The phonetic and social conditioning of rhoticity in this dialect is extensively studied by Feagin (1990).8.8 In most cases, while the sound symbolized by /r/ is not necessarily rhotic, it is distinct from other glides (except in one case, the confusion of /r/ with /y/ as [ ~ ], according to Foley 1972:36).
The inventory of r-gliding vowels is quite different from other dialects. This subsystem may be analysed as containing four heights of vowels in both the back and the front. This would present a counterexample to theories of English phonological structure like the one presented in the present work, in which only three heights may be represented. In all other cases examined here, it has been found that only three heights are necessary to represent English vowel structure, so an attempt is made now to reconcile this quite strong position with the facts of the Alabama Vr subsystem.
I will present the relevant contrasts first, and then show how another plausible analysis removes vowels from this subsystem, allowing the Vr subsystem to be analysed as having only three heights. First, it should be pointed out that the NURSE class, with a ``pure'' rhotic vowel /:/ may be analysed as in Phonological Preliminaries as a vowel containing no underlying Nucleus, but only a rhotic (or [low]) Glide. The Nucleus is then inserted by a rule that ensures the well-formedness of syllables, and r-colored, perhaps by ``backward-gemination'' (C.-J. Bailey 1985). With the NURSE class excluded, the question now posed is, What are the underlyingly specified nuclei preceding tautosyllabic /r/?
Alabama English appears to have four back vowels before /r/. First, the lexical sets CURE and FORCE are distinct here, as in Chicago and L.A. Chicano, but unlike Jamaican or Rhode Island,8.9 or possibly Appalachian English. However, unlike Chicago or Los Angeles, and like Jamaican, the sets, FORCE and NORTH, remain distinct, in this dialect,8.10 with such minimal pairs as morning/mourning, horse/hoarse, or/ore, war/wore, etc. (the first of each pair is in the NORTH class). Finally, there is a low-back unrounded nucleus in the START set, which remains distinct from the above sets. The back vowels are written as /ur, or, r, r/, with corresponding lexical sets CURE, FORCE, NORTH, START.
The front vowels before /r/ may be analysed in various ways as having from three to five heights. I will consider two contexts separately, before intervocalic /r/ and before non-prevocalic /r/.
Before intervocalic /r/ there are as many as five heights: /rV, erV, rV, ærV, arV/ are distinguished in weary, Mary, merry, marry, wiry. We may reduce the number of vowels that must be represented in the Vr subsystem to four by considering the distinction between /erV/ and /rV/. This troublesome distinction may be eliminated in one of two ways, by merger, or by reanalysis.
In Foley's data, 6 lower-class whites and 3 younger black and upper-class white speakers use // in dairy and Mary; thus apparently merging /er/ with /r/ (that is, with the sound in cherry, merry) in prevocalic position. This social distribution is reminiscent of a merger in progress, led by the socially intermediate group of non-cultured whites.8.11 For those who distinguish the Mary, merry word classes, the proposal made in Phonological Preliminaries must be extended to this dialect. There the Mary class of words was excluded from the Vr subsystem by supposing that there is a syllable boundary (written ``$'') before the /r/, as in /ey$rV/. C.-J. Bailey (1985:167) writes Mary as [mir] for older r-less Southern States speakers, a phonetic form which suggests that the /ey$r/ analysis of this class of vowels is correct, at least for those speakers. It may be that this syllable boundary is lost in younger and r-ful speakers, and that this loss is a trigger for a historical merger between /erV/ and /rV/ (Mary/merry).
We may apply the same analysis to the /arV/ class. This vowel corresponds to /ay/ in other dialects, and in Phonological Preliminaries it was argued to be unnecessary in Trager and Bloch's (1941) Vr vowel-system because the /r/ in Irish, spiral, wiry, etc., belongs to the following syllable. Thus we are left with three heights among the front vowels before intervocalic /r/: /irV, rV, ærV/.
Next consider the front vowels that occur before non-prevocalic /r/. There are again as many as five: ear [i], Ayre [e] : heir  : air [æ] (assuming r-ful pronunciations).8.12 These correspond to the vowels before intervocalic /r/ and present similar apparent difficulties for the 3-height proposal.
Ayre, which rhymes with mayor and they're, would seem to be /ey/ plus syllabic //; analyzing it as bisyllabic reduces the troublesome five by one.
Heir and air together correspond to the SQUARE lexical set. This distinction, between  and [æ], seems to be a low-level rule, applying to words of the same class, rather than a distinction in the inventory of surface phonological classes. Thus in measurements of F1 and F2 in James H's speech, tokens of there are distributed across the entire range occupied by measurements of SQUARE words, and /ærV/ words are at both the high and low ends of the range. Thus barrel (n=1), carry (n=2) are relatively high and adjacent to mulberry (n=2) and cemetery (n=1), while married (n=2) and rarity (n=1) which impressionistically contain [æ] are low on the chart and adjacent to a token of there. there (n=23) is itself distributed from highest to lowest positions on the chart, which is consistent with the treatment of this difference as a variable phonetic alternation.
A difference does exist in James H.'s speech between there and their, but it is not the one suggested: their sounds like  (n=2), while there varies among [e ~ ~ æ]. This may be derivative of the syntactic positions of the words: the determiner their is in a prosodically weaker position, and thus undergoes assimilation of the nucleus to the /-r/ glide.
C.-J. Bailey suggests that in certain contexts the /ær/ class is modified in a similar way by the rule of ``backward gemination'', in which the /r/ quality bleeds back into the onset of the preceding syllable, and the nucleus consequently raises, giving /ær/ 14#14 [r]. This is consistent with the information I have, but James H.'s two relevant tokens have undergone backward gemination to an extreme, with the nucleus becoming entirely rhotic.
It seems unnecessary to add a distinction between phonological classes /r/ vs. /ær/, if the alternation between them can be accounted for by a low-level phonetic rule. So I will assume there is a single phonological class including all the SQUARE vowels, and write them as /ær/, reflecting the traditional Southern States transcription.
The argument so far has reduced the front vowels before /r/ to just three. But the four back vowels remain problematic for the three-height theory.
Three further points are necessary to complete the argument that the Vr subsystem can be analysed as having only three heights. The point of these steps is to enable the back vowel contrasts to be reduced by one.
First, /ær/, as in SQUARE, can be analysed as a phonologically mid-front vowel. Second, /ar/, as in tire, can be analysed as similar to the same set of words in Reference American, as /ay$r/. Third, /r/, as in the set, START, can be analysed as a front vowel. If these conclusions are accepted, then the Vr subsystem will have a three-heights-and-backness structure which is consistent with the theory of English vowel structure proposed in this work: there are only three phonological vowel heights.
Consider the points in order. The SQUARE set contains a mid vowel in Reference American and all the other dialects analysed here. As pointed out above, the pronunciation of this vowel with [æ] is by no means consistent in this dialect. Foley's reports that in addition to the predominance of 59#59æ59#59 pronunciations, there are a large number of mid pronunciations in words of this class, which he classifies as belonging to the 59#5959#59 class. /ær/ in SQUARE does not occur in some related Southern dialects: in the self-conscious speech of a few emigrant Southerners from Texas, Arkansas, and Georgia, I have found a mid-front nucleus in SQUARE words. Finally, and most importantly in the present context, the speaker analysed here frequently produces a mid nucleus in these words, ranging from  to [e], rarely as low as [æ]. (Even /ærV/-class words like carry, barrel, have [ ~ e] in his speech, though rarity and married, with impressionistic [æ], are exceptions). Thus the phonetic form of this class of vowel is perhaps more likely to be mid than low in the dialect under consideration. Certainly the case for analysing the SQUARE set as containing a phonologically mid nucleus is just as strong if not stronger than the case for analysing it in the traditional way, as low.
The second point can be posed as a clear question: Is the coda of tire /ar/, /r/ or /ay$r/?8.13 Some rural speakers, such as James H., analysed here, appear to rhyme wire and (gui)tar, thus merging this class with the START class, here written /r/. This may be a rural feature. /ay/ is monophthongized categorically in the words fire, tire, wire, etc. The sound is said to have a low-front nucleus, as [a:()]+/r/ (although James H.'s single token of this class, wire, is not front, but central: [w]). This vowel is lower than /ær/ (SQUARE) and presumably in general in the South, if not in James H's speech, it is to the front of / r/ (START), and thus can be analysed as the low-front nucleus before /r/.
Where wire, tire, etc., are not merged with the START class, there are some reasons to analyse this complex vowel as bisyllabic /ay$r/. First, some pronunciations are bisyllabic (Foley, 1972:36),8.14 although these words may also be pronounced as a single syllable, even in highly-monitored speech. Second, /ay/ may still be analysed as an underlying diphthong, for a couple of reasons. Foley's transcriptions of /ay/ include no monophthongs.8.15
As discussed above, although all speakers monophthongize in certain environments (see C.-J. Bailey 1980:171), the monophthongization of /ay/, when it occurs before voiceless obstruents, is (or was) a stigmatized variable. This might be taken to suggest that the monophthongal form is not yet the underlying form. Further, while the monophthongal form can be derived from the diphthong by glide-deletion, it would be impossible to derive high-front-gliding forms in those cases that do have these glides, without an underlying feature that specifies that a high-front glide is to be added; but this amounts to having an underlying /-y/ glide as part of the representation for this class. Therefore, despite the frequent (in some environments, categorical) occurrence of the stereotypically Southern phonetic form [a:], we must analyse this vowel in this community as an underlying diphthong, /ay/, rather than as a monophthong /a:/ (though this seems to be undergoing phonological change at this time). Third, the nucleus of tire, wire, etc. (where these have distinct front [a] rather than central , merging with /r/) is both phonetically similar and historically identical to the /ay/ sound. These three reasons suggest plausibly (though they do not make for certainty) that tire, wire, etc., may be phonologically diphthongal, and bisyllabic, written as /ay$r/. If so, then since vowel sub-systems contain only the contrasts among vowels within single syllables, this sound-class does not belong in the Vr subsystem. Similarly words like spiral, with /r/ in the onset of the following syllable may be analysed as having an intervening syllable boundary and thus need not be included in this subsystem. Thus either through merger or a bisyllabic analysis, the tire, wire words need not be distinguished within this Vr sub-system.
To complete the 3-height analysis for Vr vowels, the last point is that /r/, as in START, can be analysed as a front vowel. Acoustic and phonetic vowel space is an inverted triangle, with  in the bottom corner. This is the location of the nucleus of /r/. Since this point is at the intersection of the front and back edges of the phonetic vowel space, it is both front and back at the same time. The phonetic form doesn't determine the phonological treatment of the vowel at the low corner of the triangle. So analysing it as phonologically front or back is an arbitrary decision with respect to the phonetics, which may be made according to the phonologically simpler structure. Thus if the phonology has a gap in the low-front slot and there is no space among the back vowels, it is justifiable to analyse // as structurally a front vowel.8.16
To summarize, we may reasonably suppose that the Vr subsystem contains the following classes:
// is analysed as a [rhotic] Glide without a Nucleus specification, so it is also excluded from this subsystem, which strictly includes only vowels of the form N+G[rhotic] (that is, vowels with an underlying Nucleus position, followed within the same syllable by a [rhotic] Glide). Tire, spiral are not distinguished within this subsystem, being analysed either as bisyllables with the /r/ in the following syllable, or as merged with START. The vowels before intervocalic /r/ in the classes weary, merry, and marry/rarity correspond to the three front-vowels /ir, r, r/, while Mary (=dairy), if not merged with the merry class, is analysed as having a syllable boundary before the /r/.
I have shown that the r-gliding vowels in this dialect may be analysed as retaining a 3-way height distinction among both front and back vowels. Thus despite the differences from Reference American in the lexical distribution of sounds (namely that FORCE and NORTH constitute two distinct phonological classes), and despite the important phonetic differences (e.g., the optionally lowered pronunciation of the vowel in the lexical set, SQUARE), the same static phonological structure of three heights and backness is retained.
The phonetic form of the /r/-glide itself does not rearrange this abstract phonological structure, though the phonetics needs to be fleshed out considerably. ``R-less'' and ``r-ful'' refers in most cases not to the presence or absence of a distinct phonological glide element, but to the phonetic form of that glide. This glide, which is abstractly characterized as /r/, varies impressionistically among [, , , :] (Foley 1972:34), where the ``r-less'' forms [, ] are ``usual in socially prominent white speech and Negro speech, [and] occasional in other older white speech'',8.17 and the ``r-ful'' form  occurs elsewhere (p. 47, 48). These results are consistent with Feagin's (1990) interpretation that r-lessness is itself being lost in this Southern dialect. Varying realizations of /r/ are present in all Vr classes, excepting START words, which may have monophthongal [:] for r-less speakers (p. 33). If we analyse the r-less and r-ful forms that co-occur in this dialect as derived from a single phonological system, then START-class words must have underlying /r/, whether realized as  or [:]. Thus the structural analysis may be retained, fleshed out by the particular phonetic and social distributions just presented, and those described in the discussion of acoustical measurements of the vowel system of a single older, rural, male speaker from Anniston.