Issues about tone rules in Xitshwa

This paper examines High Tone Spread (HTS) within and beyond verb words in Xitshwa, an undocumented Bantu language (S51, Guthrie 1967-1971) spoken in southern Mozambique, Inhambane province. High Tone Spread has been attested in a variety of Bantu languages (Zerbian 2006, Mutonyi 1996, Roberts-Kohno 1996, Polleto 1998, Hyman 2007, Hyman 2009). It is a phenomenon whereby a High tone is not only heard on the sponsor syllable, but may also be heard on other syllables –be it to the right or to the left. Kisseberth (1994) postulated a Domain-based Approach that accounts for the extent of High spreading. According to this author, High tone only spreads throughout a Prosodic-domain (P-domain). The critical feature in Kisseberth’s work is that whether or not a High tone spreads onto a following syllable is always and only a matter of whether the syllables are in the same P-domain. Our findings rely on Kisseberth’s description of domain sensitive HTS, especially in contexts where a complement follows the verb (Syntax-Phonology Interface). Some of our data give evidence of contexts in which, although the rule of HTS is motivated, High tone fails to spread onto the noun complement.


Introduction: a domain-based approach to phonology
The domain-based approach we wish to follow was developed by Kisseberth (1994). In this approach this author assumes that prior to the application of a rule to a representation, there is an assignment of domain-structure to the representation. In other words, if there is a new phenomenon in a language, a theory must be developed in order to account for this phenomenon (Cassimjee & Kisseberth 1998). Kisseberth defines a domain as a sequence of phonological material enclosed by a left and right bracket. Domains are of different types, and the domain that is of most important to our analysis is what Kisseberth refers to as Prosodic-domain (P-domain). In his analysis, Kisseberth claims that High tones spread rightwards inside a P-domain and that High Tone Spread (HTS) does not extend beyond a given P-domain. Kisseberth postulated three lexical rules that project P-domain brackets, namely: (i) High Tone Projection Rule -this rule says that a High-toned syllable projects a Left bracket to its left. In other words, a syllable with a High tone starts the domain and it is always at the left edge of the P-domain; (ii) Final Projection Rule -this rule says that a toneless syllable at the end of a non-verbal lexical item projects a Left bracket to its Left. In other words, a final toneless syllable in non-verbal categories cannot be included into the same domain as a preceding High (this rule is also referred to as Non-finality constraint -see Cassimjee & Kisseberth 1998, Selkirk 2009) and (iii) Pre-High Projection Rule -this rule says that a toneless syllable followed immediately by a High tone in the same word projects a P-domain bracket to its Left. That is, toneless syllable followed by a High tone in the same word will not be included in the same P-domain as the preceding High.
Our description will focus on the first two rules, namely: High Tone Projection and Final Projection. We indicate verbal stem 2 boundaries with a dash (-). We neither address nor indicate penult lengthening.
Given the above premises, we now move forward to the facts of Xitshwa.
The rest of the paper is organized as follows: in § 2, we present background information about tone in Xithswa. In § 3, we present linguistic evidence about tone in Xitshwa; in § 4, we discuss High tone spread within and beyond verb word and the conclusion in § 5 summarizes the key points discussed in our paper.

Background information on Xitshwa tone
3 Among the major questions of concern regarding Xitshwa grammar is whether or not Xitshwa is a tone language and if it is a tone language, what linguistic evidence can be adduced to prove it? The available data suggest that in Xitshwa there are some High tones that are provided in the verb by input segments (i.e. some subject prefixes, see below), while other tones are provided as part of tense-aspect-mood morphology. Unlike other Bantu languages, in Xitshwa there seems to be no lexical High tones determined by the selection of a verb root. All High tones in the verb are determined by general phonological principles and morphological information (Odden 1995).
According to Ngunga & Faquir (2011), Xitshwa displays two basic tones, namely High (H) and Low (L), common to many Bantu languages (Zerbian 2006). As it is also common in other Bantu languages, it has been shown that in Xitshwa tone plays both a lexical and a grammatical role. The Bantu literature generally assumes that the H vs L contrast can be lexically expressed as H vs Ø (Kula & Braun 2015, Hyman 2010).
2. Throughout this paper, we stick to the following terminology: 'root' refers to the barest lexical entry of a given verb while 'stem' begins with the root and extends to the end of the verb, including tense morphemes and the final vowel (Hyman & Ngunga 1994, Odden 1998). 3. The data were collected in Maputo district, southern Mozambique with two native speakers of Xitshwa. We built a list of ten verbs in Xitshwa language based on different verb root structures. Then a questionnaire was made. This questionnaire was submitted in Portuguese to Xitshwa native speakers, but respondents were instructed to answer in their mother tongue. Therefore, our target respondents were native speakers of Xitshwa, but with background knowledge of Portuguese. The researcher is also a native speaker of Xitshwa.
Several accounts on Bantu languages also suggest that Low tones do not play any role in the Phonological Component and that only H should be specified (Downing & Mtenje 2017, Hyman 2010, Hyman & Ngunga 1994, Marlo 2016, Odden 1998, Polleto 1998. Ngunga & Faquir (2011) have suggested that the contrast between H vs L in Xitshwa should be L vs Ø. That is, only L should be specified while H should remain underspecified. However, Hyman (2010) observes that the L vs Ø contrast, although confirmed for some languages, is very rare. In this analysis, we do not take a strong position on this matter, but assuming that only High tones are active in Xitshwa, we propose to specify High tone with acute accents (΄), while syllables with Low tones will be underspecified, yielding a H vs Ø system. Indeed, Xitshwa has rules of High tone spread and High tone shift, but it does not have Low spreading or Low shifting.
As referred earlier, Xitshwa lacks lexical tonal contrasts in the verb stems. Instead, tone in the verb stem is determined by some factors. Firstly, the actual tone of the verb is the result of underlying H-toned subject prefixes. Secondly, there are tones assigned to the verb by tenseaspect-mood sensitive rules. Thirdly, some tones result from tone spreading and shifting (Liphola 2001). Our discussion will focus on the High provided to the verb stem by H-toned subject prefix.
Xitshwa displays two types of subject prefixes, namely (i) H-toned subject prefixes, á-'s/he' and vá-'they', and (ii) L-toned subject prefixes, nzi-'I', hi-'we', u-'you' (sg), mu-'you' (pl). When the verb is associated with a H-toned subject prefix, this subject prefix will sponsor the High to the following toneless syllables of the verb or phrase through a general principle referred to as High Tone Spread, as in vátá-fámba 'they will go'. If the verb is associated with a H-toned subject prefix, but the High does not surface on the sponsor syllable, it will surface on other syllables within the verb through a principle referred to as High tone shifting, as in vanga-fámbi 'they must not go'. Therefore, we assume that Xitshwa is both a High Tone Spread and a High Tone Shift language. For those verb forms that display a High tone which is not associated to any underlying input segment, the High tone is assumed to be provided as part of tense-aspect morphology, as in hinga-fámba '(we) who went'. This suggests that tone may be a property of various morphemes that constitute a verb form in Xitshwa. This paper is limited to the analysis of High tone spread.
Moreover, a High tone can be added to an infinitive verb stem, yielding gerund forms (roughly equivalent to English Verb + -ing), as shown in (3) In each case, a High tones seem to appear on the initial syllable of the stem, from where it spreads to the remaining syllables. As the data in (3) show, if the verb stem is monosyllabic, the High tone will surface on this only syllable. If the verb stem has two syllables, the High tone will surface on the two syllables of the stem. If the verb has three or more syllables, the High tone also associates with all the following toneless syllables but the penult. The tonal behaviour in the infinitive can be explained by assuming that the High tone is inserted through a primary rule of association of the High tone with the initial syllable of the verb stem.
The rule that accounts for High tone insertion in the infinitive is depicted graphically in (4) as follows (Liphola 2009): This paper does not address High tone insertion in any detail. Xitshwa also has a rather marginal contour tone (HL), which is sensitive to the past perfect tense in association with a lexically H-toned subject prefix á-'s/he', as in (5). (5) HL tone in the past perfect â-gili ~ (áa-gili) 's/he had eaten' â-tili ~ (áa-tili) 's/he had come' â-tsutsumili ~ (áa-tsutsumili) 's/he had run' â-tshamili ~ (áa-tshamili) 's/he had seated' â-wili ~ (áa-wili) 's/he had fallen' â-tsalili ~ (áa-tsalili) 's/he had written' According to Selkirk (2009: 6): "in Xitsonga, all and only vowels that are penultimate in the clause are long (V:). There is no lexical vowel length contrast; this penultimate lengthening is i n t r o d u c e d b y t h e phonology". Our data seem to concur with the variety studied by Selkirk. However, in our variety of Xitshwa, long vowels can also occur in the verb-initial position, though they are not preferable. The underlying form of the verbs in (5) have long vowels /áa-/ in the initial syllable, where /á-/ is the third person subject marker and /-a-/ is a tense marker. These vowels have been fused into one vowel and this fusion has produced a single vowel with a HL tone pattern (â). Therefore, the following observations can be made: (i) in Xitshwa, there is a vowel length in verb-initial position, though this vowel length is not lexical and it is not preferable, and (ii) in Xitshwa, there is tone preservation (or tone stability)-that is, when a vowel deletes, the tones this vowel associated with are not deleted (for tone stability, see Hyman 2007& Odden 1995. Actually, in Xitshwa, there are a number of phonological processes that can be applied to avoid long vowels in verb-initial position, such as (i) glidingwhenever there is a long vowel length in word-initial, a semi-vowel can be inserted, as áwá-gíli derived from áa-gili, or (ii) contour tone as â-gili, also derived from áa-gili. Be that as it may, it is worth noting that HL contour is contrastive in Xitshwa, as in (6) Note that in (6a), HL contour does not spread. This suggests that there is no automatic contour tone spread in Xitshwa. By contrast, in (6b), the High tone of the subject prefix spreads until the penultimate syllable of the verb stem.

High tone spread in Xitshwa
As referred earlier, our analysis follows Kisseberth's (1994) analyses. We propose the existence of a rule of High Tone Spread (HTS) in Xitshwa. This rule claims that High tones spread rightwards inside a P-domain and that HTS cannot extend beyond a given P-domain. We will use verb words as our source of data, and we will claim the syllable to be the tone bearing unit (TBU). In Xitshwa, as in most Bantu languages, High tone may spread to the right onto toneless syllables in a long-distance fashion (Hyman 2009, 2007, Kisseberth 1994, Selkirk 2009). Given the three and described above in § 1, our analysis will rely on the first two lexical rules projecting P-domain brackets proposed by Kisseberth (1994), namely High Projection Rule and Final Projection Rule (see § 1 above).

Tonal pattern within a verb word
In what follows, we describe tonal patterns within verb words. Such verbs words constitute sentences on their own. In this context, the verb root is in final position in the phrase. These verb words are associated with a Htoned subject prefix vá-'they'. As the data in (7) show, the High tone is initiated by the subject prefix, but also associates with the following toneless syllables to the right.
(7) High tone in the simple past vá-gíli 'they ate' vá-tséli 'they wrote' vá-wíli 'they fell' vá-wóníli 'they saw' vá-xávíli 'they bought' vá-nyíkéli 'they gave' vá-fámbíli 'they went' vá-tsútsúmíli 'they ran' vá-hlóngólíli 'they chased' Following Kisseberth's High Projection Rule, we assume that the Htoned syllable váon the left edge starts the P-domain, then we hypothesize that this High tone spreads rightwards onto the toneless syllables of the verb stem. Note that all the examples in (7) do not bear a High tone on the last syllable. We assume that the Final Projection Rule (or Nonfinality constraint) accounts for the failure of High tone spread to the domain-final syllable of the verb. Recall (see § 1 above) that the Nonfinality constraint (Cassimjee & Kisseberth 1998) bans a High tone from spreading to the domain-final syllable of the verb. Since the High tone only spreads throughout a P-domain, we know that the final syllable (-li) of the above verbs must be outside the P-domain initiated by the H-toned third person subject prefix vá- (Kisseberth 140). Applying Kisseberth's rule on High Tone Spread we will have the following representation [váfambi] [li]. Kisseberth refers to this kind of representation as Bracket Matching Convention (BMC).
In Xitshwa, there is a rule that deletes the last syllable of the verb (-li) in the simple past tense. In other words, the last syllable is omitted in fast speech. When such a rule applies, a HL contour is produced at the end of the verb domain, e.g. vá-fámbî 'they went' derived from vá-fámbíli, as shown in (8) below: (8) HL Contour in the simple past vá-gî 'they ate' vá-tsê 'they wrote' vá-wî 'they fell' vá-wónî 'they saw' vá-xávî 'they bought' vá-nyíkê 'they gave' vá-fámbî 'they went' vá-tsútsúmî 'they ran' vá-hlóngólî 'they chased' An explanation for the appearance of a verb-final HL contour in (9) may be found by assuming that the deletion of the L-toned syllable did not mean a total loss of its tone-the Low tone was preserved and associated with preceding High toned syllable. In § 4.2 we shall see what happens to this HL contour when a complement is added to the verb.
As was observed by Kisseberth, in our Xitshwa variety monosyllabic verb stems also show a violation of the Final Projection Rule (or Nonfinality constraint) as the High tone from the subject prefix spreads onto the final syllable due to a principle referred to as minimality condition, which requires that a High tone target at least two syllables in a word (Leben 2006 andKisseberth 1994): (9) High tone on final syllable (minimality condition) vá-yá 'they are going' vá-gá 'they are eating' vá-tá 'they are coming' vá-nwá 'they are drinking' vá-wá 'they are falling' vá-fá 'they are dying' vá-khá 'they are fetching' vá-lwá 'they are fighting' It seems clear that, in this case, High tone is associated with at least two consecutive syllables, regardless of the Non-finality constraint which excludes a High tone on the final syllable.
In what follows, we shift our analysis to the contexts where a complement noun follows the verb. In Bantu literature this is referred to as Syntax-Phonology Interface. We shall see that High tone spread also has some limits in this context.

Tonal pattern where a complement follows the verb
Let us first establish, following Kisseberth's (1994), the existence of the interaction of verbal tone with a following noun complement. According to Kisseberth, when a verb ends with a High tone, this High tone extends throughout the following noun complement as far as the penultimate syllable. Our findings concur with Kisseberth's analysis. The phrases under scrutiny all consist of verbs with a lexically H-toned prefix váand toneless nouns.
But first we would like to resume examples described in (8) where a verb-final contour tone was produced due to a process of syllable deletion. Recall that when the final syllable of a verb form such as váfámbíli 'they went' is deleted, the Low tone of the deleted syllable is preserved so that it associates with the preceding syllable and a contour tone is produced, as in vá-fámbî. The question that we would like to answer is whether this contour tone is preserved when a noun complement is added do the verb. Notice that the nouns in (10) are toneless.
(10) vá-gí nyáma 'they ate meat' vá-tsé bhúku 'they wrote a book' vá-xáví tímánga 'they bought peanut' vá-nyíké páhla 'they gave cloth' The examples in (10) show that the rule of last syllable deletion still applies even when a noun complement follows the verb. However, the HL contour is not maintained in these contexts due to a rule of interaction of verbal tones with the following noun complement.
In the examples below, we compare tone pattern in minimal pair phrases. As the data in (11) show, in the first examples, High tone spreads as far as the penultimate syllable of the noun; in the second examples, High tone fails to spread across the verb. Here, the examples were taken from Kisseberth, with some difference in orthography as per our variety.
(11a) vá-xává nyáma 'they are buying meat' (11b) vá-xáva nyama 'they buy meat' (11c) vá-xává tíngúwu 'they are buying clothes' (11d) vá-xáva tinguwu 'they buy clothes' (11e) vá-kúmá xíhlámbétwána 'they are getting a cooking pot' (11f) vá-kúma xihlambatwana 'they get a cooking pot' (11g) vá-tísá xíhótlóvíla 'they are bringing a giant' (11h) vá-tísa xihotlovila 'they bring a giant' (11i) vá-ríválélá zvíphúkúphúku 'they are forgiving fools' (11j) vá-ríváléla zviphukuphuku 'they forgive fools' The important fact to note in (11) is that in the first examples of each pair (a, c, e, g, i) the High tone spreads onto and beyond the final syllable of the verb throughout the noun as far as the penultimate syllable. The final syllable of the noun complement is placed in a different P-domain, yielding the following representation for (i): [vá-rivalela zviphukuphu] [ku]. However, the second examples of each pair (b, d, f, h, j) exhibit a different tone pattern. Here, High tone spread does not go beyond the penultimate syllable of the verb and therefore all the syllables of the noun complement remain toneless. This second series of examples shows that, in this context, there is a lack of interaction between the verb and its complement noun. Given the form vá-ríváléla zviphukuphuku, the last syllable of the verb and the rest of the noun complement are set off into a separate P-domain. This would yield the following representation for (j): [vá-rivale] [la zviphukuphuku]. The data in (11) suggest that the rule accounting for interaction of verbal tone with a following noun complement is not fulfilled in all contexts.
Kisseberth has observed that when a verb is followed by a double object such as vá-xávélá xíphúkúphúku fole 'they are giving a fool tobacco', the High tone does not spread to the second object. This observation is confirmed by our data in (12): (12a) vá-xávélá xíphúkúphúku fole 'they are giving a fool' (12b) tobacco' vá-xávéla xiphukuphuku fole 'they give a fool tobacco' (12c) vá-xávélá xíkóxa pahla 'they are buying an elder clothes' (12d) vá-xávéla xikoxa pahla 'they buy an elder clothes' (12e) vá-nyíká xíphúkúphúku mali 'they are giving a fool money' (12f) vá-nyíka xiphukuphuku mali 'they give a fool money' The examples in (12) show the same tonal pattern as observed in (11). In the first series of examples (a, c, e), the High tone only fails to spread to the second object, while in the second series of examples (b, d, f), the High tone fails to spread beyond the verb, which establish, once again, that the verb and its following noun complements are set in a different Pdomain. Two observations can be made as regarding High tone spread: (i) In Syntax-Phonology Interface, High tone spread is limited to the first object and (ii) morphosyntactic factors (such as TMA) may dominate the rule of High tone spread.

Conclusion
This paper represents an initial study about tone in Xitshwa, one of the undescribed languages in Southern Mozambique. Generally, research about tone remains one of the least developed studies as regards Bantu languages spoken in Mozambique (Liphola 2009). The present study showed that Xitshwa is a tone language and has two main tonal levels.
Occasionally, a contour tone may be found -though more detailed work is needed to fully understand this phenomenon. The study also has shown that the Final Projection Rule (or Non-finality constraint) is highly motivated-though it may be violated due the size of the verb stem. High tone spread in Xitshwa seems to always apply from left to right, being different from Kimwani (a language spoken in northern Mozambique), which spreads High from right to left (Liphola 2009). Kisseberth has claimed that the critical condition on High tone spreading (HTS) is whether the sponsor syllable and the syllable affected by HTS are in the same P-domain, so that any rule that accounts for High tone spread must operate within the P-domain. The study has shown that the notion of Pdomain varies according to different Syntax-Phonology interface contexts.