On the adequacy of the Competition Model

REVIEW ARTICLE On the adequacy of the Competition Model Edward Gibson Carnegie Mellon University* The crosslinguistic study of sentence processing. Edited by Brian MacWhinney and Elizabeth Bates. Cambridge: Cambridge University Press, 1989. Pp. xviii, 489. Cloth $37.50. 1. Introduction. This book consists of a collection of thirteen papers describing a psycholinguistic model of language acquisition and use—the Competition Model (Bates & MacWhinney 1982, MacWhinney et al. 1984)—along with tests of the model in a number of different languages. The most extensive tests of the model come in the areas of sentence comprehension and firstlanguage acquisition, although evidence is also presented on sentence production , second-language acquisition, and sentence comprehension by speakers with language deficits. The book is intended as a major contribution to the field ofpsycholinguistics in two ways: first, the authors attempt to provide a uniform framework for explaining all manifestations of psycholinguistic effects, ranging from language acquisition to sentence processing to sentence production; and second, they try to test this framework in a variety of languages, something not previously attempted to this degree. It is commendable that MacWhinney & Bates (hereafter M&B) and their colleagues seek to examine numerous languages simultaneously, but the success or failure of their endeavor hinges upon the success or failure of the model upon which it is based. My comments, therefore, will consist primarily of an examination of the underlying model. Under the Competition Model, the language processing system consists of a direct mapping in the form of a connectionist network linking the form of an input sentence (an acoustic or visual signal) to the function óf that form (its meaning in context). In order to determine the function of a given input sentence, according to the model, the listener relies on cues in the input stream that correlate with meanings of the word string. Cues to meaning include such characteristics as word order, semantic plausibility, and morphological agreement among words in a sentence. For example, word order is reported to be the most important cue in the assignment ofmeaning in English, while in Italian, * This article was written during 1991-1992 while the author was at MIT in the Department of Brain and Cognitive Sciences, supported by a postdoctoral fellowship from the McDonnell-Pew Center for Cognitive Neuroscience. My thanks to the following people for their comments on earlier drafts: Sergey Avrutin, Bob Carpenter, Stephen Crain, Dan Everett, Paul Gorrell, Howard Kurtzman , Gary Marcus, Neal Pearlmutter, Steve Pinker, David Poeppel, Sandeep Prasada, Brad Pritchett , Gregg Solomon, Sally Thomason, William Snyder, and Ken Wexler. All remaining errors are my own. 812 REVIEW ARTICLE813 word order is a weaker cue than semantic plausibility (MacWhinney et al. 1984). Hence the English input the pencil kicks the cow is most often interpreted with the pencil as actor of the kicking action, but the Italian translation of the same sentence is most often interpreted with the cow as actor. Furthermore, proponents of the Competition Model assume that the same mechanisms are involved in both first-language acquisition and second-language acquisition, as well as in the acquisition of numerous other cognitive skills. In particular, they assume the existence of a back-propagation connectionist network that updates cue strengths according to how well each cue correlates with the appropriate interpretation of an input sentence. Thus a cue which is successful much of the time will receive positive reinforcement, while an unsuccessful cue will be negatively activated. The Competition Model contrasts sharply with nativist theories of first-language acquisition, grammatical competence, and use. Like the Competition Model, the nativist approach assumes that there is much for first-language learners to induce on their own (e.g. the lexicon and inflectional paradigms); but this approach also assumes the existence of a constraint-based Universal Grammar (UG) that the child uses in the acquisition of the target human language (see e.g. Chomsky 1965, 1986, Wexler & Culicover 1980, Pinker 1984, Wexler 1990, and Crain 1991). The arguments in favor of the nativist approach usually rely on the observation that children acquire certain components of language without having adequate evidence for them in the environment. This is known as the 'poverty of the stimulus' argument. For example, while knowing a language involves knowing...