The Universal Constraint Set: Convention, not Fact

All languages make the same phonological generalisations. This is the remarkable claim of Optimality Theory (OT). In early generative phonology (Chomsky & Halle 1968), phonological gen-eralisations were expressed by ordered rewrite rules. Each language, however, required its own set of rules as well as its own ordering. Later, underspeci-cation phonology (Archangeli & Pulleyblank 1989, 1994) emphasised default rules. Universal tendencies in the rules were apparent, but characterising all languages with a single set of rules remained an unreachable dream. In OT, phonological generalisations are expressed as ranked defeasible constraints. Ranking provides so many distinct but plausible grammars that it seems feasible that a universal set of phonological generalisations could account for the diversity of phonological systems. The question we face is no longer whether the assumption of such a universal set is theoretically tenable, but whether it is justiiable. There are two senses in which such an assumption could be justiied: either as a fact or as a convention. If a fact, it claims that all language users objectively instantiate the same set of generalisations. If a convention, it encourages phonologists to describe languages using an agreed but arbitrary system of gen-eralisations. In this interpretation, the universal constraint set is as arbitrary, but as useful, as the international phonetic alphabet (IPA). This chapter examines seven kinds of argument for one or other status of the universality of phonological constraints. These are the arguments from empirical evidence (section 2), restrictiveness (section 3), simplicity (section 4), universal markedness, acquisition (both section 5), learnability (section 6), and convention (section 7). Close examination nds all but the last of these arguments to be wanting. The conclusion that remains is that universality, like the IPA, makes a better convention than fact. It should be used rather than believed. Optimality Theory (Prince & Smolensky 1993, for an introduction see Archangeli & Langendoen 1997) is rst and most frequently applied to phonology 1. In this I would like to thank the following people for their comments on an earlier version of this paper: and one anonymous reviewer. I would also like to thank the editors for their patience and interest. 1 The formalism has also been applied to morphology (e.