Dialect differentiation in Lexical
Phonology: the unwelcome effects of underspecification
Introduction
The investigation of the Scottish Vowel Length Rule raises important issues for the modelling of sound change in Lexical Phonology, a topic to which we shall return. However, it also relates very directly to a synchronic question we have touched on several times already, namely the degree to which different dialects of the same language can vary. Of course, SVLR is a process specific to Scottish varieties; this kind of variation in the form, order and inventory of phonological rules is already familiar from Standard Generative Phonology. But we have departed from the SGP line in also allowing dialectal divergence in the underlying representations: for instance, various vowel oppositions (such as the RP and GenAm Sam ~ psalm and pull ~ pool pairs) are simply neutralized in toto in Scots and SSE; and we saw that the father vowel should be analyzed as underlyingly front in some varieties of English, and back in others.

We might regard these minor, scattered examples as still compatible with a generally panlectal approach to phonology, however, I shall propose a far more general and more radical underlying dialectal difference, involving the dichotomizing feature(s) which establish the structure of the whole vowel system. I shall argue that in some varieties of English only [± tense] is underlyingly relevant; in others, only length; and in still others, both. This approach is clearly incompatible with a panlectal analysis; but as we shall see, rejecting panlectal phonology is no great loss. Less obviously, this analysis is relevant to underspecification theory, which I ruled out, though essentially without argument, I shall show that arguments against underspecification are accumulating in many phonological models, and that its effects are particularly serious and unwelcome in Lexical Phonology. Most notably, whereas a constrained LP will enforce an analysis of each variety in its own terms, implying quite far-reaching underlying divergence, the use of underspecification means that abstract ness can be readily reintroduced, and that shared underliers can be permitted in cases where they are not warranted.