Hostname: page-component-745bb68f8f-d8cs5 Total loading time: 0 Render date: 2025-02-12T00:50:08.779Z Has data issue: false hasContentIssue false
  • English
  • Français

Phonological phrasing and ATR vowel harmony in Akan*

Published online by Cambridge University Press:  03 August 2015

Frank Kügler
Frank Kügler*
Affiliation:
University of Potsdam
*
E-mail: kuegler@uni-potsdam.de.
Rights & Permissions [Opens in a new window]

Abstract

This paper examines phonological phrasing in the Kwa language Akan. Regressive [+ATR] vowel harmony between words (RVH) serves as a hitherto unreported diagnostic of phonological phrasing. In this paper I discuss VP-internal and NP-internal structures, as well as SVO(O) and serial verb constructions. RVH is a general process in Akan grammar, although it is blocked in certain contexts. The analysis of phonological phrasing relies on universal syntax–phonology mapping constraints whereby lexically headed syntactic phrases are mapped onto phonological phrases. Blocking contexts call for a domain-sensitive analysis of RVH assuming recursive prosodic structure which makes reference to maximal and non-maximal phonological phrases. It is proposed (i) that phonological phrase structure is isomorphic to syntactic structure in Akan, and (ii) that the process of RVH is blocked at the edge of a maximal phonological phrase; this is formulated in terms of a domain-sensitive CrispEdge constraint.

Type
Articles
Information
Phonology , Volume 32 , Issue 1 , May 2015 , pp. 177 - 204
Copyright
Copyright © Cambridge University Press 2015 

1 Introduction

This paper examines phonological phrasing in Akan, a Kwa language with two tones spoken in Ghana (Dolphyne & Kropp Dakubu Reference Dolphyne, Dakubu and Dakubu1988).Footnote 1 The general assumption in prosodic phonology is that the presence or absence of phonetic cues or phonological processes, for instance phrase-final lengthening, may provide a means of diagnosing the presence of prosodic phrasing (cf. Nespor & Vogel Reference Nespor and Vogel1986, Selkirk Reference Selkirk1986, Reference Selkirk, Goldsmith, Riggle and Yu2011). This paper explores a hitherto rarely described process of regressive vowel harmony (RVH) between prosodic words as a tool for determining the placement of phonological phrase (φ) boundaries. RVH has been observed in a few other languages (Casali Reference Casali2008), but has not yet received a great deal of attention. In this paper I argue that the edge of a maximal phonological phrase (φmax) blocks the general process of RVH. The data is analysed using Match theory (Selkirk Reference Selkirk, Goldsmith, Riggle and Yu2011) for the formation of prosodic phrase structure. Any lexically headed syntactic phrase is mapped onto a φ. I assume domain-specific CrispEdge constraints (Itô & Mester Reference Itô, Mester, Kager, van der Hulst and Zonneveld1999, Selkirk Reference Selkirk, Goldsmith, Riggle and Yu2011) that ban multiple linking of features across prosodic domain edges. In particular, I propose a domain-sensitive constraint of the CrispEdge family to account for the fact that [+ATR] cannot spread regressively across the edge of a φmax. This proposal assumes recursive phonological phrasing (cf. Selkirk Reference Selkirk, Goldsmith, Riggle and Yu2011, Elfner Reference Elfner2012, Ito & Mester Reference Ito, Mester, Borowsky, Kawahara, Shinya and Sugahara2012). A φmax is defined as a φ that it is not dominated by any further φ.

Vowel harmony is central to the analysis of phrasal phonology in Akan. The vowel system consists of nine oral and five nasal vowels, as shown in (1) (Dolphyne Reference Dolphyne1988). Oral vowels show a three-level height distinction, and a two-way front–back distinction. Only high and low vowels have nasal counterparts.Footnote 2

  1. (1)

As in many other West African languages, in particular Niger-Congo and Nilo-Saharan languages (cf. Casali Reference Casali2008), vowel harmony is a feature of the phonology of Akan (e.g. Stewart Reference Stewart1967, Dolphyne Reference Dolphyne1988, Casali Reference Casali2012). The harmonising feature is [ATR], and the vowels form two harmonic sets, as shown in (2). The advanced vowel series consists of the high and mid vowels, which have retracted counterparts. In addition, the low vowel /a/ belongs to the [—ATR] group.Footnote 3 According to Dolphyne (Reference Dolphyne1988), vowel harmony applies within the word, and word stems are underlyingly associated with either [+ATR] or [—ATR].

  1. (2)

Akan harmonic vowel classes have been the object of articulatory and acoustic studies, with the aim of understanding the exact underlying mechanisms of the [±ATR] distinction (Lindau Reference Lindau1975, Reference Lindau1979, Hess Reference Hess1992, Tiede Reference Tiede1996). Articulatorily, the ‘relative size of the pharynx controls the phonological vowel harmony’ (Lindau Reference Lindau1975: 80), which was confirmed using ultrasound measures by Tiede (Reference Tiede1996). The articulatory studies showed that there is almost no difference in degree of primary constriction, which means that the tongue dorsum maintains its height in the oral cavity, with the crucial difference arising through a decrease in the size of the pharyngeal cavity in the case of [—ATR] vowels.

Acoustically, an increase in the size of the pharyngeal cavity results in a raising of the first formant, confirmed independently for Akan (Lindau Reference Lindau1979, Hess Reference Hess1992, Tiede Reference Tiede1996). The F1 dimension mirrors the degree of tongue height, which may be why the features [ATR] and [tense] are often used interchangeably in phonological theory. Given that Lindau (Reference Lindau1975) and Tiede (Reference Tiede1996) have shown articulatorily that in Akan the vowels differ in terms of pharyngeal size, Akan unambiguously exhibits a case of [ATR] harmony.

The paper is organised as follows. Word-level vowel harmony in Akan is given as background in §2, and data illustrating RVH between words is presented in §3, as well as a first analysis in terms of a prosodic interword markedness constraint that licenses RVH and a word-level CrispEdge constraint that prohibits multiple linking of the [ATR] feature between prosodic words. §4 sets out the basic phrase-level analysis of RVH, and presents the empirical puzzle that RVH occurs in certain contexts while it is blocked in others. I propose an analysis in terms of syntax–phonology Match constraints to arrive at the relevant prosodic phrasing, such that each lexically headed syntactic phrase is matched with a φ. In addition, I propose an analysis of recursive prosodic structure to distinguish between different levels of φ's, φmax and φnon&max, which are shown to be relevant for the application or blocking of RVH. The remainder of §4 discusses data where RVH applies or is blocked, to support the central analysis, while §5 analyses RVH as a domain-sensitive process. I argue that a domain-sensitive CrispEdge constraint referring to φmax is required to account for the blocking of RVH. §6 summarises the paper.

2 ATR harmony within the word

The goal of this section is to illustrate bidirectional root-controlled word-level harmony in Akan, and to show that, from a point of underspecification theory, the [ATR] opposition is represented underlyingly. I propose that the vowel-harmony process can be accounted for by the interaction of a markedness constraint Harmony and a positional faithfulness constraint Ident[ATR]Root.Footnote 4

Vowel harmony in Akan has been the subject of a number of phonological studies (Stewart Reference Stewart1967, Reference Stewart1983, Schachter & Fromkin Reference Schachter and Fromkin1968, Clements Reference Clements and Goyvaerts1985). According to Dolphyne (Reference Dolphyne1988: 16), vowel harmony is ‘a property of the word, that is, it characterises a whole word at a time’. That is, all syllables within a word agree with the [ATR] feature of the word stem, as shown by the data in (3).

  1. (3)

In the verbal morphology of Akan, tense, aspect and mood markers, as well as pronominal suffixes, are cliticised to the verb stem (Paster Reference Paster2010). The [ATR] feature spreads leftwards from the stem through the whole word, including all affixes. (3) illustrates the agreement of [ATR] across two prefixes, the subject pronoun and a tense marker (perfective in (a, b), and future in (c, d)). In (3a), the [+ATR] feature of the word stem /di/ affects both the perfective tense, which is realised as vowel lengthening, and the 1st person singular pronoun. The same holds for the [—ATR] feature of the word stem /tɔ/ in (b). The future tense in (c, d) is realised with the prefix /bE/, which does not merge with the preceding pronoun.Footnote 5 Like other prefixes, the future tense marker varies as a function of the [ATR] specification of the verb stem, as in (c) ([+ATR]) and (d) ([—ATR]).

As well as spreading across all affixes, the harmony process extends from the stem vowel(s) bidirectionally over the whole word (Stewart Reference Stewart1965, Schachter & Fromkin Reference Schachter and Fromkin1968, Dolphyne Reference Dolphyne1988). In the case of the completive aspect, which expresses past tense (cf. Osam Reference Osam, Beermann and Hellan2003), the aspect marker is realised as a suffix. Relevant examples from Dolphyne (Reference Dolphyne1988: 16) are given in (4). The [+ATR] of the verb stem /hunu/ in (a) spreads leftwards to the subject pronoun and also rightwards to the completive aspect marker. Spreading of [—ATR] is shown in (b).Footnote 6

  1. (4)

Root-controlled word-level vowel harmony in Akan can be accounted for by the interaction of the markedness and positional faithfulness constraints in (5). The markedness constraint in (a) ensures that all vowels in a given word agree in their [ATR] feature. In contrast, the faithfulness constraint in (b) prevents a word stem that is specified for [ATR] from changing its feature value.

  1. (5)

The tableau in (6) illustrates the constraint interaction with a disharmonic verb root /bisa/ ‘to ask’.Footnote 7 The input consists of the verb stem, a 3rd person singular subject pronoun /O/ and a completive aspect marker /I/. The optimal candidate, (a), satisfies the positional faithfulness constraint, but violates the markedness constraint, since the disharmonic word stem shows an instance of two adjacent vowels with different values for the harmonic feature. Note that completely harmonising forms such as (3) and (4) do not show any violations of the two constraints. If the word stem harmonises either to [+ATR], as in (b), or to [—ATR], as in (c), positional faithfulness requiring root identity is violated. In both cases, the markedness constraint also incurs one violation, since the word stem and one of the affixes differ in their [ATR] values. If, on the other hand, the word stem does not change its [ATR] specification, obeying the positional faithfulness constraint, but has only [—ATR] affixes, the markedness constraint is violated twice, as shown for candidate (d); there are two instances of two adjacent vowels with a change of the harmonic feature within the word. Similarly, (e) illustrates that, with two [+ATR] affixes, Harmony is violated twice.

  1. (6)

Such disharmonic stems (or words) illustrate that the faithfulness constraint is ranked above the markedness constraint, which, as shown in (7), is ranked higher than a general faithfulness constraint requiring input–output identity for the feature [ATR], which plays no role here.Footnote 8

  1. (7)

According to Casali (Reference Casali2003, Reference Casali2008), a root-controlled vowel-harmony language like Akan has [+ATR]-dominant harmony. Criteria for the [+ATR] feature being dominant involve a [+ATR] allophonic variant of /a/, as in (3) and (8) (note that for independent reasons, /ɔ/ lowers to /a/ in (8)), spreading of [+ATR] within a lexical compound (cf. (8)), spreading from a [+ATR] affix to a [—ATR] root morpheme and spreading of [+ATR] across word boundaries (cf. §3) (Casali Reference Casali2003: 321). Except for [+ATR]-affix spreading, these criteria have also been observed in Akan (Dolphyne Reference Dolphyne1988).Footnote 9

  1. (8)

3 ATR harmony between words

According to Dolphyne (Reference Dolphyne1988: 24), vowel harmony occurs also between words, as a consequence of an assimilation process. The structural prerequisite of this process is that two adjacent words differ in their [ATR] specification; more specifically, a word with [—ATR] vowels is followed by a word with [+ATR] vowels. The general rule is that, as the dominant feature, [+ATR] spreads regressively across the word boundary to the immediately preceding syllable, and delinks and reassociates the [ATR] specification of that vowel. The process is limited to the final syllable of the [—ATR] word, as shown in (9), and RVH does not extend beyond that syllable. The effect of RVH is thus that, on the surface, polysyllabic prosodic words may be disharmonic.

In (9a) and (b) (from Dolphyne Reference Dolphyne1988: 24), the [—ATR] unrounded mid front vowel of the verb stems /pε/ and /frε/ becomes [+ATR], since the following object noun has [+ATR] vowels. This change of vowel feature is henceforth indicated by underlining. Acoustic analysis of vowel formants by Hess (Reference Hess1992) confirms Dolphyne's observation that the effect of vowel harmony across word boundaries is regressive, and that it affects only the last vowel of the preceding word. The example in (9c) illustrates that only the last vowel of a preceding disyllabic stem /tɕɪrε/ is affected. Hence, independent of morphological complexity and stem complexity, [ATR] harmony between words is bounded, and affects only the immediately preceding adjacent vowel.

  1. (9)

The basic pattern of RVH is illustrated in (10).

  1. (10)

To account for RVH, I propose the prosodic markedness constraint in (11). The constraint drives the interword harmony process, penalising a configuration in which a sequence of [—ATR][+ATR] words does not show regressive association of the [+ATR] feature to the left-adjacent syllable. RVH occurs frequently in Akan, as subsequent sections will show, and I assume that it is a general phonological process in the grammar. The effect of the interword harmony process is that the change in the last vowel of a preceding word violates word-level harmony in that word.Footnote 10

  1. (11)

This markedness constraint is similar in its effects to the constraint proposed by Copperbelt Bemba by Kula & Bickmore (Reference Kula and Bickmore2015). In Interword Doubling in Copperbelt Bemba, a H tone on a word-final TBU of one word spreads onto the first TBU of the following word. This process differs from Akan in both the direction of the feature spread and the type of autosegment that spreads: [+ATR] spreads regressively in Akan, while a H tone spreads progressively in Copperbelt Bemba. In both languages, crucially, an autosegment spreads across a word edge.Footnote 11

To account for the fact that [+ATR] does not spread in an unbounded fashion throughout the whole [—ATR] word, I assume the markedness constraint in (12), which penalises the occurrence of the feature [+ATR], similar to Kula & Bickmore's constraint which penalises the occurrence of H tones in Copperbelt Bemba.

  1. (12)

The fact that [+ATR] is dominant in Akan (Casali Reference Casali2008, Reference Casali2012) may speak in favour of an analysis where the constraint in (12) is ranked higher than a corresponding constraint penalising the occurrence of [—ATR]. The effect of the constraint in (11) in combination with *[+ATR] in (12) limits the spreading of [+ATR] to the immediately left-adjacent syllable.

The fact that *[+ATR] penalises all occurrences of advanced vowels may seem problematic at first sight, as pointed out by a reviewer. Indeed, *[+ATR] and Harmony need to be dominated by Ident[ATR]Root, which ensures root identity with respect to the [ATR] feature. This ranking does not affect the basic word-level harmony pattern. Assume a [+ATR] word, e.g. /kube/ ‘coconut’, which surfaces as [kube]. Although *[+ATR] would be violated twice, any candidate with a [—ATR] vowel would fatally violate the higher-ranked positional faithfulness constraint, since the [ATR] specification of a root vowel is no longer identical to its input. The fact that *[+ATR] also stops interword harmony from iterating through the next word is in line with the Emergent Non-iterativity Hypothesis of Kaplan (Reference Kaplan2008). In his discussion of RVH in Nez Perce (2008: 284ff), however, Kaplan employs a Non-finality constraint, which clearly cannot account for the Akan data: in Nez Perce, both [—ATR] and [+ATR] optionally spread regressively across the word boundary, and Non-finality refers to the presence of prosodic heads which are required not to be associated with the final syllable. This requires both values of the [ATR] feature to spread, which is not the case in Akan; therefore the lack of [—ATR] spreading must be accounted for. Thus, for the time being, I will assume that *[+ATR] prevents iteration of RVH.

The tableau in (13) illustrates the violation of the lower-ranked word-level harmony constraints if RVH applies (candidates (a, c)). The application of RVH thus produces a disharmonic word. If RVH does not apply, as in (b), *Interword[—ATR][+ATR] is fatally violated. The fact that the markedness constraint *[+ATR] is violated more often when the regressive spreading of [+ATR] affects more than the immediately left-adjacent vowel accounts for the bounded nature of RVH (candidate (c)); the more [+ATR] vowels arise through RVH, the more violations of *[+ATR].

  1. (13)

The resulting constraint ranking is given in (14). Since RVH is a general phonological process in Akan, the interword markedness constraint licensing this process must necessarily be higher-ranked than the word-level harmony constraints.

  1. (14)

According to Dolphyne (Reference Dolphyne1988), only [+ATR] spreads regressively (cf. Casali Reference Casali2003). Hence, in (15) the [—ATR] of the object /εmʊ/ does not cause the [+ATR] of the verb stem /di/ to change into [—ATR]. Dolphyne (Reference Dolphyne1988: 24) also notes that vowel harmony between words can only be regressive, as in (9). The context of (15) is identical to (9), where RVH occurs between the verb and object.Footnote 12

  1. (15)

The form in (9c) showed that the process of RVH is bounded in Akan. There are other languages that exhibit the phonological process of RVH; for most of them, however, its distribution remains unclear, because of a lack of data.Footnote 13 One exception is Nawuri, a Kwa language spoken in Eastern Ghana (Casali Reference Casali2002: 25ff). As opposed to Akan, Nawuri shows unbounded RVH across word boundaries. In the Nawuri example in (16) (Casali Reference Casali2002: 25), all the underlying [—ATR] syllables of the verb become [+ATR]. In addition, due to other phonological processes, the stem-final /ɪ/ is deleted, with compensatory lengthening of the nominal class marker /ɑ/. Note also that word-initially the low vowel /ɑ/ has a [+ATR] allophone.

  1. (16)

Spreading of an autosegment, whether unbounded or bounded, is not restricted to the feature [ATR], but has also been shown for H tones (cf. Hyman Reference Hyman, Riad and Gussenhoven2007 for an overview). The Bantu languages Ekegusii (Bickmore Reference Bickmore1999) and Copperbelt Bemba (Kula & Bickmore Reference Kula and Bickmore2015), for instance, exhibit both bounded and unbounded H-tone spreading.Footnote 14

The examples in Dolphyne (Reference Dolphyne1988: 24) (cf. (9)) show that regressive [ATR] harmony in Akan affects the last syllable of the verb. The following sections will examine further cases of RVH, and illustrate its distribution with respect to the phrasal organisation of the sentence. Further structures show regressive [ATR] harmony between words, and illustrate that this process occurs quite generally in Akan. The central issue is that, despite its broad distribution, in Akan, the RVH process is blocked at the edges of what I will argue are φmax's.

4 Distribution of regressive vowel harmony in the sentence

4.1 Regressive vowel harmony across word boundaries

In simple SVO sentences, RVH occurs within the VP, as in (17) below, but does not extend from a verb that contains [+ATR] vowels to a preceding subject that contains [—ATR] vowels, as in (18). The data in these two examples show transitive sentences with a subject DP, a verb and an object.

In (17a, b), the structural requirements for the process of RVH are met by the [+ATR] vowels of the object and [—ATR] vowels of the preceding verb. Hence the specification of the monosyllabic verb changes to [+ATR], as shown in (9). (17a) also shows that RVH does not apply iteratively from the changed specification of the verb to the preceding subject, which contains [—ATR] vowels. The fact that RVH fails to iterate from the verb to the subject in (17a) is due to its limitation to just a single syllable; also, as the data in (18) show, [ATR] does not spread from the verb to the preceding subject. Comparing (17b) with (c) shows that [ATR] harmony does not spread progressively from a subject with [+ATR] vowels to a verb with [—ATR] vowels (cf. (15)). The application of RVH is indicated by ‘←’.

  1. (17)

Although the examples in (18) contain a [—ATR][+ATR] word sequence, the final vowel of the subject DP does not change its vowel quality. Unlike (17), the verb in (18) has an [+ATR] feature. The blocking of RVH even though its structural context is met is indicated by ‘↚’.

  1. (18)

If vowel harmony within a sentence is a general process (cf. Dolphyne Reference Dolphyne1988), we would expect that the final [—ATR] vowel of the subjects in (18) should become [+ATR]. However, RVH is blocked between an NP and a VP.Footnote 15

RVH, however, does takes place between NPs, in particular VP-internally in the case of ditransitive verbs, as in (19). The direct objects contain [+ATR] vowels, while the preceding indirect objects contain [—ATR] vowels. Vowel harmony spreads regressively from the direct object to the indirect object, turning the final vowel of the indirect object into [+ATR].

  1. (19)

To account for the difference between the application of RVH between NPs and the blocking of RVH between NP and VP, I will assume a difference in prosodic phrase structure that arises through the matching of syntactic structure to prosodic structure according to Match theory (Selkirk Reference Selkirk2009, Reference Selkirk, Goldsmith, Riggle and Yu2011). A subject NP and following VP would be distinct φmax's, not dominated by any further φ, whereas an indirect object and following direct object NP would each be φ's, embedded in a higher φ. The basic assumption of Match theory is that a universal set of faithfulness constraints maps syntactic constituency into prosodic constituency. The relevant constituents are the syntactic domains of word, phrase and clause, which are mapped into the corresponding prosodic domains of prosodic word (ω), phonological phrase (φ) and intonational phrase (ɩ). Given this mapping, Match theory predicts systematic recursion on all prosodic levels that is isomorphic to recursive syntactic structure. For the purposes of RVH in Akan, we need to establish the facts for φ formation, leaving the constituent structure of the word/ω and clause/ɩ aside.

Selkirk (Reference Selkirk, Goldsmith, Riggle and Yu2011) defines two versions of Match constraints: a general one matching any syntactic constituency of type α into a phonological output constituent of type π, and a lexically specific one that restricts the relevant syntactic input to phrases that are lexically headed. The Akan data suggest that only lexically headed phrases are required to correspond to φ's. We will see below that functionally headed phrases above VP in the clause appear to be ignored in the generation of prosodic constituent structure. The distinction of lexical and functional projections and its relevance for the syntax–phonology interface has also been argued for in edge-based theories (Truckenbrodt Reference Truckenbrodt1999).

  1. (20)

Given the syntactic structure of a ditransitive sentence in (21a), the Match constraint in (20) produces two maximal phonological phrases in prosodic phrase structure, as in (21b). The first φmax is mapped from the subject DP, which appears in the specifier of TP (Saah Reference Saah1994: 26, Boadi Reference Boadi2005: 9, Marfo Reference Marfo2005: ch. 2), showing that a syntactic phrase boundary not dominated by any further lexical phrasal projection. The second φmax is mapped from the VP, which shows recursive phonological phrasing of non-φmax's mapped from the embedded object DPs. These two φmax's are sisters dominated by the ɩ.Footnote 16 Given the syntactic representation in (21a) (cf. Saah Reference Saah1994: 33), the MatchPhrase constraint in (20) provides phonological phrasing which corresponds to lexically headed syntactic phrases.

  1. (21)

Following Ito & Mester's (Reference Ito, Mester, Borowsky, Kawahara, Shinya and Sugahara2012) proposal on recursive prosodic structure, the boundary between the object DPs in (21) is that of a non-maximal constituent, since each NP is dominated by a further phrasal projection, the VP. The boundary between the subject DP and the VP is that of a maximal constituent, since neither the subject DP nor the VP is dominated by any further φ. A φnon&max is thus recursively embedded in prosodic structure, and prosodic structure reflects syntactic structure (Selkirk Reference Selkirk, Goldsmith, Riggle and Yu2011, Ito & Mester Reference Ito, Mester, Borowsky, Kawahara, Shinya and Sugahara2012). Recursion in prosodic structure refers to the dominance relations of a category κ, as formulated in (22).

  1. (22)

(22a) represents the largest projection of a prosodic category, and (22b) the smallest projection. This dominance view gives a three-way distinction between maximal, non-maximal (or non-minimal; cf. Elfner Reference Elfner2012), and minimal prosodic constituents. The distinction between φmax and φnon&max is relevant for RVH in Akan.

Given the different subtypes of φ in (21), the proposal of this paper is that the blocking contexts involve φmax edges, while phrase edges in the non-blocking contexts are non-maximal. Elfner (Reference Elfner2012) argues for a similar approach to recursion-based prosodic subcategories for Connemara Irish, which, however, makes reference to φnon-min. The prediction of the present proposal is that, given any prosodic phrasing produced by the Match constraint in (20), the presence of a φmax edge blocks the application of RVH. The following sections examine whether various syntactic structures and their prosodic phrasing with respect to the presence or absence of φmax support the prediction of the proposed analysis of RVH.

4.2 VP-internal regressive vowel harmony

The examples given in (9), (17), and (19) show the occurrence of the process of RVH within the VP. This section gives some more examples of VP-internal structures which support the analysis that, within the VP, no φmax should arise as a result of the application of the Match constraint in (20). Hence I predict that the process of RVH should generally apply within the VP.

The harmony process applies in sentences with ditransitive verbs between an object and a verb (23a), between a direct object NP and an indirect object NP containing a modifier (23b), between a modified direct object NP and an indirect object NP (23c) and between two modified object NPs (23d, e).

  1. (23)

Assuming the syntactic structure and the corresponding prosodic structure in (21), the prediction that RVH applies within the VP is borne out. An embedded object DP, whether modified or not, forms a syntactic phrase, satisfying the Match constraint. Note that NPs are right-branching in Akan, i.e. the modifier follows its head noun (Saah Reference Saah1994, Boadi Reference Boadi2005, Aboh Reference Aboh, Aboh and Essegbey2010). The NP corresponds to a φ, which is non-maximal, given the domination by the φ corresponding to the VP. RVH applies, as it is not blocked by φmax edges. In (23a), the final [—ATR] vowel of the verb becomes [+ATR], due to RVH from the following object noun. In (23b), the [+ATR] feature of the direct object spreads regressively to the final vowel of the preceding adjective that modifies the indirect object. The fact that the direct object is modified by an adjective does not influence the general process of RVH. In (23c), RVH takes place between the objects, independent of the complexity of the direct object. The data in (23d, e) show that RVH also applies when both object NPs are modified; the [+ATR] feature spreads regressively from the direct object to the modifier of the indirect object. Furthermore, (23e) shows that RVH may occur more than once within the VP, though not iteratively. In addition to between-object harmony, the indirect object carrying [+ATR] vowels affects the final vowel of the verb, as in (17) and (23a).

(24) presents a case where RVH occurs between an adverb and a preceding object DP. The manner adverb contains [+ATR] vowels, and the preceding object noun [—ATR] vowels, which allows the [+ATR] feature to spread regressively. According to Saah (Reference Saah1994: 42), the manner adverb is ‘base-generated VP-internally’, resulting in a phrase break between the object DP and AdvP. As was the case for the two objects in ditransitive sentences, manner adverbials occur within the VP, as shown in (24b). By MatchPhrase(LexP, φ), the resulting prosodic structure contains φ's which are embedded in a higher φ, corresponding to the VP. Thus the φ's formed by both the object and the adjunct are non-maximal.

  1. (24)

4.3 NP-internal regressive vowel harmony

This section examines NP-internal RVH, and shows that, because there is no φmax edge within an NP, the analysis in §4.1 correctly predicts that RVH occurs between ω's within NPs. (25a) illustrates the structure of an NP in Akan, where <N> represents the trace of N which has been head-moved. Because a complex NP itself corresponds to a φ (25b), any φ internal to it is necessarily non-maximal. So RVH is predicted to be found within an NP.

  1. (25)

(26) illustrates that this prediction is confirmed; if a [—ATR] noun like /adamfʊ/ is followed by a [+ATR] modifier, RVH applies. The modifier is an adjective in (26a) and a quantifier in (26b).

  1. (26)

(27a, b) show that RVH applies within a complex NP embedded in a sentence, whether as a subject or object DP. As (27c) shows, RVH applies between a complex object and a preceding verb. This is because satisfying Match results in a recursive prosodic phrasing for the VP and its component phrases, with a resulting φnon&max status for the modified object, which is dominated by the φ of the VP.

  1. (27)

Postnominal modifiers within a complex NP in Akan follow a strict order, i.e. the adjective is closest to the head noun, followed by numerals, the determiner and finally quantifiers (Saah Reference Saah1994, Boadi Reference Boadi2005). DP-internally, the Match constraint produces embedded φ's, which are isomorphic to the corresponding syntactic structure. These φ's are thus non-maximal, and RVH is predicted to occur within the complex DP.

  1. (28)

As predicted, in the case of complex DPs that contain more than one modifier, RVH applies between the modifiers (29a–c), as well as between the first modifier and the head noun (29d) (cf. (26)).

  1. (29)

The word order of a possessive phrase (30) is possessor–possessee (Aboh Reference Aboh, Aboh and Essegbey2010). Note that the DP represents a φnon&max if the DP is embedded in a further φ, but a φmax if it forms its own non-embedded phrase, e.g. as a subject DP.

  1. (30)

In the case of a [+ATR] possessee, the final vowel of the possessor is affected by RVH, as shown in (31).Footnote 17

  1. (31)

4.4 Time adverbials

Time adverbials do not trigger RVH, although manner adverbials do show RVH, as in (24) above. According to Saah (Reference Saah1994: 37), ‘the time adverbial is a syntactic adjunct to TP’. It is not contained within the VP, and is moreover not dominated by any further lexical projection. This syntactic configuration results in separate φmax's corresponding to VP and AdvP, by MatchPhrase(LexP, φ), and in the prediction that RVH will be blocked between the time adverbial and the VP.

  1. (32)

Consider the example in (33), with a time adverbial following an SVO sequence. The adjacent object and adverbial differ in their [ATR] specification, meeting the criteria for RVH. However, as predicted by the analysis of recursive prosodic phrasing, the [+ATR] feature of the adjunct does not spread regressively to the object, and hence the final vowel of the object does not become the advanced vowel [e]. The time adverbial is not embedded, and thus forms an independent φmax.Footnote 18

  1. (33)

4.5 Serial verb constructions

Serialisation in Akan is manifested in a multiverb construction, which according to Schachter ‘consists, on the surface at least, of a subject noun phrase followed by a series of two or more VPs, each containing a finite verb plus, possibly, the complement(s) of that verb’ (1974: 254).Footnote 19 In what follows, it is shown why RVH does not apply between words from adjacent VPs in a serial verb construction.

Serial verb constructions have been recognised as a typical property of Akan since Christaller (Reference Christaller1875). Osam (Reference Osam, Beermann and Hellan2003) distinguishes the clause-chaining type and the integrated serial verb construction as two main categories of serial verb constructions. The clause-chaining type combines constituents referring to a series of events; they can be separated by the conjunction /ná/ ‘and’, which allows for integration of adverbs between the first argument DP and the following verb. The integrated serial verb construction type, on the other hand, constitutes a case of the metaphoric meaning of the events expressed by the individual verbs. For instance, to express that one wants to play with an entity, one ‘takes that entity and eats it’. In other words, one ‘eats a game’ (cf. (36a) below). A similar example would be the use of the construction ‘eat sadness' to express the fact that somebody is sad.

According to Baker (Reference Baker1989: 524), the two VPs in the syntactic representation are sisters within a higher functional projection, i.e. vP, AspP or T, and each of the verbs functions as a head of its own VP (Baker Reference Baker1989, Osam Reference Osam, Beermann and Hellan2003, Kambon Reference Kambon2012). The general assumption is that tense and aspect markers are copied to the head of a VP. Since the verbs of a serial verb construction share these features in Akan, Baker argues that all of them function as heads, as illustrated in (34) (from Baker Reference Baker1989: 523f; glosses adjusted).

  1. (34)

Crucially, there are two adjacent VPs, which are not dominated by any further lexical projection, as shown in (35). Applying MatchPhrase thus results in two separate φmax's, each containing a verb and its complement. RVH is predicted to be blocked between the two VPs.

  1. (35)

As predicted, although the structural prerequisite of two adjacent words differing in [ATR] specification is met in (36a), i.e. the first object /kɔtɔ/ and the second verb /di/, the mid back vowel [ɔ] of the object is not affected by RVH. (36a) is an example of the integrated serial verb construction type.

  1. (36)

The example in (36b) illustrates that RVH is also blocked in the case of a reduced serial verb construction, an example of the clause-chaining type. Structurally, (36b) is identical to (36a): the two VPs are adjacent to each other and express a sequence of events, and are sisters syntactically. In (36b), however, the second VP consists only of its head, and there is no overt argument DP. The two VPs share the object. In the case of inanimate objects such as /εmʊ/, the second object DP may be empty overtly; in the case of an animate object, the object DP must be realised as a resumptive pronoun, which is indexed in (37) (cf. Osam Reference Osam, Beermann and Hellan2003, Duah Reference Duah2013).

  1. (37)

5 An OT analysis of regressive vowel harmony in Akan sentences

The examination of the data above has shown two things. First, RVH is a general phonological process in Akan, driven by the interword featural markedness constraint *Interword[—ATR][+ATR] in (11). Second, prosodic phrase formation using Match theory (Selkirk Reference Selkirk2009, Reference Selkirk, Goldsmith, Riggle and Yu2011), in combination with a theory of recursion-based prosodic subcategories (Ito & Mester Reference Ito, Mester, Borowsky, Kawahara, Shinya and Sugahara2012), provides the basis for the characterisation of the context where RVH is blocked at the edge(s) of a φmax. What remains to be made explicit in the OT analysis of RVH and phonological phrasing is the markedness constraint that is responsible for the blocking of RVH at the edge(s) of φmax.

To account for the blocking of RVH, I employ a constraint of the CrispEdge family. Such constraints have been proposed to avoid multiple linking of a feature between prosodic constituents (Itô & Mester Reference Itô, Mester, Kager, van der Hulst and Zonneveld1999, Selkirk Reference Selkirk, Goldsmith, Riggle and Yu2011); instead, a feature should be crisply aligned with the edge of a given prosodic domain. The relevant prosodic domains are ω (CrispEdgeω), φ (CrispEdgeJ) and ɩ (CrispEdgeƖ). Assuming that the blocking of RVH occurs at certain φ's would suggest that the relevant constraint is CrispEdgeJ. However, as we saw in the discussion of VP-internal and NP-internal RVH above, [+ATR] may spread across the edge of a φ; cf. (23). Thus the data suggest that a CrispEdge constraint must be able to refer to recursion-based subcategories like φmax. This constraint, formulated in (38a), disfavours linking of [+ATR] across the edges of a φmax. It thus necessarily outranks the general phrasal and word-level CrispEdge constraints, as shown in (38b).

  1. (38)

(39) shows that the optimal candidate, (a), violates the interword markedness constraint, but obeys the necessarily higher-ranked domain-sensitive CrispEdge constraint in (38a), which restricts RVH in the context of a φmax. Based on the assumption outlined above that functionally headed projections do not produce φ's, the subject DP forms its own φmax (cf. (21)). A further φmax arises from the mapping of the VP. The optimal candidate thus consists of two adjacent φmax's. The domain-sensitive CrispEdge constraint acts on this prosodic structure to ban RVH, and penalises any candidate, such as (b), that exhibits RVH across a φmax edge. If a candidate (e.g. (c)) exhibits a different prosodic structure, MatchPhrase is violated, showing that prosodic and syntactic structure are isomorphic in Akan. Theoretically, there could be a candidate with an incorrect phrasing pattern and no RVH. However, this candidate would also violate the MatchPhrase constraint, and would thus be ruled out, given that MatchPhrase is higher-ranked than any of the structure-sensitive constraints.

  1. (39)

Given the distinction between φmax and φnon&max, the domain-sensitive CrispEdge constraint in (38) does not apply VP-internally in (40), because the boundaries are non-maximal.

  1. (40)

Given that syntactic and prosodic structure is isomorphic, RVH applies. Thus the general CrispEdgeJ constraint may be violated, showing that the phrase and word versions of the CrispEdge constraint need to be lower-ranked than the interword markedness constraint, as in (41).

  1. (41)

Consider now the case of time adverbials in (33), whose structure is given in (42). The syntactic structure suggests that the time adverbial occupies a position outside the VP. In particular, Saah (Reference Saah1994) argues that time adverbials occupy the adjunct position of TP, taking scope over the whole sentence. On the definition in (22), the VP constitutes a φmax, since it is not dominated by any other φ within the ɩ. The AdvP also constitutes a φmax, as there is no further lexical phrasal projection above AdvP. In other words, all daughters below TP which contain a lexically headed phrase are mapped into a φmax. The presence of the edge of a φmax of the VP thus prohibits RVH, with the domain-specific CrispEdge constraint being active.

  1. (42)

Finally, consider the case of the serial verb construction in (36), whose structure is given as (43). The syntactic structure in (35) above suggests, according to Baker (Reference Baker1989: 524), that the VPs are in a sister relationship. This means that each VP forms its own φmax, since there is no further lexical phrasal projection dominating the VPs. As before, the domain-specific CrispEdge constraint thus prevents RVH across the edge of a φmax, i.e. between the two VPs.

  1. (43)

6 Summary and conclusion

This article has dealt with regressive [+ATR] vowel harmony across word boundaries, which provides a hitherto unreported cue for phonological phrasing. Only a few languages so far described display RVH (Casali Reference Casali2008). This very unusual process behaves similarly to other autosegmental phenomena such as H-tone spreading in Bantu languages (cf. Hyman Reference Hyman, Riad and Gussenhoven2007, Kula & Bickmore Reference Kula and Bickmore2015), but has not yet received a great deal of attention.

In Akan, vowel harmony is a feature of the word (Stewart Reference Stewart1983, Clements Reference Clements and Goyvaerts1985, Dolphyne Reference Dolphyne1988, Casali Reference Casali2012). In addition, Dolphyne (Reference Dolphyne1988) reports cases of vowel harmony between words where the dominant feature [+ATR] spreads regressively onto the immediately adjacent [—ATR] vowel of a preceding word. RVH in Akan is limited to the immediately adjacent vowel of a preceding word, and is thus bounded, as it is in Gwa Nmle (Obeng Reference Obeng1995), for instance. Other languages exhibit unbounded RVH, for example Nawuri (Casali Reference Casali2002). RVH thus behaves like H-tone spreading in Bantu languages, where both bounded and unbounded spreading has been reported (Hyman Reference Hyman, Riad and Gussenhoven2007, Kula & Bickmore Reference Kula and Bickmore2015). In particular, the Bantu languages Ekegusii (Bickmore Reference Bickmore1999) and Copperbelt Bemba (Kula & Bickmore Reference Kula and Bickmore2015) exhibit both types of H-tone spreading. The data shown in §4.2 and §4.3 suggested that RVH is a general feature of the grammar of Akan. However, §4.4 and §4.5 examined cases where RVH does not apply, even though the structural requirement of two adjacent words differing in their [ATR] value is met.

The present paper has argued that RVH in Akan is sensitive to prosodic structure. In particular, it has been shown that a φmax edge blocks the application of RVH. In other words, the process of RVH applies between words if the two adjacent words belong to the same φmax, independent of further non-φmax boundaries. The basis for this analysis of the domain-sensitivity of RVH is that it is blocked at a particular set of syntactic boundaries, in particular between a subject and a VP (18), between a VP and time adverbial (33) and between two VPs in a serial verb construction (36). However, RVH is not blocked at syntactic boundaries within a VP, for instance (19), (23) and (24).

The analysis makes reference to the syntax–phonology interface and its consequences for prosodic phrasing. Match theory (Selkirk Reference Selkirk, Goldsmith, Riggle and Yu2011) accounts for φ formation in terms of universal Match constraints. As Akan obeys Match constraints, prosodic phrase structure is isomorphic to syntactic structure in the default case. Within the syntactic clause, the phrasal complement of the complementiser head of CP is the TP, which is mapped onto an ɩ (21). In addition, all lexically headed daughter constituents of TP are mapped onto φ's. The analysis proposed here employs structure-sensitive constraints which make appeal to recursion-based subcategories such as φmax and φnon&max (Ito & Mester Reference Ito, Mester, Borowsky, Kawahara, Shinya and Sugahara2012). Each lexically headed daughter of a TP corresponds to a φmax. According to Ito & Mester (Reference Ito, Mester, Borowsky, Kawahara, Shinya and Sugahara2012), each such phrase is maximal, because it is not dominated by a further φ, but by an ɩ (cf. (21)).

The distribution of RVH in the sentence, in particular its blocking, can be simply characterised, given this approach. The fact that not all φ boundaries predicted by MatchPhrase block RVH can be expressed in terms of the subtypes of φ's proposed by Ito & Mester (Reference Ito, Mester, Borowsky, Kawahara, Shinya and Sugahara2012). The syntactic blocking contexts form a natural class in terms of the prosodic structure that is assigned to these configurations. Only the edge of a φmax prohibits the [+ATR] feature from spreading regressively. Recursively embedded phrases such as double objects or the components of complex DPs allow RVH, since in these contexts the phrase boundaries are non-maximal.

Finally, the data presented in this paper extend the analysis of phonological phrasing in Marfo (Reference Marfo2005), which is mainly concerned with prosodic restructuring at the level of ɩ's. Marfo claims, on the basis of certain tone rules, that topicalised and focused elements in Akan form separate ɩ's. Based on vowel-harmony patterns, this paper has shown that φ's below the ɩ level, specifically the organisation into φmax's, plays a role in the distribution of vowel-harmony features in the sentence. The distinction between φmax and φnon&max is made in terms of a recursive prosodic structure, which matches up with syntactic structure in the default case. Whether or not tonal rules are sensitive to phonological phrasing, in particular the maximal–non-maximal distinction, remains to be shown in future research.

Footnotes

*

The valuable help of Kofi Dorvlo and Charles Marfo during a fieldwork visit in 2012 is greatly acknowledged. Many thanks to Monica Amoah Appenteng and Reginald Duah for sharing their native speaker intuitions with me. Parts of this paper were presented at the ‘Syntax–phonology interface from a cross-linguistic perspective’ workshop at the Zentrum für Allgemeine Sprachwissenschaft, Berlin in 2012, and at colloquium series at Frankfurt am Main (2012), Stuttgart (2013), Potsdam (2014) and ZAS (2014). Comments from participants were greatly appreciated. For useful discussion, I would like to thank Ingo Feldhausen, Caroline Féry, Susanne Genzel, Shin Ishihara, Fabian Schubö and Sabine Zerbian. In particular, I sincerely thank the editors of this thematic issue, Lisa Selkirk and Seunghun Lee, for their invaluable support and discussion on earlier versions of this paper. This work was supported by the Deutsche Forschungsgemeinschaft (grant SFB 632 ‘Information structure’; project D5).

1 Akan refers to the largest ethnic group in Ghana (Nkansa-Kyeremateng Reference Nkansa-Kyeremateng2004) and is spoken by about 8·3 million people in Ghana and the Ivory Coast (Christaller Reference Christaller1933, Lewis Reference Lewis2009). Akan has a number of dialects, some of which are more mutually intelligible than others (Schachter & Fromkin Reference Schachter and Fromkin1968). The dialects differ both at segment and at tone level (cf. Cahill Reference Cahill1985, Dolphyne Reference Dolphyne1988, Dolphyne & Kropp Dakubu Reference Dolphyne1988, Abakah Reference Abakah2000, Reference Abakah2005, Abakah & Koranteng Reference Abakah and Koranteng2007, among others). Asante Twi is one of the three largest dialects, and its influence is such that it is becoming the lingua franca of Ghana (cf. Osam Reference Osam, Beermann and Hellan2003). The data in this study is based on Asante Twi; I will use Akan as a cover term throughout the paper. The data was elicited from two native speakers during a fieldwork visit in Accra and Kumasi (Ghana) in September 2012. Three further native Asante speakers reading the sentences were recorded in Kumasi. Some further data was elicited in June 2013 during a research stay by one of my informants in Potsdam.

2 Throughout the paper, the data appear in IPA transcription. Orthographic representations are also given in italics. Akan orthography uses standard Latin script in addition to the IPA characters E and O for the half-open front and back vowels (Dolphyne Reference Dolphyne1988: 7). In total, the Akan orthography has seven vowel and 16 consonant symbols. Note that the characters e and o are ambiguous with respect to [ATR]; e can be pronounced either as unadvanced front high [ɪ] or as advanced front mid [e], o as unadvanced back high [ʊ] or advanced mid back [o].

The following abbreviations are used in glosses: compl=completive; det= determiner; fut=future; imp=imperative; nc=noun class; perf=perfective; pl=plural; prn=pronoun; prog=progressive; red=reduplicant; sg=singular.

3 The low vowel /a/ does not take part in a systematic [ATR] alternation, and some researchers have argued that it behaves as a neutral vowel. However, there are instances of /a/ surfacing in [+ATR] surroundings as low front unrounded [æ], e.g. /aburo/ [æburo] ‘maize’.

4 Since this paper is about vowel harmony between words, the facts on word-level harmony are presented only briefly. Extended work on vowel harmony in OT can be found in Baković (Reference Baković2000); for an overview see Walker (Reference Walker2012). Ballard (Reference Ballard2010) proposes an OT approach to Akan vowel harmony in terms of alignment constraints, which at the word-level captures the harmony patterns correctly. However, he assumes that RVH applies at the level of the ɩ, and not at the level of the φ, as proposed here.

5 As shown in (3), pronouns, as well as tense and aspect markers, vary according to the [ATR] specification of the stem. I therefore assume, with Clements (Reference Clements and Goyvaerts1985), that pronominal prefixes, tense, aspect and mood markers, as well as nominal affixes, are underspecified for [ATR]; their vowels are represented by capital letters in underlying representations.

6 These verbs usually occur with an overt object in Akan; if not, the resumptive pronoun /ε/ is suffixed to the verb. This pronoun is not affected by vowel harmony, and remains [—ATR], giving (a) /o-hunu-i-ε/ and (b) /ɔ-kɔ-I-ε/. In case of an overt object, the final vowel of the verb is lengthened, as in /o-hunuu kube/ ‘s/he saw a coconut’ and /ɔ-kɔɔ fie/ ‘s/he went to the house’.

7 In addition to completely harmonising words, disharmonic stems exist in Akan, e.g. /sika/ ‘money’ and /bisa/ ‘to ask’. In such cases, affixes receive their [ATR] specification from the corresponding adjacent stem syllables, as in (i).

  1. (i)

Almost all disharmonic stems have /a/ as the [—ATR] vowel, except for the word /¿insεn/ ‘be pregnant’ (cf. Dolphyne Reference Dolphyne1988). Stewart (Reference Stewart1983: 126–127) argues that all disharmonic stems originate from lexicalised compounds (cf. also Obeng Reference Obeng2000 on vowel harmony in lexicalised toponyms in Akan). The fact that disharmonic stems show an asymmetry between the [ATR] specification of a prefix and a suffix provides an argument that [—ATR] is represented underlyingly in Akan.

8 Although the analysis shown here follows the assumption that any verbal affixes are underlyingly underspecified for the feature [ATR] (cf. Clements Reference Clements and Goyvaerts1985), the question of underspecification is irrelevant, because positional faithfulness requires stem identity only. For instance, assuming a [+ATR] input for the personal pronoun prefix in (6) would result in the same violations as shown for candidate (d). In addition, the lower-ranked Ident[ATR]IO constraint would be violated if affixes were specified for [ATR]. Thanks to a reviewer for pointing this out.

9 The dominance of [+ATR] can be captured by a *[+ATR] constraint that is ranked higher than a corresponding *[—ATR] constraint. *[+ATR] is implemented below, showing that RVH is bounded in Akan.

10 Lisa Selkirk raises the question whether the interword markedness constraint needs to refer to both prosodic word edges, as the formalisation in (11) suggests. Assuming a situation where a monosyllabic function word, which in many languages does not constitute a prosodic word on its own, spreads its [+ATR] feature to a word on its left, there would be no left prosodic word edge for the [+ATR] word. Since there is no data of this sort available, and I have not carried out a thorough analysis of the status of monosyllabic function words, the question of the presence or absence of a left prosodic word edge in (11) cannot be resolved. For the time being, I assume that both word edges need to be specified, since this formalisation captures the data considered in this paper.

11 The prosodic markedness constraint in (11) has a certain stipulative character, as a reviewer correctly points out. It remains unclear why it is only [+ATR] that undergoes RVH, and why RVH fails to operate in both directions. In particular, contrary to the structurally similar constraint in Copperbelt Bemba (Kula & Bickmore Reference Kula and Bickmore2015), it remains unclear how the fact that [+ATR] only spreads regressively should be formalised. A tentative speculation about the first question would be Casali's (Reference Casali2003) proposal arguing for dominant features; in Akan, it is [+ATR] which is dominant. The interword effects would then restricted to this dominant feature. However, I leave these issues for future research.

12 Progressive spreading of the [ATR] feature can be prohibited by means of a positional faithfulness constraint requiring word or stem-initial syllables to maintain their input [ATR] specification (cf. Kaplan Reference Kaplan2008: 289). Hence, a candidate such as (ɔpε)ω1 (kʊbe)ω2 would violate positional faithfulness with the CrispEdge and *[ATR] constraints, since the first syllable of the second prosodic word changes the [ATR] specification. The main point of the present paper, however, is to characterise the distribution of RVH within the sentence, and thus any further discussion of progressive spreading is beyond the scope of the current analysis.

13 Casali (Reference Casali2008) lists eleven such languages. Among these are further Kwa languages such as Chumburung (Snider Reference Snider1985, Reference Snider1989), Gwa Nmle (Obeng Reference Obeng1995) and Nawuri, as well as the Gur languages Deg (Crouch & Herbert Reference Crouch and Herbert1997) and Dilo (Jones Reference Jones1987), the Bantu language Kinande (Mutaka Reference Mutaka1995), the Kru language Vata (Kaye Reference Kaye, van der Hulst and Smith1982), the Nupoid language Igbira (Scholz Reference Scholz1976) and the Central Sudanic language Bongo (Kilpatrick Reference Kilpatrick, Wise and Watson1985). Two further Kwa languages also show RVH, Tafi (Bobuafor Reference Bobuafor2013) and Efutu (Obeng Reference Obeng2008). Thanks to Kofi Dorvlo for drawing my attention to Tafi and Efutu. Kaplan (Reference Kaplan2008) mentions the American language Nez Perce, which shows optional bounded regressive vowel harmony across word boundaries, although for both [ATR] values.

14 It remains an open question whether the proposed interaction of *[+ATR] and *Interword[—ATR][+ATR], which accounts for bounded spreading in Akan, would also account for unbounded spreading of [+ATR] in Nawuri, and whether Akan exhibits a particular subcase of a general unbounded RVH process.

15 The small amount of data on RVH in Gwa Nmle shows RVH between a verb and subject (Obeng Reference Obeng1995); however, the subject–verb sequence is the only syntactic structure for RVH between words discussed in Obeng (Reference Obeng1995). Therefore, we cannot draw any further conclusions for Gwa Nmle here.

16 The ɩ is assumed to be the result of CP mapping (Selkirk Reference Selkirk, Goldsmith, Riggle and Yu2011). Since ɩ's are not relevant for the current discussion, I omit this level of prosodic phrasing in all examples.

17 In Akan, there is no overt morphosyntactic marking of the nouns which reflects their syntactic relation. There is, however, an extra H tone which functions as an associative marker between a pronoun and a noun, and between two nouns in some cases (Cahill Reference Cahill1985: 45, Dolphyne Reference Dolphyne1988: 69ff, Abakah & Koranteng Reference Abakah and Koranteng2007, Abakah Reference Abakah2010). In (31a) and (c), downstep is indicated by an arrow. The associative marker, a floating H tone, delinks the lexical L tone of the head noun /ǹsúó/, which becomes a floating tone. This floating L tone causes following H tones to be downstepped (Stewart Reference Stewart1965, Dolphyne Reference Dolphyne1994, Genzel & Kügler Reference Genzel, Kügler, Lee and Zee2011).

18 Most time adverbials consist of [—ATR] vowels, e.g. /εnε/ ‘today’, /εnnεra/ ‘yesterday’, which do not meet the structural requirements for RVH.

19 Serial verb constructions have been defined using various criteria; for an overview with respect to Akan see Osam (2004). For our purposes, Schachter's (Reference Schachter1974) structural definition suffices to establish the fact that a serial verb construction consists of at least two VPs.

References

REFERENCES

Abakah, Emmanuel Nicholas (2000). A closer look at downstep in Akan. Afrika und Übersee 83. 253–275.Google Scholar
Abakah, Emmanuel Nicholas (2005). Tone rules in Akan. Journal of West African Languages 32. 109–134.Google Scholar
Abakah, Emmanuel Nicholas (2010). Tone and the associative construction in Akan. Journal of West African Languages 37:2. 57–79.Google Scholar
Abakah, Emmanuel Nicholas & Koranteng, Louise (2007). The interaction of tone, syntax and semantics in Akan. Studies in the Languages of the Volta Basin 4:2. 63–86.Google Scholar
Aboh, Enoch O. (2010). The morphosyntax of the Noun Phrase. In Aboh, Enoch O. & Essegbey, James (eds.) Topics in Kwa syntax. Dordrecht: Springer. 11–38.Google Scholar
Baker, Mark C. (1989). Object sharing and projection in serial verb constructions. LI 20. 513–553.Google Scholar
Baković, Eric (2000). Harmony, dominance and control. PhD dissertation, Rutgers University.Google Scholar
Ballard, Lee W. (2010). Akan vowel harmony in Optimality Theory. MA thesis, University of Florida.Google Scholar
Bickmore, Lee S. (1999). High tone spread in Ekegusii revisited: an optimality theoretic account. Lingua 109. 109–153.Google Scholar
Boadi, Lawrence A. (2005). Three major syntactic structures in Akan: interrogatives, complementation, and relativisation. Accra: Black Mask.Google Scholar
Bobuafor, Mercy (2013). A grammar of Tafi. PhD thesis, University of Leiden.Google Scholar
Cahill, Michael C. (1985). An autosegmental analysis of Akan nasality and tone. MA thesis, University of Texas at Arlington.Google Scholar
Casali, Roderic F. (2002). Nawuri ATR harmony in typological perspective. Journal of West African Languages 29:1. 3–43.Google Scholar
Casali, Roderic F. (2003). [ATR] value asymmetries and underlying vowel inventory structure in Niger-Congo and Nilo-Saharan. Linguistic Typology 7. 307–382.Google Scholar
Casali, Roderic F. (2008). ATR harmony in African languages. Language and Linguistics Compass 2. 496–549.Google Scholar
Casali, Roderic F. (2012). [+ATR] dominance in Akan. Journal of West African Languages 39:1. 33–59.Google Scholar
Christaller, J. G. (1875). A grammar of the Asante and Fante language. Basle: Evangelical Missionary Society.Google Scholar
Christaller, J. G. (1933). Dictionary of the Asante and Fante language called Tshi (Twi) . 2nd edn. Basle: Evangelical Missionary Society.Google Scholar
Clements, G. N. (1985). Akan vowel harmony: a nonlinear analysis. In Goyvaerts, Didier L. (ed.) African linguistics: essays in memory of M. W. K. Semikenke. Amsterdam & Philadelphia: Benjamins. 55–98.Google Scholar
Crouch, Marjorie & Herbert, Patricia (1997). Deg phonology. Tamale: Ghana Institute of Linguistics, Literacy and Bible Translation.Google Scholar
Dolphyne, Florence A. (1988). The Akan (Twi-Fante) language: its sound systems and tonal structure. Accra: Ghana Universities Press.Google Scholar
Dolphyne, Florence A. (1994). A phonetic and phonological study of downdrift and downstep in Akan. Paper presented at the 25th Annual Conference on African Linguistics, Rutgers University.Google Scholar
Dolphyne, Florence A. & Dakubu, M. E. Kropp (1988). The Volta-Comoé languages. In Dakubu, M. E. Kropp (ed.) The languages of Ghana. London: Kegan Paul. 50–90.Google Scholar
Duah, Reginald A. (2013). Force dynamics and causation in Akan. PhD thesis, University of Ghana.Google Scholar
Elfner, Emily (2012). Syntax–prosody interactions in Irish. PhD dissertation, University of Massachusetts Amherst.Google Scholar
Genzel, Susanne (2013). Lexical and post-lexical tones in Akan. PhD thesis, University of Potsdam.Google Scholar
Genzel, Susanne & Kügler, Frank (2011). Phonetic realization of automatic (downdrift) and non-automatic downstep in Akan. In Lee, Wai-Sum & Zee, Eric (eds.) Proceedings of the 17th International Congress of Phonetic Sciences, Hong Kong 2011. Hong Kong: University of Hong Kong. 735–738.Google Scholar
Hess, Susan (1992). Assimilatory effects in a vowel harmony system: an acoustic analysis of advanced tongue root in Akan. JPh 20. 475–492.Google Scholar
Hyman, Larry M. (2007). Universals of tone rules: 30 years later. In Riad, Tomas & Gussenhoven, Carlos (eds.) Tones and tunes. Vol. 1: Typological studies in word and sentence prosody. Berlin & New York: Mouton de Gruyter. 1–34.Google Scholar
Itô, Junko & Mester, Armin (1999). Realignment. In Kager, René, van der Hulst, Harry & Zonneveld, Wim (eds.) The prosody–morphology interface. Cambridge: Cambridge University Press. 188–217.Google Scholar
Ito, Junko & Mester, Armin (2012). Recursive prosodic phrasing in Japanese. In Borowsky, Toni, Kawahara, Shigeto, Shinya, Takahito & Sugahara, Mariko (eds.) Prosody matters: essays in honor of Elisabeth Selkirk. London: Equinox. 280–303.Google Scholar
Jones, Peggy (1987). Collected field reports on the phonology of Dilo. Legon: Institute of African Studies, University of Ghana.Google Scholar
Kambon, Obadele B. (2012). Serial verb nominalization in Akan. PhD thesis, University of Ghana.Google Scholar
Kaplan, Aaron F. (2008). Noniterativity is an emergent property of grammar. PhD dissertation, University of California, Santa Cruz.Google Scholar
Kaye, Jonathan D. (1982). Harmony processes in Vata. In van der Hulst, Harry & Smith, Norval (eds.) The structure of phonological representations. Part 2. Dordrecht: Foris. 385–452.Google Scholar
Kilpatrick, Eileen (1985). Bongo phonology. In Wise, Mary Ruth & Watson, Richard L. (eds.) Occasional papers in the study of Sudanese languages No. 4. Juba: Summer Institute of Linguistics. 1–62.Google Scholar
Kula, Nancy C. & Bickmore, Lee S. (2015). Phrasal phonology in Copperbelt Bemba. Phonology 32. 147–176 (this issue).Google Scholar
Lewis, M. Paul (ed.) (2009). Ethnologue: languages of the world. 16th edn. Dallas: SIL International. Available at http://www.ethnologue.com.Google Scholar
Lindau, Mona (1975). [Features] for vowels. UCLA Working Papers in Phonetics 30. Available (February 2015) at http://escholarship.org/uc/item/7gv6z0vq.Google Scholar
Lindau, Mona (1979). The feature expanded. JPh 7. 163–176.Google Scholar
Marfo, Charles O. (2005). Aspects of Akan grammar and the phonology–syntax interface. PhD thesis, University of Hong Kong.Google Scholar
Mutaka, Ngessimo M. (1995). Vowel harmony in Kinande. Journal of West African Languages 25:2. 41–55.Google Scholar
Nespor, Marina & Vogel, Irene (1986). Prosodic phonology. Dordrecht: Foris.Google Scholar
Nkansa-Kyeremateng, Kofi (2004). The Akans of Ghana: their customs, history and institutions. Accra: Sebewie.Google Scholar
Obeng, Samuel Gyasi (1995). Vowel harmony in Gwa Nmle. Afrikanistische Arbeitspapiere 41. 143–152.Google Scholar
Obeng, Samuel Gyasi (2000). Vowel harmony and tone in Akan toponyms. Studies in the Linguistic Sciences 30:2. 173–183.Google Scholar
Obeng, Samuel Gyasi (2008). Efutu grammar. Munich: Lincom.Google Scholar
Osam, E. Kweku (2003). An introduction to the verbal and multiverbal system of Akan. In Beermann, Dorothee & Hellan, Lars (eds.) Proceedings of the workshop on multi-verb constructions (Trondheim Summer School 2003) . Available (February 2015) at http://www.ling.hf.ntnu.no/tross/osam.pdf.Google Scholar
Paster, Mary (2010). The verbal morphology and phonology of Asante Twi. Studies in African Linguistics 39:1. 77–120.Google Scholar
Saah, Kofi K. (1994). Studies in Akan syntax, acquisition, and sentence processing. PhD thesis, University of Ottawa.Google Scholar
Schachter, Paul (1974). A non-transformational account of serial verbs. Studies in African Linguistics. Suppl. 5. 253–270.Google Scholar
Schachter, Paul & Fromkin, Victoria A. (1968). Phonology of Akan: Akuapem, Asante and Fante. UCLA Working Papers in Phonetics 9. 1–268.Google Scholar
Scholz, Hans-Jürgen (1976). Igbira phonology. Dallas: Summer Institute of Linguistics.Google Scholar
Selkirk, Elisabeth (1986). On derived domains in sentence phonology. Phonology Yearbook 3. 371–405.Google Scholar
Selkirk, Elisabeth (2009). On clause and intonational phrase in Japanese: the syntactic grounding of prosodic constituent structure. Gengo Kenkyu 136. 35–73.Google Scholar
Selkirk, Elisabeth (2011). The syntax–phonology interface. In Goldsmith, John, Riggle, Jason & Yu, Alan C. L. (eds.) The handbook of phonological theory. 2nd edn. Oxford: Wiley-Blackwell. 435–484.Google Scholar
Snider, Keith L. (1985). Vowel coalescence across word boundaries in Chumburung. Journal of West African Languages 15:1. 3–13.Google Scholar
Snider, Keith L. (1989). Vowel coalescence in Chumburung: an autosegmental analysis. Lingua 78. 217–232.Google Scholar
Stewart, John M. (1965). The typology of the Twi tone system. Bulletin of the Institute of African Studies 1. 1–27.Google Scholar
Stewart, John M. (1967). Tongue root position in Akan vowel harmony. Phonetica 16. 185–204.Google Scholar
Stewart, John M. (1983). Akan vowel harmony: the word structure conditions and the floating vowels. Studies in African Linguistics 14. 111–139.Google Scholar
Tiede, Mark K. (1996). An MRI-based study of pharyngeal volume contrasts in Akan and English. JPh 24. 399–421.Google Scholar
Truckenbrodt, Hubert (1999). On the relation between syntactic phrases and phonological phrases. LI 30. 219–255.Google Scholar
Walker, Rachel (2012). Vowel harmony in Optimality Theory. Language and Linguistics Compass 6. 575–592. http://onlinelibrary.wiley.com/doi/10.1002/lnc3.340/epdf.Google Scholar