Representation and variation in substance-free phonology: a case study in Celtic

Abstract

This thesis presents a comprehensive analysis of the phonological patterns of two varieties of Brythonic Celtic in the framework of substance-free phonology. I argue that cross-linguistic variation in sound patterns does not derive solely from differences in grammars (implemented as Optimality Theoretic constraint rankings). Instead, I adopt the substance-free framework, based on the principle of modularity and autonomy of the phonological component, to account for cross-linguistic phonological and phonetic variation. Phonological representations in substance-free phonology are built up without regard to the physical implementation of phonological units, on the basis of the system of contrasts and patterns of alternation. Although this insight is not new when couched in terms of a language-specific assignment of a set of universal phonological features, I argue that the mapping between phonology and phonetics is also not universal and deterministic, and reject the universality of the feature set. Instead, I argue for a rich interface between phonology and phonetics.

Based on this understanding of the nature of variation, I provide a holistic analysis of the sound systems of two closely related languages: Pembrokeshire Welsh and Bothoa Breton. I propose an account in terms of a rich representational theory. Among other proposals, I defend the need for surface ternary contrasts, which I propose to implement using feature geometry. I also show that the substance-free approach, which decouples phonological representation from phonetic realization, strikes the correct balance between innatist and emergentist approaches to phonological markedness; I demonstrate this by way of an extensive case study of laryngeal phonology, which leads to a reinterpetation of the approach known as 'laryngeal realism'. I also argue that the phonological component of grammar should allow constraints with prima facie undesirable factorial consequences, if such constraints are needed to account for functionally unmotivated sound patterns, and discuss the consequences of this approach for the substance-free nature of phonological computation.