Inherent variability and the obligatory contour principle

English coronal stop deletion is constrained by the preceding segment, so that stops and sibilants favor deletion more than liquids and nonsibilant fricatives. Previous explanations of this constraint (e.g., the sonority hierarchy) have failed to account for the details, but we show that it can be comprehensively treated as a consequence of the obligatory contour principle (OCP). The OCP, introduced to account for a variety of categorical constraints against adjacent identical tones, segments, and so forth, can be generalized as a universal disfavoring of sequences of like features: *[aF] [aFJ. Therefore, coronal stop deletion, which targets the set of segments /t,d/ defined by the features [-son, -cont, +cor], is favored when the preceding segment shares any of these features. But this requires adopting the assumption of inherent variability and interpreting the OCP as a probabilistic constraint with cumulative effects (the more shared features, the greater likelihood of deletion). This suggests an attractive theoretical integration of categorical and variable processes in the grammar. The systematic patterns of variation revealed by sociolinguistic studies of language use are generally modeled in theoretical frameworks that rely on the assumption of "inherent variability" (Labov, 1969; Weinreich, Labov, & Herzog, 1968); that is, the hypothesis that the human language faculty necessarily accommodates and generates variation, and that the workings of grammar have a quantitative, noncategorical, and nondeterministic component. This assumption underlies the variable rule model (Cedergren & Sankoff, 1973; Labov, 1969), which is given a more general formulation as a probabilistic generative model by Sankoff (1978). One of the attractive features of such models is that they offer an integrated account in which social and stylistic dimensions of variation can be modeled along with linguistic dimen. sions. Thus, in a standard variable rule analysis, social differences between individuals in a speech community are represented by their characteristic values of an input parameter, whereas independent linguistic parameters describe favorable and unfavorable linguistic contexts for the occurrence of particular variants. Stylistic variation is another independent parameter, which