Abstract
This thesis aims to demonstrate that the treatment of Saisiyat phonology after the loss of the flap L can be grounded in two different approaches—whether the moraicity of coda consonants should be taken into consideration or not. The approach that suspends the role of the moraicity of coda consonants, that is, a non-moraic-coda approach, is more advantageous in terms of explanatory unification. Saisiyat is one of the Formosan languages with two branches: the northern dialect is spoken in Hsinchu County, and the southern dialect is spoken in Miaoli County. A recording of Saisiyat vocabulary conducted early by Li (1978) reveals that a flap L had been existent in Saisiyat’s consonant inventory but was lost afterwards. The aftermath triggered by the loss of L is observable in contemporary Saisiyat phonology such as the surface long vowels resulting from compensatory lengthening (CL), the increase in the number of onsetless syllables and vowel clusters, glide formation, where a high or back vowel is resyllabified into the coda of the previous syllable, resyllabification, where a coda consonant is resyllabified as the onset of the following syllable, and so on. The moraic-coda approach, on the one hand, will inevitably be confronted with the concomitant explanatory difficulties given the theoretical framework on hand. The reason why coda consonants can be viewed as mora-bearing units in Saisiyat is because the standard Moraic Phonology developed by Hyman (1984, 1985) and Hayes (1989) regards CL as a phenomenon resulting from the conservation of the mora left by the lost coda consonant. However, given the concept of mora conservation, this model predicts, though rather implicitly, that resyllabification would lead to CL. But the data in Saisiyat suggest that CL only takes place in cases involving coda loss, rather than in cases involving resyllabification. In terms of Optimality Theory ((OT) Prince and Smolensky 1993, McCarthy and Prince 1993), although one-level OT can predict that resyllabification won’t trigger lengthening, it can’t resolve CL that is opaque in nature. Two-level OT, under the framework of Derivational Optimality Theory (Rubach 1997b, 2000a, 2000b), can resolve CL by ranking the constraint MAX-μ high at the second level, the constraint in question nevertheless needs to be ranked low at the same level in evaluating data related to resyllabification cases so that lengthening won’t be falsely predicted. That is to say, if a moraic-coda approach were adopted, cases of resyllabification and those of CL triggered by the loss of the flap in Saisiyat would always cause conflicting analyses under existing theoretical frameworks. On the other hand, if we don’t resort to the moracity of Saisiyat coda consonants, since there is no other independent evidence other than CL supporting this moraic-coda assumption, this conflict caused by the ranking of a particular constraint will not arise. Thus, a more unified analysis can be put forward. What is crucial in this non-moraic-coda approach is that following Kavitskaya (2002b), CL resulting from consonant loss diachronically doesn’t involve any transfer of length or weight, but can be attributed to the inherent vowel length contiguous to certain coda consonants, which is later reanalyzed as phonologically long after consonant loss. However, if coda consonants are proved to be truly moraic in one language where both resyllabification and CL take place due to consonant loss with the former showing no lengthening, the conflict between the analysis of resyllabification and that of CL still demands some modifications of modern phonological theories, as is also pointed at the end of the thesis.