UC Berkeley

The aim of this dissertation is to investigate the mechanisms involved in phonetic learningof an unfamiliar accent, focusing on understanding what processes underlie changes to phonetic category structure, how such learning affects subsequent online lexical processing, and whether the same mechanisms that underlie learning for a single speaker are also responsible for generalization of learning to novel speakers with a similar pronunciation.

The first set of experiments (chapter 2) investigates the mechanisms that underpin thechanges in phonetic category structure (lexically-guided phonetic recalibration), following exposure to a novel artificial accent. This chapter focuses on two possible adaptive strategies that listeners may use which have been suggested in the literature: phonetic category shift and phonetic category expansion. Under the first hypothesis, listeners make targeted adjustments to a category boundary based on the specific accent they encounter, whereas under the second, they utilize a more general expansion of that category in perceptual space. Results of two experiments suggest that listeners rely on a (nonuniform) category expansion strategy that is constrained by acoustic similarity to sounds involved in the exposure accent.

The experiments in Chapter 3 focus on the relationship between changes in category structureand more online measures of speech processing. It is often assumed in the literature on perceptual learning for speech that changes in phonetic category boundaries (recalibration) following exposure to an atypical pronunciation underlie improved comprehension and/or processing of accented speech. The experiments in this chapter test whether exposure to such an artificial accent can facilitate subsequent processing of accented words, and whether the same mechanisms that constrain category boundary changes found in Chapter 1 also obtain for lexical processing. Results suggest that accent exposure does result in changes to lexical processing, and results provide tentative support for a form of category expansion as the mechanism for such changes.

The last set of experiments (Chapter 4) examine how perceptual learning may generalize2to novel speakers and novel sound contrasts. The goal of this set of experiments is to test whether the same mechanisms that are responsible for category boundary changes in a single speaker are also applicable to novel speakers. Previous literature suggests that phonetic learning of certain speech sounds (e.g., fricatives) may be speaker-specific — it does not transfer to a novel speaker, possibly because fricatives contain spectral properties that cue speaker identity. However, the results of Chapter 1 indicate that transfer to a novel sound contrast can occur within a single speaker, suggesting that there is room for transfer of learning to a novel speaker even if their pronunciation is acoustically distinct from that of the exposure speaker. Results of these experiments show transfer of phonetic learning to both novel speakers and novel phonetic contrasts. While results are mixed, depending on speaker and contrast, there is tentative evidence that listeners may use a more targeted mechanism when generalizing learning to novel speakers.

Together these experiments indicate that lexically-guided phonetic learning is flexible enoughto accommodate differences between familiar and novel contexts, suggesting that it may be a viable mechanism for adapting to variability in speech. However, while results provide evidence for some degree of generality in the underlying perceptual learning mechanisms, they also show that such generalization is constrained by acoustic similarity to previous experiences, whether the latter involves novel sound contrasts, novel speakers, or both. This supports a view of the perceptual system as one that is dynamic but that must still balance stability and plasticity