[Author Login]
Title:Gradient Phonotactics in Muna and Optimality Theory
Authors:Joe Pater, Andries W Coetzee
Comment:Superseded by ROA-842.
Abstract:Muna is a western Austronesian language spoken in Indonesia (Berg 1989). Like Semitic (Greenberg 1950), Javanese (Uhlenbeck 1949), and many other languages, Muna has restrictions on the co-occurrence of consonants with the same place of articulation: words like [kaga] are much rarer than words like [kaba]. As in these other languages, the strength of the restriction depends on the particular place of articulation; it is weaker in coronals than other places. It also parallels these other languages in that the strength of the restriction depends on whether the consonants are alike in other ways. Muna is unique in that voicing plays a dominant role in determining co-occurrence rates: pairs of homorganic consonants agreeing in voicing (e.g. /m-b/) are rarer than those that disagree (e.g. /b-p/).

We analyze the Muna data using an expanded set of relativized OCP constraints (Padgett 1995). As Frisch et al. (2004) point out in their discussion of similar facts in Arabic, the main analytic challenge is to capture the gradience in the degree to which these constraints are obeyed in the lexicon. Some constraints hold absolutely, like the constraint against multiple voiced dorsals (OCP-DOR-[VCE]), some hold of all but a few words (e.g. the constraint against multiple voiced labials OCP-LAB-[VCE]), and some admit many exceptions, but fewer than would be expected if the consonants were subject to no restriction (e.g. the constraint against multiple voiced coronals OCP-COR-[VCE]). We suggest that native speakers� knowledge of the relative strength of these constraints can be expressed by ranking: OCP-DOR-[VCE] >> OCP-LAB-[VCE] >> OCP-COR-[VCE]. Lexically specific faithfulness constraints interspersed between these constraints allow words to violate the markedness constraints, and help to provide a way of calculating relative acceptability.

We also analyze Muna in terms of Frisch et al.'s (2004) similarity metric. While some aspects of this approach are compatible with the Muna data, neither the strength of the voicing effect, nor the extent to which the co-occurrence restrictions are weakened amongst the coronals is predicted by this theory. This suggests that the greater predictiveness of this theory relative to a theory with rankable OCP constraints, held by Frisch et al. to be an asset, may be a liability.
Type:Paper/tech report
Article:This article has been withdrawn.