Bootstrapping lexical acquisition: The role of prosodic structure

Speech runs continuously, without pauses between individual words. As yet there is little firm evidence that the speech signal contains usable information as to the location of word boundaries, therefore current psycholinguistic models of lexical access are based upon this assumption and rely solely on phonemic information (that is, they consider speech as an uninterrupted string of phones). In these models, segmentation into words is a by-product of lexical identification. Babies, who do not possess a lexicon to start with, have to use independent mechanisms in order to first build their input lexicons (a list of word-forms). After reviewing three potential sources of information (distributional regularity, phonotactics, and lexical biases) that may allow infants to start acquiring an input lexicon — to bootstrap lexical acquisition — we propose that prosodic cues may be used by infants, and by adults, in order to segment the speech stream in prosodic units smaller than sentences, but bigger than words. Lexical acquisition, as well as lexical access, would be performed on the basis of this prosodically segmented pre-lexical representation. In addition, such a representation would be useful for the acquisition of phonology and syntax. In this paper, we present some experimental evidence that favors the existence of a prosodic segmentation strategy, using both adult and infant subjects. In addition, we discuss the implications of the hypothesis for models of speech processing — lexical access in adults, as well as lexical acquisition by children. 1. The preparation of this paper was supported by a grant from the Direction des Recherches, Etudes et Techniques to the first author (n 8780844). The Human Frontiers Scientific Program, as well as the Human Capital and Mobility Project, also supported the reported research. We would like to thank Jaques Mehler, John Morton, Geoff Hall, Brit van Ooyen and an anonymous reviewer for useful comments on the manuscript. Address for correspondence: MRC Cognitive Development Unit, 4 Taviton Street, London WC1H OBT, UK. The Linguistic Review 13 (1996), 383-412 0167-6318/96/0013-383