Violable is variable: Optimality theory and linguistic variation

Publisher Website
Google Scholar
Abstract
Optimality theory (OT) (McCarthy & Prince, 1993; Prince & Smolensky, 1993) has been proposed as a constraint-based theory of phonology in which the phonological facts of each language are accounted for by a language-specific ordering of a universal inventory of constraints. The constraints, expressing desirable (i.e., optimal) phonological states, evaluate possible candidate forms, selecting the optimal output. Any constraint may be violated by a surface form if it is overridden by a higherranked constraint; the ordinal sequence of constraints provides a weak quantification of constraint effects. Variability has been treated within OT by varying constraint orders. This model is analogous in several important respects to the variable rule model (VR) of Labov (1969) and Cedergren and Sankoff (1974). In VR, variable constraints express desirable phonological states which are variably realized on the surface, when not overridden by other constraints; constraints are probabilistically quantified. This article compares the OT and VR models, arguing that the VR model is superior on theoretical and empirical grounds: constraint effects in VR are stable, transparent, and learnable. Moreover, the probabilistic treatment of constraint effects allows VR to model successfully cases in which multiple violations of a single constraint lead to a cumulative reduction in likelihood of a form; such cases cannot be efficiently treated in OT.