On segmental factorability in second language learning.

de Jong, K.J. and Silbert, N. (Linguistics and Cognitive Science, Indiana University)

404 Memorial Hall, 1021 E. 3rd St, Bloomington, Indiana 47405, USA, kdejong-at-indiana.edu

Speech perception models differ concerning the role of token-specific information versus general phonological attributes in the identification of speech events. While some models rely heavily on encoded token-specific information, others rely on generalized abstractions extracted from the listeners’ experiences. This paper examines segment identification patterns in second language learners with relatively little experience with the second language in order to adduce the degree to which accuracy in one segment indicates a generalized skill which applies to other segments. 41 inexperienced native Korean EFL learners were tested in Korea. They were presented with a sampling of 14 coronal or labial consonants produced by four speakers of Midwestern American English, and were asked to identify the consonant in a quasi-open-set identification task. The consonants were placed in various prosodic environments: initial, final, and intervocalic preceding and following a stressed syllable. Current analyses focus on the acquisition of voicing contrasts, which differ in appreciable ways from glottal contrasts in Korean, and anterior non-sibilant manner contrasts, which do not exist in Korean. Analyses of accuracy rates indicate a variety of contingencies between prosodic location, point of articulation, manner, and voicing, suggesting the usefulness of segment-specific identification skills. However, further analyses of the corpus provide evidence that the listeners are developing perceptual skills which generalize across segments. For example, the accuracy of manner labeling in one prosodic condition correlates with accuracy in the other prosodic conditions, indicating that acquirers vary in general ability to distinguish stops from fricatives. J-factor analyses featural and segmental accuracy rates indicate that, for sounds not found in Korean, segmental accuracy is neatly predicted by the independent contribution of accuracy in each featural contrast. Taken together, these results indicate a model of adult second language learning which works initially in developing generalized featural contrasts. [Work supported by the NIH and NSF.]