The effects of following /l/ on vowels is perhaps the strongest and most regular VC coarticulation effect found in this data. However, the effects differ across dialects. Consider the plots in Figure , which display for six speakers the statistically significant effects of following /l/ on vowel nuclei.
The arrows displayed represent statistically significant effects. The head of the arrow is located at the mean of the distribution of instances of that particular vowel which precede an /l/, while the tail of the arrow is at the mean of the distribution of all other instances of that vowel.
Arrows with large arrowheads represent differences that are significant at the p19#190.001 level. Medium sized arrowheads represent a significance of p19#190.01, while small arrowheads represent those effects with a two-tailed significance of p19#190.1.
The basic dark-l coarticulation pattern is that seen most clearly in the CWE chart of Rita. The database of measurements for her speech was much larger (4470 tokens) than those of the other speakers, and thus the significant effects are generally more numerous and more significant. /l/ has a strong backing effect on all vowels, the strongest effects are on the high-front vowels, /iy, ey, /, where F2 is lowered by as much as 1000Hz. These differences are both statistically significant, and above the difference limen for formant frequency perception, so these are real phonetic effects, by the rule of inference aregued for in Chapter 4.
This basic pattern is repeated in the speech of all the mainland American speakers, applying to all the significant effects for Judy, James H., and Vince (as well as Jim, not shown), with only a few exceptions in which backing co-occurs with rather extreme raising or lowering. Some of these exceptions may be attributed to other processes, either phonological or phonetic, while some of the them appear to be unexplained except as brute idiosyncrasies of the phonetic implementation system of the individual dialect. Each of these exceptions is discussed below.
The first exception involves the /l/ sequence in Chicano English, which is shown to have a slight backing effect, but primarily a very strong lowering effect on //. /l/'s effect on a preceding // is a marker of Chicano speech in the Southern California. Among the words found with this alternation are [æ]lementary, B[æ]lvedere, hims[æ]lf, r[æ]latives, w[æ]ll, t[æ]ll, etc. This phenomenon was first noticed by Metcalf (1979), and has also been discussed by Wald (1984), and Santa Ana (1991). The linguistic interpretation of this significant shift (p19#190.0048) is unclear, but it is possible to attribute it to a change of phonological category (cf. Santa Ana 1991). It is clear that the effect of /l/ in the development of the ethnic Chicano dialect was quite different when it occurred in the context of // than elsewhere, but now this effect may be phonologized. If the vowel in this environment is taken as underlyingly //, then a phonetic rule is required to lower the vowel in this context to [æ]. But if it is /æ/, then no such rule is necessary.
Another apparent exception can be understood via the interaction of several phonetic processes. Jackie from Chicago has a sharply lowering effect of /l/ on /ay/, as seen in her chart in Figure . This effect is significant at p19#190.035, even though it results from only two tokens, Niles and I'll, which are at the low back periphery of the /ay/ distribution. These tokens are in this location because of three phonetic effects: relative lowering due to stress, relative lowering due to the following voiced environment, and backing due to following /l/. Chicago (and Northeastern U.S. speech more generally) displays an alternation in which the nucleus of /ay/ is raised before voiceless consonants. Following /l/ is a non-raising -- i.e., lowering -- environment for this reason. Further, stress reduction has a raising effect on /ay/ as seen in Figure , so that stressed tokens are relatively lowered. Finally, the backing effect of /l/ does seem to apply to these tokens in addition to the other lowering effects.
The chart of the Alabama English speaker, James H., in Figure contains several phonetic anomalies. For // and /ow/, in addition to a backing effect of /l/, there are also a strong lowering and raising effects, respectively. The lowering of the nucleus of // before /l/ occurs mostly with preceding labial environments: well, twelve, fellow. The effect may be due either to the following /l/ or the preceding /w/, but in either case it is an unnatural (non-assimilatory) phonetic effect. Coarticulation with a high-back-round glide, /w-/, should have a raising and backing effect, but this effect is of lowering. Similarly the strong low-backing effect (p19#190.0001) cannot be attributed to the general effect of /l/ as found in other environments or in other dialects. The other case is the strong back-raising effect of /l/ on the nucleus of /ow/ in James H.'s speech (p19#190.0007). This also cannot be attributed to stress, which is relatively balanced (9/14 tokens are stressed), or to preceding segments (either consonants or vowels) which fall into no natural class, even in a majority of cases. The words involved are old, told, hold, and hole, in order of frequency.
Many vowel-consonant interactions in this and in other data not presented here are phonetically unnatural patterns which do not appear to be explicable in terms of more general processes. Until a better explanation can be found, the only way to represent them appears to be as an arbitrary, language-particular stipulation in the system of phonetic implementation, such that these particular contexts have these particular arbitrary phonetic effects. Examples of this kind could be multiplied ad infinitum. It may be hoped that further exploration of this class of facts may lead to deeper understanding of the interaction of various processes of phonetic implementation.
A deeper pattern is evident in the effects of /l/ on the vowels in Jamaican Creole. In contrast to the mainland speakers, the Jamaican Creole speakers have a qualitatively different pattern of effects. In their charts, shown in Figure , the usual backing effect of /l/ appears to apply to the back vowels, but the front vowels do not show this effect, and have raising instead, where the more significant effect, in Juba's chart, includes some fronting as well. The long low vowel /aa/ for Roasta has a quite significant, though small, low-fronting effect, opposite to the backing effect in other dialects. This fronting effect may derive from an interaction: most tokens are of the words all, always, and 29/34 are stressed; stress has a small fronting effect on this vowel for Juba, though not for Roasta (cf., Figures , ).
The evident explanation for the lack of a backing effect on the low and front vowels is the quality of the /l/ itself. Jamaican /l/ is quite different from the /l/ found in other dialects, reflecting the differences in phonetic realization of this phonological category. In Jamaican, even tautosyllabic /l/ is clear, non-velarized, while in the other dialects studied here, tautosyllabic /l/ is dark, velarized, often vocalized. Because the realization of tautosyllabic /l/ in mainland dialects includes a velar constriction involving the body of the tongue, it has a strong effect on preceding vowels, which also derive their quality from movements of the tongue-body. Where /l/ is not velarized, there is no such effect, as with the high-front and the low vowels in JC. /l/ appears to be redundantly velarized in the context of high-back vowels, which involve a high-back (that is, velarized) tongue position to begin with. Velarization would seem to be a redundant phonetic feature added to the form of the realization of /l/ in JC in the context of high-back vowels, which in turn has a high-backing effect on the vowel nuclei.
How are these effects to be accounted for in linguistic theory? /l/ is phonologically a lateral continuant in English dialects. The /l/'s of Jamaican, of Chicago White English, of Los Angeles Chicano English, and of Alabama Southern White English are all lateral continuants. An underspecified representation for /l/ would contain solely the feature [lateral], which distinguishes it from /r/; other features, [continuant], [voice], etc., are predictable and redundant. Despite these underlying phonological similarities between the /l/ of mainland American dialects and the /l/ of Jamaican, their phonetic realizations are quite different. Are their phonological forms also different? One may imagine that the /l/ of the mainland dialects has a redundant phonological feature, [velar] or the like, while Jamaican Creole has no such feature. However, Fujimura and Sproat (1989) demonstrated that clear-l and dark-l in English (in some mainland U.S. dialect) is a gradient phonetic difference that depends probabilistically on syllable structure. The clear-l, dark-l distinction in this dialect is a probabilistic, gradient, therefore phonetic effect, rather than a categorical, discrete, phonological alternation. Therefore there is no need to include a categorical phonological feature like [velar] or [dark], etc., in the surface phonological representation of mainland /l/, since /l/'s degree of ``darkness'' is necessarily determined by factors in the phonetic implementation system.
This gradient phonetic effect does not appear in Jamaican Creole. Instead, another effect appears, where /l/-darkening (which I assume is measured by its backing effect on preceding vowels) is determined by the quality of the preceding vowel. The phonetic interpretation of /l/ in Jamaican as light or dark in the environment of high-back vowels may again be either a phonetic or a phonological process. But in either case, the phonetic process which, according to Fujimura and Sproat, implements /l/ in mainland dialects as more-or-less dark depending on syllable structure, is absent in Jamaican Creole. It follows, then, that phonetic implementation processes exist in some dialects that are absent in others. Furthermore, if the dark-l which occurs after high-back vowels in Jamaican Creole is also conditioned by a strictly phonetic rule, as is plausible since it is so conditioned in mainland English, then it would seem that there are two distinct phonetic processes by which lateral continuants are implemented as sounds.