Kidz in the 'Hood: Syntactic Bootstrapping and the Mental Lexicon

This paper explores the limits of syntactic bootstrapping and demonstrates that the use of syntactic structure to build verb meanings is constrained to operate only within 'frame neighborhoods,' i.e., complement types that antecedently share formal and interpretive features. The results suggest that inferences over change in number of arguments are easier than inferences over change in type of arguments. This kind of finding establishes the limits within which the 'syntactic bootstrapping' paradigm for verb learning can operate, and also has implications for whether we should think about the architecture of the lexicon in projectionist or constructionist terms. Comments University of Pennsylvania Institute for Research in Cognitive Science Technical Report No. IRCS-01-01. This technical report is available at ScholarlyCommons: http://repository.upenn.edu/ircs_reports/20 1 KIDZ IN THE ‘HOOD: SYNTACTIC BOOTSTRAPPING AND THE MENTAL LEXICON Jeffrey Lidz, Henry Gleitman and Lila Gleitman University of Pennsylvania Recent findings and theorizing on child language acquisition suggest that the verb lexicon is built by an arm-over-arm procedure that necessarily constructs the clause-level grammar of the exposure language on the fly as it acquires individual items (Gleitman, 1990; Gillette, Gleitman, Gleitman, and Lederer, 1999). This active learning process is likely to play a causal role in determining the adult linguistic representations, particularly at the interface between the lexicon and clause structure. This paper presents two experiments designed to explore this idea: The first experiment examines an aspect of child verb learning, its sensitivity to structural information. The other examines adult representations of the same items that figured in the child experiment. More specifically, the experimental work explores grammatical architecture in regard to relations between clause structures (or “frames”) and classes of verb meanings. An apparently paradoxical fact about such verb-to-frame relations is that verbs are very choosy about the structures in which they appear, but at the same time they are “coercible,” that is, understandable in brand new structures. For example, the sentence Horace thought the stick to James is both bad and good. It sounds “wrong,” to be sure. Yet one easily understands it as a case of psychokinesis: Horace is Syntactic Bootstrapping 2 causing the stick to move to James via an act of thinking. Our aim is to understand the scope and limits of such comprehensible innovation in adults, limits that apply in analogous ways to constrain the child’s structure-dependent verb learning. In the introductory remarks below, we attempt first to lay out the motivations behind recent theories of the verb learning process that emphasize its sensitivity to linguistic-structural information. Second, we introduce two well-known linguistic proposals of the syntax-lexicon interface that have the potential to account both for the learning facts and for adult representations at this level: the lexical projection hypothesis and the frame semantics hypothesis. After reporting the experimental work, we will try to adjudicate between these proposals in terms of the findings. The role of syntactic structure in verb learning The extralinguistic contingencies for the utterance of a word are extraordinarily variable. Indeed, Chomsky’s analysis of the relatively “stimulus-free” character of language use was a major causal factor in redirecting inquiry into the psychology of language in the present generation (Chomsky, 1959). What goes for language use undoubtedly goes for language learning as well. If the adult community cannot be trusted to narrowly circumscribe the conditions under which they will utter a word, then an unaided word-to-world pairing procedure (as described in, e.g., Locke 1690; Hume 1738; Pinker 1989) cannot provide sufficient evidence for vocabulary acquisition. Recent findings document that this problem is particularly severe for the case of verbs. Verb use even by mothers of very young children lines up very poorly with ongoing events (Lederer,