Summary: |
Vowels tend to be reduced in words that are semantically predictable from context, an effect amenable to talker- or listener-oriented accounts of speech production. This study explored the role of perception in these accounts by testing for effects of semantic predictability on vowel production in the face of impaired speech perception (but otherwise normal hearing) -- namely, in a patient with pure word deafness. Analysis of the patient’s English vowels in read speech showed no effect of semantic predictability on vowel duration, but the expected effect on vowel dispersion: vowels tended to be less dispersed in predictable than in unpredictable words. Overall, these findings contradict listener-oriented accounts of reduction relying on stored exemplars or online perceptual modeling, suggesting instead that reduction arises due to talker-centric factors related to activation of long-term, abstract representations. |