Codificação estatística das categorias fonéticas: vestígio da dinâmica da fala na fonotaxe lexical

Abstract

This paper presents a new, intriguing finding and discusses some ofits theoretical implications: phonetic categorization, including majorclass membership, is entirely predictable from phonotactic biases inthree Brazilian Portuguese word databases. The predictors are logfrequencies of ‘VC, ‘CV and V’CV sequences consisting of the 7stressed vowels combined with the 19 onset consonants, plus the 5Southeastern pre-stressed vowels. Correct vowel categorizationarises through discriminant analysis of ‘VC and ‘CV data. Correctconsonant categorization arises through discriminant analysis of V’CVdata. Results are consistent across databases and, thus, strongly suggestthat statistical biases in the lexicon can be so stable as to code phoneticcategories. The findings and their corollaries bear on the issue of therelationship of lexical phonotactics to speech dynamics.