2. The Nivaĉle phonemic inventory and basic phonotactics

Nivaĉle [niβak͡lé] is a Mataguayan language spoken in the Argentinean and Paraguayan Chaco, with approximately 14,768 speakers in Paraguay (DGEEC 2012) and 500 in Argentina (INDEC 2004–2005). The Chorote, Maká, and Wichí languages, which also belong to the Mataguayan family, occupy the nearby regions. The location of the Mataguayan languages and peoples spans Northeastern Argentina, Southeastern Bolivia, and Southwestern Paraguay, in the region known as the Gran Chaco. The work described in this article builds on a larger documentation project that took place in the Nivaĉle communities of Uj'e Lhavos and Santa Teresita (Paraguayan Chaco) between 2009 and 2013.

The phonemic inventory of Nivaĉle is quite stable across all of its varieties, with six vowels and twenty-one consonants. The vowels are /i e a ɑ o u/. The central vowel /a/ patterns with the front vowels /i/ and /e/ in processes of palatalization. In addition to these plain vowels, Stell (Reference Stell1987: 97) postulates a phonemic contrast with the glottalized vowels /ḭ ḛ a̰ ɑ̰ o̰ ṵ /. I propose instead that Nivaĉle glottalized vowels are underlying sequences of /Vʔ/ and, given that there is a consistent correlation between glottalized vowels and the locus of stress, that a postvocalic glottal stop is underlyingly moraic. If the glottal segment is aligned with the right edge of the syllable domain, the glottal stop will be parsed directly to the syllable node as a coda (3a). On the other hand, if there is another consonant intervening between the glottal stop segment and the right edge of the syllable, then the mora (and its associated /ʔ/ features) will be parsed directly into the nucleus of the syllable (3b). In other words, only if the coda position is already filled by another consonant will the glottal stop be parsed into the nucleus; a complex nucleus emerges at the expense of not creating a complex coda (an illicit syllable structure in this language).

There are thus two phonetic manifestations of underlying /Vʔ/ sequences; (i) a vowel-glottal coda sequence [Vʔ] (3a), and (ii) creaky [V̰] (3b).

Because the variable prosodic parsing of the Nivaĉle postvocalic glottal stop has effects on the syllable structure (i.e., creating either closed syllables or complex nuclei), in this article I maintain the separate transcriptions above of these two types of phonetically-glottalized vowels. I provide both phonemic and phonetic transcriptions wherever necessary.

Table 1 presents the 21 phonemic consonants of the language. The segments in square brackets represent allophonic variants of the segments to their left. Transcriptions in this article generally follow IPA conventions, but primary stress will be represented with an acute accent and secondary stress with a grave accent.

Table 1: Nivaĉle consonants

Like other Mataguayan languages, Nivaĉle has a two-way laryngeal distinction in non-continuant obstruents (plain vs. ejectives) – except for the complex segment /k͡l/ – but no voicing contrast. Nivaĉle also displays a typologically unique lateral system. There are two lateral obstruents: the lateral fricative /ɬ/ and the complex segment /k͡l/, but no lateral approximant. It has been argued (Gutiérrez Reference Gutiérrez2019) that the complex segment /k͡l/ is neither an affricate (there is no fricative release and the sequence of two phases does not agree in voicing), nor a consonant cluster. This complex segment consistently simplifies to [k] in coda position.

The core syllable structures in Nivaĉle are CV, CVC, CCV and CCVC. There are no onsetless syllables in the language, either word-initially or word-medially, and the glottal stop serves as the preferred epenthetic consonant.

All Nivaĉle consonants may appear as singleton onsets or codas, except for the ejectives /p’ t’ k’ t͡s’ t͡ʃ’/ which occur only in onsets. Several authors (Itô Reference Itô1986, Itô and Mester Reference Itô, Mester, Merchant, Padgett and Walker1994, Lombardi Reference Lombardi1995) have pointed out a crosslinguistic restriction against Laryngeal and Place occurring jointly in coda position. In that regard, Nivaĉle ejectives neutralize to their plain counterparts in this syllabic position.

Onset clusters (at most two consonants) can occur in word-initial position; coda clusters are never attested. Tables 2 and 3 show the permissibility of consonant clusters in word-initial and word-medial position. These data were taken from Seelwische's (Reference Seelwische1990) dictionary and my own fieldwork. The set of segments heading the rows indicates the first element of the cluster (C1), and the set of segments heading the columns indicates the second element (C2). While empty cells indicate attested clusters, grey cells indicate unattested clusters. In Table 2, the (+) sign indicates that the sequence is only attested across a morpheme boundary.

Table 2: Initial CC Clusters in NivaĉleFootnote ²

Table 3: Medial CC Clusters

Nivaĉle initial clusters do not involve traditional branching onsets with a rise in sonority (e.g., tj, tw), but they never show a sonority fall either; there are no sonorant-obstruent (*SO) sequence (i.e., *Nasal/Glide-Obstruent). However, there can be obstruent-obstruent sequences, as illustrated by the fricative (C1) + stop/ejective/affricate/fricative (C2) sequences. In fact, fricatives are the least constrained members of CC-initial clusters.

The following examples show word-initial clusters in alienable nominal roots – that is, roots that do not require the presence of an obligatory possessive prefix – and predicative verbs:

Note that examples with initial #pC, #fC, #xC are provided for completeness here, but that these in fact are extremely rare clusters. The pervasive generalization is that C1 of an initial CC cluster is Coronal, consistent with Morelli (Reference Morelli1999, Reference Morelli, Féry and de Vijver2003).

Ejectives and affricates cannot occur as the first member of an initial cluster (except for [t͡s], which can precede [x]). However, as seen in (4e), [fk'at͡sáx] ‘wide’, ejectives can occur as the second member of an initial cluster. Given that CCC clusters are not allowed, the initial onset cluster [fk’] provides evidence against treating ejective consonants as C+ʔ sequences.

Further, a major set of distributional generalizations of word-medial clusters that plays an important role in my analysis is that where there is a word-internal coda, the following onset is always of equal or lesser sonority; that is, Obstruent(O)-Sonorant(S) sequences are not attested in the Stem1 (St1) domain. I consider the St1 to consist of the root and derivational suffixes.

The notion of sonority thus provides a window into the fact that different layers of affixes define different prosodic domains. Example (2b), fìn-kɑ-náx ‘smoker’, consists of a single St1, while example (2b′), *fin-ɑk-náx, illustrates that an obstruent-sonorant sequence across a syllable boundary within this domain is actively and systematically avoided within the phonology of Nivaĉle (as I will further explore in section 3.2.2). However, there are other contexts where such contact persists without being subject to metathesis (or any other repair strategy):

1. (5) ɬ-tḛ́ʃ=ji

   2.s-say-1.o

   ‘you tell me’

In (5), the palatoalveolar fricative /ʃ/ in the coda precedes the palatal glide /j/ in the following onset, leading to a bad syllable contact. I hypothesize that in cases where such sequences persist, they are permitted to do so because there is a stronger prosodic boundary between them. Specifically, in (5), the first-person object pronoun is a clitic. The generalization governing metathesis is that it operates within an inner prosodic domain, identified as the St1, but it does not apply in the outer domain; that is, the higher Prosodic Word domain that contains clitics.Footnote ³

Finally, Nivaĉle has a quantity-sensitive iambic stress system. There is a consistent correlation between bimoraic weight (tautosyllabic /Vʔ/) and stress prominence. In addition, primary/secondary stress patterns reflect competing edge-alignment constraints where prosodic foot domains align with internal morphological category (MCat) edges, specifically Root (Rt), Stem1 (St1), Stem2 (St2), and Morphological Word (MWd). For a fuller explanation of Nivaĉle stress assignment domains, see Gutiérrez (Reference Gutiérrez2015).

2.1 The problem: alternating vs. non-alternating forms

As mentioned above, the phenomenon of metathesis is observed in a variety of contexts throughout the Nivaĉle grammar. It occurs in both nominal and verbal domains, characteristically resulting from the affixation of a consonant-initial suffix to a consonant-final stem. The focus of this section is pluralization in the nominal domain.

Pluralization of nouns in Nivaĉle exhibits a considerable degree of allomorphy, where the choice of allomorph is lexically determined rather than phonologically conditioned. The basic noun plural allomorphs are /-s/ ~ /-j/ ~ /-k͡l/ (the latter surfaces as [k] in word-final position). The data presentation below is organized into four sets (section 2.1.1 through section 2.1.4) that illustrate the different patterns of phonological alternations in nominal stems suffixed by these plural allomorphs.

2.1.1 Noun plurals: Non-alternating V-final noun stems

The V-final stems in (6)–(11) illustrate the plural allomorphy in Nivaĉle: each of the three suffixes occurs after all vowel qualities (the list in (6)–(11) is not exhaustive).

Further, even though only looking at a few pairs of data in Nivaĉle would suggest a correlation of -s with masculine nouns and -j with feminine nouns (see (12)–(13) below), the range of data presented in this article establishes that all three suffixes occur with both masculine and feminine roots. The epenthetic vowel in (13) is shown in square brackets. Non-human nouns are not marked for gender; feminine and masculine gender is marked on the singular determiner forms. In contrast, plural determiner forms differentiate between ‘human’ ([-pi], as in (12–13)) and ‘non-human’ ([-βa], as in (6b–11b)).

There is also a -CVC plural suffix /–wot/ (14b) which is restricted to kinship terms. However, its usage is currently undergoing attrition, as it is starting to alternate with other plural suffixes:Footnote ⁴

2.1.2 Noun plurals: Glottal-final stems

The following data show that noun stems with a final glottal stop systematically lose that glottal stop when suffixed by the plural. Note that this deglottalization is triggered by all allomorphs of the plural suffix.

The deletion of the glottal is morphologically conditioned. In (3b), it was shown that in an underlying VʔC root sequence (e.g., /k͡loʔp/), the glottal can be parsed to the nucleus of the syllable and thus be realized as a glottalized vowel: [k͡ló̰p]. With this parsing, there is no violation of the complex coda restriction: *[k͡loʔp]. In contrast, the glottal does not get realized as a glottalized vowel when a consonant-initial plural suffix is attached to the root (15b–17b). The consistent pattern is that it cannot occur in either non-metathesized (15–17) or metathesized forms (see (20) below), under nominal plural suffixation.Footnote ⁵ Interestingly, deletion of stem-final glottals in the context of nominal pluralization is also regularly attested in two other Mataguayan languages. For example, in Chorote we see awoʔ ‘worm’, awo-s ‘worms’ (Gerzenstein Reference Gerzenstein1983: 50); and in Maká we find witkinxeʔ ‘hip/side’, witkinxe-j ‘hips/sides’ (Gerzenstein Reference Gerzenstein1999: 231).

2.1.3 Noun plurals: Metathesis in C-final stems

As illustrated in (18)–(22), the forms exhibiting metathesis are all consonant-final in their unsuffixed singular form: the particular stem-final consonants known to participate in the metathesis process are /t t͡ʃ ʃ k x/, although, on the basis of the hypotheses advanced in section 3 below, it is predicted that any stem-final obstruent would participate in metathesis, under the appropriate triggering conditions. The metathesis itself can be characterized as follows: the final vowel and consonant of the unsuffixed stem in (18a)–(22a) switch their linear order when the plural suffix is attached in (18b)–(22b). Schematically, then, the segments V₁  and C₂  are reordered with the addition of the plural consonantal suffix -C: V₁C₂-C → C₂V₁-C. For example, the plural of ‘my job’ in (18a) is not *βàt-ku.mét-s, but βàt-kumté-s. The starred examples show the ungrammatical (non-metathesized) forms with unpermitted complex codas.

The reordering of segments within the phonological string triggers some featural changes such as deglottalization of ejectives (19) and glottalized vowels (20), and the spirantization of the dorsal stop present in the nominalizer suffix (22).Footnote ⁶

2.1.4 Noun plurals: C-final stems with Vowel epenthesis

The final set of data shows C-final noun stems which are not subject to metathesis, unlike the examples seen in section 2.1.3; rather, a vowel is epenthesized between the final C of the stem and the plural suffix. The first starred forms show the ungrammatical forms with complex codas, and the second starred forms show the potential, but not attested, metathesized forms.

These forms raise the following analytic question: what differentiates the C-final forms in (18–22) from the C-final forms in (23)–(28)? While this question will be discussed at greater length in section 3, for now it is sufficient to note that although metathesis would function to repair an ill-formed complex coda in word-final position (as shown in the first starred forms in (23–28)), it would result in the creation of an ill-formed consonant cluster (as shown in the second starred forms in (23–28)).Footnote ⁷ Vowel epenthesis thus emerges as an alternative repair mechanism.

As shown in the examples above, the epenthetic vowel most commonly used with the -s and -k allomorphs is [i]. There is yet another pattern of epenthesis manifested in a very small subset of the data in my field corpus, where the epenthetic vowel matches the last vowel of the root.

Harmonic epenthetic vowels are much less frequent than [i] or [e] epenthesis, and they occur most frequently with the [j] plural allomorph. Another case of harmonic epenthetic vowel can be found with the kinship plural suffix /-wot/, as in (31b), which is restricted to a limited set of terms and is currently subject to both variation and attrition.

While it is not possible on the basis of the available data to predict which epenthetic vowel will appear, the basic generalization that holds across all sets of data is that (i) plural suffixation creates an ill-formed consonant cluster, and (ii) metathesis cannot serve as the repair mechanism in (23)–(30) because, in each of these cases, it would result in the creation of a different ill-formed consonant sequence. Thus, vowel epenthesis functions as an alternate strategy to repair these cluster violations. Another potential repair mechanism for illicit consonant clusters, namely consonant deletion, is hardly ever present in the Nivaĉle data.Footnote ⁸ The examination of noun plural forms in this section suggests that the constraint Max-IO (no deletion) is highly ranked in this language, sometimes at the expense of consistency in the linear sequencing of segments (Linearity-IO) and sometimes at the expense of introducing vowels that are not part of the input representation (Dep-IO-V).

The data examined here have also shown an essential interplay between the phonological processes of metathesis and epenthesis in relation to well-formedness constraints on syllable structure.Footnote ⁹ The next sections will elaborate more fully on the roles that prosodic constraints on cluster sequences play vis-à-vis the segmental phonological system of Nivaĉle.

3. Driving forces behind metathesis and vowel epenthesis

In this section, the driving forces behind the metathesis (section 3.2) and vowel epenthesis (section 3.3) processes in Nivaĉle are presented through the analysis of two types of affixation processes: nominal pluralization, and derivation. The major hypothesis advanced in this article is that metathesis is a phonological process motivated by syllable structure constraints, namely, the avoidance of marked structures in the language: complex codas in some cases (section 3.2.1) and bad syllable contacts in others (section 3.2.2). Before moving on to my analysis, I briefly consider a previous proposal.

3.1 A note on a previous analysis

From a synchronic perspective, the Nivaĉle stem alternations presented in (18)–(22) have been regarded as VC-metathesis (Stell Reference Stell1987).Footnote ¹⁰ From a diachronic perspective, Campbell and Grondona (Reference Campbell and Grondona2007) have argued that historical vowel (and glottal) deletion is involved. The authors apply internal reconstruction to the forms presented in the singular and plural noun examples in (18)–(22) and posit several sound changes in the history of Nivaĉle. In Table 4, which is adapted from Campbell and Grondona's (Reference Campbell and Grondona2007) work, a vowel that is present in the singular form is missing from the related forms in the plural column. The information in the table has been reorganized with headings, and the phonetic symbols [y] [ts] [ɸ] are re-transcribed as [j], [t͡s], and [f].

Table 4: Vowel-alternation examples (adapted from Campbell and Grondona Reference Campbell and Grondona2007: 5)

Following the above morpheme parsing, Campbell and Grondona assume that the Nivaĉle roots underwent “a change which deleted a vowel when a vowel-initial suffix was added” (Reference Campbell and Grondona2007: 6); this change did not affect the singular suffixless words. A reconstruction is then postulated through the following vowel deletion sound change:

However, it is not very clear why this vowel deletion rule occurs in the presence of a “vowel-initial suffix”. On the one hand, a double-sided open syllable is a common environment for syncope; on the other hand, if syncope is what is involved here, there is no reason to expect the vowel of the root and the initial vowel of the suffix to have been identical, unless a historically prior vowel harmony process was involved (section 5). However, the authors neither propose nor discuss potential vowel harmony in Nivaĉle.Footnote ¹¹

Further, the VC ~ CV-C alternation, which I analyze as a metathesis process, is not restricted to the plural suffixation pattern that Campbell and Grondona analyze. As will be shown in section 3.3, there are sets of data in which a range of sonorant-initial derivational suffixes trigger the same VC-metathesis. Therefore, not only plural suffixes but also derivational suffixes would need to contain vowels identical to the ones that are deleted in the final syllable of the root. This is an inelegant analysis, because it proposes several allomorphs. The proposal advanced here argues that all these observed data receive a more coherent and integrated interpretation under a prosodic analysis of syllable structure constraints.

3.2 Metathesis

The following sections consider the motivations for, and an analysis of, the repair of impossible consonant sequences by metathesis.

3.2.1 Against complex codas

As noted in section 2, complex codas are not licit syllable types in Nivaĉle, and plural suffixation on C-final nouns would therefore create an illicit CC coda cluster. The proposal advanced here is that metathesis functions as a repair strategy, avoiding a *ComplexCoda violation while preserving the segmental identity of the vocalic and consonantal segments in the input. Further, as observed earlier, the manifestation of an underlying glottal stop as creakiness in the preceding vowel is lost, as shown in (20) above and (33) below, and spirantization of the velar stop may occur, as in (34).

Based on the two observations above – namely, that there are no complex codas in Nivaĉle and that there is a change in the linear order of the final vowel and consonant of the root – the following basic constraints are proposed, along with the ranking in (37):

1. (35) Linearity-IO: No metathesis (McCarthy and Prince Reference McCarthy and Prince1995:123)

   ‘S1 is consistent with the precedence structure of S2, and vice versa’

2. (36) *ComplexCoda: Codas are simple (*CC]_σ) (Kager Reference Kager1999)

3. (37) *CC]_σ » Linearity-IO

The following tableau shows how the ranking in (37) gives the correct output form for finxas ‘crabs.’

1. (38)

Under the hypothesis that in the grammar of Nivaĉle it is preferable to violate Linearity than ComplexCoda, (38b) emerges as the optimal candidate.

3.2.2 Syllable Contact Law

The examination of derivational suffixes shows that constraints (35) and (36) are not sufficient to explain metathesis. A range of sonorant-initial derivational suffixes trigger the same VC-metathesis phenomenon in a preceding root/stem with a final obstruent, as seen in the following examples.

Here VC-metathesis is not functioning as a repair strategy to avoid a *ComplexCoda violation, which suggests that a different prosodic markedness constraint – one which optimizes the relative sonority of consonants across a syllable coda-onset sequence – is at play.

In contrast, even though sonorant-final stems are rare, the examples in (46)–(48) show licit sonorant-sonorant sequences across morphemic boundaries.

There is no metathesis or epenthesis when the syllable contact does not involve an Obstruent-Sonorant contact across a morpheme boundary, as opposed to the data illustrated in (39)–(45), which do show metathesis. In these cases, suffixes with an initial sonorant – specifically, /m/, /n/, /j/ and /w/ – are attached to an obstruent-final stem. Rather than the expected linear concatenation (e.g., *fin-ɑk-met͡ʃ) the final obstruent of the stem metathesizes with the preceding vowel: fìn-kɑ-mét͡ʃ. Schematically, where O stands for an Obstruent and S stands for a Sonorant, the linear segmental sequence is reordered as follows: *V₁O-SV → OV₁-SV. Moreover, metathesis results in a new coda-onset sequence inside of the stem, [n.k], but this S.O cluster does not exhibit bad syllable contact. In the analysis that follows, I argue that there is a constraint against an obstruent-sonorant coda-onset sequence (i.e., *O.S). In these cases, metathesis functions as a strategy to repair what would otherwise be a prosodically non-optimal sequence.

Specifically, I argue that the driving force behind this second type of metathesis is the Syllable Contact Law (SCL) proposed by Murray and Vennemann (Reference Murray and Vennemann1983) and Vennemann (Reference Vennemann1988) in order to explain syllabification patterns and sound change at syllable boundaries. Basically, the consonantal strength of the coda should exceed or be equal to the consonantal strength of the following onset. Vennemann (Reference Vennemann1988: 8) defines consonantal strength as “a phonetic parameter of […] unimpeded (voiced) airflow” and proposes that sounds are organized in a universal ordering known as the Consonantal Strength Hierarchy, tracing back to Sievers (Reference Sievers1881) and Brugmann (Reference Brugmann1897).

Some discussions of the SCL (Parker Reference Parker2002, Reference Parker2012; Gouskova Reference Gouskova2004) have replaced consonantal strength with sonority, a concept that has been widely invoked as an explanatory principle in several different types of phonological analyses, but that also has been the object of extensive debate and controversy. A number of cross-linguistic tendencies with respect to the distribution and sequencing of segments have been made with reference to sonority hierarchies. When major natural classes are considered, the generalized sonority hierarchy in (49) is commonly assumed.

1. (49) Vowels > Glides > Liquids > Nasals > Obstruents

   (Clements Reference Clements, Kingston and Beckman1990, Kenstowicz Reference Kenstowicz1994)

Only if the Syllable Contact Law is interpreted in terms of the Sonority Hierarchy in (49) do the data in (39)–(45) show cases where the concatenation of morphemes results in bad syllable contact. Specifically, suffixation creates an obstruent-sonorant heterosyllabic sequence, where the sonority of an obstruent coda is lower than the sonority of a following onset. I thus hypothesize that (i) this type of sonority reversals like this are not tolerated in Nivaĉle, and (ii) metathesis functions as a repair strategy that optimizes an otherwise illicit syllable transition. As a result, a vowel-sonorant transition emerges at the site of morpheme concatenation, thus optimizing Syllable Contact. Note at the same time that the (underlying) stem-final obstruent is shifted by metathesis into an Onset position; for example, the stem-final [k] of (41), *fin-ɑk-met͡ʃ → fìn-kɑ-mét͡ʃ, now surfaces as an Onset rather than a Coda (and the [n.k] sequence does not violate the Syllable Contact Law). As a result, CV obstruent-vowel transitions are created. Because the most reliable perceptual cues to place of articulation in a stop depend on the formant transitions into a vowel, VC-metathesis also optimizes the perception of non-continuant obstruents.

On the basis of my Nivaĉle data, I assume the following sonority scale, where adjacent categories of Glides-Nasals and Fricatives-Affricates-Stops are conflated into the single categories of Sonorant and Obstruent, respectively.

1. (50) Vowels > Sonorants > Obstruents

In an Optimality Theoretic analysis, the Syllable Contact Law represents a family of constraints, which can be instantiated for Nivaĉle in the following terms:

1. (51) SYLLABLE CONTACT LAW (SCL) (*^[-son]]_{σ σ}[^[+son]): Sonority should not rise across a syllable boundary (from an obstruent to a sonorant).

The interaction between the SCL constraint and the previously proposed Linearity-IO constraint is illustrated in the following tableau. I defer the discussion of another potential candidate, *[fi.nɑ.ki.met͡ʃ], which involves epenthesis, until section 3.3.

1. (52) SCL, *ComplexCoda » Linearity-IO

2. (53)

The most faithful candidate to the input, (a), fatally violates SCL and is thus discarded. Candidate (b) surfaces as the optimal output because it violates the lower ranked Linearity-IO once, whereas (c) violates *ComplexCoda and Linearity-IO twice.

In essence, the hypothesis I am proposing is that syllable contact markedness constraints are highly ranked in Nivaĉle and will trigger metathesis, a Linearity-IO violation. Under this proposed analysis, an interesting question arises: What happens if suffixation of a sonorant-initial suffix to an obstruent-final stem should trigger metathesis in order to avoid violating the SCL, but the linear reordering of the final vowel and consonant of the root would itself incur a violation of a higher ranked constraint (for instance, *ComplexCoda)? In these cases, vowel epenthesis takes place, an issue I discuss in the following section.

3.3 Vowel epenthesis

When the linear reordering (metathesis) of the final vowel and consonant of a stem would incur a violation of a higher ranked constraint, epenthesis emerges as a repair mechanism. In (54) below, for example, if metathesis were applied to avoid a bad syllable contact […k-w…] (54a), a complex (and illicit) onset [p'k] would result (54b).Footnote ¹² Complex onsets do exist word-initially, but they never arise as the result of metathesis or another syllable repair mechanism. Further, deletion of the final consonant is not observed as an alternative strategy (54c). Faithfulness to consonantal identity of lexical representation is highly ranked in the Nivaĉle grammar. Finally, (54d) shows that metathesis cannot occur across morpheme boundaries.

Since the first rescue strategy – metathesis within the first morpheme, as in (54b) – does not result in an acceptable syllabic parse, the best repair strategy in this case is vowel epenthesis, a Dep-IO-V violation (54e). In tableau (59) we can see the relative ranking of Dep-IO-V and Max-Seg, and metathesis and epenthesis jointly “conspiring” (Kisseberth Reference Kisseberth1970) to eliminate bad syllable contact sequences, namely, SCL violations. *ComplexOnset and *ComplexCoda are the conditioning factors that give rise to the variation between one process and the other. The ranking in (58) crucially establishes that Max-Seg is higher ranked than Dep-IO-V: vowel epenthesis is a better repair strategy than deletion of the final consonant of the stem. In contrast, the relative ranking of Dep-IO-V with regards to [_σCC is not clear, given that complex onsets do occur (word-initially) in Nivaĉle.

While candidates (59a) and (59c) get discarded because they violate SCL and Max-Seg, an interesting contrast can be seen between (59b) and (59d). Each candidate incurs one violation from the crucially unranked constraints *_σ[CC and Dep-IO-V.Footnote ¹³ Importantly, candidate (59b) also incurs one violation of Linearity-IO, so candidate (59d) emerges as the optimal output. In sum, whenever possible, metathesis is applied to avoid a bad syllable contact. If metathesis creates an illicit consonant cluster, vowel epenthesis is applied.

Vowel epenthesis is also observed when C-initial plural allomorphs are attached to stems with a medial cluster. In these cases, Dep-IO-V must be also crucially outranked by an undominated syllable markedness constraint, namely *ComplexCoda. As mentioned above, it was already established that while complex onsets occur in Nivaĉle, they never emerge as a result of metathesis; *[_σCC is crucially unranked with respect to Dep-IO-V, and they are both ranked higher than Linearity-IO. The emergence of complex codas and complex onsets is thus avoided through vowel epenthesis. Let us consider the following example:

As in the example discussed in (54), metathesis does not occur in (60) because it would either result in a derived complex onset (60d) or a complex coda (60e). Deletion of the final consonant of the stem, as a strategy to avoid the emergence of a complex coda in (60f), is also not permitted. Consider the candidates in (60) in the following tableau (62), along with the following proposed ranking of constraints (61):

1. (61) *CC] _σ, Max-Seg » Dep-IO-V, * [_σCC » Linearity-IO

2. (62)

Candidate (62e) emerges as the optimal output: vowel epenthesis is the best strategy to avoid the emergence of a complex coda (62a, 62c), or a complex onset due to metathesis (62b) when the root has a medial CC cluster /t͡ʃint͡ʃeʔx/ (as opposed to a medial singleton C as in /finax/). Deletion of a segment is worse than epenthesizing a vowel, so candidate (62d) gets discarded.

4. Domain of metathesis

It is often the case in languages where metathesis is driven by the Syllable Contact Law that the consonants across a syllable or morpheme boundary are the ones that metathesize. For instance, we find in Sidamo (Gouskova Reference Gouskova2004: 228–229): /hab-nemmo/→ [han-bemmo] ‘we forget’; /has-nemmo/→ [han.semmo] ‘we look for’; in Old Spanish (Holt Reference Holt2004: 52) /kad.nado/ → [kan.dado] ‘lock’; and in Leti (van Engelenhoven Reference van Engelenhoven2004: 91): lau ‘civet’ + nama ‘tongue’ → [lanwama]. However, in Nivaĉle, the ill-formed *O-S consonant sequence across the root-suffix boundary (63b) does not metathesize. Rather, metathesis in Nivaĉle is root-bound; it affects the final consonant of the root and the preceding vowel (63a).

In other words, although the trigger for metathesis is the fact that the initial consonant of the suffix is more sonorous than the final consonant of the root, the repair mechanism for this violation of the Syllable Contact Law does not involve segments from two different morphological domains (63b). On the contrary, the segments that metathesize are exclusively within the morphological domain of the root; Nivaĉle metathesis respects morpheme boundaries. For purposes of clarity, the examples in (63) are repeated in (64); the right edge of the root is indicated, and the ordering reversals of candidates (64a) and (64b) are marked with crossing lines.

1. (64)

As indicated by the circled segment in (64b), if metathesis were to reverse the order of the final consonant of the root and the initial consonant of the suffix, then the suffix consonant (the circled [j]) would intrude between the contiguous [e] and [t͡ʃ] segments of the root.

In sum, the Nivacle pattern is in line with proposed cross-linguistic tendencies for metathesis. Mielke and Hume (Reference Mielke, Hume, Hume, Smith and de Weijer2001) have argued that metathesis disrupts word and root recognition, so ordering reversals tend to involve adjacent segments at the middle or right edge of the root.

5. Metathesis and pseudometathesis

Having established the broad range of properties that characterize and constrain the VC-metathesis process in Nivaĉle, one of the remaining issues is to consider how this case fits the broader cross-linguistic typology of metathesis.

Under the theoretical umbrella of Evolutionary Phonology (Blevins Reference Blevins2004), historical, non-teleological, and phonetic explanations are posited for synchronic sound patterns. Blevins and Garrett (Reference Blevins, Garrett, Hayes, Kirchner and Steriade2004) propose four categories of phonetically natural metathesis processes (examples are taken from their article): (i) perceptual metathesis: features with elongated phonetic cues are reinterpreted in non-historic positions (e.g., laryngeal metathesis in Cayuga); (ii) compensatory metathesis: within a foot, a feature in a weak syllable shifts to a strong syllable (e.g., V́₁CV₂  → V́₁V₂C in Rotuman); (iii) coarticulatory metathesis: the overlap in gestures of two adjacent segments – that is, CC coarticulation – results in a reinterpreted ordering (e.g., Mokilese kp → pk); and (iv) auditory metathesis: the sibilant noise present in a sequential speech stream is decoupled from the speech stream (e.g., sibilant-stop, stop-sibilant metatheses).

Importantly, Blevins and Garrett state that other synchronic alternations, such as VC > CV or CV > VC-metathesis that lack the phonetically natural properties listed in the above typology, are actually cases of pseudometathesis (Mills and Grima Reference Mills, Grima and Naylor1980). More precisely, what looks on the surface like synchronic CV- or VC-metathesis may actually involve two processes (also known as telescoping (Wang Reference Wang1968)): copy-vowel epenthesis and historical vowel deletion. The challenge posed by pseudometathesis processes is that the two discrete processes might not be independently recoverable and are therefore opaque.

The Nivaĉle VC/CV synchronic alternations motivated by prosodic constraints do not fall into any of the metathesis categories proposed by Blevins and Garrett (Reference Blevins, Garrett, Hayes, Kirchner and Steriade2004). Therefore, two options could be considered. One is that these alternations are a case of pseudometathesis. The other is that the Nivaĉle data argue for an additional category in the cross-linguistic typology.

Under the first hypothesis, namely that the Nivaĉle data are best interpreted as a case of pseudometathesis, two diachronic processes could be posited: (i) epenthesis of a copy-vowel into an otherwise illicit syllable structure, and (ii) deletion/syncope of the underlying vowel of the stem that served as the base for the copy-vowel epenthesis process, as demonstrated in (65).

As a more fully fleshed-out version of Campbell and Grondona's (Reference Campbell and Grondona2007) proposal, this approach would have the advantage of reducing the number of suffix allomorphs in both the inflectional and derivational domains. Instead of a multiplicity of vowel-initial allomorphs that are required to match the preceding vowel in the stem (e.g., -is, -es, -as, -ɑs, -os, -us, -ij, -ej, aj, etc.), the lexical identity of the allomorphs would be simply consonantal (i.e, -s, -j, -k͡l).Footnote ¹⁴

Some evidence for copy-vowel epenthesis can be seen in (66)–(69) (also presented in section 2.1.4). In these examples, suffixation of the plural allomorph –j to a consonant-final stem with a glottalized vowel involves a harmonic epenthetic vowel:

In (66)–(69), vowel syncope (or metathesis) does not occur, because it would yield an illicit consonant cluster: either a complex coda or a word-medial complex onset in ɑft.k͡lej/ ɑf.tk͡lej (66), an unattested (and derived) complex onset cluster xt (67), or a medial cluster s.ʃ (68). It is not clear, though, why *[jistɑj] is not an optimal output for (69). I can only hypothesize at this point that [s] was in fact, at some stage, in variation with [t͡s], as originally observed by Hunt (Reference Hunt1924).

In sum, pseudometathesis (copy-vowel epenthesis and vowel syncope), could explain the Nivaĉle alternating form phenomena that are here analyzed as synchronic metathesis. However, this approach is problematic in at least one respect. As discussed in section 2.1.4, the most regular Nivaĉle epenthetic vowel is [i]. What would be the motivation, then, for a subset of nouns to undergo harmonic vowel epenthesis while another, larger set adopts [i]-epenthesis? While all the forms listed above have glottalized vowels, it is also true that [i]-epenthesis also applies to roots with glottalized vowels (e.g., k͡ló̰p ‘winter’, k͡lopís ‘winters’). This is still a puzzling topic; I have not yet reached an adequate explanation.

With regards to vowel syncope, which is neither a regular process nor as productive as vowel epenthesis, it sometimes involves vowel identity between the vowel of the stem and the suffix, as in (70)–(71), but other times does not, as in (72)–(73):

The fact that vowel deletion, [ʔ] deletion, and both harmonic and non-harmonic vowel epenthesis are attested in Nivaĉle points to the range of phonological processes that Nivaĉle stems undergo in the context of suffixation. Two challenges thus seem to exist for the pseudometathesis account: (i) the existence of non-harmonic vowel epenthesis, and (ii) the fact that [i]-epenthesis and “copy-vowel epenthesis + syncope” (i.e., synchronic metathesis) seem to be in a principled, complementary distribution.Footnote ¹⁵ More specifically, if epenthesis were a historically earlier change than syncope, why would epenthesis have involved a copy-V in exactly those environments that would later be subject to syncope, and a non-copy-vowel [i] in those environments where, it turned out, the eventual syncope would not happen? In any case, the possibility that VC-metathesis arose through the reanalysis of what originated diachronically as copy-vowel epenthesis and (unstressed) vowel syncope sound changes definitely needs to be further explored within Nivaĉle and across the Mataguayan languages.

Whatever its historical origin, I propose that synchronic VC-metathesis functions as a phonological strategy to avoid violation of syllable structure constraints and to optimize the sonority cline of heterosyllabic consonant clusters. As Crowhurst and Trechter (Reference Crowhurst and Trechter2014:148) point out in their analysis of vowel-rhotic metathesis in Guarayu, phonological factors may contribute to the diffusion and the generalization of metathesis as the innovative pattern. Moreover, the elimination of complex codas and bad syllable contacts is structure-preserving: neither complex codas nor obstruent-sonorant sequences are ever attested as well-formed structures in the Nivaĉle language. What is more, the correlation between metathesis and structure preservation has been suggested by Hume (Reference Hume2004: 221): “any order of two segments is a potential output of metathesis, provided that the reordered sequence forms an attested structure in the language”. I argue that what I analyze as synchronic VC-metathesis in Nivaĉle thus conforms to a model that is different in its perspective from Evolutionary Phonology. Furthermore, I have observed in the context of the proposed analysis that VC-metathesis is blocked when the output of metathesis would incur violations of high-ranked syllable markedness constraints such as *ComplexCoda; instead, vowel epenthesis occurs. Metathesis and vowel epenthesis can be regarded as two conspiring mechanisms driven by the avoidance of marked structures. A detailed analysis of the interrelation between vowel deletion, epenthesis and metathesis in Nivaĉle, and within the Mataguayan family, constitutes an issue for future investigation.