Abstract

Detailed study of data in Dixon's (1977) grammar indicates that previous analyses of Yidi# have erred in supposing that the synchronic pattern of the language continues the historical pattern, whereby various nominal stems have their underlying final vowels deleted when no suffix follows. Instead, it appears that the system has undergone a radical reanalysis, whereby the suffixed forms are now projectable by general principles from the isolation forms. More precisely, a pattern of multiple predictability has developed: the form of suffixed allomorphs is largely predictable from the isolation allomorphs, but the older pattern, whereby isolation allomorphs can be predicted from the suffixed allomorphs, also persists. From this descriptive result, three principal theoretical consequences are developed: (a) Yidi# possesses a fully-productive pattern of alternation that is not driven by markedness-faithfulness interactions; (b) the phonological constraints that are active in Yidi# likely include some that are quite unlikely to be members of a universal inventory; (c) there are more relations of predictability among surface forms in Yidi# than can be treated by the normal method, namely that of deriving all the surface allomorphs from a single underlying representation. A tentative suggestion is made for how Optimality Theory might be extended to treat cases of this sort, by means of a class of "Anticorrespondence" constraints. Phonological Restructuring in Yidi# and its Theoretical Consequences 1.

Citations