next up previous
Next: Phonetic Implementation of Reference Up: Reference American Vowel Structure Previous: Apparent Problems in the

Features in Reference American

What is the featural composition of the present system? Statically, there are three vowel heights, low, mid, and high, plus two degrees of backness, front and back, within the nucleus. Additionally, there must be structures which accomodate both length and for gliding.

Treatments of vowel length in the early structuralist literature argued over the relative merits of representing long i, for example, as /i:/ vs. /ii/ vs. /iy/. In more recent phonology (SPE, for example), a static feature was used to distinguish the ``short'' and ``long'' vowels, so that what here are long vowels was distinguished from their short counterparts as [peripheral], [tense], [long], or some other feature. In autosegmental theory, there are a number of kinds of structures which are used to represent vowel length and complexity. Where length alone is the concern, the complex of features which represents the vowel's quality becomes linked to multiple timing slots. E.g., a long low vowel, like the ``broad A''(/a:/, PALM) of father is contrasted with a short low-front vowel, like ``short A'' (/æ/, TRAP) of blather by associating the same vowel quality features to two timing slots instead of one. Where vowels are phonetically complex, as in diphthongs, the syllable structure again licenses multiple vocalic timing slots, but instead of linking them to a single static feature-complex, they are separately linked to distinct feature-complexes that specify vowel quality. Thus the features for the // in /y/ are linked to the timing slot of the syllabic constituent in the syllable (the nucleus), and the /y/ features are linked to the timing slot of the non-syllabic constituent (the glide).

In this proposal, we have argued for the existence of a glide position within the syllable which may be specified for features that distinguish front, back, and rhotic glides. This glide position is absent in syllables containing the short vowels, and the constraint that stressed rhymes branch is met in that coda consonants occur with stressed short vowels. Given this arrangement, which accounts for the V, Vr, Vy, Vw classes, and given the class of long vowels to be represented in this system, it is unnecessary and undesirable to use some static feature, such as [long], to distinguish them. Instead, it is preferable to use the independently motivated temporal structure, namely the nucleus and glide positions of syllables. This dovetails nicely with autosegmental theory, which deals with vowel length via multiple timing slots.

The proposed structure of Reference American also lends itself to economical representation in terms of features. Three degrees of height and two of backness must be distinguished for nuclei; glides may freely be unspecified (for long vowels) or specified as rhotic, or specified as front or back with combinatory restrictions on the nuclei they may co-occur with (/y, oy/, /w/ are the only vowels with underlying front and back glides). A number of featural analyses for this structure are possible, as already discussed above in the section on static vowel features.

The solution proposed here for the problem of [+high,+low] incompatibility is to have [high] and [low] be privative features which occur on a height tier characterizing syllable nuclei. If the feature that distinguishes front from back may also be taken as a privative feature, then we may have a uniform formal treatment of vowel-quality features, which is desirable if it can be maintained.

I make the reasonable proposal that central vowels are phonetically unmarked in the front-back dimension -- reasonable, since they lie in the middle of the front-back dimension. In English, the vowels with central nuclei /, / are to be classified with the back vowels, in opposition to the front vowels. That is, there is a full set of front vowels, but the back/rounded and the central vowels are in complementary distribution in the phonological structure, as I will now show.

Consider the pattern of the specification of rounding on the non-front vowels for Reference American. Backness and rounding are related in a hierarchy of both height and vowel complexity: the [round] feature is (redundantly) applied to the high, mid, and low long back vowels (/u:, o:, :/), to the high and mid back vowels before glides (/ur, or, oy/ but not /r, y, w/), and to the high back short vowel (/U/ but neither of /, /3.50). In short, the more complex the sub-system, the more heights have rounding, where long vowels are more complex than vowels with glides. (British English is somewhat different, since the short, low vowel here labeled //, as in LOT, is [round], as short [].) This implicational pattern is shown in Table [*].

[round] applies to vowels above the line.

Table: Implicational constraints on the feature [round]
Long Gliding Short
u: ur u
o: or oy  
: ar ay w  

Within the different classes of sub-systems in Table [*], back, rounded nuclei occur in complementary distribution with central and unrounded nuclei. Because of this complementary distribution, the back/rounded and the central nuclei of Reference American are underlyingly represented with the same value on the front/back dimension, and an implicational rule describing the pattern in Table [*] is assumed to be responsible for the redundant marking of the back/round vowels with the features [back] or [round] or [back/round]. Chapter 2 showed that tongue-backing and lip-rounding have the same acoustic effect in the acoustic-phonetic front-back dimension. If acoustic dimensions are taken as the fundamental to vowel quality in English, then backing and rounding are mutually reinforcing articulatory aspects of the same acoustic-phonetic feature.

Chapter 2 also shows that phonetic [], the resonant sound of a uniform acoustic tube closed at one end, is a reasonable approximation to the average vowel quality for a number of English speakers, and suggests that other vowels are to be understood acoustically as deviations from this average position. Thus the mid-central articulatory position whose sound is written as [] (this is different from the usually raised position of the realization of unstressed //, which may sometimes be written as []) is ``unmarked'' in a phonetic sense as well: it is the center from which other vowels are understood as deviating.

Since the central vowels can therefore be understood as phonetically unmarked, and since the back vowels are in complementary distribution with the central vowels, I will consider the central and back/round vowels to be underlyingly unmarked.

The underlyingly marked set of vowels on the front-back dimension are thus the front vowels. The privative feature used to make the underlying distinction between the front and central or back vowels is then assumed to be [front], rather than the traditional [back] feature of SPE, and other generative work. At a lower level, the feature [back] or [round] or [back/round] is added by the assumed implicational rule to a subset of the unmarked vowels.

I thus represent the static structure of Reference American with the three privative features [high], [low], and [front]. This structure is compatible with theories of markedness. The unmarked short vowel is central (i.e., non-front) and mid, namely //. The choice of [front] as the marked value for the front-back dimension allows //, defined as the unmarked stressed vowel, to be phonologically identified with //, as my intuitions and as monitored pronunciations in my dialect suggest. In SPE, on the other hand, where [back] is the front-back feature, the same formal step would lead // to be identified with as the unstressed allophone of //, an unlikely result. In this proposal both //, and // are phonologically unmarked, which seems quite sensible, and // is the unstressed allophone of // in Reference American. (Note that //=// is not true of British Received Pronunciation. Also, in Southern U.S. dialects which distinguish // and //, there are further complications.) In this proposal the cross-linguistically infrequent sound, /æ/, has a marked frontness value, as it should. After the redundant feature [round] is added to the specifications for the relatively higher non-front vowels, then /u; ur, or, oy; u:, o:, :/ are all marked for roundness (as discussed above), while /, , / are unmarked in the front vs. back/round dimension, which, again, makes phonetic sense since they are central vowels. To summarize, the six vowel types which form the basis of each Nucleus-Glide sub-system have the underlying feature specifications given in Table [*].

Table: Feature specifications for nuclei.
  Front Central/Back
High [high,front] [high]
Mid [front]  
Low [low,front] [low]

Thus, three privative features with seven total occurrences in the structure are used to specify six vowel qualities, which is the maximum possible economy here.

Now let us examine glides. The glide position may freely (i.e., in combination with any of the nuclei) be unspecified (for long vowels) or specified for a [rhotic] feature of some kind (for Vr vowels). The featural representation of the restricted front- and back- glides in /w/(MOUTH) and /y, oy/ (PRICE, CHOICE) is the last step. In order to distinguish /y/ from /w/ in dialects where they have the same nucleus, we must distinguish front glides from back glides. If we use the same features as are used for the front-back differences among nuclei, then the glides in /y, oy/ are marked by the feature [front], while the back-glide /w/ does not need to be specified for backness.

However, if the glide /w/ is completely unspecified, then there will be no difference between the underlying representations of /:/ and /w/. These both have non-front, low nuclei, and both would contain a glide which is completely unspecified. This is, of course, the underlying representation for length: a glide without static features. If the long vowel /:/ has no features specified in its glide slot, then the diphthong /w/ must be specified in some way to mark the contrast; we may use the feature [high] for this purpose. The feature-combinations [high,front] and [high] instead of [front] and [   ], are therefore used to distinguish the front and back glides from each other and from the long vowels, which have unspecified glides. Thus the possible underlying glide structures are as in Figure [*].

Figure: Possible glide structures

Because the /-w/ glide is high, it falls in the class of redundantly marked [round] vowels of Table [*]

Some of these glide structures can occur with all six nucleus values in Reference American, each of them forms what I call, following Labov, a vowel ``sub-system'', sets of vowels which have important roles in the patterns of misunderstanding (Labov 1990) and sound change (LYS).

Summarizing, the following formal system will generate the possible vowels of Reference American.

(1) Vowel 14#14 Nucleus (Glide).
(2) Nucleus licenses [front] and the height features [high] and [low].
(3) Glide licenses [front], [high], and [rhotic].

Rhoticity and lowness

Notice that the feature set, [front], [high], [low], is rather similar to the feature set, [front], [high], [rhotic], and also that the features [low] and [rhotic] occur in complementary distribution within the vowel: one occurs only in nucleus position and the other occurs only in glide position. This suggests the proposal that the two features are underlingly just one -- [low] -- and that the [rhotic] feature which characterizes the phonetic form of these glides may be added later by redundancy rules. This proposal is strengthened by a number of observations: 1) The vowel [] has a non-high tongue-body position; 2) In ``r-less'' dialects, the reflex of [] is a non-rhotic, non-high inglide; 3) in these dialects (e.g., British Received Pronunciation) the reflexes of Vr sequences with a non-high nucleus are phonetically monophthongs.

Under this proposal, the difference between rhotic and non-rhotic dialects is in whether or not the redundant [rhotic] feature is added to the underlying forms. The underlying forms are identical. Note further that the Vr sequences in r-less dialects that are monophthongs -- those with mid (that is, relatively low) nuclei -- can be explained naturally if /r/ is simply the feature, [low]: the glide and the nucleus are both low, and in a process much like the redundant specification of vowel quality features on the glide in long vowels, the vowel quality features of the nucleus are inherited by the glide, with which it is already entirely compatible. Thus the nucleus and glide merge. (This argues for the specification of [rhotic] preceding the nucleus-to-glide inheritance of other quality features.)

Finally, consider the issue of phonetic implementation of glide features. Glides do not uniformly attain a particular degree of phonetic height. The glide in /y/ may reach [], while that in /oy/ reaches [i]. /w/ is commonly [o] rather than [u]. Hence a glide which is [high] is relatively high. This analogy may be applied to r-glides: if they are underlyingly [low], then this is relative: high nuclei may glide inward and downward to a phonetic height of mid, rather than low. This makes 1) and 2) compatible with a feature [low] rather than [-high] or some such other, merely non-high, feature specification: [low] in glide position simply means relatively low.3.51 (3) above thus becomes:

(4) GLIDE licenses [high], [front], and [low].

(4) in turn may be collapsed with (2) by the two following statements:

(5) Nucleus and Glide are vocalic.

(6) [front], [high], and [low] are vocalic features.

The first of these may be eliminated since it simply restates the conclusion of Section [*]: Vowel14#14 Nucleus (Glide). The second is perhaps the fundamental stipulation of vowel phonology; it might be better stated in terms of phonological licensing: The vowel constituent within the syllable licenses the feature [front] as well as the height tier with privative features [high] and [low].

A later rule redundantly marks [low] postnuclear glides as [rhotic], in r-ful dialects.3.52 These rules give formal featural representations to the structure in Table [*] (page [*]), as repeated in Table [*]. As a convenient linear representation for the multi-linear structure of vowel features occupying Nucleus and Glide slots, I will sometimes use the notations N[...], and G[...] to denote features [...] in the Nucleus and Glide positions in the syllable, respectively. V, V:, Vr, Vy, Vw are by now familiar as names of sub-systems.

Table: Feature structure of the Reference American vowel system
  N N+G N+G[low] N+G[high] N+G[high,front]
N [front] [ ] [front] [ ] [front] [ ] [front] [ ] [front] [ ]
[high] i u i: u: ir ur 21#21 21#21 21#21 21#21
[ ] e   e: o: er or 21#21 21#21 21#21 oy
[low] æ   a: : ær r 21#21 w 21#21 y
  (V) (V:) (Vr) (Vw) (Vy)

This proposal has a number of formal virtues. It is extremely simple in terms of the number and type of underlying features: [high], [low], [front]. It is also has relatively few gaps; none except in Vy and Vw combinations, plus the restriction of /ær/ (that is, N[front, low]+G[low]) to the environment where an unstressed syllable follows. The degree of underspecification is fairly extreme: the total number of features (counting the presence of a glide as a privative feature), is 52, for representing all of the 21 stressed vowels in Table [*]. At the same time, the degree of abstractness is not unreasonable.

The proposal also has a number of substantive virtues. First, it makes it easy to state the traditional short-long distributional dichotomy: short vowels lack a glide, long vowels have one.3.53 This, in conjunction with the dictum that stressed rhymes branch, explains the well known distributional fact discussed earlier: stressed short vowels occur only in checked syllables.

Like any set of vowel classes, the present set allows the comparison of English dialects in terms of mergers and splits of vowel classes in the set; unlike proposals in which the differences are expressed as different phonemes, the phonetic comparisons between corresponding classes are here usually (though not always) specified by rules that specify redundant phonological and phonetic details for an underlying phonological structure that is relatively uniform across dialects.

A problem with many abstract, pan-dialectal systems is that they are static, and do not contain within them the seeds of their own change. This is not the case for the phonological structure of Reference American, since many mergers, shifts, etc., are made sense of, phonologically, in this system. Thus /w/-fronting in the South makes sense as a movement into a gap in the Vw system. The merger of Mary, merry simplifies a difficult-to-represent distinction. The raising of /æ/ is seen as initially due to a phonological shift of this historically short vowel into the long vowel slot emptied by the loss of the broad A (PALM) distinction. Most importantly, when coda consonants undergo phonetic change to become vocalic, the principle that only one glide can occur per syllable predicts severe restrictions on the possible phonological outcomes when the segment is analysed as a phonological glide. It is important to note that while these structural changes can be accounted for as changes in the abstract phonological pattern, there are also a great many low-level sound shifts in which the rules of phonetic implementation show social and historical variation. The system of abstract categories is one important level at which sound change occurs, but it is not the only one, as we will see in the phonetic patterns shown in later chapters.

The similarity of different dialects in the phonological structures presented here helps to explain the mutual intelligibility of English dialects. The problem of perception of other dialects has at least two facets. First, when a phonological comprehension error is noticed, the listener must reconstruct the intended sound. How this cognitive procedure works, and to what extent it does work, is an important and interesting topic of study, partly examined in the Cross-Dialectal Comprehension studies of Labov, et al. Second, in learning to understand a different but related dialect, listeners must be able to relate the sounds of the different dialect to the phonological categories of their own dialect. Listeners may map their own phonological structure onto a new phonetic space. How these processes proceed, and what are their limitations, are questions that go beyond the scope of this thesis; nonetheless, it is important to point out that these are important issues in studying the structure of phonologically related dialects.

The proposal accomodates both sides of the issue of whether to represent long vowels as /V:/ or /VV/. A vowel with an unspecified glide position may reasonably be written as /V:/, while the derived representation of the same vowel, after the glide has inherited vowel-quality features from the nucleus, may reasonably be written as /VV/. We have therefore removed from the horns of a false dilemma.

For convenience of reference and transcription, there is a simple symbolic representation of each of the vowels. Weird symbols are reduced to ``a'' symbols: , , æ. Since the length-versus-glide argument is decided in favor of both (length 26#26 presence of a glide), people who prefer one or the other of /i:, e:, o:, u:/ and /ii, ee, oo, uu/ and /iy, ey, ow, uw/ as convenient representations for the same vowel classes, can remain justified in their preferences, which amount to decisions about which level to transcribe in symbols. This entirely extra-theoretic, arbitrary decision can of course be made in any way one wishes to. I will occasionally use /iy, ey, ow, uw/ for /i:, e:, o:, u:/; ambiguity does not result.

This proposal provides a uniform surface phonological structure (the output of the lexical phonology, in the sense of Kiparsky 1982) which can be applied to dialects of English. Differences of inventory are to be stated in terms of mergers and splits relative to this inventory. Phonetic differences between the same categories across dialects can be stated in terms of the phonological rules that fill in these relatively underspecified feature-structures. It has considerable formal simplicity and elegance; it explains a number of interesting substantive patterns; it is not so opaque that it cannot be used for other purposes than construction of phonological theory. I will assume this phonological framework as a basis for the phonological and phonetic analyses of the remainder of this thesis.

next up previous
Next: Phonetic Implementation of Reference Up: Reference American Vowel Structure Previous: Apparent Problems in the
Thomas Veatch 2005-01-25