Phonetic implementation rules are an important part of the grammatical system. They are part of linguistic performance, but the competence/performance distinction is not very important at this level of the linguistic system.3.54 The mapping from phonological specifications to distributions of F1-F2 values in actual speech is undoubtedly very complex. In theories such as Lindblom's, phonetic forms are a balance between maximizing auditory distinctness, which is a very complicated, non-linear, context-sensitive system of constraints, and maximizing articulatory ease, which is another complicated, non-linear, context-sensitive system of constraints. Developing an explanation for the patterns of surface distributions of vowel nuclei, patterns which result from the joint optimization of both of these systems, may be beyond our capabilities, since the workings of this optimization are only partly known.
Part of the research program of this thesis is to attempt to be clear about what these distributions themselves are. As an effective discovery procedure, one needs an approach that leads to interesting regularities. Without attempting to explain the entire system, it is important to describe the surface distributions and to try to find what the surface patterns are. This approach could well be wrong, but there appear to be symmetries and patterns, and describing those patterns explicitly will lead to the regularities that do exist, in addition to a clear descriptive picture of the surface patterns. Some may object that we can't expect to find anything, and they could be right. But if we don't look, we won't know.
In this section I discuss a mapping from the minimally specified post-lexical phonological structure to a more fully specified, default phonetic represention for this hypothetical dialect, which will be used for comparison with the phonetic forms in the particular dialects studied in later chapters. The phonological specifications of the vowels in Reference American are quite underspecified. This allows many of the differences between dialects to be stated in terms of differing rules for filling in the phonetic details from the cross-dialectally similar, underspecified post-lexical forms given here. Much work remains to describe the phonetic forms correctly, and to characterize the system that generates them.
Numerical phonetic implementation rules come in two theoretical flavors. First is the n-ary phonological rule, where categorical phonological rules are considered to operate on numbered categories (as in Labov, Yaeger, & Steiner 1972: 167ff). A quite different kind of rule is the kind of phonetic implementation rules found in Liberman and Pierrehumbert (1984), where phonologically specified targets are mapped onto F0 values according to general parameters such as pitch range, and the F0 contour is interpolated between the targets. The latter class of rules assumes some speaker-specific bounds on a continuous phonetic space, with continuous mathematical relationships (rather than discrete relationships among small-n sets of categories) describing the mapping of phonological structures into acoustic dimensions (which not coincidentally are derived from articulatory configurations that are continuously variable and subject to conscious control). The phonetic implementation rules to be described here are of the latter type.
I propose the following set of simple and rather general phonetic rules for specifying the phonetic forms of the vowels of Reference American. The proposal assumes that phonetic vowel space is a triangle (reflecting the acoustic facts), and that an important step in specifying the grammar of phonetic form is the specification of target vowel qualities for phonological categories in this phonetic space. Further steps involve phonetic timing and duration patterns, the effects of phrasal stress, and the coarticulatory effects of nearby consonants. Beginning with the features specifications in Table , then, the rules in Figure apply. English-specific phonological rules operating on discrete abstract categories are marked with an asterisk (*), while the rest are proposed as universal phonetic implementation rules.
This set of phonetic rules generates the geometrical configuration in Figure , which corresponds closely to the phonetic qualities in Table . The rules need not be ``linguistic'', in the sense of being language-particular; however, they do describe part of the system of phonetic implementation.
This system generates phonetic forms from the post-lexical phonological specifications that are the input to the phonetic implementation system. This system, or one functionally like it, is implicitly assumed to map the phonological structure to phonetic performance. The reason for proposing such a system explicitly is the usual reason for formalism in linguistic theory: It is easier to see where the system is wrong, when it is made more explicit. Thus this rule-system is proposed in order that further linguistic and phonetic research may build on it, by developing arguments for or against particular features of the system. This proposal represents progress in at least this sense.
Note that a phonetic form for /a:/ is not listed in Table , because I am not familiar with the phonetic forms for this vowel class in dialects that distinguish it. The remaining vowel qualities given are reasonably close to observable vowel qualities in some dialects.
This proposal provides an illustration of the division of labor between systems of phonetic implementation and phonological structure. Phonetic implementation is an independently necessary part of the system of speech performance. As argued at various points in this thesis, it also must be partly linguistic, because dialects differ in some of the details of phonetic implementation, and therefore speakers who acquire the speech forms of particular dialects must learn these differences as part of their knowledge of the dialect. Since this system is independently required, and partly linguistic, it makes sense to attribute some of the descriptive load to the structure of this system, thereby simplifying the phonological structure. One of the main lessons of the systems of phonetic implementation of pitch contours argued for in Liberman & Pierrehumbert (1984) is that n-ary pitch levels may be derived from an underlying binary phonological contrast. In this system, the front-back dimension is underlyingly represented by a single privative feature which may be present or absent. But on the surface there are seven distinct degrees of frontness, derived through the system of phonetic implementation.
The output phonetic forms of this implementation system in Figure and Table form the reference phonetic forms which I will use as a baseline when describing vowels' phonetic realizations in particular dialects as having ``shifted'' in one direction or another.
An important difference between the phonetic forms of Table and those that are often considered as the basic realizations of these vowels in ``General American'' has to do with the long, non-low vowels. In Reference American as defined here, and furthermore in most other English dialects, the non-low long vowels /i:, e:, o:, u:/ are phonetic diphthongs. As pointed out to me by William Labov, this contradicts Lehiste & Peterson's (1961) declaration that the long high vowels in ``American English'' are monophthongs. ``Within the rather homogeneous dialect analysed, ... the syllabic sounds of American English can be described in terms of fifteen syllable nuclei, subdivided into short and long. ...[The] simple long nuclei are tense monophthongs...[i]...[u].'' Examination of spectrograms given in their article shows that there are in-glides for the realization of /i:/ in bead, bean, but this is so far from the expected forms that they were not discussed. It turns out that their subjects were speakers of a conservative white dialect in the Chicago area. In Wisconsin and Minnesota, to the north and west of Chicago, not just /i:/ and /u:/ but also /e:/ and /o:/ are in fact commonly monophthongs and sometimes even inglides. In Chicago, /i:, u:/ are commonly monophthongs. However in most dialects of English, all of these long vowels are up-gliding diphthongs. Lehiste & Peterson's work is classic and often-reprinted, but it suffers from narrowness of dialectal coverage. Phonetic facts from single particular dialects should not be overgeneralized and attributed to the imagined homogeneous national dialect of ``General American.'' The more general American facts are therefore represented in this Reference American system, with phonetic glides on the high-front and high-back vowels (depending on whether the vowel is front or back), as well as on the long mid vowels.
Notice that the glides are defined in relative rather than absolute phonetic terms. Thus a [high] glide will glide from the nucleus to a phonetically higher position, but it will not glide to phonetic high position unless the nucleus is at least mid. The [high] glides of [, o, oi, iy, uw, ei, ou]3.55are each a step above their nuclei. Similarly when r-glides occur on vowels, the r-glide itself retains the color of the nucleus:  in [i] is higher and fronter than the  in .
It is important to note that the nuclei of these long vowels are generally more peripheral and higher than the nuclei of their lax counterparts /i, e, , u/,3.56 though this is not represented in the rules above, since these relationships are dialect-specific.
As Kenyon and Knott (1953) and later scholars have pointed out (e.g. Halle & Mohanan 1985) in various terms, vowel-vowel sequences (as for example N+G) often result in a relatively raised, tense, peripheral quality on the first vowel (the nucleus). This phonetic tendency, which may be called V/_V Raising, applies to non-low Vr# sequences (beer, bear, boor, bore) but not 'VrV sequences within a word (syrup, berry, Barry, jury, sorry). It also commonly applies to the nuclei of the long vowels, consisting of NG sequences: /i:, e:, o:, u:/ all have raised, peripheral nuclei relative to /i, e, , u/ (not shown in Table ). This is the synchronic form of the historical sound-change generalization in Labov, Yaeger and Steiner (1972), that in chain shifts, peripheral (usually, long) vowels rise. A formal statement of the rule, which appears to apply under complex sets of conditions, and applies differently in different dialects and languages, is beyond the scope of the present work. Its effects, however, are pointed out in several of the discussions of the rules of phonetic implementation in later chapters.
V/_V Raising also may be understood as the driving force behind the widespread sound change which raises lengthened /æ/ in many American dialects. If /æ/ becomes a long vowel (after /a:/ merges with other classes, presumably, since /a:/ occupies the long, low, front slot in Reference American), then it fits the structural description of the V/_V raising rule, and the phonetic raising that operates on the nucleus of this vowel may be seen as deriving from V/_V Raising.
Harris gives /æ/-tensing rules for Belfast, Philadelphia, New York City, considered to be deep, lexical phonological rules. Features like [+tense] cannot be added to lexical representations in lexical phonology because of structure preservation. The present analysis suggests that the tensing rule is instead a lengthening rule, with the peripheral and raised quality of ``tensed'' forms being analysed as phonetic features added later by the V/_V Raising rule.
Also, the tensing and raising of /I, , U/ in the ``Southern shift'' (Labov, Yaeger, & Steiner 1972) can also be attributed to V/_V Raising, if these vowels are treated as phonetically lengthened.
Actual systems of phonetic implementation are quite complex and subject to numerous influences, of stress, consonant context, etc. The rule-system provided here does not incorporate these influences, though it may be extended in order to do so.
The phonetic implementation rules given here are proposed as universal rules: they are not dependent on the linguistic phonetic patterns and processes that occur in different dialects. That is, up to this point the rule system proposed is consistent with the assumption of a universal phonetic implementation system. It will be seen in the descriptions in later chapters that this assumption must be modified, in order to generate the phonetic forms characteristic of particular dialects. The description of the interactions with adjacent segments, and other effects within the phonetic implementation system, what I sometimes call the phonetic grammar, is largely a matter for future research, with some steps in this direction taken in the following chapters. The regularities and patterns found in surface phonetic forms may lead to a deeper understanding of the phonological and phonetic system. It is to be hoped that this section provides a useful starting-point for investigation of these patterns.