Abstract
This research investigates three Atayal dialects, namely Pyanan Squliq, Skikun C□uli□ and Mtla□an C□uli□ in Ilan prefecture, northwest Taiwan. The purpose is two-fold. One is to present new data and the other to provide formal accounts. Due to the under-described status of Ilan Atayal dialects in the literature, this research first of all generalizes the consonantal distribution of each dialect and argues that although there are underlying glides, not every surface glide has its underlying status. What follows is the discussion of the long term debate on the phonological or phonetic status of the schwa between consonants, i.e., galu□ + an ‘sympathize + LV’--> [g□lu□an], (Egerod (1965a), Li (1980), Rau (1992) and H. Huang (2006d)). Following Huang (2006d), this thesis argues that the schwa actually has its phonological status; what really happens outside the rightmost foot is vowel reduction, rather than vowel deletion. Such an empirical fact of vowel reduction is argued to be best explained by *Complex (against onset clusters) >> Foot-Bin >> *Full Vowel (Foot) (against full vowels outside the foot) >> Parse-Syllable >> *(□) (against schwas within the foot) in the framework of Optimality Theory (McCarthy and Prince (1993)). The implication of such a constraint ranking is that a cripple vowel, or a schwa is necessary between the otherwise consonant clusters of polysyllabic input. A further prediction is that to best satisfy Parse-Syl, total vowel deletion is also possible if doing so does not lead to violations of phonotactics, which is fully supported by data with vowel-glide sequences. Extended issues involve vowel coalescence and syllable loss. In monomorphemic data, although word-final off-glides remain intact from coalescence, word-internal off-glides however undergo coalescence without exceptions, due to Max-IO-C, Max-IO-μ and *Comp-PV (against complex postvocalic segments). In polymorphemic data where vowel sequences occur at morpheme boundaries, high-vowel-ended stems either undergo gliding or coalescence in these dialects, whereas a-ended stems undergo coalescence in Mtla□an. Further changes of /aj, aw/ to [ej, ow] in Mtla□an are due to the dialect-specific constraint that bans sharp falling sonority in the nucleus domain. Corresponding data in Skikun, instead of undergoing coalescence, presents /l/ preservation (or /l/ insertion) to avoid the complication of vowel sequences at morpheme boundaries, and Pyanan behaves as an in-between case, suggesting the diversity of repair strategies of Onset violations for vowel-vowel sequences. Syllable loss is argued to result from the constraint reranking: Parse-Syl >> Max-OO and there are three factors that govern the application of this new grammar, namely the sonority distance between the onset and the nucleus, phonetic cues and the markedness hierarchy of place features of the onsets. Although not every prepenultimate syllable undergoes syllable loss, every lost syllable stands outside the prosodic foot, suggesting that there are positional faithfulness constraints that can properly protect the segments inside the foot after footing.