Four tones and downtrend: a preliminary report on pitch realization in Mambila, a language with four level tones

Bruce Connell
Institute of Social and Cultural Anthropology
Oxford University
51 Banbury Road
Oxford OX2 6PE
fax: +44 1865 274 630

revised version of a paper presented to the 27th ACAL, University of Florida, Gainesville, 29-31 Mar 1996 and the LAGB, University of Sussex, Brighton, 11-13 Apr 1996.

comments welcome

1. Introduction

One of the more developed areas in pitch research, on both tone and non-tonal or intonation languages, has been concerned with F0 downtrends over the course of an utterance. Particularly in European languages this research has frequently focused on declination. However when considering tone languages of the sort generally found in Africa, declination is not the only source of F0 lowering: phenomena such as downdrift, (e.g. Hombert, 1974), downstep - automatic or non-automatic (Stewart, 1983), downglide and final lowering (Welmers, 1973) have all been discussed in the literature. One of the fundamental areas of debate is the relative importance of the roles of phonetics and phonology in determining downtrend phenomena (see Snider & van der Hulst, 1993) for discussion). One means of finding evidence to resolve the debate - or at least further it - is through experimental work. However, detailed experimental studies of tone languages are still relatively few: pitch realization has been examined in a small number of languages with two or three level tones, most notably Igbo (Laniran, 1992b; Liberman, Schultz, Hong & Okeke, 1993) (Lindau, 1986; Inkelas and Leben 1990), and Yoruba (Connell & Ladd, 1990; Laniran, 1992a), though to date there have been almost no instrumental studies of languages with four lexically distinctive tone levels. Instrumental analyses have been done on Bamileke Dschang (Bird & Stegen, 1993) as having both H and L tones subject to downstep, giving four tone levels, and Anlo Ewe (Clements, 1981), presented has having four phonetic tone levels. Over the past two decades, researchers concerned with intonation languages (e.g. Gussenhoven & Rietveld, 1988; Ladd, 1984; Ladd, 1990; Pierrehumbert & Beckman, 1988), have increasingly adopted the view that it is discrete targets or levels (i.e. High and Low tones) that are relevant in these languages, rather than contours or configurations. Much of the impetus for this redirection came from early work on African tone languages having two tone levels. The existence of languages with multiple tone levels raises some interesting questions with respect to pitch realization, the nature of the (phonetic) tonal space, and the phonological specification of tone.

Before going further, it may be useful to clarify what is meant by 'downtrend', as there has been considerable overlapping or ambiguous usage of terminology in discussing tone within the Africanist literature, and the cross-fertilization mentioned above has, if anything, complicated the terminology even more. For the purposes of this paper, the following terms are used:

Declination: the gradual modification of the phonetic backdrop of F0 over the course of an utterance.

Automatic downstep: the lowering of F0 in sequences of alternating Hs and Ls (sometimes referred to as downdrift).

Non-automatic Downstep: F0 lowering in tonal sequences with no obvious conditioning environment (sometimes referred to simply as downstep). Crucial to the characterization of non-automatic downstep is the setting of a new lower height for subsequent tones within the same phonological phrase, a 'lowering of the ceiling'.

Final lowering: a more abrupt lowering restricted to the ends of utterances.(Liberman & Pierrehumbert, 1984) and (Pierrehumbert, et al., 1988) report final lowering in English and Japanese respectively, and suggest it is restricted to (or extends over) the last 250 ms or so of an utterance. Welmers (1973) suggests it may be characteristic of discrete level languages in Africa.
Downtrend is non-specific, an umbrella term for all of the above.

Other terms which may usefully be defined here are:

Register: Clements (1990: 59) defines register as the "frequency band internal to the speaker's range which determines the highest and lowest frequency within which tones can be realized at any given point in an utterance" - to distinguish it from an individual's normal pitch 'range'. Potentially, the register can be raised or lowered, expanded or narrowed. One of the major areas of debate in the current literature on pitch realization centers on the relation between downstep and the register.

Tonal space: Ladd (1992) uses this as loosely equivalent to Clements' 'register'; here I define tonal space by adapting Clements' concept of register such that it applies to individual tones: 'the frequency band which determines the highest and lowest frequency within which a given tone can ordinarily be realized at any given point in an utterance'. That is, each tone is assumed to have its own space within the register.

Terrace-level language vs discrete level languages: As the name suggests, terracing languages show a stepping or staircase phenomenon, this effect typically resulting from downstep; and as the name implies for discrete level languages, each tone has its own space which in principle does not impinge on other the space of other tones. While the distinction in practice has long been recognized as fuzzy, it can be useful, particularly if the notion of 'discrete' is adopted in a weak form.

This paper provides a preliminary report on pitch realization in Mambila, a language with four lexically distinctive tone levels. The report is mainly descriptive; in the absence of any prior experimental studies of Mambila tone, the research was designed in the first instance to give some idea as to what might be its potentially most interesting aspects, thereby providing a foundation for further, more detailed, work. Given the preliminary nature of the study and especially that it is based on the speech of only two speakers, there is no attempt to formalize or model the data.

The questions addressed are basic:

1) Considering the types of downtrend discussed above, which if any, are present in Mambila?

2) Assuming any of these are present, are their effects similar with respect to the different tones? E.g., in the case of declination, are all tones affected to the same degree?
3) If there is declination, is there any correlation between rate of decline and utterance length? Or between the height of the initial tone and utterance length?

4) Are there any other observable interactions between different adjacent tones?
- if so, does each tone interact to the same extent with each of the other tones?

5) Does each tone have its own unimpingeable space? In other words, in the absence of downstep, can Mambila be described as a discrete-level language?

There are of course numerous other questions which could and eventually should be addressed. From a phonological perspective it is now common to analyze languages with multiple tone levels as having two registers, upper and lower; is there evidence for such an analysis for Mambila? (And what would constitute such evidence from the phonetic standpoint?)

Mambila is also interesting from a diachronic standpoint. Though there are other languages in the general area having four and even five tone levels, nearby and (presumed) closely related languages have either three or two tones. The four tone system of Mambila is almost certainly a reflex of an earlier two tone system; the three tone systems conceivably represent an intermediate step or may represent a diachronically independent development. Developments internal to the four tone system, e.g. the proliferation of contour tones, is also of interest from this perspective.

Finally, and from a methodological perspective, given that much of the debate in this area has focused on the relative roles of phonetics and phonology, it is important to be able to separate out at the beginning those effects which can clearly be attributed to one or the other. Where in the signal to measure F0, e.g. to avoid consonant perturbations, is by now a relatively well understood problem. Apparently less attention in this vein of research has been paid to the need to account for differences in intrinsic pitch of vowels, which phonetic research shows can be quite substantial (as much as 25 Hz between high and low vowels; (Ohala, 1978; Laver, 1994). Differences in pitch correlating with vowel height are likely normalized by listeners during the perception process (Hombert, 1978), but are a potential source of confusion for the pitch researcher. To date there has been very little experimental work done on intrinsic pitch of vowels in tone languages. (Hombert, 1977; Laniran, 1992a) for Yoruba show a significant height effect, though more so for Hs than Ls; (Zee, 1978) reports a similar tendency for Taiwanese. It is not clear whether, and if so to what extent, this is an aspect of tone production which can be over-ridden by phonological concerns in the same sense that declination (which is also assumed to be a phonetic effect) can be.

Similarly, the possibility of tonal coarticulation is too infrequently addressed. (Gandour, Potisuk & Dechongkit, 1994) report an investigation of tonal coarticulation in Thai; work from a similar perspective has not been carried out on African tonal systems; though automatic downstep and the kind of local effects reported below in this paper are obviously related phenomena.

2. Mambila: Mambila is a cluster of dialects located in the Nigeria-Cameroon borderland. It is generally considered to be Bantoid (see Blench 1993 for discussion). Phonological research on Mambila tone has been done by Perrin (1974, n.d.). In her analysis, three units are required to describe tone in Mambila: the syllable, (phonological) word, and span. Every syllable bears at least one of four level tones which, in Perrin's system, are numbered from 1 to 4 (High to Low). There is sufficient lexical contrast based solely on tone that it is not problematic to find minimal pairs; triplets and quadruplets are less common, but do exist. Examples of these are given in (1).

1. Examples of lexical tonal contrasts:

Possible tonal patterns - combinations or sequences of tones within the syllable - are limited to 2-1, 3-1, 4-1, 1-4, 2-4, 3-4, 2-1-4. These patterns appear to define the phonological word, along with others introduced through the addition or influence of grammatical tones, whether a word is of one, two, or three syllables. Three syllable words in Mambila are uncommon, with the majority of words being monosyllabic. The few exceptions to these patterns that have been found also prove to be phonologically deviant in other ways.

Little mention is made by Perrin of pitch or tonal behaviour with respect to the span; the span however, is said to correspond to the grammatical clause, so many of the phenomena under examination in this paper can be seen as properties of, or corresponding to, the span.

As well as having a lexical function, tone also fills grammatical (e.g. negation, imperative) and semantic (deictic) functions in Mambila (e.g. motion toward can be indicated by replacing the inherent tone with T4). Perrin notes that there is little in the way of tone sandhi, and makes no mention of any dialect of Mambila exhibiting downstep as is frequently found in other (two tone) languages of the wider area.

3. Field recordings. The data used in this paper are from the variety of Mambila (referred to locally as Ba) spoken in the village of Somié, Adamawa Province, Cameroon.1 It differs little from the Atta/Sonkolong varieties (also Ba) discussed by Perrin. A corpus of sentences was constructed to allow for an examination of various aspects of the Mambila tone system. This comprised: a) like-tone sentences (i.e. all T1, T2, T3, or T4) of varying length; b) mixed tone sequences, where for the most part all but one tone were the same. The odd tone was placed in some cases early in the sentence, and other cases late in the sentence. Sentences ranged in length from two to nine syllables (the same range was not possible for all four tones). In all, 30 sentences were used. This corpus was recorded by two male speakers, both in their early twenties and both of whom grew up in the village of Somié, in adjacent compounds. The recordings were done in 'village conditions' using a DAT recorder. The resulting recordings are therefore basically of high quality, but for a variety reasons a number of tokens had to be discarded. The list was recorded five times by each speaker. Note that while Mambila has an orthography, and there has been some bible translation and school primers prepared, it is essentially an unwritten language. The younger generations (including the speakers used in this study) have had instruction in Mambila in primary school, but this has not been extensive. The reading therefore required rehearsal, and was somewhat uneven. Measurement of F0 was done following standard procedures (cf. Connell & Ladd, 1990; Bird & Stegen, 1993).

2. Tone Combinations used in the dataset: spaces between tone numbers reflect word boundaries. Some sentence types had more than one token.

1 1 1
1 1 1 1 1
111 1 11 1 1 1

4 4
44 44 44

1 1 4 1
1 4 1 1
1 4 1 1 1

4 4 11 1
4 4 1 1 1
4 4 4 1
4 4 4 4 1

2 2 2
2 2 22 2

21 2 2 222
21 21 22 22 1

4 2 2 222

3 3
33 3 33
33 3 3 3
3 3 33 33
33 3 3 33 33

3 1 33 33

4 33
4 3 4 44
4 44 4 3 44
4 44 4 1 3 44

4. Results

The results obtained are at once both straightforward and complex. It is apparent that whatever downtrends do exist in Mambila are minimal; it is not so clear, however whether certain effects should be attributed to declination, final lowering, or both.

4.1. Like tone sequences: Table One presents mean F0 values and standard deviations for the initial (I), penultimate (P), and final (F) syllables, and a second measurement taken near the end of the final syllable, final lowering (FL),2 for the four tones.
4.1.1. Declination: Comparing values for I and P, these figures suggest that for all tones there is little that can be called declination in Mambila, and certainly none on the scale reported by e.g., Hombert (1974) for Shona or Lindau (1986) for Hausa. In fact, for JE there is only a slight decline for T2 and T3, while for speaker NM there is no drop in F0 between I and P for any tone.

Table One: Mean values and standard deviations of F0 for like tone sentences. I = initial syllable, P = penultimate syllables, F = final syllable, and FL a measurement near the end of the final syllable. The number of tokens on which the measurement is based is given in parentheses after the tone number.


The form of reporting in Table One is useful in testing for declination when collapsing results from sentences of varying length, but it may also obscure other factors of potential interest, e.g., whether a given effect applies relatively uniform throughout an utterance or occurs at only one place and persists.

For T1, the figures show a rise from I to P for JE. This actually occurs early in the sentence: by the second syllable for shorter sentences and the third syllable for longer sentences, from where F0 tends to remain relatively constant until the end of the utterance. NM was not consistent across sentences in this respect; but tended to show the opposite result: shorter sentences followed the trend in evidence in Table One for T1, i.e. a gradual rise throughout the utterance, while longer sentences tended to reach their peak on the first or second syllable and then remain level. Note, however, that since in all cases the rise, in absolute terms, is on the order of a few Hertz, and well within the amount of variation found, the significance of this result is not clear. A similar tendency for High tones, however, has been reported for other languages, e.g. Liberman et al. (1993) for Igbo, and (Leben, 1996) for Baule.

For JE, in T2 utterances there is a small drop in F0 which showed a greater tendency to occur in longer sentences, though on average less than one semitone. Speaker NM showed no fall in F0 for T2 except in the longer sentences, where it was only slight, and less that observed for JE. On balance, then, it can be argued that effectively no declination exists for T2, despite an apparent slight tendency in this direction for longer sentences. Note also that there is no raising of the initial tones in these sentences, a phenomenon often observed in the literature to correlate with length of utterance in regard to declination (e.g. ('t Hart, 1979), but see Ladd & Johnson, 1987 for discussion). However, given the small tendency for a drop in F0 in longer sentences, this is an area which bears further investigation.

T3 shows a similar discrepancy between the two speakers: JE has a relatively consistent, albeit slight, tendency for a drop in F0 across all utterances, while again for NM only the longest utterances showed any decline. Other than this finding for NM, there is no indication from T3 that declination correlates with utterance length. For both speakers, the average drop in F0 was again equivalent to less than one semitone.

Due to an unnoticed problem in constructing the sentences used to investigate T4, the amount of data available for this tone is less than for other tones.3 However, results for T4 in fact show greater consistency across the two speakers than was found for the other three tones. For both speakers, F0 remained relatively constant until the penultimate syllable, from where it falls off more sharply, suggestive of the fall often associated with final Low tones.

4.1.2 Final Lowering: The evidence for final lowering (comparing now values for P, F and FL in Table One), is stronger for some tones than others. Specifically, there appears to be a case for final lowering of T4 for both speakers and of T2 and T3 for NM.

T1 does not show any appreciable drop in F0 over the last two syllables for either speaker. For T2 and T3, both speakers show a tendency for a drop in F0 over the final part of the utterance, which is somewhat stronger for NM. For both speakers, the decline is steepest in, though not restricted to, the last half of the last syllable, i.e. within the last 40 - 80 ms.4 There are, however, differences between the two speakers. For NM, the only downtrend in T2 and T3 sentences was over these last two syllables; for JE, on the other hand, it is not entirely clear that what in isolation (i.e. looking only at the difference between P and FL) appears to be final lowering can in fact be separated from the drop in F0 that occurred earlier in the utterance.

It might be argued that the two speakers are here employing different strategies to achieve essentially the same end - a certain degree of F0 lowering over the course of an utterance (i.e. declination), although this might be seen as an unlikely and unwanted solution. Rather, given the slender evidence for declination in T2 and T3 utterances for JE, it may be more plausible to suggest that the main (or only) effect, for both speakers, is one of final lowering. This solution, though, is still somewhat wanting in support; it is discussed further after examining results for T4.
Only T4 shows a substantial decrease in F0 over this final stretch of the utterance, dropping an average of 18 Hz for JE (equivalent to approximately three semitones) and 29 Hz for NM (4.5 semitones). Given the absence of any declination effect in T4 utterances, it would seem reasonable to regard this as final lowering. The behaviour of T4 is also noteworthy, in that the sharpest decline in F0 starts earlier than for other tones, and is relatively consistent across the last two syllables. In both these findings the behaviour of T4 agrees generally with that of Low in Yoruba (Connell, et al., 1990; Laniran, 1992a).

As mentioned, there is cause to see the late drop in F0 in T2 and T3 utterances as final lowering. However, it is different from the lowering observed for T4, not only in being essentially confined to the last half of the final syllable, but also in being less steep. For JE, in both cases, the drop is just under one semitone, while for NM, T2 falls approximately one semitone and T3 1.7 semitones. It is conceivable that the tonal space for T2 and T3 within the register is restricted, inhibiting a further drop. On the other hand, it is also interesting to observe that when the overall downtrend (i.e. from I to FL) is compared for the two speakers in terms of semitones, there is a very close match: approximately one semitone for T2 and 1.7 semitones for T3. This lends support to the analysis suggested above, that there may be declination associated with T2 and T3, though realized at different rates for the two speakers. If this is the case, it presents a situation where declination applies differently to different tones, or for some tones it may not apply at all. This seems in any case to be unavoidable, since no downtrends have been observed for T1. Moreover, it would not be the first reported case of a downtrend applying differentially to different tones: Connell & Ladd (1990) suggest this with respect to Yoruba, and Clements (1981) observes a tendency for the two (phonetic) mid tones of Anlo-Ewe to demonstrate greater downdrift (though L does show downdrift as well).5 If this analysis is accepted, a remaining question is whether the final lowering associated with T4 should be reanalyzed as declination, and its much steeper drop attributed to the widely reported tendency for final Low tones to fall. On the other hand, typological studies of tone systems (e.g. Hombert, 1974) indicate the status of T4 may be crucial in understanding T2 and T3. It appears from this work that declination (automatic downstep, downdrift) in a higher tone implies declination (automatic downstep, downdrift) in a lower tone; if the downtrend in T4 is analyzed as final lowering it would consequently provide some evidence that T2 and T3 are also subject to final lowering (i.e. not subject to declination); on the other hand, if T4 is analyzed as undergoing declination, this status provides no evidence either way for the status of T2 and T3.

4.2. Mixed Tone Sequences. We look first at utterances combining T1 and T4, with T4 placed early or late in a sequence of T1, then with a T1 placed late in a sequence of T4. Subsequently interactions of T2 and T3 with T4 are examined.

4.2.1. Effect of early T4 on T1: Three test sentences allowed for examining the effect of an early placed T4 on T1; mean F0 values and standard deviations for these are given in Tables 2 - 4. These show clearly that a preceding T4 has a lowering effect on T1, but that this is also very much a local effect. There is no persistent lowering as one finds in cases of downstep, as T1 normally returns to its inherent level within two syllables.6 Both speakers agree in this. Further support for interpreting this is a local phonetic effect comes from variations in speaking rate, particularly for NM, which showed there was a greater tendency to reach the inherent T1 target with more slowly read sentences. T4 generally appears not to be affected in these environments.

4.2.2. Interaction of late T1 with T4: The figures in Table 5 show a final T1 reaching its expected range after a sequence of T4. This confirms both that any influence of T4 on T1 is a local effect only, and that T1 is not subject to final lowering. It is interesting to note in these sentences that T4 shows a tendency to downtrend. It is not immediately clear how this is to be interpreted, though it may be an indication that T4 is in fact subject to declination, in contrast to what appeared in the like-tone sequences above.

4.2.3. Interaction of early T4 with T2: The values given in Table 6 show a similar interaction (or absence of) between early T4 and a subsequent sequence of T2; for both speakers T2 reaches its normal level on the first succeeding syllable, i.e., unlike T1 it appears unaffected by T4.


4.2.4. Interaction of T3 and T4: The interaction between tones shown in Table 7 is conceivably more complex than those reported above. First, it is clear in this environment T3 does not reach its normal range, but is nestled firmly in T4 space. For NM, it is in fact lower than the succeeding T4. Following what is described above for T1, this behaviour is interpreted as a local effect, non-attainment of the target in the case of JE, or late attainment in the case of NM. Further work, using a sequence of T3 would confirm or refute this conclusion. Finally, the drop in F0 on the second occurrence of 'nuar', in this utterance bears comment. This is related to the second potential problem with T4 referred to in fn. 1, the fact that 'nuar bele nuar' translated literally as 'person deceives person', is a prosodic constituent in its own right, whether seen as a compound word or phrase. The drop in F0 on 'nuar' might therefore be attributed to final lowering; since it is not utterance final, it may not be expected to reach as low as otherwise expected. On the other hand, it would then be expected that 'co', being at the beginning of a new prosodic constituent, should remain unaffected by the preceding T4. Since this is not the case it may be better to view this drop as related to declination, thereby providing further evidence (in addition to Table 5) that T4 may indeed be subject to this form of downtrend.


5. Discussion

To summarize, with the exception of the observed local lowering effect, T1 shows no convincing evidence of downtrend of any sort. With respect to T2 and T3, there is evidence to support either or possibly both of declination or final lowering. The difficulty in interpreting this lies largely in the different results obtained for the two speakers, with NM's results being more suggestive of final lowering. However, the possibility of attributing the difference to variable rates of decline in a declination component has not been ruled out. The similarity in overall decline for the two speakers as expressed in semitones lends support to this interpretation. Regardless of which interpretation is accepted, the effect is slight; if declination, it is at a considerably lower rate than reported elsewhere, and if final lowering, it starts later and falls less than what is observed here for T4 or is reported elsewhere in the literature. More important, the effect is that there is little or no overlap of the respective tonal space of the two tones. There is a similar ambiguity concerning results for T4, though in this case it is a result not of disagreement between the two speakers, but of their different responses to different sentence types. The most reasonable interpretation, i.e. that based on the like-tone utterances, is that there is final lowering but no declination for T4.

Further research is expected to resolve these ambiguities. Despite these unanswered questions, these findings appear to parallel in a general sense what has been reported for at least some other languages with relatively complex tonal systems, e.g. Yoruba.

The examination of mixed tone sequences suggests that automatic downstep, also apparently plays no role in Mambila tone realization. Rather, the main instance of downtrend in these sentences was a local interaction of tones, especially between T1 and T4, where T4 had the effect of temporarily lowering the immediately following T1, or possibly two succeeding T1s. Nonetheless, since our dataset did not include longer sequences of alternating T1 and T4 (i.e. T1 T4 T1 T4 T1 T4 T1), the possibility of automatic downstep as an accumulation of local effects cannot be entirely ruled out at this stage. A second and more problematic finding was the tendency in some sentences for T4 to exhibit possible declination, contrary to what was seen for the like-tone sequences.

Mambila appears to be a discrete-level language, in that any impingement of one tone in another's space is rapidly corrected. In some respects, therefore, tone realization in Mambila seems relatively straightforward. In particular, we seem not to be faced with the question of how to model the phonetic effect of downstep carrying on through subsequent tones, e.g. whether there is a lowering or a narrowing of the register. If further study shows automatic downstep does exist, it can be attributed to local phonetic effects. The reason for this relatively uncomplicated aspect of Mambila tone may seem obvious; it is an account based in the phonology of Mambila, and on the same the same track as that used to explain the typological observation referred to above: with the register being relatively crowded, having four level tones and seven combinations of these at the lexical level, and with these being potentially subject to modification at the grammatical level, there is accordingly less room for maneuver on other planes. There is, in other words, a trade-off between complexity and simplicity at different levels of the linguistic structure; the only phonetic perturbations the phonology permits are sufficiently small that they do not endanger phonological contrast.

The conclusions here are preliminary, especially given that the work is based on only two speakers. If correct, languages such as Mambila may have little to contribute to the downstep/register debate. Nevertheless, the tonal register in Mambila is still subject to fluctuation which needs to be understood. It is noteworthy for example that the overall variation observed, particularly for JE, e.g. as reflected in the SDs in Table One and those mixed tone sentences involving T1 and T4, suggest there is greater flexibility for T1 and T2 than for T3 or T4; the tonal space of the top tones can be manipulated, without the lower tones varying by a corresponding amount. This may provide evidence that in Mambila one form of register manipulation consists of narrowing or expanding, as opposed to lowering or raising, the register. Further research involving different speaking modes (questions vs statements, planned comparisons utilizing variations in speech rate and loudness) will shed further light on this aspect of Mambila pitch realization. In addition, given the evidence for local effects, research to determine the extend of tonal coarticulation needs to be undertaken.


Bird, S. & Stegen, O. (1993) Principles of F0 measurement. Poster paper, Fourth Conference on Laboratory Phonology. Oxford.

Bird, S., & Stegen, O. (1995). The Bamileke Dschang associative construction: instrumental findings (EUCCS/RP-66). Centre for Cognitive Science, University of Edinburgh.

Clements, N. G. (1981) Downdrift in a tone language with four tone levels: a report on some preliminary measurements. Unpublished ms. (Published in York Papers in Linguistics, Vol. 15 1991)

Clements, G. N. (1990) The status of register in intonation: comments on the paper by Ladd and by Inkelas and Leben. In J. Kingston & M. Beckman (Ed.), Papers in Laboratory Phonology I: Between the grammar and physics of speech (pp. 58-71). Cambridge: Cambridge University Press.

Connell, B. A. & Ladd, D. R. (1990) Aspects of Pitch Realization in Yoruba. Phonology, 7 1, 1-29.

Gandour, J., Potisuk, S. & Dechongkit, S. (1994) Tonal coarticulation in Thai. Journal of Phonetics, 22 4, 477-492.

Gussenhoven, C. & Rietveld, A. (1988) Fundamental frequency in Dutch: Testing three hypotheses. Journal of Phonetics, 16 355-369.

't Hart, J. (1979) Explorations in automatic stylization of F0 curves. IPO Annuak Progress Report, 14 61-65.

Hombert, J.-M. (1974) Universals of Downdrift: their phonetic basis and significance for a theory of tone. Studies in African Linguistics, Suppl. 5 169-183.

Hombert, J.-M. (1977) Consonant types, vowel height, and tone in Yoruba. Studies in African Linguistics, 8 2, 173-190.

Hombert, J.-M. (1978) Consonant types, vowel quality, and tone. In F. V. (Ed.), Tone: a linguistic survey (pp. 77-111). New York: Academic Press.

Ladd, D. R. (1984) Declination: a review and some issues. Phonology Yearbook, 1 53-74.

Ladd, D. R. (1990) Metrical representation of pitch register. In J. Kingston & M. Beckman (Ed.), Papers in Laboratory Phonology I: Between the grammar and physics of speech Cambridge: Cambridge University Press.

Ladd, D. R. (1992) An introduction to intonational phonology. In G. J. Docherty & D. R. Ladd (Ed.), Papers in Laboratory Phonology II: Gesture, segment, prosody. Cambridge: Cambridge University Press.

Ladd, D. R. & Johnson, C. (1987) 'Metrical' factors in the scaling of sentence-initial accent peaks. Phonetica, 44 238-245.

Laniran, Y. (1992a) Intonation in a tone language: the phonetic implentation of tone in Yoruba. Ph.D., Cornell,

Laniran, Y. (1992b) Phonetic aspects of tone realization in Igbo. Progress Reports from Oxford Phonetics, 5 35-51.

Laver, J. (1994) Principles of Phonetics. Cambridge: Cambridge University Press.

Leben, W. R. (1996) Prosodic domains and the theory of tone. Paper presented to the 27th ACAL, Gainesville, FL.

Liberman, M. & Pierrehumbert, J. (1984) Intonational invariance under changes in pitch range and length. In M. Aronoff & R. Oehrle (Ed.), Language and Sound Structure (pp. 157-233). Cambridge MA: MIT.

Liberman, M., Schultz, J. M., Hong, S. & Okeke, V. (1993) The phonetic interpretation of tone in Igbo. Phonetica, 50 147-160.

Lindau, M. (1986) Testing a model of intonation in a tone language. Journal of the Acoustical Society of America, 80 757-764.

Ohala, J. J. (1978) The production of tone. In V. Fromkin (Ed.), Tone: a linguistic survey (pp. 5-39). New York: Academic Press.

Perrin, M. J. (1974) Mambila. In J. Bendor-Samuel (Ed.), Ten Nigerian Tone Systems (pp. 93-108). Jos: Institute of Linguistics.

Perrin, M. (n.d.) The tone system in Mambila: some further comments.

Pierrehumbert, J. & Beckman, M. (1988) Japanese Tone Structure. Cambridge MA: MIT Press.

Snider, K. & van der Hulst, H. (1993) Issues in the representation of tonal register. In H. van der Hulst & K. Snider (Ed.), The Phonlogy of Tone (pp. 1-27). Berlin: Mouton de Gruyter.

Stewart, J. M. (1983) Downstep and Floating Low Tones in Adioukrou. Journal of African Languages and Linguistics, 5 57-78.

Welmers, W. (1973) African Language Structures. Berkeley: University of California Press.

Zee, E. (1978) The interaction of tone and vowel quality. UCLA Working Papers in Phonetics, 41 53-67.