Preface

The aim of the present dissertation is to make a contribution - modest as it may be - to the development of a working model of phonological representation. Following the mainstream, I take the autosegmental framework as a given. I will argue that unary features are more adequate in representing melodic oppositions than binary or scalar features. A very limited type of skeletal structure will be assumed, strictly alternating C and V positions, laying the burden of expressing syllable structure on the relationships between these positions. The two relationships involved are licensing and government, with a novel definition introduced for the latter, both of which are local and unidirectional forces. I will show that there is much to gain by repartitioning the skeleton, that is, by claiming that it is made up not of CV units - as all researchers who accept this minimal representation assume - but of VC units. I am also going to examine branching onsets, with the result that these consonant clusters are in many respects similar to those created by syncope in English, especially word-medially, but no firm conclusions will be reached here. The evolving theory will be shown at work in predicting the sites where lenition occurs and its direction too, as well as in explaining phonotactic restrictions affecting consonant clusters. Finally, I will outline some basic assumptions that appear to be important in devising a theory of melodic representations, without taking this last enterprise to an end.

The ultimate goal of linguists is to capture the essence of natural language, that is, to construct models that not only describe the way language works, but also provide theoretically plausible explanations for why things could not be otherwise. This, of course, is not an easy goal, in fact, it probably is an ideal craved for but never to be attained. It is common knowledge that the most fruitful method of approaching this goal is to posit very few axioms and locate the set of phenomena that cannot adequately be accounted for by these. Only as a last resort should new presuppositions be accepted. It follows from this method that there are always bound to be things which the theory cannot predict or which it predicts to be nonexistent. The desirable state of affairs is one in which it is the infrequent things that stick out, while the everyday phenomena are easily accounted for. Thus the fact that a certain theory is unable to provide an obvious means of dealing with, say, branching onsets can even be seen as a merit, given that the existence of word-initial rising sonority clusters (to the exclusion of falling sonority clusters) is quite a unique property of a subset of human languages. There always remain fuzzy edges in theories; locating the problem cases, not trying to incorporate them into the set of explainable phenomena by all means is often the best one can do.

One of the main principles underlying the theory I am developing in this thesis is theoretical minimalism. I try to reduce the conceptual machinery to as little as possible. Such gambles always involve some give-and-take; by reducing syllabic constituency there will inevitably emerge an intricate system of relationships between skeletal positions; by reducing the number of primes, the interpretative conventions are bound to become more complex. It is often useful to take such risks if only to experiment with the extremes: some insightful observations can occasionally be made this way.

Most of the natural language data I present in this theory are from English, the theoretical claims ought, nevertheless, to be applicable to other languages too. Some parametric regularities are noted, but the task of testing the theory, tailored by and large for English, against further systems is one I do not undertake here.

Throughout the thesis I use the symbols of the IPA to indicate the melodic content of strings. To highlight the characters somewhat I typeset them with a sans serif font, which also makes enclosing brackets or solid! unnecessary. The sometimes arcane paragraph numbering is to help those who possess the previous version of this text and also to ease internal referencing.