Can the concept of ``vowel'' be defined on syllable structure? (2) above constitutes a theory of the structure of the vocalic part of the syllable: each of the three classes, nucleus, glide, and /r/ fall in a separate slot in the syllable. However, we have now replaced (2) by the simpler scheme in (5):
(5) Nucleus (Glide)
This theory is not immediately compatible with the standard view of syllable structure, in which the rhyme consists of a nucleus preceding a coda. The nucleus is an accepted syllabic constituent, with the same name. But the glide position per se is not an accepted constituent of the syllable. Consider the widely accepted syllable structure in (a).
Postnuclear glides, which are not explicitly located in this structure, must occur somewhere within the rhyme. I have shown that glides must occur in a single, non-branching location within the syllable. The fact that it is non-branching may be emphasized by calling this location the glide ``slot''. This may be easily expressed as the claim that the glide slot is a labelled, non-branching constituent within the English syllable. The glide node cannot be identical with the coda, because it is non-branching, while the coda can contain multiple segments, in English.3.34 Nor can it be identical with the nucleus, which precedes it. Therefore the glide node must be within the nucleus as in (b) or within the coda as in (c); or conceivably as a daughter of the rhyme node itself, as in (d).
Various ways of renaming the nodes in (b) and (c) are possible, but the labels are not material to the arguments below. Names like peak, satellite, head, tail, vowel, etc., are available. In (b), for example, I would like to relabel the node nuc' as vowel, since this would give a maximally simple formal specification of the target of study of this thesis: Vowels are nucleus-glide combinations. Others may prefer to relabel nuc as peak, and nuc' as nucleus. In (c), the node coda could be renamed tail, and coda' renamed coda. The ``X-bar''-like notation used here is intended to highlight the arbitrariness of node-labelling.
Which of the structures (b, c, d) is correct? This is an important issue in phonological theory. Since stating the generalizations found in this chapter is somewhat simpler using (b) than with (c), I will examine some relevant arguments below. However, the main arguments of this chapter could be restated in any of the three forms.
Goldsmith (1990:109) states that arguments for a branching nucleus constituent within the syllable are unconvincing. If so, then, we should prefer (c) or (d). The flat rhyme structure in (d) is an unhappy compromise. (d) would allow statements in which glide and nucleus are grouped together to have equal status with those in which glide and coda are grouped. If there were some large number of rules that refer to glide and coda together and an equally large number of rules that refer to nucleus and glide together, economy will still be served if one of the pairs is put into a single constituent. Then at least half of the references in these rules will be simplified, which is a definite improvement over the flat structure, (d), in which none of them would be simplified. In short, the unhappy compromise of (d) amounts to lost generalizations relative to either (b) or (c), and is thus inferior. Next I propose some arguments in favor of the branching-nucleus, (b).
What is the structural significance of the branching, tree-like structure of the syllable? The application of constituent-structure analysis to syllabic constituents suggests that the tree structures within syllables should have the same structural significance as syntactic tree structures. If the trees in syllable-structure have the properties of context-free grammars, then constraints across constituents, which amount to context-sensitivity, should be impossible. The definition of ``context-free'' is that elements in distinct constituents cannot influence each other, while elements within a constituent may. Syllable structure certainly is not perfectly context-free, since there do exist constraints between onsets and nuclei, between nuclei and codas (e.g., the pen/pin merger in the Southern U.S. See also Borowski 1988), etc. Nevertheless, to the extent that syllable-structure is context-free, combinatory constraints within the constituents of the syllable are acceptable, but across constituents, they are not.
Severe restrictions on combination of vowels and glide are common in the world's languages (cf. Maddieson 1984). On the other hand, vowels typically combine quite freely with other post-nuclear consonants. Since the context-free grammar view of syllable structure restricts combinations of elements within constituents rather than across constituents, the existence of pervasive restrictions on nucleus-glide combinations, along with the rarity of restrictions on nucleus-coda combinations, would be evidence that nucleus and glides form a constituent distinct from the nucleus-coda constituent (the rhyme). This suggests alternative (b) above.
Another argument for (b) comes from historical phonology. The vowel classes that generally act as a unit in historical change are units formed from the nucleus and glide together. Jamaican Creole provides a number of mergers in which nucleus-glide sequences underwent changes as a unit, as may be seen in transcriptions provided in Wells (1982, vol. 3, p. 576). In Jamaican Creole, a number of Wells' lexical sets merged into one, while their counterparts without the rhotic glide did not merge. Thus NEAR and SQUARE (/ir, er/) merge, while FLEECE and FACE (/iy, ey/) remain separate, as do KIT and DRESS (/I, /). Similarly CURE and FORCE (/ur, or/) merged, with no effect on their counterparts GOOSE, GOAT (/uw, ow/) or FOOT, STRUT (/U, /).
The low vowels (including /oy/) merged into three classes according to their glides: front-gliding, back-gliding, and neither. Thus the nucleus-glide sequence of the low, front-gliding diphthongs of PRICE, CHOICE, (/y, oy/) merged into JCE /ai/; the long or /r/-gliding vowels of THOUGHT, CLOTH, PALM, BATH, NORTH, and START (/:, :, a:, a:, r, r/, all merged into JCE /a:/, and MOUTH (/w/), the only back-gliding low vowel, remained distinct. Besides the mergers of LOT and TRAP (/, æ/) and of the unstressed vowels in LETTER, COMMA (/, /), this amounts to a fairly thorough description of the phonological changes in the genesis of Jamaica Creole.
We have seen that it is the nucleus and glide together which define the units which undergo the merger, and which define the units which result from the mergers. In this and many other examples, the nucleus and the glide act together as a unit in historical change. Stating these changes is made simpler when these are treated as a single unit, as in structure (b) above.
Finally, the nucleus and glide positions contain segments which are quite similar in sonority and in phonetic and phonological substance (cf. page ), while the coda position is less sonorous and distinguishes different kinds of features.
In order to retain the structure in (c), we must reject fundamental assumptions about the significance of branching structure in the syllable: it is nothing like context-free grammar. We must accept increased complexity in historical phonological description. And we must believe the counter-intuitive claim that elements in distinct constituents may be phonetically more similar to each other than elements within a single constituent. For these reasons, despite Goldsmith's claim to the contrary, the structure in (b) seems superior.