Celtic initial mutation: pattern extraction and subcategorisation

Abstract

In this paper I argue that initial consonant mutation in the Celtic languages does not involve synchronically derived phonological alternation, nor is it the product of full lexical listing of alternant wordforms. Rather, Celtic initial mutation involves associations of consonants represented in the lexicon which relate a specific initial consonant of a radical form to its associated mutation reflexes. Together with subcategorisation, which ensures that the correct mutation reflex of a wordform appears in the correct environment, the appropriate initial consonant is selected from an association of consonants extracted by the speaker from the recurring patterns in the language. It is these consonant associations, not full alternant wordforms, that are listed in the lexicon. The advantages to this approach are threefold: broad patterns are identified which full lexical listing obscures, the participation of neologisms in mutation is accounted for straightforwardly, and some of the motivation for diachronic reanalysis within the mutation system is made more apparent.