臺灣海陸客語二字詞協同發音聲學研究

This project aims to conduct an acoustic-phonetic study of tonal coarticulation on disyllabic words in Taiwan Hailu Hakka. The following questions will be addressed. First, does Hailu Hakka, having a seven-tone system, exhibitanticipatory and/or regressive tonal coarticulation? Then, what is the correlation between the directionality of coarticulation and the effect type, i.e., assimilation and dissimilation? Next, based on the magnitude and temporal extent of contextual effects of the Hailu tones on neighboring tones, we will investigate which tones are better effect triggers and better effect targets for tonal coarticulation. To elicit ditonal sequences, 63 disyllabic words have been created by using 9 word-initial syllables carrying 7 different lexical tones plus 2 sandhi forms of T/35/ and T/55/ in combination of 7 word-final syllables carrying 7 different lexical tones. Furthermore, we repeat these 63 disyllabic words five times and arrange them randomly to generate eight different reading lists, each with 315 disyllabic words. We plan to recruit 4 male and 4 female Hailu Hakka native speakers to produce Hailu ditonal sequences by asking them to read one of the 8 different reading lists. This will lead to a total of 2,520 (9 word-initial tones x 7 word-final tones x 5 repetitions x 8 speakers) ditonal sequence, and 5,040 tonal samples. To establish the acoustic data of the duration and contour shape of each tonal sample, we will first use Wavesurfer and Matlab to compute the raw duration of each tone, and the F0 values extracted at 21 equidistant sampling time points along the tonal trajectory of each tonal sample. Then, for each individual speaker, each of their raw F0 data will be converted into a log z-score value by the log z-score normalization method so as to reduce cross-speaker tonal variations. With the acoustic data, raw and normalized, of the duration and contour shape of each tonal sample, we will be ready to answer our research questions using different statistical models. Pearson correlation coefficient will be adopted to detect the tendency of the directionality, effect type, and the magnitude and temporal length of the tonal coarticulatory effects for Hailu Hakka.A mixed method of quadratic functions and the Individual Growth Curve statistical model will be employed to simulate the characteristics of the tonal height and contour shape of each normalized tone sample, locate the sources and targets of coarticulatory effects, and explain in detail the pattern of tonal coarticulation of Hailu Hakka. A multi-layered contribution of our project results is expected: an empirical corpus of multi-speaker ditonal data in Hailu Hakka will be compiled, on which a scientific acoustic analysis will be conducted. The coarticulation pattern of Hailu Hakka will be better understood, and compared with the general findings of tonal coarticulation found in other tone languages. Last but not least, our project results will provide feedback on the theory of typology of tonal coarticulation.