2.1 Syllable structure

According to some analyses (Dell & Elmedlaoui Reference Dell and Elmedlaoui1985, Reference Dell and Elmedlaoui1988, Reference Dell and Elmedlaoui2002), Tashlhiyt has syllables without vocalic nuclei, resulting in particularly long consonantal sequences (but see Coleman Reference Coleman, Durand and Laks1996, Reference Coleman, Ohala, Hasegawa, Ohala, Granville and Bailey1999, Reference Coleman2001 for an alternative account).

In addition to the typologically common syllables with vocalic nuclei V, CV, VC and CVC, the language has been described as having consonant-only syllables C̩, CC̩, C̩C and CC̩C.Footnote ¹ According to Dell & Elmedlaoui (Reference Dell and Elmedlaoui1985, Reference Dell and Elmedlaoui1988, Reference Dell and Elmedlaoui2002), all consonant types are allowed in syllable nucleus position, even a voiceless stop, as exemplified in (1) (syllable boundaries are indicated by dots).

1. (1)

   

The analysis results from a syllabification algorithm that goes through the sonority scale from highest to lowest, assigning the syllable nucleus to the most sonorous segments in a sequence, according to the following scale: low vowels > high vowels > liquids > nasals > voiced fricatives > voiceless fricatives > voiced plosives > voiceless plosives (cf. Dell & Elmedlaoui Reference Dell and Elmedlaoui1985).Footnote ² In addition to sonority requirements, the competition between segments is governed by well-formedness conditions, such as a dispreference for complex subsyllabic constituents (onset, coda) and onsetless syllables.

The main sources of evidence for syllabification are judgements about well-formedness in versification (Dell & Elmedlaoui Reference Dell and Elmedlaoui2002, Ridouane Reference Ridouane2008), morphological regularities that can be captured straightforwardly if this syllabification is assumed (Dell & Elmedlaoui Reference Dell and Elmedlaoui2002) and native-speaker intuitions (Dell & Elmedlaoui Reference Dell and Elmedlaoui2002, Ridouane et al. Reference Ridouane, Hermes and Hallé2014). However, words like those in (1) may surface phonetically with one or more vocoids that have a schwa-like quality. The above authors treat these vocoids as phonetic artefacts, resulting from the articulatory realisation of the consonantal sequences they appear to break up (Dell & Elmedlaoui Reference Dell and Elmedlaoui1985, Reference Dell and Elmedlaoui2002, Reference Dell and Elmedlaoui2008, Ridouane Reference Ridouane2008, Ridouane & Fougeron Reference Ridouane and Fougeron2011). Support for this analysis comes from the observation that vocoids frequently surface next to voiced segments and rarely surface within homorganic sequences, which have unreleased consonants (but see Dell & Elmedlaoui Reference Dell and Elmedlaoui2008 for exceptions). Thus they are assumed to be largely predictable from the laryngeal and supralaryngeal specification of the consonantal environment.

However, this analysis has been challenged (e.g. Coleman Reference Coleman, Durand and Laks1996, Reference Coleman, Ohala, Hasegawa, Ohala, Granville and Bailey1999, Reference Coleman2001, Angoujard Reference Angoujard1997, Louali Reference Louali1999, Louali & Puech Reference Louali and Puech2000). For instance, Coleman (Reference Coleman2001) affords vocoids a phonological status, treating them as epenthetic vowels that occupy an empty nucleus position, leading to the syllabification of words such as in (1) as /təs.χəft/ rather than /ts̩.χf̩t/. In the remainder of this paper, we refer to this account as the ‘nuclear schwa’ account.

While arguments in favour or against the different analyses are based on phonotactic and metrical evidence, it might be informative to look at other levels of prosodic organisation, as suggested by Gordon & Nafi (Reference Gordon and Nafi2012). They suggest that there may be two types of vocoid, one intrusive, playing no role in the phonology (see Hall Reference Hall2006 for extensive discussion), and one with a phonological role in terms of its tone-bearing qualities. Although in Tashlhiyt these vocoids do not contribute to syllable weight for versification in traditional songs, they can nonetheless align with musical notes (Dell & Elmedlaoui Reference Dell and Elmedlaoui2008, Dell Reference Dell, Cairns and Raimy2011): ‘schwas may not be metrically relevant … but they are nonetheless relevant in singing, as they participate in the mapping of the text onto the melody’ (Dell & Elmedlaoui Reference Dell and Elmedlaoui2008: 152).

For example, consider a disyllabic word (e.g. /ad.rar/ ‘mountain’), produced in a song with a sequence of two musical notes X and Y. The musical notes are aligned to the two vowels in a one-to-one mapping. However, in words with no vowel but with sonorant consonants, the musical notes can either be realised on the sonorant consonants or on transitional vocoids within the consonant cluster (e.g. /n.∫r̩k/ [n∫rk] or [nə∫rək] ‘we share’). Dell & Elmedlaoui (Reference Dell and Elmedlaoui2008) report free alternation between these two forms. Furthermore, if the word contains a syllable with neither a lexical vowel nor a sonorant consonant (such as the second syllable of /tn̩.db̩t/ ‘you regret’), there are three options: (i) the second musical note is not realised at all, (ii) it is realised on a vocoid or (iii) it is realised earlier, resulting in two notes aligned to the first syllable /tn̩/.

Dell & Elmedlaoui argue that vocoids serve as carriers of musical notes in singing and are thus relevant for a description of musical form, suggesting that they could be relevant for the description of intonational tones too. However, Dell & Elmedlaoui maintain that despite their role in singing, they do not need to be considered as phonological entities (2008: 177–182), since they do not play a role in syllabification or in metre (2008: 55). We return to this in the interpretation of the results in §3.2.5.

2.2 Stress and intonation

In terms of stress and intonation, Tashlhiyt Berber is particularly challenging. Due to its rare phonotactic patterns, the phonetic opportunity afforded for the execution of intonational pitch movements is exceptionally limited. While research on the language has mainly concentrated on elements lower in the prosodic tree such as the syllable, little is known about its metrical structure. Dell & Elmedlaoui point out that stress accent is likely to be a property of units larger than words, rather than being part of the lexical representation. Even before this, Applegate (Reference Applegate1958: 9) noted in his grammar of Tashlhiyt:

Tonal aspects of Tashlhiyt are reported to interact with its phonotactic patterns. Based on preliminary observations and native-speaker intuitions, Dell & Elmedlaoui (Reference Dell and Elmedlaoui1985) provide an analysis of optional syllabification in intonation phrase final position which goes hand in hand with the placement of intonational tones. For example, when the word /igidr/ ‘eagle’ is at the end of a phrase with interrogative intonation, a rising F0 contour can occur on the final /r/, in which case it is analysed as a syllable nucleus and thus a tone-bearing unit (resulting in /i.gi.dr̩/). Alternatively, within this analysis, the final consonant can ‘lose’ its syllabic status, and be annexed to the previous syllable (resulting in /i.gidr/). In this case, the F0 peak is on the second vowel /i/. They propose an optional rule of prepausal annexation, turning trisyllabic /i.gi.dr̩/ into disyllabic /i.gidr/, with a final complex coda. This analysis is supported by Elmedlaoui's intuitions about syllable count and acceptability of F0 peak placement. According to Dell & Elmedlaoui, this alternation is only observable when the word has a final sonorant consonant, which would either form the nucleus of a light syllable or be part of a complex coda. They state that ‘similar observations can be made with other intonations’ (Dell & Elmedlaoui Reference Dell and Elmedlaoui1985: 119).

The question arises as to whether this constitutes variation in the phonological association of the intonational tone or whether it should be seen as variation in phonetic alignment. We take association to be discrete, referring to a phonological linking of a tone with a tone-bearing unit, and alignment to be continuous, referring to the exact position of a tone in relation to a landmark in the speech signal, such as the edge of a constituent (see e.g. Ladd Reference Ladd2008). Dell & Elmedlaoui (Reference Dell and Elmedlaoui1985: 119) explicitly claim that the intonational tone is ‘assigned to the last tone-bearing unit’, which they define as a syllable with a sonorant nucleus (vowel or sonorant consonant), indicating that there may be evidence for a discrete location of tones, and thus a phonological association. Furthermore, they also discuss the importance of sonority and syllable weight, both of which play a relevant role in other areas of the linguistic system of Tashlhiyt (e.g. Dell & Elmedlaoui Reference Dell and Elmedlaoui2002).

In many languages, tonal association is determined by word stress (e.g. Ladd Reference Ladd2008). For instance, in many dialects of Arabic a pitch accent is typically associated with a lexically stressed syllable (Chahal & Hellmuth Reference Chahal, Hellmuth and Jun2014). As pointed out above, the very existence of word stress in Tashlhiyt has been called into question. Nonetheless, Gordon & Nafi (Reference Gordon and Nafi2012) have recently found evidence for word-level prominence. Their investigation of three acoustic correlates of potential phonological prominence (F0, intensity and duration) revealed that word-final syllables have greater intensity than their counterparts in the penultimate syllable of a word. Moreover, as they point out, phrase-final and, to a lesser degree, word-final nuclei were consistently lengthened, although it is difficult to tease apart the effect of word- or phrase-level prominence from word- or phrase-level lengthening (also referred to as domain-final lengthening; e.g. Beckman & Edwards Reference Beckman, Edwards, Kingston and Beckman1990). Interestingly, Gordon & Nafi also found that phrase-final syllables were associated with a higher F0 than other syllables, and interpret this as a reflex of a high pitch accent. Predictably, this H tone was phonetically more salient when associated with sonorant nuclei (vowel or sonorant consonant) than with voiced obstruent nuclei. Phrase-final high pitch can be a result of a final rise to a high boundary (analysed as a boundary tone H%), which is common in list intonation, for example.

Gordon & Nafi report that phrase-final syllables with voiceless obstruent nuclei often surfaced with interconsonantal vocoids that provided a phonetic context for a F0 peak. Moreover, there were more vocoids in the final syllable (both at word and phrase level) than in the penult. They speculate that the presence of these vocoids might be due to the necessity to realise a communicatively relevant pitch movement in that position. This fits in with the observations by Dell & Elmedlaoui (Reference Dell and Elmedlaoui2008) and Dell (Reference Dell, Cairns and Raimy2011) on singing discussed above, in which vocoids are said to bear musical notes.

Nonetheless, we cannot rule out an alternative explanation for the presence of vocoids towards the end of words and phrases: at least in English, phrase-final position goes hand in hand with a reduction in overlap of articulatory gestures (Edwards et al. Reference Edwards, Beckman and Fletcher1991). This reduction in overlap results in more space for open transitions (and thus vocalic elements) between the oral release and subsequent constriction. This is the case regardless of accentuation or any requirement to produce a particular pitch movement.

Gordon & Nafi's study provided evidence that certain elements of the utterance are highlighted through a combination of segmental prominence and pitch movement. Whether this is a manifestation of word stress or not is unclear, partly due to the nature of the corpus (isolated words and words with a following adverb), making it difficult to tease apart whether a tone is purely delimitative (in terms of an edge tone) or prominence-lending (in terms of a pitch accent) and whether an apparent segmental prominence is a result of modifications in gestural organisation at the edges of phrases.

A first instrumental investigation on intonational tones in slightly longer sentences was conducted by Grice et al. (Reference Grice, Roettger, Ridouane and Fougeron2011), who investigated the nature and distribution of an intonation phrase medial H tone in declarative sentences for three speakers. The target word was placed in a constant carrier phrase and was implicitly contrasted with other target words in the same experiment (/inna target baHra/ ‘he said target a lot’). They obtained a F0 peak on the penultimate syllable of the utterance (i.e. /baH/) followed by a fall to low pitch (represented as a HL boundary sequence). Unlike Gordon & Nafi, they did not find a final rise in pitch at the end of the phrase. Instead, they found an F0 peak earlier in the phrase (represented as a H tone). The distribution of this phrase-medial H peak was highly dependent on the segmental make-up of the target word: the F0 peak was consistently located on a sonorant nucleus in the target word, if one was available. Moreover, if two sonorant nuclei were available in the target word (e.g. /fl̩.kl̩f/ ‘on skin patches’) the F0 peak occurred on the rightmost of the two. If the target word contained no sonorants, the location of H was highly variable: it was either on a vocoid within the word or on a lexical vowel before the target word (i.e. on /inna/). Recall that similar strategies have also been reported for musical notes (which can align with either a vocoid in the syllable associated with the note or with the preceding syllable; see §2.1).

Grice et al. (Reference Grice, Roettger, Ridouane and Fougeron2011) thus showed (i) that F0 peaks corresponding to H tones can occur utterance-medially, (ii) that the location of the H is affected by the segmental makeup of the target word, (iii) that the H may be located on a transitional vocoid (or even on a vowel before the target word) and (iv) that the H might be used to mark contrast or some other kind of prominence. However, since these recordings were made for an independent articulatory experiment (Hermes et al. Reference Hermes, Ridouane, Mücke and Grice2011), this study was exploratory in nature, and lacked the necessary control to tease apart the different factors affecting F0 peak placement.

In the current study we investigate tonal placement, controlling for segmental make-up and introducing an explicit contrast. To investigate the intonation patterns with respect to the function they perform, we also recorded two different sentence modalities, in the form of polar questions and contrastive statements. We explicitly tested the validity of Dell & Elmedlaoui's observations that intonational tones are associated with tone-bearing units, and that sonority and syllable weight affect this association.

3.1 Method

3.1.2 Speech materials

We controlled the segmental properties of the target words, so as to investigate the interaction of sonority and weight as determining factors in tonal association (cf. Dell & Elmedlaoui Reference Dell and Elmedlaoui1985). In addition, we manipulated sentence modality, using the structures most frequently produced in the semi-spontaneous elicitation task described in Appendix A, as shown in (2).

1. (2)

   

We recorded 28 pairs of disyllabic target words, which varied in the sonority of the nucleus of both syllables. Corresponding members of a pair differ in the syllable weight of the final syllable (CV vs. CVC). Stimuli were selected to cover a wide range of segment types. Two different words were used for each combination of syllable types. Table I shows the make-up of the syllabic nucleus in the penultimate and final syllables, with examples (see Appendix B for the full list of target words).Footnote ³

Table I Examples of target words and conditions, each condition consisting of two item pairs. The members of each pair differ in syllable weight. V=vowel, S=sonorant consonant, N=nasal, L=liquid, O=obstruent ([+voice] and [−voice]). Examples are given in the Latin script used here (see note 3) and in phonemic transcription.

3.1.3 Procedure

Recordings were made in a soundproof booth at the Laboratoire de Phonétique et Phonologie in Paris for three of the subjects (S1, S3 and S4), and in a quiet room for S2. The reading was recorded using a Marantz PMD 660 solid-state recorder at a sampling rate of 44·1 kHz, and a Shure SM10A head-mounted microphone. Before recording began, participants were asked to read a word-list to the experimenter, to ensure that they were familiar with all the items in the experiment. They were then seated in front of a computer screen, and read out orthographically presented materials containing the target words along the lines of (2), as pairs of slides like those in Fig. 1.

Figure 1 Form of elicitation materials (in translation) presented to subjects on consecutive slides for reading aloud.

Slide sequences were presented in random order. In slide 1 a target word (A) is in a polar question. In slide 2 the same target word is presented in a negative statement, followed by a different target word (B) in a contrastive statement. Because of the previous negative statement, the second target word is explicitly contrasted (corrective focus). The negative statement served merely as a context to trigger this contrast, and is not analysed further. In both the polar question and the contrastive statement, the target word is located at the right edge of the intonational phrase. Each target word appeared in both positions three times, resulting in 336 tokens per speaker (the 14 phonotactic conditions in Appendix B×2 words per condition×2 syllable weights (light vs. heavy)×2 sentence modalities (statement vs. question)×3 repetitions).

4.1 Factors determining F0 peak placement

Our results point to four competing factors affecting the location of the F0 peak. They can be stated as in (3).

1. (3)

   

The competition amongst these factors leads to a probabilistic distribution of F0 peak placement across the different conditions.

It is possible to account for the tendency to place F0 peaks late in the phrase and on the most sonorous syllable in terms of phonetic grounding. A pitch event near the end of the phrase is salient, due to recency effects (e.g. Healy et al. Reference Healy, Havas and Parkour2000). Furthermore, a F0 peak later in the phrase may better highlight the target word (that is phrase-final in our case): F0 peak delay has a similar effect to an increase in peak height (Gussenhoven Reference Gussenhoven2004), which is often used for emphasis (Ladd & Morton Reference Ladd and Morton1997). In terms of sonority, a pitch target or contour on a vowel is easier to perceive than one on a consonant. Even if the consonant is voiced – clearly a prerequisite for pitch – a pitch target is easier to perceive on nasals and liquids than on fricatives and stops (e.g. Zhang Reference Zhang, Hayes, Kirchner and Steriade2004, Barnes et al. Reference Barnes, Brugos, Veilleux, Shattuck-Hufnagel, Campbell, Gibbon and Hirst2014).

The different preferences in polar questions and contrastive statements might also be considered as functionally motivated, in this case to help distinguish sentence modality. A final rise in pitch is cross-linguistically common in polar questions (Ultan Reference Ultan, Greenberg, Ferguson and Moravcsik1978, Ohala Reference Ohala1983, Reference Ohala1984, Gussenhoven Reference Gussenhoven2004, although see Rialland Reference Rialland, Riad and Gussenhoven2007). However, it is also fairly common for a rise to occur on the accented syllable and for a fall to follow it post-accentually (Grice et al. Reference Grice, Ladd and Arvaniti2000). In this latter case, the fall occurs on a less prominent part of the phrase, making the rise the salient feature. The later rise in polar questions might thus be seen as ensuring that the rise is the predominant pitch movement over the word. The variability across the two sentence modalities in this dataset may also reflect the redundant nature of F0 peak location: both morphosyntactic devices (question particle) and global pitch cues (cf. §3.2.1) could already be sufficient for marking sentence modality.

As discussed in §3.2.2–§3.2.4, the variation in F0 peak placement found in this study supports Dell & Elmedlaoui's (Reference Dell and Elmedlaoui1985) observation that alternation in tonal association is determined by the identity of the nucleus and the weight of the final syllable. We now turn to another of their observations, that variability in the placement of intonational tones can be related to syllabification in intonation phrase final position. Recall that Dell & Elmedlaoui propose an optional rule of prepausal annexation that transforms a syllabic consonant into a coda. Prepausal annexation allows them to make the generalisation that the last tone-bearing unit in the utterance bears the tonal event. Under this assumption, a word such as /tugl/ would be monosyllabic if the F0 peak is on /u/, and disyllabic if it is on /l/. Even though this analysis reflects native-speaker intuitions on syllable count and acceptability judgements on the placement of the H tone, it is not sufficient to explain some of our observations in the present study.

First, disyllabic words containing two vowels can show the same type of alternation. For instance, in /ti.ri/ ‘she wanted’, the H tone can occur on either of the vowels, at least in contrastive statements. For native speakers, words with two vowels are disyllabic, regardless of the position of the H tone. Thus there appear to be cases in which the position of the H tone is not related to a different syllabification of the word.

Second, disyllables with a final heavy syllable, e.g. /tu.glt/, showed F0 peak alternation too. According to Dell & Elmedlaoui (Reference Dell and Elmedlaoui1985: 120), prepausal annexation ‘requires the prepausal syllable to be an open one’. This means that resyllabification cannot account for these cases either. Thus, although the generalisation that the last tone-bearing unit in the utterance bears the tone reflects a clear tendency, it is not absolute.

4.2 Tonal association

We have discussed the representation of F0 peak location in terms of phonological association of a H tone. The tone-bearing units were identified as syllables with sonorant nuclei. However, we have also observed that not every such syllable is associated with a tone, leading to the question as to what exactly determines association in this language.

Gordon & Nafi (Reference Gordon and Nafi2012) argue that tones are associated with metrically strong syllables. They interpret increased intensity and duration on word-final syllables as reflective of metrical strength, and an accompanying F0 peak as a H pitch-accent tone. The results they obtain point clearly to this analysis, as the position of the tonal peak was constant across the different words. The question arises, then, as to why there was so much variation in peak position in our study and so little in theirs. A close examination of the word-set used in their study provides an explanation: of the 22 target words, 14 had a heavy final syllable (a factor leading to a preference for a final F0 peak in our data). Of the remaining eight target words, five had an obstruent in penultimate position and a sonorant in the final syllable (again, this would also have led to a final peak in our data). It is therefore unsurprising that the peak was consistently on the final syllable (although there was some degree of alternation; Matt Gordon, personal communication).

The variability in peak placement found in the productions of all four of our speakers is difficult to reconcile with an analysis of the H tone as a pitch accent on a lexically specified metrically strong syllable. Instead, our results provide support for Dell & Elmedlaoui's suggestion that tones are ‘a property of units larger than words’ (2002: 14). Thus they could be seen as associated with postlexical prosodic structures. In this view, not only tone but also metrical strength is determined outside the lexicon. This is consistent with an account in which the H tone is associated with a constituent higher than the prosodic word, since we have seen that, if the target word contains no sonorants, the tone can be placed on a syllable outside the target word. The semi-spontaneous recordings indicate that polar questions and contrastive statements have edge tones at different levels in the prosodic hierarchy. Further recordings are needed to confirm this, however.

The fact that there is a general rightmost tendency in both questions and statements points to an analysis of the H tone as a right peripheral tone – an edge tone. However, this tone can be associated with specific syllables (if the nucleus of the syllable is sonorant). In line with the proposed analysis of intonation phrase medial tones in Grice et al. (Reference Grice, Roettger, Ridouane and Fougeron2011), one could account for this association as a secondary association of a phrasal edge tone to a tone-bearing unit (Pierrehumbert & Beckman Reference Pierrehumbert and Beckman1988, Grice et al. Reference Grice, Ladd and Arvaniti2000). The F0 peak in both contrastive statements and polar questions could thus be analysed as final edge tones seeking a secondary association to a syllable. This association has been shown to lead to increases in intensity and duration (Katsika et al. Reference Katsika, Krivokapić, Mooshammer, Tiede and Goldstein2014), explaining Gordon & Nafi's observations and their subsequent analysis involving a pitch accent.

In Grice et al.'s (2000) analysis of the ‘Eastern European Question Tune’, edge tones with the possibility of a secondary association to a tone-bearing unit are referred to as phrase accents. These tones share many of the properties of pitch accents if they are associated with a tone-bearing unit. Phrase accents are at once edge tones (‘phrase’) and prominence-lending tones (‘accent’). This analysis incorporates the insights from Gordon & Nafi as to the phonetic characteristics of syllables bearing these tones, and captures the fact that these tones tend to occur near the right edge of phrases. However, more compelling evidence for this account comes from cases where a target word has neither vowels nor sonorant consonants. In §3.2.5 we illustrated three possible patterns that may be obtained. The first pattern has no F0 peak at all. In the second pattern there is a F0 peak on a vocoid within the target word. The fact that the F0 peak is actually on the vocoid means that we cannot a priori rule out an association of tones with syllables containing these vocoids.

However, the third pattern provides evidence against treating syllables with vocoids like syllables with lexical vowels and sonorants. In this pattern the F0 peak is anticipated; it is placed on the word before the target word. Our results point to strong restrictions on the distribution of this pattern: the F0 peak can only be anticipated if the target word has no lexical vowels or sonorant consonants. The presence of a vocoid next to a voiced obstruent does not necessarily cause the peak to occur on the target word, making the analysis of the vocoid as epenthetic (as in the ‘nuclear schwa’ account) problematic.

If the F0 peak is anticipated in words made up entirely of voiceless obstruents, there cannot be a vocoid on the target word at all. Thus, even though tonal placement cannot be explained in the ‘nuclear schwa’ account, an adequate treatment of the prosodic structure needs to make reference to the vocoid: a peak before the target word and a vocoid breaking up a voiceless cluster in this target word are mutually exclusive. Thus the vocoid is relevant for the intonational description of the language, even if it might be the result of prosodically conditioned gestural organisation.

The question arises as to how the placement of the edge tone can be accounted for without an association to a tone-bearing unit (i.e. a syllable with a vowel or a sonorant consonant) in the target word. There is evidence supporting an analysis in which it would simply align with an element with enough voicing or energy to make it audible. This analysis is also compatible with Gussenhoven's (Reference Gussenhoven, Broe and Pierrehumbert2000) treatment of non-peripheral placement of edge tones. In his treatment, a distinction is made between ‘association’, equivalent to what we refer to as secondary association to a tone-bearing unit, and OT ‘alignment’, where a tone is placed close to another tone or a phrase edge, but without reference to any particular syllable. In Gussenhoven's approach, a target word with no lexical vowel or sonorant would require the H tone to simply align with voiced material as close to the edge as possible.

The observed patterns of tonal association need to be considered in the light of language typology. Although free alternation of secondary association is so far unattested, some degree of alternation is observed in a number of languages, depending on the properties of syllables at or near the phrase edge. For instance, in Standard Greek, we find secondary association of a phrase accent to a lexical stress if there is one available; otherwise the edge tone is close to the phrase edge. Moreover, in Cypriot Greek the phrase accent is always at the edge (Grice et al. Reference Grice, Ladd and Arvaniti2000, Arvaniti et al. Reference Arvaniti, Ladd and Mennen2006, Arvaniti Reference Arvaniti, Cohn, Fougeron and Huffman2012).

Cross-linguistically, there are stress and tone systems that are prominence-driven (e.g. de Lacy Reference de Lacy2002, 2007). Many languages have weight-sensitive word stress or tone systems (see Gordon Reference Gordon2006 for an overview). Sonority is a well-known determinant of phonotactic patterns and syllable structure. However, languages in which sonority determines structures at higher levels of the prosodic hierarchy are typologically rare. While sonority certainly plays a role in many lexical tone systems, the sonority of segments can also determine word-stress patterns, although this is less common (cf. de Lacy Reference de Lacy and de Lacy2007 for an overview). Thus, both syllable weight and sonority are known to have an impact on prosodic regularities within word-prosodic domains. An impact of these factors on prosodic constituents higher in the prosodic hierarchy has also been reported in the literature, as in Japanese, which allows secondary association of edge tones to sonorant moras, but not to non-sonorant moras (Pierrehumbert & Beckman Reference Pierrehumbert and Beckman1988). Nonetheless, no language has so far been reported to have such a high degree of variability in the extent to which all of these factors together affect the placement of tones.