Although gradient phonotactics, phonological generalizations that are statistical rather than categorical, are a ubiquitous feature of human languages, current models of gradient phonotactics do not address the typological or diachronic aspects of these generalizations—why some phonotactic patterns are more common than others, and how and why these patterns change over time. In this dissertation I propose that the statistical properties of the lexicon are shaped in part by unconscious phonotactic preferences on the part of language users—biases that affect a word’s chance of becoming established among a community of speakers, or remaining in use once established.
The dissertation is devoted to establishing two main claims: first, that phonotactic preferences are real, and second, that the mechanism that drives these preferences consists of competitions among words during speech production. In support of the first, empirical, claim, I present three main examples of phonotactic preferences: (1) a bias in favor of /b/ over /d/ in the development of Latin into French, which is motivated by articulatory ease, (2) a gradient OCP effect in English, motivated by processing ease, and (3) a correlation between tautomorphemic and heteromorphemic phonotactics in several languages, which I argue is motivated by the structure of the human phonotactic learning algorithm.
In support of the second, theoretical claim, that phonotactic preferences can result from competitions among synonyms, I present a spreading activation model of speech production (Dell 1986) which consists of a network in which lexical items which match concepts the speaker wishes to express are activated by those concepts, and then in turn activate their constituent phonological subparts. Synonyms, words that represent the same concept, are simultaneously activated and race to reach an activation threshold—the winner of this race is selected and used by the speaker to express the concept. Properties that allow a word to be accessed more quickly thus confer an advantage to words that have those properties. Words with phonotactic patterns that facilitate lexical access will tend to be used more than words without those patterns, leading over time to a lexicon in which these “good” patterns predominate.