There are seven apparent problems with this base-6 system, which I will discuss and overcome in turn.
One apparent problem is the large number of gaps in the Vy and Vw classes. These slots may be labelled /iy, ey, æy, uy, iw, ew, æ w, ow, uw/. Several of these gaps seem motivated on phonetic grounds: the slots /iy, ey, ow, uw/ cannot be filled in a manner that would be phonetically distinguishable from the corresponding long vowels /i:, e:, o:, u:/, with which Trager & Bloch identified them. This could be taken as an argument that these four vowels are not plain long vowels, but front- and back-gliding vowels, as indeed phonetically they are, in most dialects. However, this would lead to gaps in the V: system for /i:, e:, o:, u:/, which would then be similarly unexplained. Whichever set of forms are chosen as underlying, the other set of forms would not be phonetically distinguishable; therefore there can be only one set. The choice made here is preferred because the glides are phonetically predictable, because the underlying specification of (plain) long vowels is simpler than for gliding vowels, and because the raising of nuclei of long vowels is more simply stated when /i:, e:, o:, u:/ are considered to form a natural class of long vowels opposed to /y, w, oy/.
The remaining 5 gaps are arguably legitimate, true gaps in the system, assuming that phonological gaps are often a precondition for phonetic vowel-shifting to occur. Thus the slots which may be labelled /uy, æy, iw, æw, ew/ are phonetically filled by vowel shifts in various dialects. Thus /y, oy/ undergo a chain shift to [oI, ui] in various dialects (e.g., Southampton, England). [æ, e] precede a back-glide in the fronting and raising shift of /w/ in Philadelphia and the Southern U.S. (which in mild form results in [æU] and in extreme forms can go as far as [e], merging crown with crayon). [I], which may be phonologically identified as /iw/, can result from the fronting of /uw/ in the speech of many young, white, suburban U.S. speakers. /iw/ has been argued to be the phonological form of the word dew in those dialects which retain the marginal difference between dew and do, which has almost disappeared in the U.S. Finally, [æI] occurs phonetically in the Southern U.S. where /æ/ precedes /, /; this shift might have resulted in merger if there weren't a gap in the position labelled /æy/. Thus these gaps may plausibly be seen as true gaps in the system, which are preconditions for some of the phonetic changes that do indeed occur.
Next, notice the asymmetry between the Vy and Vw paradigms, which are full of gaps, and the Vr paradigm, which has no gaps at all. This asymmetry may be explained by the different historical sources of these sub-systems: The glide /r/ descended historically from a flapped or trilled consonant. Vowels and single coda consonants normally combine quite freely; this was evidently the case for vowel-r sequences in an earlier stage of English. Thus the current vowel system simply inherited this gap-less vowel-r paradigm when /r/ became a glide.
A second problem which both the base-5 and base-6 systems have difficulty with is the 4-way (height) contrast in the set, weary, Mary, merry, marry. Many dialects (such as my own) merge the latter three classes, while other dialects merge various other subsets of the three or change elements in other ways. However, in those dialects where the difference is maintained, at least one of the four must be analysed, in the present approach, as different in temporal structure from the others. The most reasonable possibility is to put the /r/ in Mary into the next syllable, thus allowing Mary to have a tense nucleus. If /r/ is in the second syllable in Mary, the glide position in the first syllable is available to make the vowel long, and thereby phonetically tense. That is, Mary is analysed like eyrie was, in the above discussion of Trager & Bloch. The other vowels in this set would have no such boundary between the nucleus and the /r/. Phonetic analysis of these vowel classes in a dialect with the necessary distinctions (e.g., New Jersey) might support or disconfirm this hypothesis.
The fact that this set of contrasts is difficult to represent in this system might be taken as evidence against this analysis. However, one criterion for judging the costliness of a phonological structure is whether or not it is eliminated through historical changes. The complexity of structural differences should be related to their instability in linguistic history. Marginal distinctions should have only a marginal place made for them in the phonological structure. It would be quite sensible if distinctions which are phonologically costly to represent were also difficult to maintain and historically unstable. If a set of contrasts were simplified in different ways in different, independent historical changes, this would constitute evidence in support of structures in which the contrasts are difficult to represent. This is the case here: the various mergers and sound changes have had the effect of making it possible to represent the remaining contrasts in a simpler way within this system. In dialects like my own, Mary=merry=marry; in other dialects Mary=merry. In Philadelphia merry is nearly merged with Murray.3.47. Far from being negative evidence, the facts of these various mergers and sound shifts, which have the effect of simplifying this difficult-to-represent set of distinctions, provide support for this analysis.
Third, the nuclei of /:/ and /or, oy/ are representationally distinct, despite the fact that for some speakers of dialects with the contrasts, // 22#22 /:/ 22#22 /o:/ (with minimal pairs, cot, caught, coat), the nucleus of /or/ (court) or of /oy/ (boy) may be subjectively identified with /:/ rather than /o:/. If speakers' intuitions match their phonological structures, then this structure would not predict this identification. These speakers may have a different structural system, such as the one proposed in a later chapter for Alabama English.
Fourth, mergers and splits may modify the contrasts in this structure. For example, I myself make no distinction among /a:, :, /: the long low vowels have merged with the short low back vowel. If the above structure is taken as describing my dialect, then it now has gaps, which are not good. But phonemic mergers generally do create gaps in phonological structures. Since mergers nevertheless continue to happen, the unfortunate gaps in this structure -- in dialects which merge various phonemes -- may simply be unavoidable.
Fifth, /ær/ has a restricted distribution: it doesn't occur word-finally. /ærV/ exists (marry, sparrow, etc.) but /ær#/ does not.3.48 On the other hand, the slot for /ær/ that the base-6 system provides is an improvement on the coverage of the VrV vowels provided by the base-5 system. /æ/ has no slot in the base-5 vowel system, and /ær/ doesn't either. So this is at least an improvement, if not a perfect solution.
Sixth, there is no slot for /r/, distinct in some dialects from both /or/ and /r/. Wells gives this a separate lexical set, NORTH, as opposed to FORCE and START. In Alabama, all three appear to be kept separate. The structure of the Alabama Vr sub-system is treated at length in that chapter, and found to be compatible with a base-6 analysis. The /r/ and /or/ classes merge in most dialects, either with each other or with other Vr classes. In most American dialects, NORTH is not distinguished from FORCE, while in Jamaica, for example, NORTH is pronounced like START, and FORCE is pronounced like CURE. In the current proposal, this distinction is a difficult one to represent. The various mergers appear as natural, if not predictable, responses of the system to the representational costliness of the distinction between these vowels. Again we find that a distinction that does not fit the paradigm is lost in different dialects and in different ways. Such mergers then appear as simplifications of hard-to-represent distinctions, providing support for the structure rather than evidence against it.
The seventh apparent problem is the representation of the rhotic monophthong, // (as in NURSE). If // is taken as a sequence, /r/, then presumably the featural content of the // nucleus is nil. However Reference American already has a vowel with this structure, namely /or/. /or/ is not [high], [low], or [front],3.49 so its nucleus in this system is unspecified for vowel-quality features; its glide is specified as /r/. Thus //, as /r/, has the structural representation in Figure , identical to that of /or/.
Some other structural difference must distinguish them.
Since [high], [low], [front] are unavailable, the only underlying
structures that may be used are the ``features'' Nucleus and Glide.
Both /or/ and // contain a glide slot, so the remaining possibility
is to contrast them by presence vs. absence of the nucleus. Thus, /or/ is
in this analysis a vowel consisting (underlyingly) of a nucleus
without any features, plus a [rhotic] glide. On the other hand //
is analysed as a vowel consisting of a [rhotic] glide, with no nucleus
at all. Later in the derivation of phonetic forms (after the nucleus
of /or/ is specified for other features such as [round]), the nucleus
slot for // is inserted by rules which ensure the
satisfaction of well-formedness conditions on syllables (such as the
condition that requires a nucleus in each syllable; these are part of
the process of syllabification). Then the surface vowel-quality
features of the nucleus are then inherited from the glide's features,
making // rhotic throughout. The following derivation (for r-ful
dialects) relates the underlying form of syllabic /r/ (written //,
and represented as G[rhotic]) to its surface form, a long monophthong
that is rhotic throughout.
Indo-European is considered to have underlyingly unspecified vowels, the so-called zero-grade vowels. Yip (1987) has also argued that ``suffix-initial /i/ in English is absent underlyingly and is inserted by rule.'' These previous analyses lend further plausibility to what might seem a rather unintuitive proposal that // has no underlying nucleus.
This analysis of // is quite intriguing on a number of grounds. On the one hand, there is little direct (phonetic) evidence that // is underlyingly a sequence; // is a phonetic monophthong in most dialects, excepting Celtic dialects, where /r/ is not a glide at all, and excepting dialects similar in this respect to New York City, where a subclass of //, /'V/ (as in hurry), contains . On the other hand, // has often been analysed as a sequence of +r, perhaps because both of these segments are independently motivated in nucleus and glide positions, and because /r/ is not otherwise motivated as a nuclear segment. The present proposal accomodates both views. Since the nucleus links to the vowel-quality features of the glide, both positions have identical phonetic content, thereby formalizing the fact that // is a monophthong. But we retain the claim that rhotic features are derived from, or underlyingly restricted to, the glide position, since in the underspecified underlying form, the nucleus itself is absent. Thus what appears to be a problem in the representation of // leads to a tidy solution in which formerly opposing views of the nature of this vowel are simultaneously accomodated.
This analysis of // also helps us to understand the merger of fir, fur, (trans)fer, and (unstressed) for. These short vowels before /r/ lost their nuclear vowel-quality features, and -- in this analysis -- lost even the Nucleus specification. This analysis is useful because it explains why it should be that the resulting merged class of // vowels no longer fits into the base-6 set of nuclei. The nucleus of // is neither [high] nor [low] nor [front], nor is it even ``none of the above'' (as is the mid-central vowel, //). // is something else entirely, namely [rhotic]. This analysis makes it explicit and formal that // is not part of the base-6 system. Since it has no nucleus, it cannot be a part of any system of nuclei.
The seven apparent problems have evaporated, led to plausible and interesting analyses, or metamorphosed into strong support for the analysis. None remain serious enough to compel the abandonment of the base-6 analysis of Reference American proposed here.