Psycholinguistics James Myers May 21, 2004 Nativism and modularity in language development OVERVIEW: 1. "Nature vs. Nurture": a false dichotomy 2. Nativism and modularity 3. The role of the environment 4. Nativist models 5. Positive arguments for nativism ============================================================= 1. "Nature vs. Nurture": a false dichotomy 1.1 The problem: We talk to our dogs all the time, and yet they never learn to talk; thus language is part of "human nature" and is thus innate; it is "programmed by our genes" and "hard-wired" in our brains. Babies born in Taiwan begin to speak Chinese, while babies born in the US begin to speak English; thus language must be "nurtured" in us by experience; it's a "learned behavior". 1.2 Two extreme positions: Nativism: emphasizes innate factors in cognitive development. Empiricism: emphasizes environmental factors in cognitive development. The wrong way to look at it: "What percentage of X (e.g. language) is innate, and what percentage is learned from experience?" The right way: "How do innate and environmental factors interact in (language) development?" 1.3 A classic example: the development of visual perception Of course the ability to see is innate, right? Yet many experiments have proven that experience with visual stimuli is necessary for the development of vision: e.g. cats raised in a visual environment containing only vertical lines become "blind" for horizontal lines. (Blakemore and Cooper 1970) Thus, vision is "hard-wired" but we also "learn to see". Experience shapes development by "laying down the wires"; the genes merely constrain how experience can affect this process. 1.4 What do genes do, anyway? Genotype: the genetic makeup of an organism (living thing) (i.e. the DNA: deoxyribonucleic acid) Phenotype: the overt properties of an organism that are produced by an interaction between the genotype and the environment. There is NO simple correspondence between genotype and phenotype: phenotypic complexity does not correlate with genotypic complexity. Apparently "simple" mammals often have many more genes than humans. In particular, the genetic difference between humans and chimpanzees is very very small (something like 2%). All the differences in size, hairiness, etc all have to be in there, in addition to any cognitive differences (like language). 1.5 So in what sense might language be "innate"? In principle, the genotype may interact with the environment at many different levels during the development of the brain: [discussion from Elman, et al. (1996)] 1.5.1 At one extreme is the level of Representation; that is, if language is innate, it is represented explicitly in the microcircuitry of the neurons in the brain; all experience can do is "fine-tune" this circuitry. "The grammar genes would be stretches of DNA that code for proteins, or trigger the transcription of proteins, in certain times and places in the brain, that guide, attract, or glue neurons into networks that, in combination with the synaptic tuning that takes place during learning, are necessary to compute the solution to some grammatical problem." (Pinker 1994:322) Problem: Are there enough genes to encode all the differences that humans and chimpanzees have in language, if so much linguistic knowledge is really innate at the Representational level? 1.5.2 At the other extreme is the level of Timing; genes that merely speed up or slow down development may not seem very important, but they can cause dramatic changes throughout the whole system (the "butterfly in China" effect). One example outside language: the human face. Which looks more like a person: an adult chimpanzee (with a face that points out, large brow, sloping forehead) or an infant chimpanzee (flatter face, small brow, rising forehead)? An example within language: the Elman (1992) connectionist model (see below). Problem: Without a detailed model of how Timing alone could give rise to complex grammar, this suggestion still feels more like wishful thinking. 1.5.3 Thus we can describe the nativism "debate" as concerning the level where genes have their greatest effect: nativists (e.g. Pinker) tend to believe in Representational innateness, while empiricists tend to believe in innateness at deeper, more general levels -- which automatically makes empiricists anti-modularist as well. 2. Nativism and modularity 2.1 These concepts are separate, and sort of independent: Innate things need not be modules (e.g. if the genes control brain development at the Timing level) Modules need not be innate: Nichelli, et al. (1994): Specific parts of the brain become active during specific parts of a chess game! 2.2 Classic evidence that language is a module (Pinker 1994): Double dissociations: A and B are distinct if one can find A without B and also B without A. 2.2.1 Nonlinguistic cognition with damaged linguistic cognition: Isolated children, e.g. "Genie", kept isolated from human language and contact for her first 13 years: Applesauce buy store. Genie have Momma have baby grow up. I like elephant eat peanut. [However, it's still unknown if Genie's nonlinguistic cognition is completely normal] Aphasia: brain-damage causes consistent language deficits, but nonlinguistic cognition may be spared. 2.2.2 Normal linguistic cognition with impaired nonlinguistic cognition: William's syndrome: caused by a defective gene on chromosome 11 which regulates calcium, affecting the brain, skull and internal organs. It causes significant mental retardation (IQ = 50), so they get easily lost, can't tell left from right, etc. Yet language is completely normal, even "hyper-grammatical": This is a story about chocolates. Once upon a time, in Chocolate World there used to be a Chocolate Princess. She was such a yummy princess.... [Notice that even discourse structure seems to be OK!] 2.3 "Innateness" and "modularity" can interact: a property X affecting language may be innate, but X need not be itself part of the language module. Example: *rats-eater (no regular plurals allowed inside compounds) Gordon (1986): even 3- to 5-year-old English speakers "know" this constraint; their parents virtually never use "N-Ver" compounds at all, so they couldn't have learned it (the "poverty of the stimulus" argument; see below); therefore this constraint must be innate. But WHAT is innate? Is it a specific part of the language module (as Pinker 1994 suggests), or is it an innate, but non-modular cognitive principle? Operating principles (Slobin 1973): children's preferred ways of operating on information (not just language), e.g. "Pay attention to the ends of words." Thus the contrast between "rat-eater" and "rats-eater" is harder to perceive than "rat-eater" vs. "rat-eaters"; even if the parents violated the constraint, kids may still ignore the internal "s" due to a general cognitive bias. 2.4 Modularity in language development: 2.4.1 Is linguistic development EXPLAINED by general cognitive development? (That is, is nonlinguistic cognition sufficient to explain it?) Or is general cognitive development merely one of many necessary but not sufficient requirements for language development? 2.4.2 Example: "object cognition" and language development. Object permanence (Piaget): the cognitive ability to know that objects continue to exist even when they cannot be perceived. 4 months old: When an object is hidden from view, babies quickly forget about it and do not try to search for it. 9+ months old: Will actively search for a hidden object. 18-24 months: Passes the "invisible displacement" test: child can find object even after it has been moved while out of view (e.g. when passed from hand to hand). Object permanence seems to precede the development of nouns: Early nouns refer to the "here and now" (directly perceivable) Later nouns refer to objects that are not immediately present. The use of nouns to name entire classes requires object permanence (otherwise nouns are merely naming sensations). But does object permanence EXPLAIN noun development? That is, does nonlinguistic cognition directly affect language development? NO, because other experiments have found that babies have object permanence as young as three months (Baillargeon 1987): Basically, you show babies a "magic trick": Show the babies a solid screen rotating forward and backward until they stop paying attention. Then show them an object being placed behind the screen. If the screen is rotated again, it should be stopped by the object: the visual stimulus is different from before, but this stimulus doesn't interest the babies. HOWEVER, if you do a "magic trick" so that the screen can now rotate all the way back like before, the visual stimulus is the same, but now the babies are very interested. So they must "know" that objects don't normally disappear when hidden. In other words, Piaget's tests (searching, invisible displacement) are too hard, especially for memory; hence infants are ready to learn object names by 3 months, but they still don't; hence language development has its own schedule, independent of nonlinguistic cognition; hence language is modular... (maybe...???) 3. The role of the environment 3.1 But isn't it obvious that parents teach their kids language? In one sense, YES: children get lots of positive evidence about what is grammatical in their parents' language. Moreover, child-directed speech ("motherese") has some properties that may be helpful to children: Exaggerated intonation (highlights speech as special information; helps with "prosodic bootstrapping" into syntax?) Clear articulation (helps with learning phonological forms?) Shorter sentences (helps with learning basic syntax?) Concrete references (helps with semantic bootstrapping: child's experience directly related to linguistic forms?) However, this just begs the question. If language development REQUIRES that child-directed speech have these properties, how do parents KNOW that they must talk this way to children??? So perhaps motherese itself is innate! Fernald (1992): "the special form of speech used by human mothers with infants has evolved as a species-specific parenting behavior." Evidence: (1) universal across human cultures; (2) domain-specific: it's beneficial to parents and infants in their early interactions and not in communication with others; (3) an innate ability to use motherese could contribute to reproductive success, and so it may have been selected by evolution. Another hint: Shatz and Gelman (1973) found that even 4-year-olds knew to use "babytalk" with 2-year-olds, even when they did not have a younger brother or sister at home! Thus, if you try to avoid nativism in babies by saying that parents teach their children, you really just push the nativism mystery back to the parents. 3.2 In one important sense, though, parents do NOT teach their children: (almost) no negative evidence (evidence about what is ungrammatical). 3.2.1 There's a big difference here between L1 and L2 acquisition: When learning a second language, negative evidence is of course crucial. With first language, however, children ignore negative evidence even on those rare occasions where parents give it (real example from Braine 1971): Kid: Want other one spoon, Daddy. Dad: You mean, you want "the other spoon." Kid: Yes, I want other one spoon, please, Daddy. Dad: Can you say "the other spoon"? Kid: Other ... one ... spoon. Dad: Say "other." Kid: Other. Dad: Spoon. Kid: Spoon. Dad: Other ... spoon. Kid: Other ... spoon. Now give me other one spoon? 3.2.2 Is positive evidence sufficient to acquire language just from experience? Chomsky's POVERTY OF THE STIMULUS argument says NO. This argument claims that there is not enough evidence in the linguistic environment ("the stimulus") for kids to develop the linguistic knowledge that they have; therefore, that knowledge must be innate. Without negative evidence, kids cannot learn if something is (a) totally disallowed, (b) optional, or (c) required but they just haven't heard it yet. Example: *"rats-eater" again Positive evidence: rat-eater, mouse-eater, mice-eater How can kids therefore conclude that *rats-eater is illegal? This conclusion is not logical. (It's the old induction problem.) Therefore, some aspect of the knowledge of this constraint must be innate (though perhaps not specific to language, as we saw above). 3.2.3 Learnability theory (invented by Gold 1967): a mathematical theory of "formal grammars" which makes simplified assumptions about human language that then allow for some dramatic proofs about the role of experience and innate knowledge in language acquisition. The key idea is the logical problem of induction, noted above: you can't learn a universal generalization (i.e. a grammar) from finite observations (i.e. your parents' utterances). Gold (1967) proved that without negative evidence, it is impossible for children to get the right grammar in many cases (though they aren't very realistic in Gold's original model). Wexler and Culicover (1980) proved that with simple innate knowledge about syntax, a child can learn any human language, but ONLY IF the input data includes sentences with at least two levels of embedding, e.g.: [this is the cat [who chased the rat [who ate the cheese]]] Lightfoot (1989) argued that learning is possible even with input data that has no embedding, but ONLY IF there's much more innate knowledge, specifically, that certain properties of language always go together: the kid only has to hear evidence for one property to automatically learn the other. Morgan (1986) proved that if the input data includes prosodic marking of syntactic boundaries (so allowing prosodic bootstrapping), then much less syntactic knowledge needs to be innate. 3.2.4 There is also the inverse of the poverty of stimulus argument (which doesn't have a name): The stimulus may have helpful information that children ignore, which implies that language is innately programmed to mature at its own rate (e.g. showing a plant how to grow will not help it grow!) Examples above: noun development is not "helped" by object permanence; kids reject negative evidence. Meier (1991): iconicity doesn't help kids acquiring ASL. 3.2.5 But maybe it's really the "poverty of the imagination" argument? (this joke comes from Robert van Valin, I think) Maybe the stimulus is actually rich with information, but we researchers are just too stupid to see it. 3.3 In a sense, "nativism" is the null hypothesis in the study of how learning works: it says that learning NEVER happens, since we're born already knowing everything. It thus may block the discovery of how learning REALLY works. 3.3.1 An example of the inherent philosophical danger of nativism: Fodor (1981) Premise A: In order to learn a new concept, we must describe it in terms of concepts we already know. Premise B: Semantic feature theory doesn't work; there is no finite set of innate "atomic" concepts out of which all other concepts can be built. (E.g. "CD player" cannot be built out of innate semantic features, as far as we know.) Therefore, ALL concepts must be innate (including "CD player" -- before CD players were invented, we already had the concept, just waiting for the real thing to come along!) How to avoid this absurd conclusion...?? 3.3.2 An example of challenging nativism: how a simple assumption about the development of memory can undermine Gold's proof. Elman (1992): a simple connectionist computer program for learning syntax WITHOUT any built-in assumptions about syntactic constituents or traces Network was given sentences that were grammatical in a toy grammar of English (e.g. no articles). Some had no embedding (e.g. (1)-(3)) and some did (e.g. (4)-(5)): (1) Boy chases dog. (2) Boys chase dog. (3) Mary walks. (4) Boy who dogs chase feeds cat. (5) Girl who cases dogs hits cat. [Note the long-distance dependencies in (4) and (5): subject-verb agreement.] Sentences were presented one word at a time; the network guessed what the next word might be; the experimenter then trained the program on the correct answer. Result: the network was unable to learn the syntactic rules correctly, just as Chomsky and Gold would have predicted: not enough innate knowledge. But wait.... Next, the network was given a gradually increasing working memory capacity. At first, it was designed only to process two to three words at one time; hence it could process all of (1)-(3), but only parts of (4)-(5). Gradually, working memory capacity was increased. Result: the network WAS able to learn all of the syntactic rules. That is, if you gave it "Boy who dogs chase ..." it would correctly guess that the next word would have to be a singular verb (e.g. "feeds", not "feed"). That is, it learned about long-distance dependences! Apparently, this is because the network first learns simple word associations, and then after these become strong, it's ready to go to more complex structures. Interpretation: it is possible to learn complex linguistic rules solely from experience if we make a simple assumption about the innate properties of nonlinguistic cognition (i.e. memory): nativism without modularity again. 3.3.3 Some responses by anti-connectionists (e.g. Marcus 1998, a student of Pinker): The training method for the connectionist model uses negative evidence! The model cannot generalize to new cases. E.g. if it is only trained that "Boy chases girl" is grammatical, it will not know that "Girl chases boy" must be grammatical too. 3.4 The biology of environmental influences: Critical period: a hypothetical biologically determined period of time when the environment can influence cognitive development. 3.4.1 E.g. critical period for the development of vision under influence of visual stimuli (the cats): first 2-5 months of life; after this, the cats will forever have impaired vision. 3.4.2 Birds have critical periods in learning birdsongs (Nottebohm 1970): Male chaffinches (European bird) raised in isolation sing songs of normal duration and pitch, but show no phrasing and lack a complex ending (like syntax!) The critical period is the first 10 months of life; after a song is established during that time, it will not change. This critical period depends on levels of testosterone (male hormone): it is thus determined by biological factors, not by experience. 3.4.3 Better term: "sensitive period": this emphasizes that there is no abrupt end, but rather a gradual dropping off of sensitivity to experience. 3.4.4 Is there a critical/sensitive period for human language? Isolated children imply that the critical period ends around adolescence: When "Genie" returned to normal human life at 13.5 years, she could not acquire English completely, having particular trouble with syntax and morphology (e.g. few grammatical morphemes). [Curtiss 1977] When "Isabelle" escaped from her isolation (her brain-damaged mother never talked) at 6.5 years, she acquired perfectly normal English [Tartter 1986] Similar evidence from signers of ASL (Meier 1991): if you acquire ASL at too old an age, your grammar will not be quite right, especially morphology. 3.5 If language is innate, can it be created practically out of nothing? Pidgins and creoles. Pidgin: a supplementary language that arises when speakers of mutually unintelligible languages are in close contact: not much grammar (e.g. no syntactic structure). It borrows the words mostly from the politically dominant language, the lexifier language. Hawaiian Pidgin (Bickerton 1983) "No, the men pau work, they go, make garden. Plant this, cabbage, plant potato, like that." [pau = "finished"] Creole: a distinct form of language that emerges in the second generation of pidgin-speakers: fully grammatical (e.g. it has syntactic structure, uses grammatical morphemes) Hawaiian Creole (Bickerton 1983) "When work pau da guys they stay go make garden for plant potato an' cabbage...." Note the use of grammatical morphemes: "stay go" means "going to" "for" marks a tenseless clause "when" marks a subordinate clause In other words, there is a sudden increase in complexity and an "invention" of grammar in a single generation. Creoles are not exotic. Some confirmed or possible examples include: Yiddish (German + Hebrew) Modern Hebrew (derived from an artificially simplified version of ancient Hebrew) Modern German (from many German-speaking tribes) Japanese (a Korean-like language + an Austronesian language??) ASL and other sign languages? Evidence that ASL is a creole that continues to be "reinvented" by children (Singleton and Newport 1993 [cited in Pinker 1994]): "Simon", a young deaf boy, was born to deaf parents. The parents had learned ASL after adolescence, and thus their grammar was poor (see "critical period" above) Simon learned ASL only from his parents. Nevertheless, his ASL grammar turned out better than his parents' did! So perhaps Simon was really "inventing" ASL, based on his innate default grammar. If the nativists are right, then actually all languages are creoles: children don't "learn" language but rather "invent" a "new" language based on the small amount of evidence that they hear, using a lot of innate assumptions. 4. Nativist models These are some models that assume that language is innate at the level of Representation; they are therefore surely wrong; but they're also famous. 4.1 Parameter setting: language experience as mere "trigger" (like the sun triggers the growth of a plant) Claim: children only use linguistic experience to choose among a very narrow range of possible hypotheses given to them by the innate language module. Language is built out of parameters: properties of a grammar like switches with a very small set of "settings" (e.g. NPs are right-headed vs. left-headed). Are there "default" settings for parameters? (like the "factory settings" in a computer program) 4.1.1 Parameters immediately solve the learnability problem: if there are a finite set of parameters, each with a finite number of settings, then the set of possible human languages must also be finite, which makes all languages learnable in principle, since kids only have to go through the finite space of languages, and can never get totally lost. Hence this view is also called "Strong Nativism". Amazingly, using the mathematics of learnability theory, Osherson et al. (1989) proved that Strong Nativism MUST be true if three reasonable assumptions are made: Learning is possible even if the input data is "noisy" (i.e. contains some ungrammatical sentences). Children have bad long-term memory: they only remember the most recently heard sentences. Children follow some principle of grammatical simplicity, i.e. an innate assumption that given data that fits more than one grammar, they should pick the simpler grammar. 4.1.2 The most famous such principle is "subset principle," which also provides a logical argument for the existence of innate defaults: Reminder: language learning requires induction: hypothesize a finite grammar that will generate an infinite number of structures, using only a finite amount of data. Thus language development involves hypothesizing a series of grammars; at first quite different from your parents, and then gradually your hypotheses get better and better. Definition: Grammar A is a "subset" of Grammar B if B generates all of the sentences that A generates, but A does not generate all the sentences that B generates. ("superset" is the opposite case.) Kids could hypothesize a superset grammar (i.e. one that generates sentences that are not produced in their parents' language). But if there is no negative evidence, how could they ever learn that the "extra" sentences they produce are ungrammatical?? So it's safer to hypothesize a subset grammar (i.e. one that generates sentences that are also found in the parents' language). Then, as development proceeds, kids could make their grammar "larger and larger", generating more and more sentences, but always a subset of their parents' sentences. That's the subset principle. 4.2 Defaults in child language development? 4.2.1 The fixed/free word-order parameter (Pinker 1995): Some languages have very strict word order (English "Dogs hate cats", *"Cats dogs hate"), while others have freer word order (e.g. Russian). A language with strict word order is a subset of one with free word order (looked at the level of syntactic categories): English = {NVN, ...} Russian = {NVN, NNV, ...} If Russian kids obey the Subset Principle, they should begin with a strict word-order grammar like English, and this does seem to be true. 4.2.2 The null-subject parameter: In null-subject languages (e.g. Spanish, Mandarin) overt subjects are optional. Thus null-subject grammars generate all sentence types generated by obligatory-subject languages, plus some more; thus the subset grammar is the non-null-subject grammar (e.g. English). Prediction: English-learning kids will not drop subjects. False! Kids acquiring English at first produce many null-subject sentences: "Play it", "Eating cereal", "See window", etc. Explanation (Bloom 1990): Kids are not dropping subjects syntactically, but phonologically: there is a statistically significant correlation between the length of the verb phrase and the probability of subject omission, suggesting that children omit subjects as a function of overall processing load. [For some mysterious reason, Bloom's reasonable and empirically supported analysis is mostly ignored by researchers interested in syntax acquisition -- I suspect because it's not as much fun for them!] 4.3 Bioprogram hypothesis: humans have an entire default grammar that only emerges when language experience is extremely impoverished (as with creoles) [Bickerton (1981)] Some similarities that Bickerton claims are shared by all creoles, in spite of their being historically unrelated: Subject + Verb + Object order Verbs always precede complements Inflectional particles are always in the order tense + mood + aspect 5. Positive arguments for nativism 5.1 The logical problem of nativism As mentioned above, the main problem with nativism is that it is a negative claim: it says that learning some things is impossible, so those things must be innate. But logically, a negative cannot be proven with evidence (e.g. you can't prove that UFOs definitely do NOT exist). The positive claim of nativism is: DNA encodes some language abilities. Recently researchers have collected some evidence for this claim. 5.2 FOXP2 is the name given to a gene recently identified on chromosome 7 which has been linked to a particular language disorder (Lai et al. 2001). People born with damage in this gene will have problems with linguistic abilities like speech articulation, phonemic awareness, use of morphological rules, and understanding sentences. As with most genes, its function was identified by finding a special group of people who seemed to have inherited this language problem, a situation that's very rare. However, you can't say that FOXP2 is "the" language gene. First of all, the language effects are caused by damaging it -- this doesn't tell us exactly what it does when it's not damaged. Also, the protein built by FOXP2 also has other uses in the body. Probably this gene affects language by preventing other genes from being "turned on" during brain development. Moreover, some researchers (e.g. Elman et al. 1996, who looked at early evidence) are skeptical about how language-specific the gene's effects really are (IQ is also affected somewhat). Certainly the gene's effects aren't very "modular": damaging it affects many aspects of language processing, including control of mouth muscles! However, it is still important that there are key differences between the human FOXP2 gene and the version found in other animals (including chimpanzees), and significantly more of these differences are found in parts of the gene that affect protein structure, suggesting that the differences were selected by evolution (Enard et al. 2002). The importance of FOXP2 is that it opens a door into the complex chain by which DNA affects language. I just hope drug companies and other evil capitalists don't screw everything up.... 5.3 Stromswold (2001) is a recent meta-analysis (i.e. combining the data from many studies and doing statistics on the whole thing) that has found that identical twins (who share all their genes) show a higher correlation in language abilities than fraternal twins (who only share half their genes): both the degree of language disorder, and also the varying linguistic abilities of non-impaired people, show this genetic effect. The study also found some evidence for modularity within language: different types of linguistic abilities may have different genetic bases (e.g. lexical processing vs. syntactic processing). REFERENCES Baillargeon, R. (1987). Object permanence in 3.5- and 4.5-month-old infants. Developmental Psychology, 23, 655-664. Bickerton, D. (1981). Roots of language. Karoma. Bickerton, D. (1983). Creole languages. In W. S-Y Wang (Ed.) The emergence of language: Development and evolution (pp. 59-69). W. H. Freeman. Blakemore, C., & Cooper, G. F. (1970). Development of the brain depends on the visual environment. Nature, 228, 477-478. Bloom, P. (1990). Subjectless sentences in child language. Linguistic Inquiry, 21, 491-504. Braine, M. D. S. (1971). On two types of models of the internalization of grammars. In D. I. Slobin (Ed.) The ontogenesis of grammar: A theoretical symposium. Academic Press. Curtiss, S. (1977). Genie: A psycholinguistic study of a modern-day wild child. Academic Press. Elman, J. L., Bates, E. A., Johnson, M. H., Karmiloff-Smith, A., Parisi, D., & Plunkett, K. (1996). Rethinking innateness: A connectionist perspective on development. MIT Press. Elman, J. L. (1992). Grammatical structure and distributed representations. In S. Davis (Ed.) Connectionism: Theory and practice (pp. 138-178). Oxford University Press. Enard, W., Przeworski, M., Fisher, S. E., Lai, C. S. L., Wiebe, V., Kitano, T., Monaco, A. P., & Paabo, S. (2002). Molecular evolution of FOXP2, a gene involved in speech and language. Nature, 418 (Aug. 22), 869-872. Fernald, A. (1992). Human maternal vocalizations to infants as biologically relevant signals: An evolutionary perspective. In J. H. Barkow, L. Cosmides, & J. Tooby (Eds.) The adapted mind: Evolutionary psychology and the generation of culture (pp. 391-428). Oxford University Press. Fodor, J. A. (1981). The present status of the innateness controversy. In J. A. Fodor (Ed.) Representations. MIT Press. Gold, E. M. (1967). Language identification in the limit. Information and Control, 16, 447-474. Gordon, P. (1986). Level-ordering in lexical development. Cognition, 21, 73-93. Lai, C. S. L., Fisher, S. E., Hurst, J. A., Vargha-Khadem, F., & Monaco, A. P. (2001). A forkhead-domain gene is mutated in a severe speech and language disorder. Nature (4 Oct.), 413, 519-523. Lightfoot, D. (1989). The child's trigger experience: Degree-0 learnability. Behavioral and Brain Sciences, 12, 321-375. Marcus, G. F. (1998). Can connectionism save constructivism? Cognition, 66, 153-182. Meier, R. P. (1991). Language acquisition by deaf children. American Scientist, 79, 60-70. Morgan, J. L. (1986). From simple input to complex grammar. MIT Press. Nichelli, P., Grafman, J., Pietrini, P., Alway, D., Carton, J. C., & Miletich, R. (1994). Brain activity in chess playing. Nature, 369, 191. Nottebohm, F. (1970). Ontogeny of bird song. Science, 167, 950-956. Osherson, D. N., Stob, M., & Weinstein, S. (1989). Learning theory and natural language. In R. J. Matthews & W. Demopoulos (Eds.) Learnability and linguistic theory. Kluwer. Pinker, S. (1994). The language instinct: How the mind creates language. William Morrow and Company. Pinker, S. (1995). Language acquisition. In L. R. Gleitman & M. Liberman (Eds.) Language: An invitation to cognitive science, Vol. 1 (2nd edition) (pp. 135-182). MIT Press. Shatz, M., & Gelman, R. (1973). The development of communication skills: Modifications in the speech of young children as a function of listener." Monographs of the Society for Research in Child Development 38 (5, Serial No. 152). Slobin, D. I. (1973). Cognitive prerequisites for the development of grammar. In C. A. Ferguson & D. I. Slobin (Eds.) Studies of child language development (pp. 175-208). Hold, Rinehart, and Winston. Stromswold, K. (2001). The heritability of language: A review and metaanalysis of twin, adoption, and linkage studies. Language, 77(4), 647-723). Tartter, V. C. (1986). Language processes. Hold, Rinehart and Winston. Wexler, K., & Culicover, P. (1980). Formal principles of language acquisition. MIT Press.