This program of research seeks to understand how listeners recognize phonological variants of spoken words as the word intended by the talker. As a first step in achieving this goal, research will focus on developing a clear and accurate description of phonological variation by collecting, transcribing, and analyzing a 100, 000-word corpus of conversational speech. Two types of analyses will be performed. Phonological analyses will center on developing a picture of phonological variation that the recognition system encounters on a regular basis, calculating the forms of variation, their frequency, and the contexts in which they occur. Acoustic analyses of variants will be carried out to develop a detailed description of variation that is tightly linked to the speech signal itself. The results from both lines of work should provide a solid foundation on which the recognition of phonological variants can be profitably studied in future work.
