Probabilistic relations between words: Evidence from reduction in lexical production

Abstract

The ideas of frequency and predictability have played a fundamental role in models of human language processing for well over a hundred years (Schuchardt, 1885; Jespersen, 1922; Zipf, 1929; Martinet, 1960; Oldfield & Wingfield, 1965; Fidelholz, 1975; Jescheniak & Levelt, 1994; Bybee, 1996). While most psycholinguistic models have thus long included word frequency as a component, recent models have proposed more generally that probabilistic information about words, phrases, and other linguistic structure is represented in the minds of language users and plays a role in language comprehension (Jurafsky, 1996; MacDonald, 1993; McRae, Spivey-Knowlton, & Tanenhaus, 1998; Narayanan & Jurafsky, 1998; Trueswell & Tanenhaus, 1994) production (Gregory, Raymond, Bell, Fosler-Lussier, & Jurafsky, 1999; Roland & Jurafsky, 2000) and learning (Brent & Cartwright, 1996; Landauer & Dumais, 1997; Saffran, Aslin, & Newport, 1996; Seidenberg & MacDonald, 1999). In recent papers (Bell, Jurafsky, Fosler-Lussier, Girand, & Gildea, 1999; Gregory et al., 1999; Jurafsky, Bell, Fosler-Lussier, Girand, & Raymond, 1998), we have been studying the role of predictability and frequency in lexical production. Our goal is to understand the many factors that affect production variability as reflected in reduction processes such as vowel reduction, durational shortening, or final segmental deletion of words in spontaneous speech. One proposal that has resulted from this work is the Probabilistic Reduction Hypothesis: word forms are reduced when they have a higher probability. The probability of a word is conditioned on many aspects of its context, including neighboring words, syntactic and lexical structure, semantic expectations, and discourse factors. This proposal thus generalizes over earlier models which refer only to word frequency (Zipf, 1929; Fidelholz, 1975; Rhodes, 1992, 1996) or predictability (Fowler & Housum, 1987). In this paper we focus on a particular domain of probabilistic linguistic knowledge in lexical production: the role of local probabilistic relations between words.