain development, Brain damage, Childhood aphasia

All reasonable scholars today agree that genes and environment interact to determine complex cognitive outcomes. So why does the nature-nurture controversy persist, and why does it play such an important role in the study of language and language development? First, the controversy persists because it has practical implications that cannot be postponed (i.e., what can we do to avoid bad outcomes and insure better ones?). This is certainly the case for speech/language pathologists, who must deal on a daily basis with children who have failed to develop language on a normal schedule. Second, the controversy persists because we lack a precise, testable theory of the process by which genes and environment interact—particularly within the fields that study language, a behavioral domain with no obvious analogues in the species that have been studied in such detail by neuroscientists and molecular biologists. Address correspondence to Dr. E. Bates, University of California, San Diego, 9500 Gilman Drive, La Jolla, CA 92093-0526. Phone: (619) 534-3007; Fax: (619) 534-7190; E-mail: .

J. COMMUN. DISORD. 32 (1999), 195–205 © 1999 by Elsevier Science Inc. All rights reserved. 655 Avenue of the Americas, New York, NY 10010

0021-9924/99/$–see front matter PII S0021-9924(99)00015-5

196

BATES

As my colleagues and I have argued elsewhere (Elman, Bates, Johnson, Karmiloff-Smith, Parisi, & Plunkett, 1996), an explicit and detailed interactionist account of behavioral development may be just around the corner, for three reasons. First, developmentalists have begun to make use of insights from nonlinear dynamics (Elman et al., 1996; Thelen & Smith, 1994). This is the latest and perhaps the last frontier of theoretical physics, offering insights into the processes by which complex, surprising, and apparently discontinuous outcomes can arise from small quantitative changes along a single dimension. Second, it is now possible to simulate behavioral change in multilayered neural networks, systems that embody the nonlinear dynamical principles required to explain the emergence of complex solutions from simpler inputs (Elman et al., 1996; Rumelhart & McClelland, 1986). Third, students of behavioral development are becoming aware of some remarkable breakthroughs in developmental neurobiology (Edelman, 1987; Elman et al., 1996; Johnson, 1997). These neurobiological results are bad news for yesterday’s nativists, because they underscore the extraordinarily plastic and activity-dependent nature of cortical specialization and buttress the case for an experience-dependent account of the development of higher cortical functions. Although the desired theory is around the corner, it is still not fully articulated. In the absence of an explicit theory of gene-environment interaction, researchers often confound three theoretically and empirically independent issues: innateness—a construct that is often left undefined; localization—the belief that an outcome must be innate because it is mediated by a particular part of the brain; and domain specificity—the belief that an outcome must be innate because it is universal but so arbitrary and idiosyncratic that it is hard to understand how it could be learned. In fact, innateness and localization are not the same thing. For example, there are areas of the brain that “light up” in neural imaging studies during chess, and yet we know that chess is not innate (Nichelli, Grafman, Pietrini, Alway, Carton, & Miletich, 1994). Anything that we know, whether it is innate or acquired, must be represented somewhere in the brain, so that (in principle) it should be possible to access that knowledge and visualize its activation through some kind of neural imaging technique. Nor are we justified in concluding that an outcome must be innate because it is peculiar. Language is universal and it certainly contains many idiosyncratic features, but these features could have emerged inevitably from the peculiar nature of the problem that language is intended to solve (i.e., mapping a high-dimensional set of universal meanings onto a low-dimensional channel). To offer one example from a completely different domain, the visual cortex of most adult mammals contains a pattern of alternating stripes called ocular dominance columns, representing alternating input from each eye. This is a universal outcome in many species, but it isn’t at all obvious how it could be learned, because it does not exist “out there” in the world and its relation to

LANGUAGE AND INFANT BRAIN

197

the problem of visual perception is not at all obvious. But this does not mean that ocular dominance columns are under direct genetic control. We now know that these columns are the inevitable result of competition early in development (in the prenatal period in many species) between two eyes with overlapping visual fields (Miller, 1994; Shatz, 1996). Ocular dominance columns do not develop at all in cats if this competition is precluded by blocking the input from one eye in utero, and they never appear in frogs, because the frog’s eyes are set so far apart that the visual fields do not overlap in the brain, and no competition takes place. However, by surgically inserting a third eye between the two eyes of a developing frog, researchers have forced a competition; as a result, the frog develops ocular dominance columns similar to those seen in mammals, despite millions of years of evolution in which frogs did without them altogether (Reh & Constantine-Paton, 1985). The presence or absence of ocular dominance columns has also been simulated in artificial neural networks, by varying the degree of competition from two artificial eyes. The point is that this utterly peculiar structure, whose relationship to visual success is not at all obvious, arises inevitably whenever animals or neural networks have to solve this visual competition problem. That which is inevitable does not have to be innate. A similar story can be told for language. To tell this story, we need to do two things: define what it means for anything (including language) to be innate, and provide evidence suggesting that genetic contributions to language are very indirect.

WHAT DOES IT MEAN TO SAY THAT LANGUAGE IS INNATE? As a first approximation, we can define innateness as a claim about the amount of information in a complex outcome that is contributed by the genes (keeping in mind, of course, that genes do not act independently, and that they can be turned on and off by environmental signals throughout the lifetime of the organism). Elman et al. have proposed a 3-level taxonomy of claims about innateness, ordered from strong to weak with regard to the amount of information that must be contributed by the genes for this claim to work. Each level is operationally defined in terms that correspond to real brains and to artificial neural networks, as follows: Representational constraints refer to a direct innate structuring of the mental/neural representations that underlie and constitute knowledge, operationally defined in terms of direct genetic control over detailed patterns of synaptic connectivity at the cortical level. Strong nativist claims about language (Fodor, 1983; Pinker, 1994), physics (Spelke, 1991) or social reasoning (Horgan, 1995; Leslie, 1994) have to assume representational nativism, implicitly or explicitly, because that is the only level with the required coding

198

BATES

power for the implementation of detailed knowledge that is independent of experience. But this is not the only logically possible way for innate outcomes to be achieved. Architectural constraints refer to innate structuring of the informationprocessing system that must acquire and/or contain these representations. Although representation and architecture are not the same thing, there is no question that the range of representations a system can take is strongly constrained at the architectural level. In traditional serial digital computers, some programs can only run on a machine with the right size, speed and power. In neural networks, some forms of knowledge can only be realized or acquired in a system with the right structure (the right number of units, number of layers, types of connectivity between layers, etc.). To operationalize architectural constraints, Elman et al. break things down into three sublevels, carefully defined to correspond to operational definitions in both real brains and neural networks: • Basic computing units. This sublevel refers to neuronal types, their firing thresholds, neurotransmitters, excitatory/inhibitory properties, etc. • Local architecture. This sublevel refers to regional factors like the number and thickness of layers, density of different cell types within layers, type of neural circuitry (e.g., with or without recurrence). • Global architecture. This sublevel includes gross architectural facts like the characteristic sources of input (afferent pathways) and patterns of output (efferent pathways) that connect brain regions to the outside world and to one another. Chronotopic constraints refer to innate constraints on the timing of developmental events. In the developing brain, this would include constraints on the number of cell divisions that take place in neurogenesis, spatio-temporal waves of synaptic growth and pruning, and relative differences in timing between subsystems (e.g., differences among vision, audition, etc. in the timing of thalamic innervation of the cortex). These levels of architectural and chronotopic constraint can have massive consequences for computation, that is, for the kinds of knowledge that a system will be able to learn. However, it is very important to underscore that such constraints are not the same thing as innate knowledge. The genetic control that is required to increase the probability of learning is far less, and far less direct, than the genetic control required to specify the same knowledge without learning. Representational nativism would require a huge investment of genetic information—probably more than we have available when we keep in mind that 100,000 genes have to determine the whole body as well as a brain with approximately 100,000,000,000,000 synaptic connections. By contrast, architectural and chronotopic nativism both involve much smaller genetic

LANGUAGE AND INFANT BRAIN

199

contributions to achieve the same ends, relying heavily on information in the environment, and on the processes of learning and development. Evolutionary biologists who study the processes by which genes operate to build real tissue have pointed out that these developmental processes are highly constrained, placing powerful limitations on the range of outcomes that are possible or “evolvable” (Gerhardt & Kirschner, 1998). To get around these limitations, evolution is forced to be very conservative, reusing the same genes over and over, building new functions through minor quantitative tuning of a common body plan (and a common brain plan) shared by virtually all vertebrates, and most invertebrates as well (Gerhardt & Kirschner, 1998). If we want to understand how language evolved and how it emerges in the lifetime of an individual, we are probably going to have to cast our theory primarily in terms of architectural and chronotopic constraints.

IN WHAT WAY IS LANGUAGE INNATE? A theory of the development of brain organization for language based on architectural and chronotopic constraints can be drawn in part from research findings on adults and infants with focal brain injury. Four conclusions have emerged from our work and that of other investigators in this area: 1. Linguistic knowledge is not localized in a clear and compact form in either the infant or the adult brain. 2. Left-hemisphere injuries are not associated with aphasia if those injuries occur early in life. Instead, the infant brain is highly plastic, permitting alternative “brain plans” for language to emerge if the standard situation does not hold. 3. However, the infant brain is not a tabula rasa; it is already highly differentiated at birth, and certain regions are biased from the beginning toward modes of information processing that are particularly useful for language, leading (in the absence of focal injury) to the standard form of brain organization for language. 4. The processing biases that lead to the “standard brain plan” may be both innate and localized, in both infants and adults, but they are not specific to language. The first conclusion, that linguistic knowledge is broadly distributed even within the adult brain, is based on several recent findings in the aphasia literature (for reviews, see Bates & Wulfeck, 1989; Bates, Wulfeck, & MacWhinney, 1991), including (a) studies showing that patients with so-called agrammatic aphasia can make surprisingly fine-grained judgments of grammaticality; (b) cross-linguistic studies of adult aphasia, showing that detailed languagespecific knowledge is preserved and evident in both expressive and receptive language abilities, despite serious deficits in real-time processing; and (c) a

200

BATES

large and growing literature showing that expressive and receptive agrammatism are not specific to any particular aphasia group or lesion site—indeed, many highly specific grammatical deficits can also be demonstrated in normal listeners when they are forced to process language under stress. These results argue strongly against a localized “mental organ for grammar.” Explanations for aphasic symptoms must lie instead in the (partially localized) perceptual, attentional, motor and memory processes by which this distributed knowledge is activated and deployed in real time. Support for the other three conclusions comes from studies of infants and children with left- or right-hemisphere injuries acquired before 6 months of age (for reviews, see Bates, in press; Bates, Vicari, & Trauner, in press; Eisele & Aram, 1995; Stiles, Bates, Thal, Trauner, & Reilly, 1998; Vargha-Khadem, Isaacs, & Muter, 1994). Despite all the evidence for distributed knowledge cited above, aphasia is strongly associated with damage to the left hemisphere in adults. And yet we have known for some time that aphasia does not occur in children who suffer unilateral injuries to the left or right hemisphere early in life. Despite initial delays (which prove that the brain is not truly “equipotential” for language—see below), children with early focal brain injury almost invariably go on to achieve language abilities within the normal range. This result for human language is compatible with a large and growing literature on cortical plasticity in other species (for reviews, see Elman et al., 1996; Johnson, 1997). Although the language outcomes for children with early focal brain injury are generally very good (albeit slightly below the outcomes obtained with uninjured controls), we have found that most of these children experience moderate to severe initial delays. Perhaps more interesting and more important for our purposes here, there is evidence for specific lesion-symptom correlations if we look at the very earliest stages of language development, before a new “brain plan for language” has a chance to emerge. The nature of these initial deficits suggests that the regional specialization for language is built upon local variations in computational style—variations that are only indirectly related to language content, and do not map consistently onto lesion-behavior correlations in adults. Some surprising findings from our work on early focal brain injury include the following (Thal, Marchman, Stiles, Aram, Trauner, Nass, & Bates, 1991; ]Bates, Thal, Trauner, Fenson, Aram, Eisele, & Nass, 1997; Reilly, Bates, & Marchman, 1998): (a) evidence that very early comprehension and gesture are more affected by right-hemisphere damage (a direct contradiction of findings for adults); (b) evidence that expressive vocabulary is affected more by left temporal injuries throughout (but not beyond) the period from 10–60 months of age; (c) related evidence that expressive grammar does not dissociate from vocabulary in this age range, and is affected by the same lesions that compromise vocabulary production; (d) a brief and surprising period from 19–30

LANGUAGE AND INFANT BRAIN

201

months when injuries to either left or right frontal areas result in marked delays in vocabulary and grammar; (e) no evidence at any point in early development that left frontal areas play a privileged role; and finally (f) the complete disappearance of side—or site—specific effects after 5 years of age, when most children with early focal brain injury are performing well within the normal range. Our explanation for these complex findings draws on parallels between the deficits observed in early language and the deficits observed in the same children in visual-spatial cognition, suggesting that regions of cortex are specialized for modes of information processing that have a similar effect in linguistic and nonlinguistic domains. For example, rapid and detailed perceptual processing, both auditory and visual, is disrupted by lesions to the left temporal cortex). This is the kind of processing that children must apply in order to analyze the speech signal well enough to reproduce it themselves (“perception for production”). By contrast, early damage to the right hemisphere has its greatest impact on the integration of multiple levels of a pattern, or multiple sources of information, in both linguistic and nonlinguistic domains. This kind of analysis is not necessary for adults to comprehend a word that they have known for a long time, but it is very helpful in the first stages of word comprehension, when children have to achieve those first mappings of sound and meaning (“perception for meaning”). During the early stages of language learning, regions compete to “attract” the language task. Under default circumstances (i.e., in the absence of brain injury), particular regions are more likely to gain control over various parts of the language task. But when default circumstances do not hold, a number of different forms of brain organization are possible, and most of them appear to work very well. Clearly there has to be something innate and unique about the human brain to explain why language emerges in our species, and in our species alone. Even within an interactionist view of this kind, one has to start somewhere. However, we have proposed that these constraints are architectural and spatiotemporal in nature, and only indirectly related to the domain-specific representations for language that are ultimately acqured. This conclusion contrasts markedly with the phrenological view, and with strong nativist proposals for linguistic innateness. For example, Noam Chomsky (1975) has proposed that “linguistic theory, the theory of UG [Universal Grammar] . . . is an innate property of the human mind” (p. 34), and that we should conceive of “the growth of language as analogous to the development of a bodily organ” (p. 11). The mental organ metaphor leaves little room for learning. Indeed, Chomsky (1980) has argued that “a general learning theory . . . seems to me dubious, unargued, and without any empirical support” (p. 110). This kind of representational nativism was theoretically plausible and attractive when it was first proposed thirty years ago, but it has proven hard to defend against new results from developmental neurobiology, including results for children with early focal brain injury.

202

BATES

Because the evidence is not good for strong, representational forms of nativism, the differences that we observe from one species to another must be captured primarily by architectural and chronotopic facts. The final product emerges from the interaction between these constraints and the specific problems that an organism with such structure encounters in the world. The research described here was supported by NIDCD 2 R01-DC99216 (Cross-linguistic Studies of Aphasia), NIDCD P50 DC012389-0351 (Origins of Communicative Disorders) and NINDS P50 NS22343 (Center for the Study of the Neurological Bases of Language & Learning). Portions of this manuscript were adapted from Bates, Elman, Johnson, Karmiloff-Smith, Parisi, & Plunkett (1998).

REFERENCES Bates, E. (in press). Plasticity, localization, and language development. In S.H. Broman & J.M. Fletcher (Eds.), The changing nervous system: Neurobehavioral consequences of early brain disorders. New York: Oxford University Press. Bates, E., Elman, J., Johnson, M., Karmiloff-Smith, A., Parisi, D., & Plunkett, K. (1998). Innateness and emergentism. In W. Bechtel & G. Graham (Eds.), A companion to cognitive science (pp. 590–601). Malden, MA and Oxford: Blackwell Publishers. Bates, E., Thal, D., Trauner, D., Fenson, J., Aram, D., Eisele, J., & Nass, R. (1997). From first words to grammar in children with focal brain injury. In D. Thal & J. Reilly, (Eds.). Special issue on Origins of Communication Disorders. Developmental Neuropsychology, 13(3), 447–476. Bates, E., Vicari, S., & Trauner, D. (in press). Neural mediation of language development: Perspectives from lesion studies of infants and children. To appear in H. Tager-Flusberg (Ed.), Neurodevelopmental disorders: Contributions to a new framework from the cognitive neurosciences. Cambridge: MIT Press, in press. Bates, E. & Wulfeck, B. (1989). Comparative aphasiology: A cross-linguistic approach to language breakdown. Invited review article with peer commentary. Aphasiology, 3(2), 111–142. Bates, E., Wulfeck, B., & MacWhinney, B. (1991). Cross-linguistic studies of aphasia: An overview. Special issue on cross-linguistic studies of aphasia (E. Bates, Ed.). Brain and Language, 41(2), 123–148. Chomsky, N. (1975). Reflections on language. New York: Parthenon Press. Chomsky, N. (1980). On cognitive structures and their development: A reply

LANGUAGE AND INFANT BRAIN

203

to Piaget. In M. Piatelli-Palmarini (Ed.), Language and learning: The debate between Jean Piaget and Noam Chomsky. Cambridge: Harvard University Press. Edelman, G.M. (1987). Neural Darwinism: The theory of neuronal group selection. New York: Basic Books. Eisele, J., & Aram, D. (1995). Lexical and grammatical development in children with early hemisphere damage: A cross-sectional view from birth to adolescence. In P. Fletcher & B. MacWhinney (Eds.), The handbook of child language (pp. 664–689). Oxford: Basil Blackwell. Elman, J., Bates, E., Johnson, M., Karmiloff-Smith, A., Parisi, D., & Plunkett, K. (1996). Rethinking innateness: A connectionist perspective on development. Cambridge: MIT Press/Bradford Books. Fodor, J.A. (1983). The modularity of mind: An essay on faculty psychology. Cambridge: MIT Press. Gerhardt, J., & Kirschner, M. (1998). Cells, embryos & evolution. Oxford: Basil Blackwell. Horgan, J. (1995). The new Social Darwinists. Scientific American, 273(4), 174–181. Johnson, M.H. (1997). Developmental cognitive neuroscience: An introduction. Oxford: Oxford University Press. Leslie, A.M. (1994). Pretending and believing: Issues in the theory of Tomm. Cognition, 50(1–3), 211–238. Miller, K. (1994). A model for the development of simple cell receptive fields and the ordered arrangement of orientation columns thorugh activity-dependent competition between ON- and OFF-center inputs. Journal of Neuroscience, 14(1), 409–441. Nichelli, P., Grafman, J., Pietrini, P., Alway, D., Carton, J.C., & Miletich, R. (1994). Brain activity in chess playing. Nature, 369(6477), 191. Pinker, S. (1994). The language instinct: How the mind creates language. New York: William Morrow. Reh, T.A., & Constantine-Paton, M. (1985). Eye-specific segregation requires neural activity in three-eyed Rana pipiens. Journal of Neuroscience, 5, 1132–1143. Reilly, J., Bates, E., & Marchman, V. (1998). Narrative discourse in children with early focal brain injury. In M. Dennis (Ed.), Special issue, Discourse in children with anomalous brain development or acquired brain injury. Brain and Language, 61(3), 335–375.

204

BATES

Rumelhart, D., & McClelland, J. (1986). Parallel distributed processing: Explorations in the microstructure of cognition. Cambridge: MIT Press. Shatz, C. (1996). The emergence of order in visual system development. Proceedings of the National Academy of Sciences, 93, 602–608. Spelke, E.S. (1991). Physical knowledge in infancy: Reflections on Piaget’s theory. In S. Carey & R. Gelman, (Eds.), Epigenesis of the mind: Essays in biology and knowledge. Hillsdale, NJ: Erlbaum. Stiles, J., Bates, E., Thal, D., Trauner, D., & Reilly, J. (1998). Linguistic, cognitive and affective development in children with pre- and perinatal focal brain injury: A ten-year overview from the San Diego longitudinal project. In C. Rovee-Collier, L. Lipsit, & H. Hayne (Eds.), Advances in Infancy Research (pp. 131–163). Norwood, NJ: Ablex. Thal, D., Marchman, V., Stiles, J., Aram, D., Trauner, D., Nass, R., & Bates, E. (1991). Early lexical development in children with focal brain injury. Brain and Language, 40(4), 491–527. Thelen, E., & Smith, L.B. (1994). A dynamic systems approach to the development of cognition and action. Cambridge: MIT Press. Vargha-Khadem, F., Isaacs, E., & Muter, V. (1994). A review of cognitive outcome after unilateral lesions sustained during childhood. Journal of Child Neurology, 9(suppl.), 2S67–2S73.

CONTINUING EDUCATION Precursors to Speech in Infancy: The Prediction of Speech and Language Disorders QUESTIONS 1. The occurrence of ocular dominance columns in the visual system supports the notion that a. Fundamental aspects of brain development are “hard-wired” from birth. b. Experience with the environment can fundamentally affect the architecture of the brain. c. Genetics control most of what governs inter-species differences in the brain. d. Brain development in humans is entirely under genetic control. 2. Bates argues that the neurobiological underpinnings of language a. Are the result of brain characteristics unique to humans.

LANGUAGE AND INFANT BRAIN

205

b. Are the result of adaptation of general properties and functions of the brain. c. Prove the existence of a “mental organ for grammar.” d. Prove the infant brain starts out completely undifferentiated in structure and function. 3. The relationship between site-of-lesion and resulting deficits with early focal lesions suggests a. Both right and left hemisphere structures play important roles in the acquisition of language. b. The brain has initial biases in terms of how it will support language skills. c. Different regions of the brain may support acquisition of language rather than support language once it has been acquired. d. All of the above are true. 4. The course of aphasia in children a. Includes a relatively good long-term outcome if the lesion occurs early in life. b. Demonstrates the ability of the developing brain to reorganize itself to support language. c. Includes initial deficits that demonstrate damage does affect brain regions that support language acquisition. d. All of the above are true. 5. The phrase “chronotopic constraints” refers to the idea that a. The age at which a child learns language affects his or her success. b. Gradients in the timing of neurological events shape brain development. c. Development would take twice as long if it is entirely determined by genetics. d. There is a spectrum of cell types that can serve a specific function.