Repair strategies in labial dissimilation : Diminutive formations in Xitsonga

This paper presents phonological processes in Xitsonga diminutives. The round vowels /u/ and /o/ are changed into the glide [w] to avoid vowel hiatus. When the glide [w] is preceded by labial consonants, then other processes occur: either the labial nasal [m] corresponds to a velar, or the glide [w] deletes when preceded by labial obstruents. The seemingly disjunctive processes are argued to be responses to the OCP-LABIAL constraint in which adjacent segments with [labial] are prohibited. A rule-based analysis and a constraint-based (Optimality Theory) analysis are presented to discuss these processes. Data with affrication is introduced as an alternative repair strategy, but is not included in the formal analysis, which focuses on labial dissimilation.


Introduction
In Xitsonga diminutive formation, the juncture between the root and the suffix undergoes various phonological processes: glide formation, velarisation and vowel deletion.We argue that these seemingly disjointed processes are results of repair strategies that apply in order to avoid a sequence of labial sounds.
(3) Phonological patterns in Xitsonga diminutives Vowel hiatus motivates the deletion of (unrounded) vowels.Velarisation of the root consonant /m/ to [ŋ] and the round-vowel deletion after /f/ and other labial sounds are argued to be the result of labial dissimilation.Phonological processes in diminutive formation are found in other southern Bantu languages as well.In Northern Sotho, when the stem-final vowel is a back vowel ([o] or [u]), it becomes the labial glide [w] (cf.Kotzé and Zerbian 2008:12), just as in Xitsonga.In this paper, a rule-based analysis and a constraint-based analysis (the latter in terms of Optimality Theory (OT)) will be discussed in order to explain various phonological processes that occur in the diminutive formation in Xitsonga.
In section 2, facts about Xitsonga and the data collection methods are presented.This will be followed by a data-oriented section that aims to provide a comprehensive picture of phonological patterns of Xitsonga diminutive formation.The analyses in section 4 argue that phonological processes found in Xitsonga diminutives are repair strategies in labial dissimilation.In section 5, a short discussion about the affrication data will be presented.

Background and data collection
Xitsonga The data collection method was aided by a Xitsonga noun corpus built from a Xitsonga dictionary (Cuenod 1967).The Microsoft Access database consists of 1582 nouns, which the second author identified as being in active use in her own speech.Where there were any discrepancies between the second author's pronunciations and those of Cuenod, these were resolved by following the second author's speech form.A script that creates a reverse string list 1 The status of the velar fricative [x] has been cross-examined with native speakers of Xitsonga.In Baumbach (1987), a single ideophone hhatawu [xatawu] ('of keeping very quiet') is listed.Linguistically trained speakers of Xitsonga were unable to identify a single word containing the [x] sound in the language.Lee (2009) confirmed that his consonant chart was directly taken from Baumbach (1987) prior to any consultation with Xitsonga speakers.We conclude that the variety of Xitsonga in this paper does not have the velar fricative [x].
sorted nouns from the last character in the orthography.2Any items that we were unsure about were double-checked by the second author with other Xitsonga native speakers.

Diminutive formations in Xitsonga: Data
Various phonological processes in Xitsonga diminutives have been reported in previous studies (cf . Baumbach 1987:148-157, Ziervogel 1967:216-234), but the reported paradigm has gaps in terms of phonological environments.This section aims to fill these gaps.The root-final unrounded vowels /a, e, i/ undergo vowel deletion in most environments.As such, this section will only report on roots that end with the round vowels /o/ and /u/.

Glide formation
The root-final round vowels /o/ and /u/ correspond to the labial glide [w] in diminutive forms.
The consonant preceding the root-final vowel can be an alveolar, a palatal, a velar or a glottal.In ( 5), the root-final vowel is /o/.The collected data contains accidental gaps because no roots have the following /-Co/ forms, in which the consonant /C/ is [ʂ, ŋ, ʧ, ʤ] or breathy voiced.

Velarisation of the bilabial nasal [m]
When the root-final round vowels /o/ and /u/ are preceded by the bilabial nasal /m/, the labial corresponds to the velar [ŋ] in the surface form.In ( 7), the labial nasal is followed by /o/, and in (8) it is followed by /u/.

Deletion after the labial continuants [f], [pf], [bv] or [w]
When root-final round vowels are preceded by the oral labials (/f/, / (m) pf/, /bv/ or /w/), the round vowels delete.The data in this section fills a gap in the previous studies.

Section summary
The generalisations so far are three-fold.First, in vowel hiatus contexts, round vowels correspond to the labial glide [w] in the surface form.Non-round vowels do not usually have correspondents and simply delete before the diminutive suffix, as in (10a).Second, when the consonant preceding round vowels is an oral labial ( These generalisations will be analysed in terms of a rule-based analysis and a constraint-based analysis (OT) in section 4.

An Optimality Theory (constraint-based) analysis
The data analysis in this section involved the use of a constraint-based theory, namely Optimality Theory (OT; Prince andSmolensky 1993/2004) as the framework.In this theory, the avoidance of sequences of [labial] segments is posited as a competition between faithfulness constraints ( 14) and markedness constraints ( 15).[ʃipotoana] W * L c.
[ʃipotana] W * L The ranking when the labial nasal velarises is shown in ( 17).The faithful candidate in (17b) violates the *VV constraint.In (17c), the candidate avoids *VV violation by changing the round vowel to a labial glide, but this incurs a violation of OCP-LABIAL (in addition to the violation of IDENT(CONS)).The deletion of the input round vowel violates MAX-V[LAB] in (17d).The [ʃinomoana] W * L L c.
[ʃinomana] W * L L In the deletion cases, the faithful candidate in (18b) violates the *VV constraint.Changing the underlying /f/ to the velar fricative [x], as in (18c), violates the constraint *[x] which is motivated by the phonotactic ban on this sound in Xitsonga.The glide formation candidate in (18d) violates the higher-ranked OCP-LABIAL constraint.The optimal candidate in (18a) violates the MAX-V[LAB] constraint because the input /o/ does not have a correspondent in the output, but it satisfies other markedness constraints that are ranked higher.[ʃisefoana] W * L c.
[ʃimaljana] W * L W * So far, we have proposed two analyses of Xitsonga diminutive formation: in (i) a rule-based theory and in (ii) a constraint-based theory (OT).While both analyses can account for the surface patterns in Xitsonga diminutives, an OT analysis provides a clear reason why these processes occur -as an avoidance measure against two labials or two vowels in the surface form.
When the final round vowel is preceded by the three labial obstruents [p, m b, β], affrication is employed.This is another type of repair strategy that will be discussed in the following section.
The variation involving affrication in diminutive construction does not have a simple explanation.A detailed study of this phenomenon is left for future studies as it is beyond the scope of the current paper.

The motivation behind affrication as a repair strategy of labial dissimilation
The data in section 5.1 suggests that the process of affrication is not as uniform as the other processes described in section 3.In this section, instead of developing a formal analysis of the affrication process, we will present a short discussion thereof, drawing from cross-linguistic comparisons.
An analogous affrication pattern is attested in isiZulu.Labial consonants in the root alternate with palatal consonants in the diminutive construction (Doke 1926:127 in Bennett 2013:335).
Both in isiZulu and in Xitsonga, the last consonant of the root changes to a palatal affricate.In isiZulu, this phenomenon is also found in the passive construction, while in Xitsonga the passive does not show affrication. ( An analysis of such palatalisation is proposed in Kotzé and Zerbian (2008).In discussing palatalisation in the Sotho languages, they argue that this process is triggered either by an overt palatal glide or by a palatal glide that emerges as a result of two phonological processes, namely glide formation and labial dissimilation.In Xitsonga, it is possible to surmise that such diachronic change has occurred, where affrication would have occurred after the frication process.
Two theories can possibly account for this process.First, the Perceptual Cues theory, proposed in Ohala (1981) and Steriade (2009), would argue that affrication in Xitsonga or palatalisation in isiZulu as a repair strategy in labial dissimilation is perceptually driven.Second, the Consonant Mutation theory, reported in other Bantu languages as well, would argue that a mutation to an affricate is due to historical super high vowels (Zoll 1996).The super high vowels would have changed to an affricate at some point in the language's history.
The confounding factors regarding affrication are, firstly, that Xitsonga also has affrication where the liquid consonants themselves change to an affricate, as in ( 29).This type of affrication might also be triggered by a type of dissimilation.Secondly, the affrication of liquids behaves similarly to isiZulu palatalisation or Bantu consonant mutation because the root consonant itself changes to an affricate and the root-final vowel is deleted.

Conclusion
In this paper, we have shown that disjointed phonological processes can be understood as closely related repair strategies driven by OCP-LABIAL.A response to vowel hiatus by round vowels is that they correspond to a labial glide (/o/ or /u/ → [w]).When these round vowels are immediately preceded by labial consonants, two repair strategies occur in response to labial dissimilation: either the labial nasal /m/ corresponds to the velar nasal [ŋ] or the round vowels delete after [f, pf, bv, w].
Round vowels occasionally correspond to an affricate.This pattern may be a repair strategy for labial dissimilation, but there are many exceptions.More research is needed to understand the exact nature of the affrication process from a diachronic as well as a phonetic perspective.
[f],[pf],[bv]  or[w]), round-vowel deletion takes place; the same pattern is observed in non-round vowels.While this generalisation also applies to other oral labials ([p], [ph], [b], [ m b], [β]), the former set of oral labials undergoes affrication in addition to vowel deletion (see section 5.1 for further discussion).Third, when the consonant preceding round vowels is /m/, the round vowels correspond to [w] and the /m/ corresponds to the velar nasal[ŋ].Non-round vowels do not have such a correspondent relationship, as illustrated in (10b).(10)Phonological processes when the root-final vowel is round a. Changes in the root-final vowel i

( 14 )
Faithfulness constraints (cf.McCarthy and Prince 1995) a. MAX-V[LAB]: Assign a violation mark when a round vowel ([labial]) in the input does not have a correspondent in the output.b.IDENT(LAB): Assign a violation mark when corresponding segments in the input and the output do not have identical values for the [labial] feature.c.IDENT(CONS): Assign a violation mark when corresponding segments in the input and the output do not have identical values for the [consonantal] feature.(15) Markedness constraints a. OCP-LABIAL: Assign a violation mark to labials that are adjacent.(cf.Myers 1997) b. *VV: Assign a violation mark to vowels that are adjacent.(cf.Casali 1996) c. *[x]: Assign a violation mark to the velar fricative [x].(Phonotactic constraint) d. *Cj: Assign a violation mark to palatalised stops. 4(Phonotactic constraint) The tableau 5 in (16) shows that the two markedness constraints *VV and MAX-V dominate the IDENT(CONS) constraint.In the optimal candidate in (16a), the root-final round vowel has a correspondent [w] in the output.Vowel hiatus is repaired by changing the vowel to a glide.The faithful candidate in (16b) violates the *VV constraint.The deletion candidate in (16c) satisfies the *VV constraint, but it violates the MAX-V[LAB] constraint.(16) Glide Formation: OCP-LABIAL, *VV, MAX-V[LAB], IDENT(LAB) >> IDENT(CONS) optimal candidate violates two faithfulness constraints: IDENT(LAB) due to the change of place from [m] to [ŋ], and IDENT(CONS) because /o/ corresponds with [w].(17) Velarisation: OCP-LABIAL, *VV, MAX-V[LAB] >> IDENT(LAB), IDENT(CONS) /ʃi-nomo-ana/ OCP- (also called Changana or Rhonga) is spoken by about 2 million people in South Africa, Zimbabwe, Mozambique and Lesotho.In South Africa, Xitsonga is one of the 11 official languages and is spoken mainly in the northeastern part of the Limpopo province.Most Xitsonga speakers know English, but many of them use only Xitsonga in everyday life.It is also possible to receive education up to a tertiary level in the language.Xitsonga speakers share the linguistic sphere with Tshivenda and Northern Sotho speakers.The Xitsonga variety reported in this paper is the Xiluleke dialect spoken in the Mhinga area of Limpopo, South Africa.