Talker-specific perceptual adaptation during online speech perception

Abstract

Despite the ubiquity of between-talker differences in accent and dialect, little is known about how listeners adjust to this source of variability as language is perceived in real time. In three experiments, we examined whether, and when, listeners can use specific knowledge of a particular talker's accent during on-line speech processing. Listeners were exposed to the speech of two talkers, a male who had an unfamiliar regional dialect of American English, in which the /æ/ vowel is raised to /ei/ only before /g/ (e.g., bag is pronounced /beig/), and a female talker without the dialect. In order to examine how knowledge of a particular talker's accent influenced language processing, we examined listeners' interpretation of unaccented words such as back and bake in contexts that included a competitor like bag. If interpretation processes are talker-specific, the pattern of competition from bag should vary depending on how that talker pronounces the competitor word. In all three experiments, listeners rapidly used their knowledge of how the talker would have pronounced bag to either rule out or include bag as a temporary competitor. Providing a cue to talker identity prior to the critical word strengthened these effects. These results are consistent with views of language processing in which multiple sources of information, including previous experience with the current talker and contextual cues, are rapidly integrated during lexical activation and selection processes.

Keywords:

• Perceptual learning
• Accent
• Talker-specific
• Eye tracking

Acknowledgments

The authors would like to thank Scott Fraundorf and Stephanie Gnatek for recording stimuli, as well as Annie Tremblay and Gary Dell for their helpful comments on previous versions of this manuscript. This research was partially supported by NIH Grant T32-HD055272 to AMT and NSF Grant 10-19161 to SBS.

Notes

¹All p-values were corrected for violations of sphericity using the Greenhouse-Geisser correction. For cases in which sphericity was violated, the Greenhouse-Geisser ϵ is reported. Where no ϵ value is reported, sphericity was not violated. For clarity, the uncorrected degrees of freedom are reported in all cases.