Abstract
This thesis aims to codify phonemic segments, define a syllable and expect to obtain further evidence for both two issues by loanwords, based on newly- collected data.The significance of the presentation about all of the possible consonant clusters in Tsou includes the great violation of Sonority Dispersion Principle (Clements, 1990), and the plausibility of homorganic consonant clusters (Tung 1964, Hsin 1999). Due to onset phonotactics, vowel~glide alternation in numeral compounding and the dialectal correspondence, glides are excluded in the inventory. Besides, reduplication, stress assignment, alternate vowel deletion (Wright and Ladefoged 1994, 1997), minimal pairs, pitch (Tsai 1999) are proved to be insufficient evidence to clarify if there are underlying long vowels. ‘Contrasting pairs in regard to the occurrence and non- occurrence of stress and the position of stress’ (Tung 1964:40~41), which seems to imply that stress is sometimes phonemic are not found; stress assignment always predictable falls on the penultimate syllable. Irregular stress assignment of some words in the older generation results from the historical and dialectal /r/ ~ /e/ correspondence.The evidence Tung (1964) and Hsin (1999) use to argue for non- coda assumption, namely empirical evidence, reduplication and affixation is proved to be insufficient; empirically there are content words that end with codas and (C(C))VC. CV syllabification and the C(C)V template are compatible with the constraint ranking MAX I-O >> NO-CODA >>MAX B-R (Chen 2001), under the principle of the emergence of the unmark (McCarthy & Prince 1994a). Phonotactics, the CCVC reduplication template and language games however show that Tsou is not a language that disallows codas.Despite the fact the intention to use loanword preservation and modification to testify the issues of native inventory and plausibility of codas fails, predictable stress assignment of Japanese loanwords mixed with native words still enforces the argument that stress is not phonemic. Furthermore, truncation of loanwords in a Tsou way is newly unearthed. The observation of X- slot- based segmental preservation is also represented, although the theoretical implication is not elucidated yet.Color terms, heavy-light redupliction, criteria for multiple reduplication templates and the segmental/ tonal preservation/ modification of Taiwanese loanwords are the suggested issues for feature research.