Mapping to perceptual identification in Mandarin learners of English

• Mandarin and Korean speakers categorized 8 consonants covering the obstruent space. • Mapping between English and Mandarin and Korean can predict English identifications. • Second language labeling is deeply entangled with mapping to the native language. • Prosodic position does not heavily affect the mapping-to-identification relationship. This paper examines the relationship between cross-language segmental mapping and second language identification accuracy in Taiwan Mandarin speakers learning English, and compares this relationship with that found in previous, parallel research on Korean learners of English. Mapping and identification data were collected for English anterior plosives and non-sibilant fricatives, by means of two parallel identification experiments. Mapping data came from a 17-alternative identification task with Zhuyin Fuhao labels (phonetic script used to annotate Mandarin sounds in Taiwan), and identification data came from a 15-alternative identification task with Roman labels, both applied to the same stimuli. Mapping data were used to generate predictions about the identification performance by estimating what the performance would be, given the use of only the Mandarin categories. Like the previous Korean data, Mandarin speakers exhibited identification rates for plosives that are very close to predicted, indicating that their plosive identification performance was heavily entangled with their Mandarin system, while fricative identification performance was greatly under-predicted by the mapping data. Further analyses of category differentiation measured with d -prime estimates showed that Mandarin speakers’ manner differentiation performance was very well-predicted by the mapping data, while Korean speakers’ laryngeal differentiation was better predicted. Taken together, these results indicate that the second language identification performance and the cross-language mapping into the first language are closely entangled in a single system. The additional second language component appears in a pervasive increment in performance in the second language beyond what is predicted from the first language system, and in more unaccounted-for variance in laryngeal identification than in manner identification.