The "poverty of the stimulus argument" is a form of the problem of the under-determination of theory by data, applied to the problem of language learning. Two other well-known problems of under-determination include Willard Van Orman Quine's (1960) Gavagai example (a visitor to a foreign country sees a rabbit pass just as his informant utters the word "gavagai;" given only this evidence, "gavagai" might mean anything from rabbit, furry or nice day, isn't it? to undetached part of rabbit) and Nelson Goodman's Grue paradox (why is it that we take our experience in which all emeralds that we have thus far observed have been green to suggest that emeralds are green rather than the (equally confirmed so far) possibility that emeralds are grue, namely "green when examined before the year 2000, and blue when examined thereafter"?).
Learning a language involves going beyond the data: a child hears only a finite number of sentences, yet learns to speak and comprehend sentences drawn from a grammar that can represent an infinite number of sentences. The trouble that the child faces is thus a problem of under-determination: any finite set of example sentences is compatible with an infinite number of grammars. The child's task is to pick among those grammars.
The term "poverty of the stimulus" itself is relatively recent, perhaps first used by Noam Chomsky (1980: 34); but the argument as applied to language learning goes at least as far back as Chomsky's (1959) review of B. F. Skinner's Verbal Behavior. The exact formulation of the argument varies (Chomsky 1980; Crain 1991; Garfield 1994; Wexler 1991), but a typical version states that (1) children rapidly and, to first approximation, uniformly acquire language; (2) children are only exposed to a finite amount of data; yet (3) children appear to converge on a grammar capable of interpreting unfamiliar sentences; the conclusion is often argued to be that some aspect of grammar is innate.
Although the poverty of the stimulus argument is sometimes described in conjunction with the claim that children do not receive correction for their grammatical errors (for a recent review of the role of parental correction in the acquisition of grammar, see Marcus 1993), it is important to reject the notion that nativist explanations of language acquisition depend on the lack of parental correction. Even if parents did provide reliable correction to their children, innate constraints on the generalizations that children make would be necessary, because many plausible errors simply never occur. For instance, children never go through a period where they erroneously form yes-no questions by moving the first is to the front of the sentence. Although one can turn The man is hungry into Is the man hungry?, children never, by a false analogy, turn The man who is hungry is ordering dinner into Is the man who hungry is ordering dinner? (e.g., Chomsky 1965; Crain and Nakayama 1987). More generally, at every stage of LANGUAGE ACQUISITION -- inferring the meaning of a new word or morpheme, creating a morphological or syntactic rule, or determining the subcategorization frame of a new verb -- the child can make an infinity of logically possible generalizations, regardless of whether negative evidence exists. But children do not simply cycle through all logical possibilities and check to see what their parents say about each one; their choice of hypotheses instead must, in part, be dictated by innately given learning mechanisms. The open question is not whether there are innately given constraints, but rather whether those constraints are specific to language.
An excellent example of the poverty of the stimulus argument comes from Peter Gordon's (1985) work on the relation between plural formation and compounding. Paul Kiparsky (1983) had noted that while irregular plurals can appear in compounds (mice-infested), regular plurals sound awkward inside of compounds (rats-infested), perhaps because the design of the grammar is such that the process of compounding can only use stored (irregular) plurals as input, whereas the process of compounding serves as input to the process of regular plural formation. If irregular plurals inside compounds were common, it would be easy to see how a general purpose learning device might learn the contrast between regulars and irregulars, but in fact, as Gordon noted, plurals inside compounds are rare. Given this, one might expect that children would not be able to systematically distinguish between regulars and irregulars appearing in compounds. But Gordon found that although children allow irregular plurals inside of compounds, they systematically exclude regulars from compounds; children say things like mice-eater, but not rats-eater. As Gordon put it, "it would seem that of all the hypotheses available, there would be little to persuade an open-minded learner to choose this, rather than some other path." Instead, Gordon suggests, the child's mind is structured such that it is predisposed to learn one kind of grammar rather than another.
Recently, some scholars have tried to use CONNECTIONIST APPROACHES TO LANGUAGE learning to challenge the poverty of the stimulus argument, but connectionist models cannot obviate the need for innate constraints; instead they would simply provide a different theory of what those constraints are. Different connectionist models differ from one another in their architecture, representational schemes, learning algorithms, and so forth; each model thus differs from every other model in its innate design (Marcus 1998a, 1998b). Advocates of radical connectionism often overlook the importance of these innate design features, but such models cannot refute the poverty of stimulus argument; instead they can only show that (at most) the innate constraints are different in character than those suggested by other researchers. Moreover, such researchers have yet to provide any concrete example of a putatively unlearnable aspect of language that has been later shown to be learnable; hence their critique of the poverty of the stimulus argument is for now without much force.
Chomsky, N. A. (1959). Review of Verbal Behavior. Language 35:26-58.
Chomsky, N. A. (1965). Aspects of a Theory of Syntax. Cambridge, MA: MIT Press.
Chomsky, N. A. (1980). Rules and Representations. New York: Columbia University Press.
Crain, S. (1991). Language acquisition in the absence of experience. Behavioral and Brain Sciences 14.
Crain, S., and M. Nakayama. (1987). Structure dependence in grammar formation. Language 63:522-543.
Garfield, J. L. (1994). Innateness. In S. Guttenplan, Ed., A Companion to the Philosophy of Mind. Oxford: Blackwell.
Gordon, P. (1985). Level-ordering in lexical development. Cognition 21:73-93.
Kiparsky, P. (1983). Word-formation and the lexicon. In F. Ingemann, Ed., Proceedings of the 1982 Mid-American Linguistics Conference. Lawrence, KS: University of Kansas.
Marcus, G. F. (1993). Negative evidence in language acquisition. Cognition 46:53-85.
Marcus, G. F. (1998a). Can connectionism save constructivism? Cognition 66:153-182.
Marcus, G. F. (1998b). Rethinking eliminative connectionism. Cognitive Psychology 37(3).
Quine, W. V. O. (1960). Word and Object. Cambridge, MA: MIT Press.
Wexler, K. (1991). On the arguments from the poverty of the stimulus. In A. Kasher, Ed., The Chomskyan Turn. Oxford: Blackwell .
Copyright © 1999 Massachusetts Institute of Technology