The Domain of Phonological Processes

Abstract

In many Australian languages, stress rules appear to apply to every suffix independently, before they are subjected to word-level phrasing (1). This pattern is difficult to account for whether we derive it in Lexical Phonology (e.g. Poser 1986) or in Optimality Theory (e.g. Pensalfini 2000), since we cannot derive it by applying stress rules to incrementally larger domains which include the stem. The problem appears to lie with our assumptions about how words are built up, and how they are subjected to phonological processes. The usual assumption about word structure is that it is built up in layers from the root, like an onion. In phonology, rules are assumed to apply to domains which minimally include the root. They cannot apply to affixal domains without the root. Here, I defend an alternative view of word structure: one based on adjacent, rather than nested, domains. In this model, outputs from the lexicon are independently subjected to rules of metrical stress assignment, as in (2). The domain of metrical stress is therefore not the (grammatical/distributional) word, but each productive morpheme within the word. Note that syllabification must be applied to the whole form rather than the individual parts, since some affixes apparently begin in clusters. However, we need to allow for inter-word syllabification anyway because of other word sandhi and word syllabification effects. One example of such a pattern applying to word-initial geminates has been described for Swiss German (Kraehenmann 2001). Similar patterns are found in Oceanic languages such as Leti (Hume, Muller & van Engelenhoven 1997). One sidebranch of phonology did investigate the possibility of domains external to the root: ‘prosodic phonology’ (e.g. Nespor & Vogel 1986). This theory was developed in part as an account of 'bracketing paradoxes', where the prosodic structure did not seem to reflect the morphological derivation of a word. The Australian patterns are amenable to a prosodic phonology account. However, prosodic phonology is a declarative model — anything can be potentially declared to be a prosodic domain — and hence over-generates patterns. The model argued for here makes a stronger and more constrained claim: only morphemes which on independent grounds can be shown to be elements of the lexicon may constitute a domain for phonological processes.