What Underlies Apparent Word-specific Phonetic Details

Particular words can exhibit acoustic differences based on factors like lexical frequency and emotional valence. What underlies such patterns? In this talk, I examine the possibility of word-specific acoustic details and alternative analyses. One set of experiments tests listeners' accuracy in identifying aurally presented homophone mates based on prior exposure and the semantic characteristics of the items, e.g. emotional valence. Listeners could identify homophones with significantly above chance accuracy in several of the experimental conditions. A second set of experiments tests whether listeners will converge to word-specific shifts, e.g. lengthened VOT in the initial stop of some words and shortened VOT in others, and how changing local frequency with exposure might itself impact production. There is weak evidence that listeners might learn arbitrary word-specific acoustic details, but only with substantial exposure. I analyze these results with an exemplar model that includes dimensions linking words based on factors like emotional valence; meanings which are frequently produced with similar acoustic characteristics (e.g. higher F0 in happy words) are associated with those acoustic characteristics. These broader associations can explain most apparent word-specific effects in perception. This model does not require that acoustic details are directly associated with individual words, though it also does not exclude such associations, which may be weak but detectable under the right conditions.

Note: The colloquium will be followed immediately by a social in the Linguistics Courtyard.