<< >> TOC
1. Description of tonal depression
The term "tonal depression" refers to a distinct lowering of the fundamental frequency (F0). The set of Zulu consonants that consistently cooccur with F0 lowering on the immediately following vowel is shown in (1):
The continuants and affricates v, z, h, dl, j are clearly voiced. The stops bh, d, g are voiceless unless they are prenasalized (TKF 1987, GPT 1988). The depressor clicks gc, gq, gx are marked as post-breathy voiced (R/N 1979), which really only means that they are depressors. In terms of surface phonation, "pitch lowering is the primary and only reliable manifestation of depression in Zulu" (TKF 1987: 271).
Example (2) shows waveform and spectrogram of the word ukubheka 'to look at', as pronounced by a KwaZulu speaker from the instructional tapes accompanying R/N 1979.
As indicated by the accent mark, the F0 rise on the second vowel is perceived as a H tone by native speakers. In contrast to the "lenis voiced" (R/N 1979) k [g], the depressor consonant bh is clearly voiceless, but nevertheless induces a noticeable dip in the fundamental frequency of the following vowel. I took one F0 measurement from the peak of each vowel and two from e, since it is bimoraic due to penultimate lengthening. The tone bearing unit in Zulu is arguably the mora, since only lengthened vowels may carry full phonological contour tones.
To illustrate the degree of F0 lowering induced by depressor consonants, example (3) juxtaposes similar F0 measurements for different words with the same moraic structure (VCVCVVCV, w/ V = moraic V-slot) and the same basic H tone pattern (H on the second vowel).
While the pitch contours of the words with aspirated or lenis voiced stops pattern together, it is rather clear that the depressor consonants (here: g in ukugula and bh in ukubheka) cause a deviation towards a significantly lower pitch level.
As in other Bantu languages, H tones are mobile and do not necessarily surface on the donator morpheme, i.e. the morpheme that carries an underlying specification for a H tone. Depending on the morphosyntactic structure of the word and the tonal patterns associated with it, it can thus happen that a H tone targets a V-slot (or mora) that is also subject to tonal depression due to the segmental context. Since the pitch raising gesture of the H tone is antagonistic to the lowering gesture involved in depression, a conflict arises which is usually avoided by associating the H tone with the next available V-slot on the right. This resolution algorithm has been referred to as 'depressor induced H tone shift', and it is blocked if another depressor consonant intervenes between the initial target vowel and the next available one. In this case, H association largely overrides the depression gesture, and the target vowel surfaces with a rising onglide leading into a H tone, i.e. with a depression-induced monomoraic phonetic contour. The pitch level of such a 'depressed H tone' remains lower than that of a regular H tone. The following examples serve to illustrate this behavior (data from Rycroft/Ngcobo 1979). Consider (4) first:
With Khumalo (1987) and Clark (1988), I assume that only H tones (indicated by accents) are manipulated to yield the basic H tone melodies, whereas mid tone levels are supplied by default to TBUs without tonal specification. I tentatively follow Clark (1988) and Downing (1990) in assuming that H tone melodies are accentual, which means that morphemes can serve as "donators" and introduce H tones into a metrical structure. This structure may then be manipulated by a particular set of rules (which are not relevant here) with the result that the H tone appears to be mobile, i.e. it can surface away from its donator. In (4) above, the prefix u- as well as the root leth are donators, but the root's tone is attracted to the penultimate syllable, which yields the characteristic surface pattern.
The mobility of H tones is further exemplified in (5), where the prefix H tone surfaces on the antepenultimate syllable for metrical reasons:
(6) illustrates the effect of a depressor consonant on the tonal melody. The morphological material matches that of (4) with exception of the subject prefix zi- containing a depressor. The H tone donated by this prefix would be expected to surface in situ, as it does in (4). However, the presence of the depressor consonant causes an adjustment in the tonal melody that results in the H tone surfacing on the next syllable to the right.
The morphological material in (7) parallels that of (5) insofar as the subject prefix represents the only donator morpheme. As in (5), the metrical pattern targets the antepenultimate syllable for the H tone. However, the root initial depressor v causes an adjustment similar to the one in (6), i.e. the H tone appears to be shifted to the next syllable on the right.
The morphological material in (8) would lead one to expect the same surface tonal pattern as that of (6), since the subject prefix in both cases starts with a depressor consonant, which should shift the H tone to the right. However, the initial depressor consonant of the tense prefix in (8) seems to block this shift.
Without going into details, I do not think that any of the existing analyses can account for these descriptive facts with sufficient explanatory adequacy. Traditional autosegmental accounts manipulating H tone patterns via spreading and delinking have to make ad hoc assumptions about the timing of L tone introduction through the depressor consonants. Without such assumptions, it remains unclear why depressors seem transparent for the purposes of initial H tone displacement (cf. 7), but opaque for depressor induced H shift (cf. 8). A metrical analysis of H tone patterns can accommodate the transparency facts, since the relevant TBUs are units on the metrical grid are unaffected by specifications dependent on the skeletal tier. The opacity effects in the context of depressor induced H shift, however, pose a problem and, in my opinion, require additional assumptions about interface conditions between metrically relevant units and the segmental structure, which I hope to provide in the remainder of this paper.
2, 3 The falling contour on b˘shwa is due to a relevant underlying specification of this morpheme.