Vowel deletion and binarity in portuguese: a study based on Bayesian predictive inference

Abstract

The applications of statistical models in laboratory phonology have been shown to be productive in evaluating the predictive power of hypotheses. In this paper, the strength of the restriction FootBin was tested statistically in Brazilian and European Portuguese, on the basis of a Bayesian predictive inference model. We assumed that if the restriction FootBin is active there is a preference, in a language, for words with an even number of syllables. This sudy, based on such assumption, was conducted on the basis of acoustic and statistical analysis of data from an experiment on vowel deletion / reduction. Our statistical results show that FootBin is active in both PB and EP, although the strength of binarity is stronger in PB. The fact that FootBin is active in both varieties with a different strength in BP and in EP supports an optimality constraint-based analysis.