Committee: Lisa Davidson (chair), Renee Blake, Gregory Guy, Susannah Levi, Melissa Baese-Berk (University of Oregon)

Abstract:

Previous research suggests listeners’ expectations of the speech signal can affect their comprehension and perception of unfamiliar voices. The literature has also demonstrated, however, that people can improve their comprehension of even difficult speech through highvariability exposure. Through a series of three experiments, this dissertation contributes several key findings to the growing body of literature investigating adaptation and generalization to unfamiliar talkers. Consistent with past work (Baese-Berk et al., 2013), the first experiment found listeners who were trained on five unrelated non-native (L2) accented talkers had more accurate comprehension of both a novel talker of an accent they had been trained on, and a novel talker of an unfamiliar accent. These data provide further evidence that experience with a diversity of L2 accents facilitates generalization.

Experiment 2 tested whether this finding could be replicated with real-world exposure to accented speech, rather than the artificial laboratory training that is common to attunement research. Individuals with greater lifetime experience listening to non-native accented speech, particularly from family, friends, and in their childhood community, were found to have more accurate comprehension of two unfamiliar L2 talkers compared to listeners with less prior exposure to non-native speech.

The third experiment investigated whether exposure to multiple different accents of any kind would provide listeners with enough variability of experience to facilitate generalization to non-native voices. Training on five different native regional varieties of English did not facilitate viii comprehension of unfamiliar non-native talkers, however, and no difference was found compared to a control group trained on five speakers of American English. This suggests it is familiarity with the intra-talker variation characteristic of non-native speech that aids in generalization to other L2 talkers, and not just experience with inter-talker variation that differs from the listener’s own dialect. Finally, people with more negative biases towards non-native speech were less accurate in their transcriptions of unfamiliar L2 talkers. This pattern was observed for all participant groups in all three experiments, with the exception of individuals who had extensive prior lifetime experience interacting with non-native speakers. Though this group of listeners still exhibited the same range of biases as other conditions, biases were not found to be predictive of comprehension accuracy.

Together these results are consistent with a model of adaptation and generalization wherein listeners learn patterns common to L2 speech through exposure, including the structures that are often difficult for L2 speakers and the tendency for greater within-speaker variation. Greater knowledge of these patterns is proposed to facilitate comprehension of novel non-native talkers, rather than reliance on new speech being acoustically similar to prior accented input.