Perceptual category mining in human second language speech perception

This paper intends to clarify the process of the Perceptual Assimilation Model (PAM) predicting the patterns of human categorical speech perception in audio perception of second language (L2) speech signals. The original stage of categorizing acoustic stimuli in the L2 signal often involves assimilation from L1 categories. Whether L1 sounds will assimilate to L2 ones can be assimilated to the L2 category is decided by difference of category distance. This study, with evidence from Cantonese learners' perception of the English tr-signal, proposes that a careful mining process to find out the intended assimilating L1 category is necessary. The chosen L1 category should be tested on perceptual similarity to the L2 category in fine-grained phonetic environments, rather than just possible L1 phonological mappings from observed production errors. Two experiments with different L1 assimilating candidates are done using the AX identification and ABX discrimination paradigms. Target trVC syllables are both aligned with twVC, a representative of phonological closeness through contrastive analysis, and with chVC, a representative of perceptually similar candidate which is less mentioned in literature. Results from both identification and discrimination show that accuracy rate for twVC is almost ceiling whereas that of trVC is significantly lower. Results suggest that perceptual distance is better represented by perceptual similarity and such screening process should be applied as a pre-examination procedure instead of choosing L1 assimilator based on phonological similarity.