Stress-epenthesis interaction and defect-driven rules

Abstract

This thesis is concerned with the interaction of two phenomena: vowel epenthesis and stress assignment. Vowels are potential targets for stress, and they count for determining the stress window in systems where stress is grammaticaly predictable. Epenthetic vowels, however, are often exceptional in this respect. There is an array of languages where epenthetic vowels shun stress, or in some other way disrupt canonical stress assignment. The two conditions, unstressability and metrical invisibility of epenthetic vowels, are not universally connected. On the contrary, in a number of languages epenthetic vowels are (only) partially visible to metrical structure.

The present thesis brings together the relevant cases, and proposes a formal analysis of the attested facts. The languages under scrutiny are: Swahili (Bantu), Dakota (Siouan), Mohawk (Iroquoian), Winnebago\ Hocank (Siouan), Yimas (Sepik-Ranu), and Selayarese (Makassar). Data from loanwords in North Kyungsang Korean and Japanese are also presented, but not formally analysed, as the stress turns out not to be gramatically predictable.

The proposed analysis is framed in the defect-driven rule formalism (Frampton 2008), a serial framework where phonological rules are triggered by defects in the input string. The analysis provides an extension of Frampton's model, by considering defect-driven rule interaction and ordering.

The discussion bears on the treatment of opacity in different theories of the nature of the phonological component, as well as on the issues involving serial derivations vs. parallel evaluation, and the role of rules and constraints in modelling phonological computation. It is argued that the data considered here necessitates a serial analysis, and that within serial frameworks, the defect-driven rule formalism makes the most accurate empirical predictions for the current dataset.