Li, BeiBeiLiWong, Bruce XiaoBruce XiaoWongChen, SiSiChenZhang, CaicaiCaicaiZhangLau, PuiyinPuiyinLauDr. YANG Yike2024-03-252024-03-252023Li, B., Wang, B. X., Chen, S., Zhang, C., Lau, P., & Yang, Y. (2023 Dec 3). Talker normalisation of prosodic cues in non-native speakers. Architectures and Mechanisms for Language Processing Asia 2023, Hong Kong.http://hdl.handle.net/20.500.11861/9152https://ling.cuhk.edu.hk/amlap.asia/files/10099.pdfIn speech perception, the surface acoustic cues are susceptible to variability caused by social factors, as well as biological factors. Consequently, listeners must normalise the variable acoustic cues in speech perception. The process of listeners mapping variable acoustic signals onto the invariable underlying linguistic categories is called talker normalisation. Several talker normalisation models have been proposed (Chen et al., 2022; Shao & Zhang, 2019) and the current study aims to test the generalization of one of the latest talker normalisation models involving probabilistic parametric representation (PPR, Chen et al., 2022). Their model proposed that surface prosodic cues are represented as statistical distributions where the parameters of the distributions of the lexical tones are mentally stored, and this has been shown for both native Cantonese and Mandarin speakers in Cantonese lexical tone recognition (Chen et al., 2019, 2022). The mentally stored distributions may still play a role even if a tonal context is present for native Cantonese speakers. In the current study, we take tonal contexts into consideration in investigating the effect of tonal contexts and PPR on Cantonese tone recognition among Mandarin speakers. Thirty-four Hong Kong Cantonese speakers (17 females and 17 males) were recruited and instructed to produce the target syllable /ji/ with T3 [33] embedded in two carrier sentences, i.e., /th iŋ th iŋ/ T1 [55] (“Listen to”) __ and /tsɐu hɐi/ T6 [22] (“This is just”) __. The tones of the carrier sentences were then manipulated based on the semitone distance (Table 1) between T1, T3 and T6 estimated from 68 speakers (CUSENT; Lee, n.d.). T1 [55] in /th iŋ t h iŋ/ was manipulated to be 3.56 and 3.90 semitones higher than the target syllable (T3 [33]) for male and female speakers respectively; similarly, T6 [22] in /tsɐu hɐi/ was manipulated to be 3.56 and 3.90 semitones lower. Seven male and seven female native Mandarin speakers, with less than one-year formal Cantonese instruction, were recruited in the Cantonese tone recognition task where they were informed about the speaker gender in the stimuli. If the mentally stored distribution PPs play a more significant role in tone recognition, the target T3 would be identified as T3 regardless of its tonal context. On the contrary, if tonal context is more important, the target T3 would be identified as the low T6 in manipulated high T1 context and, mutatis mutandis, T3 would be identified as high T1 in manipulated low T6 context. Multinomial mixed effects modelling was used to test the significance of tonal contexts and mentally stored distributions on Cantonese tone recognition among Mandarin speakers. The result shows that mentally stored distributions have a significant effect on Cantonese tone recognition in both manipulated low and high tone contexts (ps < 0.001), while only the low tone context has a significant effect (p < 0.001). Therefore, the findings provide insights about the accessibility of distributional information of non-native prosodic cues with the presence of contextual information.enConference Paper