In this section I present a structural analysis of the surface phonology of English vowels. A system with a five-vowel base (i, e, , o, u) combined with various glides is discussed and dismissed; then, a ``base-6'' system (i, e, æ, , o, u) is examined and various apparent problems and actual virtues are brought forth, including a discussion of the structure of the phonological representation of postvocalic /r/ (cf. Bhat 1974).
It was shown above that RA English syllables contain a glide slot, which is occupied by at most one of /y, w, r/. Economy will be served if some distinctions among English vowels can be attributed to this independently motivated structure instead of to other vocalic features. Therefore the contrasts between Wells' classes FLEECE and KIT, GOOSE and FOOT, FACE and DRESS, GOAT and STRUT, which could be attributed to a static feature such as [tense], [peripheral], [long], etc., can instead be attributed to the (temporal) structure of the syllable, that is, the presence or absence of a glide slot. Consider the preliminary hypothesis, following Trager & Bloch, that the long vowels contain glides which are identical to the front and back glides of /y, oy, w/. Then one may put these vowels, for which the details of the feature-specifications for the glides are unclear, into a structure with the short vowels, the low diphthongs, and the r-glides. I make the basis a 5-vowel system, because the Vy and Vw vowels, joined together, fit into a space with 5 slots. This analysis is represented in Table .
Three low vowels cannot be fitted into this system: /æ, , a:/ as in TRAP, THOUGHT, and PALM,3.45 traditionally named ``short A'', ``long open O'', and ``broad A'', respectively. The unstable status of these low vowels is diachronically evident in that in different dialects they variously undergo raising, breaking, merger with other vowels, etc.
In some dialects (in particular, Reference American), these vowel classes are kept distinct, so it is important to provide a structural representation for them. Considering the set /, æ, :, a:/ as a group, we may divide them into short and long, front and back, as in Table .
The reason /:/ is considered to be long, while // is considered short, is that /:/ occurs in free position, while // does not. Further, they have historically been analysed as long and short (cf. the traditional names ``long open O'' and ``short O''). Similarly, /æ/ (``short A'') is a checked, therefore short, vowel, and /a:/ is a free, therefore long, vowel. Given that this length dimension is necessary for the low vowels, the same dimension might also be used for the mid and high vowels as well. In Trager & Bloch's system represented above in Table , /h/ signifies both length and ingliding, so the Vh vowels are their (pure) long vowels, which I will signify as V: rather than Vh. The three words idea, yeah, huh and words of those classes must be eliminated if the sets FLEECE, FACE, GOAT are to take their places in the V: sub-system. Idea seems properly classified with words like Maria, Judea, diarrhea, etc., in which the final vowel sequence contains two syllables, and may be written /iy$/.
Yeah, huh are among a quite small class of utterances in English which are difficult to analyse phonologically, including the children's teasing call, [`nænænæ`nænæ], the agreement sounds [~h~, mhm, nhn, h], and the disagreement sounds [~ ~, mm, nn, ].
Huh is a monosyllabic form similar to the bisyllabic forms in the agreement or disagreement classes. Indeed in at least one meaning it can be replaced by hmm (also acceptable as [hn], [h]). These are evidently quite different from fully specified phonological forms, since they lack specification for place of articulation or even oral closure for the syllabics. If they freely vary from forms including the default vowel  to forms including syllabic nasals at various places of articulation, it would seem that they need not be given a phonological representation in the language.
The phonological form of yeah, if there is one, is difficult to establish. In Philadelphia, I notice that African American speakers say [y:]; I believe I pronounce the word as [yæ:]. In the Chicago speech analysed below, while Rita says [ys] for yes, she has a long glide down the front of vowel space for yeah, which could be suggests either [yæ] or [y], but not [y], which would be expected if it had the vowel in DRESS ([drs]). If the form is [yæ], then it is like the teasing call, [`nænænæ`næ`næ], in that it contains an utterance-final, stressed [æ], which would seem to be ruled out by the fundamental principle of English phonology, that short vowels (like ``short A'', or /æ/) cannot occur in a stressed syllable without a coda. Instead of abandoning basic principles of English phonology to accomodate these marginal forms, we may instead classify yeah with affective forms like those discussed above, which do not have a fully specified, normal, phonological representation in the language.
With idea, yeah, huh now eliminated from Trager & Bloch's long vowel sub-system, gaps which correspond neatly to the /iy, ey, ow, uw/ classes are left empty. The lexical sets, PALM and THOUGHT, occupy the two low slots in the V: sub-system.
Next, notice that the vowels which may now be written as /i:, e:, :, o:, u:/ (as well as lengthened /æ:/ in dialects where /æ/ has become long) undergo raising of the nucleus relative to the corresponding short or simple vowels. This does not apply to the nuclei of the ``true'' diphthongs /y, w, oy/, which in T&B's system are classified with the above vowels as containing /-y, -w/. This raising relationship is sometimes referred to as the vowel-before-vowel raising rule, discussed below, which applies to the nuclei of all these long vowels, but not the true diphthongs, in various dialects. Thus it would be an improvement on T&B's system to analyse the mid and high vowels as /i:, e:, o:, u:/, that is as long vowels similar to /a:, :/, so that this rule may be stated simply, as the raising of long vowels. This would imply that the system has a six vowel base, an implication that is an important prerequisite for any system in which /æ/ is to be analysed as a short vowel along with /, , , , /. I therefore propose as the surface phonological vowel system for Reference American, the ``base-6'' structure given in Table .3.46