Listener Adaptation to Non-Native Speech and the Limits of its Generalization

Landblom, Stephanie Anne

Graduate Thesis Or Dissertation

Listener Adaptation to Non-Native Speech and the Limits of its Generalization Public Deposited

Analytics

Download PDF

Citations

Citeable URL: https://scholar.colorado.edu/concern/graduate_thesis_or_dissertations/v118rd682

Abstract

Previous research has shown that listeners undergo a process of adaptation to individual speakers by learning indexical information phonetically encoded in the speech signal and using that information to help in subsequent speech processing (Nygaard & Pisoni, 1998). Such adaptation processes can generalize across speakers and can apply to non-native speech as well (e.g. Weil, 2001). Listeners with enough exposure to several speakers from the same language background are able to process speech of a new speaker (of the same language background) more accurately (Bradlow & Bent, 2008). This study looks to expand this research by testing whether generalization can extend to speakers of different language backgrounds who share similarities with the speaker a listener adapted to. To test this, listeners were exposed to English sentences and narratives recorded by several Gujarati speakers. They were then post-tested on a series of English sentences from either a Kannada or a Russian speaker. Though different, both Kannada- and Gujarati- accented English fall into the category of Indian English for the average American listener. This was designed to test how categorical dialect representation could help with the adaptation/generalization process. The Russian speaker shared a salient feature with the Gujarati speakers, namely the substitution of /v/ for /w/. This tested whether listeners could adapt to salient features across unrelated accents. Both reaction time and accuracy scores were measured in the post-test and compared between the Gujarati listeners and a control group who heard only native-accented American English speakers. There were no significant differences in response times, but the Gujarati-trained listeners were significantly more accurate in both the Russian and the Kannada post-test, with the Gujarati trained listeners being more accurate in the Kannada post-test than in the Russian post-test. The findings in this study suggest that listeners start adapting to feature alternations and can generalize this to other speakers of similar language backgrounds or to speakers that have certain features in common in their L2.

Creator