Multinomial models of speech perception

Abstract

It is often the case that two different acoustic cues can influence perception of both of two adjacent phonemes within a syllable. In such cases, it is of interest to determine whether the cues are evaluated independently; that is, does a listener’s evaluation of one cue influence evaluation of the other cue? Another, related question is whether, as some researchers claim, phoneme decisions are dependent in the sense that, when cued by the same information, the decisions compete for the information. This study modeled data from a two factor, four category experiment in which F1 offset frequency and vowel duration cue both voicing and vowel identity in CVC syllables. A family of multinomial processing tree models [e.g., D. M. Riefer and W. H. Batchelder, Psychol. Rev. 95, 318–339 (1988)], of which the fuzzy logic model of perception for the two factor, four category design is a special case, was developed and tested to explore the independence issues. This modeling approach is highly flexible and allows one to test a wide range of hypotheses regarding speech cue processing.