<< >> TOC
2. A phonetically grounded featural basis
Tonal depression in Zulu is a highly localized phenomenon: if associated with a depressor consonant, the depression gesture is centered within that consonant (TKF: 263). The resulting F0 lowering reaches into the following vowel and may also affect the preceding one. The feature responsible for depression does show some important autosegmental characteristics (see below), but these are highly restricted.
F0 lowering can be achieved via adjustments of (a) the vocal folds (overall and/or longitudinal slacking), (b) the larynx as a whole (lowering), and (c) the subglottal airpressure (reduction) (cf. R, C-M, LSG 1997). The latter is typically used for global pitch control, i.e. in establishing the main pitch ranges in intonational contours and stress patterns (cf. Strik and Boves 1995). Both (a) and (b) are regulated by laryngeal musculature, which appears to allow for more fine-tuned adjustment than the respiratory musculature. Thus, (a) and (b) are common adjustment strategies on the local level.
The local character of tonal depression suggests involvement of laryngeal musculature. Lowering of the whole larynx via the strap muscles is clearly employed in the exaggerated pitch adjustments required in operatic singing (Sundberg 1977) or glissando enunciation (cf. R, C-M, LSG 1997). As electromyographic (EMG) data show, it is also used by Mandarin speakers within contour tones (Hallé 1994). Based on fiberoptic investigation, TKF (265) claim that strap muscles are not involved in Zulu tonal depression, since the expected lowering of the larynx was not observable. However, these visual observations alone may not be sufficient to rule out involvement of particular muscles, since the glottis was not always in plain view.
Since the vocal folds are not consistently slack in depressor consonants (/bh, d, g/ are voiceless when not prenasalized), TKF - by exclusion of all other possibilities - argue that some sort of vocal fold shortening is mainly responsible for the F0 lowering in tonal depression. In some speakers, this shortening is achieved via "an anterior movement of the arytenoids and the posterior movement of the tubercle of the epiglottis" (TKF: 265). This latter mechanism is not observable in other speakers, so TKF (265) stipulate relaxation of the cricothyroid muscle to yield the same effect.
If it is true that vocal fold shortening is the primary cause of tonal depression, this phenomenon cannot be captured by existing proposals on feature geometry without some modification, since Sagey (1990), Halle (1995), and Clements/Hume (1995) incorporate only [slack/stiff vocal folds] and [spread/constricted glottis] into their models. The shortening of the vocal folds is usually accompanied by [constricted glottis], i.e. "the arytenoids [are] almost always adducted" (TKF: 267), but at least for depressor /h/ the constriction is incomplete. It seems to be compatible with [stiff vf] as well as with [slack vf.] (TKF 271), i.e. the causal articulatory gesture is independent of the ones responsible for the primary voicing features.
Of course, one may argue that tonal features are suprasegmental and thus do not need to be captured within the segmental feature geometry. In recent proposals, tonal features are dependent of the typical tone bearing units syllable (σ) or mora (μ) (cf. Yip 1995 for a relevant model and further references), and the tonal categories L, M, H may be taken to be shorthands for relevant articulatory gestures. Similarly, extra low L (cf. Cope 1970, Laughren 1981) may be introduced as a similar shorthand. Such a strategy is unsatisfactory for several reasons: (a) it remains at the descriptive level, but - ultimately - an explanatory adequate theory of feature geometry will have to fill its shorthands with articulatory content; (b) it cannot explain why tonal depression may be triggered by a non-TBU; (c) it wrongly predicts suprasegmental (i.e. non-local on the skeletal level) spreading characteristics of depressed tones.
Shryrock (1995: 6) describes consonantal interaction with tone in Musey, a Chadic language spoken in Cameroon. In this language, the opposite effect from the Zulu case can be observed: a particular set of consonants raises F0 on the following vowel, and low tones seem to be pushed away. As in Zulu, the relevant set of consonants does not represent a natural class describable in terms of commonly assumed features, since it includes both voiced and voiceless obstruents. Like TKF, Shryrock (69-73) provides detailed phonetic evidence against casting the tonally relevant obstruents in a class based on e.g. [slack/stiff vf], [raised/lowered larynx], [fortis/lenis voice] etc. His analysis further parallels that of TKF in that he posits an increase in the longitudinal tension of the vocal folds as the primary cause of F0 raising in Musey. It has independently been shown that tensioning of the cricothyroid muscle raises F0 (cf. Hallé 1994).
Since evidence from two languages from different families (i.e. Bantu and Chadic) points to a similar articulatory gesture as an active component in the phonology of the particular language, it seems warranted to introduce a new feature into the descriptive apparatus. Since the relevant obstruents in these two languages are not tone bearing units as commonly understood, the new feature should be part of the segmental feature geometry.
Assuming the basic feature geometry from Halle (1995: 2), I propose that [decreased longitudinal vocal fold tension] vs. [increased longitudinal vocal fold tension] be added under the laryngeal node, or [lax vf.] vs. [tense vf] for short:
[tense vf] would be the active feature in Musey, whereas [lax vf] is active in Zulu.
To complete the picture, lowering and raising of the larynx should be encoded as well, since the strap muscles are available not only for conscious and voluntary formant control, but also for linguistic encoding of tone (see above). A feature [lowered/raised larynx] should, however, be distinguished from the glottalic features, since the two sets are independent from each other (cf. Trigo 1991). The following feature configuration is possible:
With this enhanced feature apparatus, tonal categories can be correlated with articulatory gestures and thus phonological features as follows:
The laryngeal area contains a large set of muscles, and - as far as I can gather from the literature - the exact function of each individual muscle is not completely understood. Correlating phonological laryngopharyngeal features with particular physiological mechanisms is also difficult, because many muscle groups have synergistic effects. The articulators in (11) thus only have suggestive value. However, even if views on the exact mechanisms of articulation may still change e.g. with better imaging techniques, it seems clear that tonal categories can and must be interpreted as cover terms for sets of articulatory gestures. These gestures may cooccur, but the presence of all of them at the same time does not seem to be required to achieve an F0 lowering or raising effect.
The importance of providing articulatory specification for tonal categories lies in the fact that the points of interaction with segmental features are illuminated. If a language exhibits consistent correlations of H tone with voiceless and L tone with voiced consonants, one may suspect that [ slack/stiff vf. ] is the dominant F0 control feature in this particular language (e.g. Ngizim, cf. Purnell 1997: 29-31). Languages in which tone is largely independent of segmental specifications (e.g. Mandarin), one would expect [lowered/raised larynx ] to control F0, and ejectives/implosives to be absent. The correlation of segmental inventories with particular tonal patterns may thus be an area worth exploring. In Nguni languages, voiceless and ejective consonants are neutral with respect to tone, which leaves only [ lax/tense vf ] as an F0 control feature that is relatively independent of segmental structure, with the notable exception of depressor consonants.
<< >> TOC