One of the most important insights of Optimality Theory (Prince & Smolensky 1993) is that phonological processes can be reduced to the interaction between faithfulness and universal markedness principles. In the most constrained version of the theory, all phonological processes should be thus reducible. This hypothesis is tested by alternations that appear to be phonological but in which universal markedness principles appear to play no role. If we are to pursue the claim that all phonological processes depend on the interaction of faithfulness and markedness, then processes that are not dependent on markedness must lie outside phonology. In this paper I will examine a group of such processes, the initial consonant mutations of the Celtic languages, and argue that they belong entirely to the morphology of the languages, not the phonology.
Celtic consonant mutationsThe initial consonant mutations of the Celtic languages are of great interest in theoretical linguistics because they appear to be (and are frequently argued to be) phonological rules that apply in morphosyntactic rather than phonological environments. Within phonological theory they are interesting also because many of the individual mutations have different effects on different classes of sounds. Thus, for example, Eclipsis in Irish voices voiceless stops (e.g. p AE b) and nasalizes voiced stops (e.g. b AE m). The Soft Mutation in Welsh voices voiceless stops (e.g. p AE b) and spirantizes voiced stops (e.g. b AE v). The processes are thus not uniform and only partially predictable. In derivational phonology, devising phonological rules to account for the mutations is challenging, but not impossible. Derivational phonology allows its rules to be arbitrary and independent of universal markedness principles. But the advent of OT has forced a rethinking of the nature of phonological processes: according to the new theory, phonological processes are predicted to result from the interaction of markedness and faithfulness. In this paper I will argue that since the morphosyntactically conditioned consonant mutations of Celtic do not result from that interaction, they cannot be phonological. Instead, the mutations are best regarded as being exclusively in the domain of the morphology, not the phonology at all. I will argue that, like inflected forms, mutated forms are listed separately in a word-based lexicon as allomorphs, and that the selection of mutated allomorphs is determined by a form of government similar to that determining the distribution of Case.