Consonant Effects

In this chapter, coarticulatory effects are investigated in different dialects, in order to show that vowels in the environment of particular following consonants are regularly found to have modified phonetic quality (relative to their average quality, in terms of F1, F2 measurements), in ways that are particular to individual dialects. The exploration of the form and structure of the phonetic system by which consonant environments exert their varied effects on vowels in the dialects studied here is beyond the scope of the present work. Instead, I wish simply to document the existence of this realm of linguistic variation. Once the existence of these dialect- and language-particular coarticulatory effects is accepted, further research may begin to characterize the structure and functioning of this aspect of the linguistic system.

In view of the distinction between ``hard'' and ``soft'' coarticulation advanced here, the system by which phonological representations are converted into speech would appear to have considerable leeway for quite complex and as yet badly-understood sets of interactions between vowels and their phonetic or phonological context. These interactions may or may not be dialect-specific, but it is important not to approach their study with the preconception that they must be universal effects. Before deciding whether a given effect is language-specific or not, we must see what some of these effects are in the first place. To this end, a study of the effects of two particular consonant environments on the F1-F2 measurements of the vowel nuclei has been carried out. These are the backing effect of following /l/ in all dialects except Jamaican, and the lowering effect of following // in Alabama as opposed to other dialects.

The statistical methods are discussed above in Section . To briefly summarize the discussion there, the effect of a given consonant on a particular vowel is measured by locating the means of two sets of F1-F2 measurements: first, those of the given vowel in the context of that particular consonant, and second, the measurements of that vowel in all other contexts. The data points are projected onto the line between the means in F1-F2 space, and statistical tests for different variances (F ratio) and different means (equal-variance t-test or unequal-variance t-test) between the two sets of measurements are carried out.

It should be pointed out that the formant trajectories for a given vowel would show even more differences across consonant contexts than are shown by the differences among nucleus measurements. Thus the differences found here are undoubtedly the tip of the iceberg, and a more complete theory of phonetic performance must account for those differences as well.

The task of interpreting these statistical effects is, in general, a difficult one. The fact that the data is unstructured natural speech means that there is no control over other interacting effects. For example, after a /w/-glide, unstressed vowels are frequently backed and rounded, so that the word with, for example, will have a nucleus far to the back of tokens of the // vowel in other contexts. Due to imbalances in the data, such an effect may masquerade as an entirely different one. Thus for example if the word with occurs very frequently, and the following consonant // otherwise co-occurs rarely with preceding //, then the strong backing effect of the preceding /w/-glide will also show up as an apparent strong backing effect of //. Thus it is a difficult matter to be certain, in any one case, whether the effect is genuine or due to other factors which are skewing the data. For this reason, the effects presented below are not based on single vowel-consonant coarticulatory or allophonic effects (here called VC effects), but are based on sets of individual VC effects which pattern similarly. If all the labial consonants (/p,b,m/), for example, have a similar low-fronting effect on a preceding vowel, it is unlikely that the low-fronting effect is due to some lexical or prosodic interaction, because an identical interaction must have coincidentally applied to all three consonants. That seems to be a rather unlikely event. These two effects, chosen from among many, are consistent in this way.

The simplifying assumption of a universal phonetic implementation system may be necessary in the absence of a necessary body of observation and theory. This assumed universal phonetic system contains solely information about such matters as acoustics, aerodynamics, vocal-tract anatomy, and physiology, and additionally a non-cognitive, non-linguistic, universal system by which phonological categories are implemented and interpreted in speech production and perception. Some previous work testing these assumptions has been done. Göstë Bruce, for example, showed in his work on Swedish tone that the implementation of accent places a pitch inflection at a point in the syllable different from the implementation of stress in English. Similarly, Laniran (1991) showed that phonologically similar tone sequences on phonologically similar segmental sequences was realized in different patterns of pitch-contours in different two different languages. In these cases the phonetic realization of similar phonological forms is different according to linguistic or other high-level factors.

I will conclude that strictly phonetic differences between dialects (and languages, for that matter) do exist. Phonetics may consider cross-linguistic and cross-dialectal differences, just theoretical linguistics includes strictly phonetic considerations.