Statistical evidence for rule ordering

Abstract

ABSTRACTA set of ordered rules for generating variants of a variable determines (a) underlying/surface distinctions among some of the variants and (b) a hierarchical classification of the variants. In the analytical framework of variable rules, frequency data on variant occurrences in context bear only on (b) and not on (a). We provide a combinatorial characterization and enumeration of the set of classifications on n variants, the set of underlying/surface configurations, and the set of rule orders. We describe the statistical and computational techniques for generalizing variable rule analysis to the inference of rule order. These procedures are applied to the problems of the reduction of syllable-final consonants <s>, <n>, and <r> in Caribbean Spanish (n = 3, 4, 6 variants, respectively). Previous analyses have tended to assume that successive weakenings occur in an intrinsic order determined by phonological strength. Our results show that aspiration and deletion can indeed be seen to be intrinsically ordered in both <s> and <r> reduction, though an unordered analysis is equally likely in the case of <s>. On the other hand, velarization and deletion of <n> are unordered, while vocalization is a subsequent process, independent of the other two. Similarly, spirantization, aspiration, and lateralization of <r> are unordered, as confirmed by data sets from both Puerto Rican and Panamanian speakers. Furthermore, with both <n> and <r>, intrinsically ordered rule schemata proved to be extremely unlikely by statistical criteria. Syllable-final consonant reduction then consists of largely independent processes, most of which occur simultaneously.