prosody (tone, tonosyntax, intonation, tonation)                                back to home page

Research on West African languages shows that we need to distinguish tones, tonosyntax (and tonomorphology), intonation, and a hybrid between intonation and tone that I call tonation. A basic synthesis of the issues follows. Some relevant articles are listed at the end of this section. In many cases it is difficult to isolate specific prosodic issues from the fuller grammar of a language. Therefore most of the material on pitch accent, tone, and tonation is in my fieldwork-based grammars of individual languages; see fieldwork and documentation.

Of the languages that I have worked on in Mali and adjacent sub-Saharan African countries, a few languages are nontonal, one has pitch accent, several have two tone levels H[igh] and L[ow], and still others have three tone levels H, L, and M[id].

nontonal:             Arabic (including Hassaniya), Fulfulde, riverine Songhay (Timbuktu, Gao)
pitch accent:       Tamashek
2 tone levels:      other Songhay languages, all Dogon languages
3 tone levels:      Bangime, cliffs Jenaama Bozo, Jalkunan, Tiefo-N, Tiefo-D

pitch accent

In Tamashek, default accent is on the antepenult of words and certain tightly-knit phrases such as preverb plus verb or preposition plus noun. A lexical (for nouns) or grammatical (for verbs) accent may occur on penults and final syllables. Only one accent surfaces on a word (or tightly-knit phrase).

tone (melodies, overlays, sandhi)

Most of the 2- and 3-level languages have contour tones on terminal syllables, e.g. <LH> rising and <HL> falling, less often <LHL> (bell-shaped). Grammatically expected <HLH> is usually absent, or if present then flattened to H. I use angled brackets for contour tones on individual syllables. In tonal languages, a bisyllabic CvCv stem may be Cv̀Cv́ (L.H), Cv̀Cv̀ (L.L), Cv́Cv̀ (H.L), or Cv́Cv́ (H.H). Some languages also allow Cv́Cv̂ (H.<HL>). Terminal heavy syllables CvL and Cvv more freely allow rising or falling contours.

Tones may be lexical or grammatical at word level. Most of the relevant languages have lexical tone melodies that are audible in some grammatical contexts. For example, lexical Cv̀, Cv̀Cv̀, and Cv̀Cv̀Cv̀ can be assigned to /L/ melody, which may be distributed over multiple syllables. Cv̂:, Cv́Cv̀, and Cv́Cv́Cv̀ (or in some languages Cv́Cv̀Cv̀) may be assigned to /HL/ melody. In other grammatical contexts, lexical melodies are modified or completely replaced by grammatically conditioned tones. In some western Dogon languages, verbs have no lexical tones, and all actually occurring verb forms have grammatically specified tone overlays.

Some word-level tonal alternations are really due to tone sandhi, typically at morpheme or word boundaries. Productive tone sandhi is phonological, rather than tonomorphological. Tone sandhi generally consists of assimilation, spreading, dissimilation, polarization, and displacement (shift). “Spreading” and “assimilation” are essentially the same thing. Examples (# = boundary, X = atonal, i.e. no prespecified lexical tone) are the following, for 2-level systems:

assimilation/spreading:        H.L.L → H.H.L (internal) and H#L.L → H#H.L (at boundary)
dissimilation:                               L.L#L → L.H#L 
polarization:                                 suffix #X in H#L and L#H
displacement:                               L.H#L.L → L.L#H.L

In 3-level systems, partial as well as full assimilation/dissimilation is possible, e.g. L.L#H → L.M#H. Likewise, additional forms of polarization are possible, e.g. M#H → L#H.

Two processes are intermediate between tone sandhi and tonomorphology or tonosyntax: floating tones and terrace formation.

floating tones and hosts

Many African languages have morphemes (affixes, clitics) that consist only of a tone. They normally arise when an originally syllabic morpheme is attritted, i.e. its segments (vowels and consonants) have disappeared, but the original tone survives. A floating tone morpheme produced in this way is realized, if at all, on an adjacent syllable in a host word. For example, in Ben Tey (Dogon), 1Sg possessor is a floating L-tone prefixed (or procliticized) at the left edge of the possessum. For example, bɛ̂yⁿ ‘beard’ has /HL/ lexical melody, but combines with 1Sg possessor L+ as bɛ᷈yⁿ ‘my beard’ with bell-shaped <LHL> tone. In cases like this, whether the best analysis is phonological (with a floating tone morpheme that attaches to a host by tone sandhi), or morphophonological (in this case, with a rule that the initial syllable of the possessum is tonally modified in a particular way), is arguably a judgement call. Phonological models with underlying tonal morphemes that trigger tone sandhi have been popular in the past, but their psychological reality is questionable.

terrace formation

The second borderline issue is terrace formation, whereby morphemes or words are fused tonally into a a tonally flat sequence. This is particularly evident in Bangime, where M-toned elements in some (but not all) combinations spread their M-tone rightward, sometimes to the end of a following word. There are also some cases of L-terraces and H-terraces. The issue is confounded by terminology, since terracing and the verb to terrace have sometimes been used in the literature to describe arrays of multiple terraced chunks, each one lower-pitched than the one to its left (by downdrift), forming a descending arpeggio. However, the noun terrace has a dictionary meaning “a level paved area or platform next to a building; a patio or veranda,” and there is no reason why this noun, as in “M-terrace,” should not be used to denote a single pitch-flattened chunk, in the absence of any suitable synonym. Terrace formation is intermediate between tone and either tonomorphology (stem plus affixes) or tonosyntax (multi-word sequences), since at least in Bangime it it occurs only in specific morphosyntactic combinations.

Terraces in Bangime are themselves subject to tone sandhi processes. In one situation, the entire M- or H-terrace is treated just like any multisyllabic M- or H-toned word in assimilations and dissimilations at boundaries with following words. For example, in Bangime a prosodically heavy H-toned sequence drops its final tone to M (or downstepped H) before a pause. Cv́Cv́ is prosodically light, so by itself it doesn’t drop prepausally. However, when a Cv́Cv́ word is joined to a preceding word in an H-terrace, e.g. Cv́Cv́#Cv́Cv́, the combination is prosodically heavy, and it is realized as Cv́Cv́#Cv́Cv̄ with final M-tone before a pause. Similarly, an L-terrace is subject to boundary tone sandhi in Bangime when followed by an L- or M-toned syllable, just as any multisyllabic L-toned word does: simple L.L#L.L → L.M#L.L, and with bracketed L-terraced chunk [L.L#L.L]#L.L → [L.L#L.M]#L.L. Note that the bracket-internal L.L#L.L does not undergo this process and become *L.M#L.L, since tone sandhi assimilations and dissimilations cannot apply to nonfinal syllables within a terrace.

The second type of interaction of tone sandhi with terrace formation is when an entire Bangime M-terrace undergoes polarization of the M#H → L#H type, as when underlying M proclitic combines with a following word as M-terraced M-M.M, then this M-terrace as a whole is dropped to L before an H-tone, as in [L#L.L]#H.

tonomorphology and tonal ablaut

Tonomorphology denotes the set of word-internal tonal processes that are clearly triggered by grammatical categories, rather than by productive tone sandhi. In Dogon languages, inflectional (mainly aspect and negation) suffixes control tonal modifications on the verb. The lexical melody of the verb stem is either partially modified, or completely replaced by a suffix-controlled overlay. Especially when the modification is partial, and limited to the left and/or right edges of the stem, one could try floating-tone analyses. For example, H.H stem plus (L+)X suffix → H.L#X. Processes of this type occur in Dogon verbal morphology. In Songhay, verbal derivational suffixes impose flat overlays {H} or {L} on the entire stem, erasing lexical melodies.

Still more challenging are cases where tonal ablaut occurs in the absence of affixes. The best developed system is again that of Bangime. For example, each noun has two tonal forms, one occurring in the absence of a preceding determiner (possessor or definite), the other occurring after a determiner. The former reflects the lexical melody: /L/, /H/, /M/, /LH/, /HL/, etc. The post-determiner form is generated from the lexical melody as follows. The leftmost tone of the lexical  melody is inverted to a polar tone (L → H, both H and M → L) and this tone spreads to the end of the noun. For example, both lexical Cv̀Cv́ and Cv̀Cv̀ become Cv́Cv́ after a determiner. A similar inversion between M and L occurs in verbal morphology. An important topic in Bangime grammar is whether the inversions in nouns and in verbs are two manifestations of a single deep semantic principle.

One would expect that tonal ablaut would precede terrace formation, which would precede tone sandhi. However, there are some puzzling cases in Bangime where unexpected rule orderings occur.

tonosyntax (controller, target, overlay)

I use tonosyntax to refer to phrasal patterns whereby one word or sub-phrase, the controller, imposes a stem-wide tonal overlay such as {L}, {HL}, {LH}, or {H} on another word or word-string, the target, inside the same phrase.  The term is especially appropriate when there is a general principle, rather than a set of idiosyncratic tonal effects of individual morphemes (which might be handled using floating-tone morphemes or affixal tonomorphology). Tonosyntax is most systematic in NPs in many Dogon languages. The overlay is indexed by superscripts on the target, on its left or right “pointing” toward the controller, which is normally adjacent. Typical atomic schemas are NL Adj and NL Dem (a modifying adjective or demonstrative controls {L} on the preceding noun), and Poss HLN (a preposed possessor controls {HL} on the possessum). That these cannot be reduced to simple category-independent phrasal prosody is shown by N Num (a noun and following numeral do not interact tonally). Contrast English, where seven apples and rotten apples have the same prosody. Furthermore, in some Dogon languages demonstratives (‘this/that’) are controllers while definites (‘the’) are not, even when the demonstratives and definites are identical in form.

It turns out that the distinction between controllers (modifying adjectives, demonstratives, possessors, and relative clauses) and non-controllers (numerals, definites, universal quantifiers, free plural morphemes, and discourse elements like ‘topic’ and ‘only’), can be reduced to a semantic distinction between reference restrictors (intersective modifiers in set-theoretic terms) and non-restrictors.

Dogon NPs of three or more words are of two tonal types. If there are no prenominal controllers, i.e. if the noun is followed by two or more modifiers, the rightmost controller controls its regular overlay on the noun and on any intervening words. For example, when N Num is followed by a demonstrative, the result is [N Num]L Dem, where the noun and numeral both occur in the target domain.

If there is both a preposed controller (a possessor) and a postposed controller, several outputs are possible. For example, the sequence Poss N Dem could, in theory, surface in a language with {HL} as possessor-controlled overlay, as a) Poss HL[N Dem], b) Poss HLN Dem, c) [Poss N]L Dem, or d) Poss NL Dem. The parameters are whether the possessor or the demonstrative “wins” and whether the target domain stops with the noun or continues to include the unsuccessful controller. In languages where these parameters are sufficient to account for all the data, an optimality theoretic model is workable.

However, there are also some Dogon languages with more complex tonosyntactic patterns, for example where a sequence like N Num Dem or N Num Def does not follow the expected tonosyntactic patterns. Instead of expected [N Num]L Dem and N Num Def, the outputs can be of the type [N Num]L+H LDem and [N Num]L+H LDef, where the determiner (if not already L-toned) drops to {L} and the preceding string is {L}-toned with a single final H-tone. Although such tonosyntax patterns probably originated as tone sandhi gone awry, they are synchronically constructional, i.e. they cannot be produced by successive application of the atomic tonosyntactic properties of numerals and determiners.

terminal intonation

I use this term to refer to optional and gradient (uncalibrated) modifications affecting the final syllable of a phrase or clause. A final low-pitch target without final-segment prolongation is usual in West African languages for indicatives that terminate a speaking turn or a paragraph-like segment in narrative. This can be interpreted as default intonation, favored by natural downdrift at the end of a prosodic unit.

Marked intonation involves a higher than modal terminal pitch level, and/or variable prolongation of the final segment (vowel or coda sonorant). These effects mark the final clause or phrase as unfinished, implying that more is to come, whether uttered by the same speaker or by the interlocutor. They can therefore signal that the speaker intends to hold the floor, or that a response is sought from the interlocutor.

Polar (yes/no) interrogatives do not always work this way in West African languages, as Annie Rialland has shown. Some Dogon languages, such as Donno So, have a polar interrogative construction characterized by terminal L-tone added to the final word of the indicative version of the clause, which is normally the verb. In Donno So this is actually tonal rather than intonational, as seen by its effects on other tones in the affected word. If the word elsewhere is Cv̀:Cv́, the added final interrogative L-tone pushes the input tones leftward, resulting in forms like Cv̌:Cv̀. Donno So also has another polar interrogative construction with final ma (likely the same as ma ‘or’), with variable pitch and often extended prolongation as ma→. Donno So therefore has both of the interrogative patterns whose distribution in West Africa has been studied by Rialland. However, the terminal L is tonal, not intonational, in its phonological effects. Furthermore, the construction with ma→ can, and probably should, be analysed as an incomplete disjunction, cf. Are you coming tomorrow, or… (aren’t you)? In this view, its terminal intonation indicates unfinished status.

tonation (lexical/grammatical prolongation, dying quails)

Tonation is my term for intonation-like phonetic effects, i.e. uncalibrated prolongation of the final vowel or sonorant with or without added pitch targets, that are lexical or grammatical in function, as opposed to marking incompleteness. The best examples are in certain Dogon languages.
Uncalibrated prolongation is lexically baked into a large subset of Dogon expressive adverbials (so-called “ideophones”). It is distinguishable from phonological length (or gemination), not only by the wide variability of its duration, but also by applying to word-final sonorants. For example, in Jamsay dém→ ‘straight’ prolongation applies to the m, not the vowel, in spite of the fact that geminate consonants are disallowed in Jamsay phonology.
If disjunctive/interrogative ma→ is attributed to true intonation, the remaining cases that I classify as tonation involve a combination of uncalibrated prolongation with a terminal low-pitch target. The slow pitch decline is audible when the phrase otherwise ends in an H-tone. I refer to this type of tonation as the “dying quail” effect (symbol  \). It is associated with certain constructions that normally combine two parallel phrases. In Jamsay and Togo Kan, dying quail is regular in NP conjunction (‘X and Y’). In Donno So and some other Dogon languages, it is associated with two-part willy-nilly conditional antecedents (‘whether it rains (or) it doesn’t rain, …’). In both cases there is a prosodic break between the two paired components.
It is interesting to see how the dying quail effect interacts with tonosyntax. The only context where this combination can occur is relative clauses with conjoined NP as (internal) head NP. Relative clauses are tonosyntactic controllers, which elsewhere drop tones of the head NP to {L}. If the internal head is a conjunction X\ Y\ where at least Y otherwise ends in an H-tone (so Y\ has a conspicuously falling pitch), there are several possibilities. One is to prioritize tonosyntax, casting an L-toned overlay over Y\ and perhaps (in spite of the prosodic break) also over X\ . Another is to prioritize tonation, so that X\ Y\ functions as a tonosyntactic island, impervious to external controllers. A third, essentially a compromise, is to apply the {L} tonosyntactic overlay only to the final word or syllable in Y\ .  The data show that the compromise pattern is most common for Jamsay, but prioritization of tonosyntax is typical for Togo Kan.

      *2016          Laura McPherson & J. Heath. Phrasal grammatical tone in the Dogon languages: The role of constraint interaction. Natural Language and Linguistic Theory 34(2):593–639.
      *2015          Dogon noncompositional constructional tonosyntax. Journal of African Languages and Linguistics 36(2):233–252.
      *2013          (J. Heath & Laura McPherson) Tonosyntax and reference restriction in Dogon NPs. Language 89(2):265-96.
                                    Project Muse:


[last update Nov 2017]

back to home page