Perceptual normalization for inter- and intratalker variation in Cantonese level tones.

Inter- and intratalker variation in the production of lexical tones may contribute to acoustic overlap among tone categories. The present study investigated whether such category overlap gives rise to perceptual ambiguity and, if so, whether listeners are able to reduce this ambiguity using contextual information. In the first experiment, native Cantonese-speaking listeners were asked to identify isolated Cantonese level tones produced by 7 talkers. Identification accuracy was significantly higher when the presentation of items was blocked by talker rather mixed across talkers. In the second experiment, listeners identified the final (target) tone of 6-syllable semantically neutral sentences with f0 patterns of the context (i.e., the first 5 syllables) altered. The same target tone was identified differently depending on the context. In the third experiment, the context portions of stimulus sentences from the second experiment were divided into 2 halves, and their f0 patterns were altered independently. In identifying the target tone, listeners relied more heavily on the f0 pattern of the second (last) half of the context. These results are discussed in relation to characteristic inter- and intratalker variations of lexical tones.