Introduction

In chapter 7 I showed that when stress is assigned to a compound or a phrase, the final empty nucleus of the first term may occupy a position where it is a potential complement of a binary foot. I claim that when such a situation occurs the final empty nucleus of the first term is incorporated within the binary foot unless it must government-license its onset. Since a metrical incorporation entails the loss of its government-licensing property, the empty nucleus is phonetically manifested, thus preventing its metrical incorporation. In conclusion, as is the case for proper government, when ‘metrical’ government is in conflict with government-licensing French opts for government-licensing. An empty nucleus fails to be properly or metrically governed when it has to government-license its onset. In terms of stress assignment, we capture the behaviour of the final nucleus of the first member of a compound, and also the difference between certain dialects of French with respect to phrases. I turn next to the alternation between schwa or zero with the vowel [ɛ], which I claim is also a consequence of stress assignment.

The alternation between schwa/zero and [ɛ]

Up to this point in the analysis we have considered cases where an underlying empty nucleus is realized either as zero or as schwa. We will now consider another manifestation of an underlying empty nucleus: its alternation with the mid vowel [ɛ]. Consider a verb such as harceler ‘to harass’, for example.

This book represents an expanded and modified version of my Ph.D. dissertation (Charette 1988). The sections of my thesis which have been published will not be found here. The modifications involve certain refinements dealing with the level at which proper government operates. The notion of licensing is now more clearly defined, especially with respect to licensing as a condition for government. My treatment of compounds now directly follows from my analysis of word-internal empty nuclei. Some problematic data concerning the alternation of schwa with [ɛ] are now accounted for in terms of the effect of a floating consonant on a word-final empty nucleus.

This book is addressed to any linguist interested in phonological theories in general and Government Phonology, the framework in which it is written, in particular. Those who are unfamiliar with Government Phonology will find a fairly detailed presentation of certain aspects of this theory in chapter 1. The reader already familiar with this framework will be interested by new proposals and the sharpening of certain notions. This book may also be of interest for its new analyses of certain phenomena in French.

Since this book represents an expanded version of my dissertation, I wish to thank those who helped me with both the original work and the preparation of this book. In particular I am grateful to Jean Lowenstamm and Glyne Piggott for their help, support, comments and suggestions.

Introduction

Descriptions of French typically assume the existence of a vowel schwa, the so-called ‘e-muet’ or ‘unstable-e’, the salient property of which is that it alternates with zero, thus making it phonologically different from all the other vowels of French. While in certain dialects of French, schwa is phonetically similar to the vowel /œ/, it is phonologically different from this phoneme. In this chapter I present the properties of schwa along with some relevant data which lead to the conclusion that schwa is phonologically different from /œ/ as well as from all other vowels of the phonemic inventory. The [ə] (i.e. schwa), or [œ], which derives from schwa, has one salient property. Under certain circumstances it alternates with zero or with [ɛ]. Unlike the [œ]s which are allophones of schwa, those which express the phoneme /œ/ do not alternate either with zero or with [ɛ]. The latter alternates with the vowel [ø] (e.g. ils veulent [vœl] ‘they want’ versus il veut [vø] ‘he wants’, malheur [malœr] ‘misfortune’ versus malheureux [malørø] ‘unfortunate’).

The alternation between schwa and zero

I begin the overview of the facts by considering the alternation between schwa and zero. I present different types of cases where such an alternation is found.

Word-internal and word-final schwas

The alternation of schwa with zero can be observed in comparing the two groups of words given in (1). In both groups of words schwa is preceded by two consonants.

Introduction

In this chapter I propose a new account of the behaviour of schwa, aiming to demonstrate that its properties follow from general principles of phonological theory. From such a perspective, the behaviour of schwa is neither accidental nor is it specific to French. My analysis can be extended to other languages where similar phenomena are found. It will be argued that in languages where a vowel alternates with zero, the properties of this segment always follow from the same principles of the theory. What distinguishes one language from another is the nature of the segment which alternates with zero. Therefore, the alternation between [∅] and zero in French, between [u] and zero in Tangale, between [ə] and zero in Khalkha Mongolian and between [ɨ] and zero in Moroccan Arabic are accounted for in terms of the same set of principles and parameters. My analysis of schwa will also lead to a better understanding and to a sharpening of the theory of government.

Interestingly, my analysis embodies proposals that have been made by Selkirk (1978) and Anderson (1982) in their respective analyses of schwa. Like Anderson I analyse schwa as the phonetic interpretation of an underlying empty nucleus, and like Selkirk I consider that a given nucleus is in a binary relation with another nucleus.

Introduction

In this chapter devoted to word-final empty nuclei, I argue in favour of the principle of ‘Coda’ Licensing proposed by Kaye (1990c), according to which a consonant occurs within a rhyme if it is licensed by an adjacent segment in the onset position. It follows from this principle that every word or more precisely every domain contains a final nucleus in its representation. While in some languages a final nucleus may be underlyingly empty (e.g. French, English, Portuguese etc.), in some other languages it may not (e.g. Dessano, Dida etc.). The ‘Coda’ Licensing Principle is universal. But in those languages where a word-final nucleus may be underlyingly empty, subject to parametric variation, those word-final empty nuclei may or may not be manifested as null. From this parametric choice, languages like Brazilian Portuguese (BP) and Japanese are distinguished from others like English and French. While in French and English a word-final nucleus can be empty and realized as zero, this is not the case in languages of the type of Japanese and BP, where words always end with a vowel. That is, while word-final empty nuclei are licensed in French and English they are not licensed in languages of the type of Japanese. In addition, in languages in which word-final empty nuclei are licensed, there is variation as to whether the empty nucleus may be manifested as null after a consonant cluster.

Introduction

Up to this point in the analysis of schwa I claim that a word-internal empty nucleus is realized as zero if it is properly governed. Moreover, in the preceding chapter I argued that a word, or more precisely a domain, universally ends with a nucleus. In French a domain-final nucleus can be underlyingly empty; if it is empty, it is licensed (i.e. it is authorized to remain without segmental content). Finally, a licensed domain-final empty nucleus is a government-licenser for its onset. These proposals seem, however, to encounter a problem with compounds. While in certain types of compounds the final nucleus of the first term is never manifested, it must be phonetically expressed in others. The question now is why the final nucleus of the first term behaves as in word-final position (i.e. it is never manifested) in one type of compound while in another type it behaves as if it were word-internal (i.e. it is realized as zero unless it is preceded by a consonant cluster). To account for these facts I argue that while the final nucleus of the first term of a compound occurs in word-final position and is therefore licensed, its behaviour is determined by stress assignment. Let us first consider the facts.

The facts

One type of compound has as its first term a word whose final nucleus is underlyingly empty and is preceded by a single consonant. The second term can either be phonetically monosyllabic or polysyllabic.

Introduction

I conclude the analysis of schwa by considering some additional contexts where schwa alternates with zero. While some of the facts discussed in this chapter follow directly from my analysis, some others do not. Some of the data do appear problematic for my treatment of schwa. In these instances I do not always have a clear understanding of the behaviour of the data, but I nevertheless suggest a direction of research which might be worth exploring. I also reconsider the controversial behaviour of schwa occurring in word-initial syllables. In attributing to a word-initial syllable a special status, I capture similarities between French and other languages and I account for the difference between Quebec and Parisian French with regard to the behaviour of empty nuclei in word-initial syllables of bisyllabic words. Finally, it follows from my proposal that the two dialects differ as far as proper government of an empty nucleus in initial syllable of bisyllabic words is concerned, but are similar in their treatment of this nucleus in initial position of polysyllabic words. I start with sequences of clitics, one of the classical problems with respect to the behaviour of adjacent schwas.

Sequences of clitics

Without going into a detailed syntactic analysis of clitics, we may roughly say that syntactically object clitics are always attached to a verb.

Introduction

In this chapter I consider word-internal empty nuclei preceded by a consonant cluster. I provide a principled account for the fact that in this context a properly governable empty nucleus fails to be properly governed and receives a phonetic interpretation. To do so I introduce a new notion: government-licensing. I argue that a non-nuclear head can govern a complement only if (i) it has the required charm value or, if charmless, the required complexity, and (ii) it is government-licensed by the nucleus governing it. Government-licensing accounts for, among other things, the phonetic realization of an empty nucleus following a governing onset (i.e. a non-nuclear skeletal point which governs a complement), or for the simplification of a consonant cluster preceding an unrealized empty nucleus. If, in order to govern a complement, a non-nuclear point must be government-licensed and if a properly governed empty nucleus may not be such a licenser, in a situation where a properly governable empty nucleus follows a consonant cluster, the two following possibilities are predicted: (i) in order to government-license the consonant, the empty nucleus will fail to be properly governed, or (ii) proper government will apply, thus preventing the non-nuclear head from governing its complement, which entails the loss of this complement or of this non-nuclear head.

For better and worse, almost every philosophical development of significance since 1800 has been a response to Kant. This is especially true on the subject of a priori knowledge. The central problem of the Critique had been the a priori, and Kant had dealt with it from the complementary perspectives of judgment and experience. His “Copernican revolution” gave him a theory of experience and a non-Platonist account of the a priori. But when the Critique was well on its way, Kant discovered the notion of a synthetic a priori judgment, and he saw in this a particularly appealing way of formulating his project as that of explaining how such judgments are possible.

The constitutive dimension of Kant's theories of experience and the a priori will figure prominently in later developments. As we shall see, one of the turning points in our story will involve a Copernican turn, though the issue it concerns will be different from Kant's.