Breit Florian. Welsh mutation and strict modularity

University College London (University of London), 2019. —348 p.The generative view of the language faculty is premised on the modularity of mind. In its most restrictive form, strict modularity, this predicts that phonological processes cannot be sensitive to syntactic information and vice versa. Initial Consonant Mutation in languages such as Welsh appear to falsify the strict modularity hypothesis in relation to the morphosyntax-phonology interface. This is because mutation involves the triggering of a phonologically regular process in a morphosyntactically determined environment. Consider the Welsh data in (1): (1) /kaθ/ 'cat' (a) [i χaθ] 'her cat' (b) [i kaθ] 'his cat' The plosive-fricative alternation in (1) cannot be purely phonological in nature, because the phonological environment in (1a) and (1b) is identical. The only differentiating factor is the gender of the possessive pronoun. This is problematic because it appears as though the phonology must make reference to information proper to morphosyntax. Virtually all preceding accounts of Welsh mutation violate strict modularity. These accounts employ ad-hoc diacritics to mark mutation environments in the morphosyntax. In most accounts, the diacritics are assumed to persist across the interface into phonology, where they can trigger phonological processes. Alternatively, it has been suggested that the diacritics trigger some form of suppletion on the mutation target. However, approaches in the latter category also necessarily appeal to phonology in a manner not compatible with strict modularity. In this thesis I propose a new model of Welsh mutation, combining the floating feature approach of Lieber (1983) with phonologically conditioned allomorphy within a Distributed Morphology framework. I show that Welsh mutation can be accounted for successfully in this way without violating strict modularity. The account is more restrictive and makes better predictions, in particular predicting that variation is trigger-dependent. I develop the first in-depth analysis of items that are exceptionally immutable, proposing that they begin with an empty CV-sequence, a structure independently motivated by weight in Welsh and initial sC(C)-sequences in Irish.