In the interests of clarity, let us consider the nucleus, glide, and coda to be mutually exclusive, so that the coda contains the post-vocalic segments other than /y,w,r/. In Section below I discuss the issue of how these three elements fit into syllable structure. For now, let us consider them as non-overlapping, distinct constituents, which occur at most once in each syllable. This will make it easy to map them onto the correct syllable structure representation later.
In the schema derived from Trager & Bloch (1941), namely Nucleus (Glide) (r), the vowels, /i, e, ɔ, o, ə, u/, are in the Nucleus, /y, w, h/ are in the Glide, and /r/ forms a slot of its own, distinct from and freely combining with the others. Let us suppose instead that:
(3) /r/ occurs in the Glide position.
(4) Glide position supports at most one segment.
Are (3), (4) compatible with the observations made above of the three degrees of phonological vowel length? The answer is no; the facts from which the third degree of length was derived will be explained by (3), (4), but the view in which Reference American vowels have 3 phonological degrees of length must be abandoned, as I now show.
The diphthongs /ɑy, oy, ɑw/ unambiguously contain a glide.3.28 It then follows from (3), (4) that /ɑy, oy, ɑw/ cannot be followed by an /r/ within the same syllable. The glide position is already filled, so by (4), /r/ cannot occupy it. By (3), /r/ cannot occur in coda or nucleus positions either. So if /r/ is to be licensed by syllable structure, it must go in a separate syllable. (It may also be deleted, as in r-less dialects where the vowel in hire is a monophthong, as [ha:]. Then, it can also be a monosyllable, which is -- not coincidentally -- compatible with the theory's constraints.) As observed above, breaking occurs in Reference American in words with glide-r sequences (dire, hour, etc.) so that the form becomes disyllabic. Thus the theory, which states that only one glide can occur in a syllable, and that /r/ and /w, y/ are glides, is consistent with the facts.
A final observation re-confirms the analysis of /r/ as a glide, distinct from the coda. Various coda consonants, even consonant clusters, do occur after /ɑy, ɑw, oy/ within the syllable (as in heist /hayst/, mound /mɑwnd/, joint /joynt/, etc.), though /r/ cannot. If /r/'s location is an undistinguished part of the coda, which evidently may contain final nasals and obstruents, then /r/ should also occur in the same environments. But /r/ is not acceptable in these environments. An additional stipulation is necessary to restrict /r/ from these contexts while allowing other coda segments to occur there. This stipulation is avoided if /r/ is analysed instead as a glide.
What remains of the short/long/longer distinction discussed above? Vowel length is here expressed as sequential structure. Thus the class of longer vowels, /ɑy, ɑw, oy/, are formally represented expressed as nucleus-glide sequences. The class of short vowels, as in pit, pet, Pat, pot, putt, put,3.29 are bare nuclei, without any glide. What about the long vowels? These are also expressed as nucleus-glide sequences when T&B write them as /iy, ey, ow, uw/, making no formal distinction of length between the long and longer classes of vowels. The next observation will show, first, how the long/longer distinction is unnecessary to account for the different behavior of the vowels before /r/, and second, how the independently motivated analysis of /r/ as a glide can be used to clarify new facts that are otherwise inexplicable.
The facts are best introduced in their historical context. In England around the beginning of the 17th century, the vowels in the words her, fir, fur, and unstressed for3.30merged into a single, continuant, rhotic monophthong3.31 without distinctive [I, U, ɛ, o] color -- that is, without distinctive height or backness. (The merger actually took place in two steps, according to Jesperson 1909:319. First the high vowels merged: fir = fur; then the mid vowel: berth = birth.) Thus the former length contrast among non-low vowels before /r/ (exemplified as fear vs. fir, hair vs. her, boor vs. burr) was lost, and is not now distinctive. The present contrast between these vowels is in whether or not they have nuclear /ɚ/.
A phonetic explanation for this response to the change may suffice: the retroflex continuant /r/ gesture ``has a relatively incompressible trajectory whose timing `slides' with respect to other gestures in a sequence'' (Boyce & Espy-Wilson 1991). The effect of this ``sliding'' is that the /r/ gesture may begin even before preceding segments begin, so as to attain a configuration that produces a fully rhotic sound at the point in the acoustic sequence corresponding to the timing slot of the /r/ itself. The temporal incompressibility of this very complex gesture (which can involve three constrictions, as discussed on page and by Ohala 1985) suggests that the rhotic quality may bleed into the preceding short vowels, merging them together, while preceding long vowels were long enough to retain their individual nuclear vowel qualities.
Today the situation remains the same: There is no contrast between long and short for mid and high vowels before /r/. This is the key: The long vowels before /r/ could just as well be categorized as short. Since there is no contrast between /iy, i/, /ey, e/, /uw, u/ before tautosyllabic /r/, we are free to represent fear, fair, boor, etc., without filling the glide position, which otherwise is used to represent the long-short contrast. This in turn allows /r/ to occur in the glide position in these words.
If /r/ is located in the glide position, the three degrees of vowel length are collapsed into two. The distinction between long and short vowels in most environments remains the same: short vowels have only a nucleus, while long vowels consist of both a nucleus and a glide. But the distinction of long vs. longer vowels preceding /r/ is now changed; it is now the same as the long-short distinction. That is, the nucleus in nucleus-r sequences (now represented as /ir, er, or, ur, ar/, as in beer, bear, bore, boor,bar, respectively) is just one element long, while the vowel in a sequence of nucleus-glide-$-/r/ is two elements long (where $ represents a syllable boundary). The phonetic tenseness of the nuclei of nucleus-r sequences can be attributed to the vowel-before-vowel raising rule, as discussed below.
In the most thoroughly studied linguistic variable in English (Labov 1966, 1968, Fasold 1969, 1972, Guy 1980, Santa Ana 1991, Patrick in progress, among others) the deletion of /-t,-d/, preceding /r/ was found to have the same phonetic effect as a preceding vowel. Among various phonetic and grammatical effects, it is generally found that preceding consonants favor deletion, while vowels disfavor it. In most dialects where it was studied, /r/ is generally found to behave like a vowel rather than as a consonant.3.32 This is perhaps the strongest of empirical support for this analysis, since it comes from numerous large-scale studies of actual speech in various English dialects. The form of this supporting argument, it should be noted, is rather unusual: the statistical structure of phonetic observations is, I argue, explained in this case by the underlying phonological structure.
To summarize: First, it was shown that though Trager and Bloch place /r/ in post-glide position, there is no need to do so. Second, in just those cases where a glide is unambiguously present (/ɑy, oy, ɑw/), /r/ cannot occur within the same syllable; this is explained if /r/ is a glide, and only one glide can occur in a syllable. Conversely, just in those cases where the glide can be considered absent (as in /ir, er, or, ur/, where there is no long-short contrast because of the earlier merger of the vowels in fir, fur, her, and unstressed for3.33), /r/ can follow the vowel nucleus within a single syllable. Thus various independent observations support the theory that /r/ is a glide.