Abstract:
This study aims to analyze segment timing in 12 Southeast Asian languages, namely Standard Thai (TH), Southern Thai (TT), Tai Yuan (TY), Mon (MN), Thai Khmer (KM), Vietnamese (VN), Burmese (BM), Sgaw Karen (SG), Standard Malay (ML), Cebuano (CB), Green Hmong (HM), and Mien (MI). Spontaneous speech from three speakers from each language was recorded. Vocalic, consonantal, voiced, and unvoiced intervals of 30 seconds of speech from each speaker were measured and analyzed using the three language typological classification models of Ramus et al. (1999), Grabe and Low (2002), and Dellwo et al. (2007). The durations of the four intervals were converted into eight parameters: 1) proportion of vocalic intervals (%V) 2) standard deviation of vocalic intervals (∆V) 3) standard deviation of consonantal intervals (∆C) 4) raw pairwise variability index of consonantal intervals (rPVI_C) 5) normalized pairwise variability index of vocalic intervals (nPVI_V) 6) proportion of voiced intervals (%VO) 7) variation coefficient of the standard deviation of unvoiced intervals (varcoUV) and 8) standard deviation of unvoiced intervals (∆UV). In addition, principle component analysis (PCA) was used to explore the relations among the parameters. The main phonetic and phonological features used to account for the values of the eight parameters are: syllable structure complexity, the existence or not of vowel length distinctions, and stress location. It was found that some aspects of the findings rejected the thesis hypotheses: 1) contrary to prediction, %V and %VO for languages with complex syllable structure (CSS) were not necessarily lower than those with simpler syllable structure (SSS); 2) similarly, ∆C and varcoUV values for CSS languages were not reliably higher than for SSS languages; and 3) ∆V values for languages which make a vowel length distinction are not always higher than those of languages not making this distinction, again contrary to prediction. However, the findings which support the hypotheses were: 1) nPVI_V values for fixed lexical stress languages were higher than those of variable lexical stress languages; 2) %VO values for languages which make a vowel length distinction were greater than those of languages not making this distinction; and 3) segment timing patterns can be used to classify languages as hypothesized. A new method of analyzing segment-timing parameters for language classification using PCA was proposed. The results from the PCA show that the 12 languages can be classified into 4 groups: 1) MN-KM 2) BM-HM 3) VN-TT-TY and 4) ML-CB. TH, SG, and MI are not explicitly clustered with the other languages. The phonetic and phonological features which seem to influence the 12-language classification are number of syllable in a word, tone, and phonation type.