Cross-Linguistic Perception and Learning of Mandarin Chinese Sounds by Japanese Adult Learners

摘要

This dissertation presents a cross-linguistic investigation of how nonnative sounds are perceived by second language (L2) learners in terms of their first language (L1) categories for an understudies language pair---Japanese and Mandarin Chinese. Category mapping experiment empirically measured the perceived phonetic distances between Chinese sounds and their most resembling Japanese categories, which generated testable predictions on discriminability of Chinese sound contrasts according to Perception Assimilation Model (PAM). Category discrimination experiment obtained data concerning L2 learners' actual performance on discrimination Chinese sounds. The discrepancy between PAM's predictions and actual performances revealed that PAM cannot be applied to L2 perceptual learning. It was suggested that the discriminability of L2 sound contrasts was not only determined by perceived phonetic distances but probably involved other factors, such as the distinctiveness of certain phonetic features, e.g. aspiration and retroflexion.The training experiment assessed the improvement of L2 learners' performance in identifying Chinese sound contrasts with exposure to high variability stimuli and feedback. The results not only proved the effectiveness of training in shaping L2 learners' perception but showed that the training effects were generalizable to new tokens spoken by unfamiliar talkers. In addition to perception, the production of Chinese sounds by Japanese learners was also examined from the phonetic perspective in terms of perceived foreign accentedness. Regression of L2 learners' and native speakers foreign accentedness ratings against acoustic measurements of their speech production revealed that although both segmental and suprasegmental variables contributed to the perception of foreign accent, suprasegmental variables such as total and intonation patterns were the most influential factor in predicting perceived foreign accent. To conclude, PAM failed to accurately predict learning difficulties of nonnative sounds faced by L2 learners solely based on perceived phonetic distances. As Speech Learning Model (SLM) hypothesizes, production was found to be driven by perception, since equivalence classification of L2 sounds to L1 categories prevented the establishment of a new phonological category, thus further resulted in divergence in L2 production. Although production was hypothesized to eventually resemble perception, asynchrony between production and perception was observed due to different mechanisms involved.