Foot-based allomorphy in Tlapanec (Mè’phàà)

2.1 Tlapanec phonology

Huehuetepec Tlapanec has the following consonants, shown in Table 1 (cf. Tiburcio Cano 2017: 41); there are still disagreements among scholars with respect to the consonant inventory, partially due to inter-dialectal variations and analyses, such as whether the nasal + stop combinations are sequences or singletons. In Table 1, in cases where the orthographic conventions used in this article diverges from the IPA symbols, the latter appears in //.

We follow Marlett and Weathers (2018) in considering labiovelars and prenasalised stops as singletons rather than sequences. In this article, we consider that palatalized consonants are also units, rather than sequences; they behave as a unit with respect to the constraint against a glottal stop before stems beginning with a cluster, as we will see in Section 3.1 and Section 3.4 (see also Uchihara 2021 on the status of such complex consonants). Marlett and Weathers (2018) argue that ʦ is an allophone of /s/; even though it is true that in some varieties certain words show free variation between [s] ∼ [ʦ], as Oropeza Bruno (2014: 74) points out, there are some minimal pairs and thus we consider that /ʦ/ is an independent phoneme, albeit marginal. According to Marlett and Weathers (2018), r is an allophone of /d/, which appears in weak syllables. It is true that its distribution is limited to weak syllables, as we will see in Section 5.1, and it is also cross linguistically common that flapping occurs in weak syllables (González 2003: 59–62). However, we would rather consider r as a marginally contrastive phoneme, since in rare cases a [d] can appear in weak syllables ([dúdiiⁿ] ‘avocado’, [dúdíʔ] ‘sal de cal’), especially in loans and compounds. The phoneme /l/ is not common and is only found in loans and in some clitics.

Tlapanec varieties have five vowel qualities, a, e, i, o, and u; additionally, vowel length and nasalization are contrastive in the strong syllable, that is the last syllable of the prosodic word. A vowel may be followed by a glottal stop: gù ⁿ ʔ ‘moon’, gùʔ.gúú ⁿ ʔ ‘raccoon’. In this article, we adopt the analysis that the glottal stop [ʔ] is a vowel feature, following Carrasco Zúñiga (2006) and Navarro Solano (2012). This is because a glottal stop cannot occur at the initial position, and when it occurs between vowels, it is parsed with the preceding syllable, rather than the following vowel, since speakers pronounce the glottal stop along with the preceding CV, rather than the following V (Tiburcio Cano 2017: 48): [rìʔ.ì] ‘flower’. Moreover, as we will see throughout this article, a glottal stop behaves differently from the rest of the consonants, in that it can be elided from the prefixes while other consonants cannot. See Weathers et al. (2012), Tiburcio Cano (2017: 47–53), and Marlett and Weathers (2018) for an alternative analysis that glottal stop is a consonant.

Tlapanec has three surface levels of tones, low (abbreviated as L, represented as à), mid (M, a) and high (H, á), as can be illustrated by the following minimal triplet:

The tone bearing unit (TBU) is the mora. On disyllabic forms (with one mora each), all of the possible combinations of these three tones (tone melodies) are attested: L.L, L.M, L.H, M.L, M.M, M.H, H.L, H.M, and H.H. In Huehuetepec Tlapanec, lexical tautosyllabic contours (on long vowels) are limited to LH, HM and ML, but LM and HL contours also occur as a result of morpheme concatenation (Tiburcio Cano 2017: 60). Contour tones can only occur on long vowels, and since long vowels can only occur on the final syllable of the strong syllable, contour tones are generally also limited to this syllable (although there are a very few cases of contour tones in non-final syllables).

2.2 Tlapanec prosodic structure

Tlapanec syllable structure is (S)(C)(r)V(V)(ʔ); no coda is allowed (a glottal stop is a vowel feature), and onsets can have up to three consonants, which are clusters of a fricative + consonant that can be augmented with an r (Marlett and Weathers 2018). Tlapanec employs iambic feet aligned to the right edge of the prosodic word (Marlett and Weathers 2018: 23; Suárez 1983: 26). The acoustic correlate of accent in Tlapanec is still unclear. On the one hand, Marlett and Weathers (2018: 23) state that “it is not clear that something comparable to stress is appropriately identified in Me’phaa”; and speakers also have difficulty in identifying which syllable is ‘strong’. On the other hand, Carrasco Zúñiga (2012) proposes that the most reliable acoustic correlate of accent is the duration of the onset consonant in the strong syllable, while Herrera Zendejas (2018) claims that the accent is assigned to the syllable which has the right-most highest tone. Our impression by working with the speakers and acoustic studies (Cornelio Tiburcio in prep.) support Carrasco Zúñiga’s (2012) hypothesis, but confirming the acoustic correlates of accent in Tlapanec is beyond the scope of this article. In order to reflect the fact that the acoustic correlate of stress in Tlapanec is still unclear, we refer to syllables as either “weak” versus “strong”, rather than “unstressed” versus “stressed” or “atonic” versus “tonic”, following Akinlabi and Urua’s (2003) terminology.

Although the phonetic correlate of stress is still unclear, Marlett and Weathers (2018: 23–24) remark that there is evidence that words and phrases are prosodically right-headed (whether this syllable has a short vowel or long vowel; thus, quantity-insensitive), since various phonological and phonetic details require reference to a weak syllable of the foot. This is also corroborated by the fact that a vowel can undergo syncope in any syllable except for this strong syllable.

Some contrasts are maintained only in the final syllable of the prosodic word; this is the phonological correlate of accent. First, nasalization is only contrastive in the strong syllable (Marlett and Weathers 2018). Following Weathers et al.’s (2012) convention, nasalized vowels are indicated with a raised n.^[1] The following minimal pair justifies that nasalization is contrastive in this position:

Secondly, vowel length is also only contrastive in the strong syllable (Carrasco Zúñiga 2006: 90; Suárez 1983:6). Additionally, since contour tones can only occur on long vowels, they are also restricted to the strong syllable. In this article, following the orthographic convention widely in use in Tlapanec linguistics, a long vowel is represented by doubling vowels; such a representation also facilitates representation of tones. Again, the following minimal pairs justify the contrast of vowel duration in this position.

Finally in Tlapanec, there is a general tendency for the strong syllable to require an onset more strictly than the weak syllable.^[2] Thus, none of the 41 monosyllabic native nouns in our database lacks an onset. On the other hand, out of the 153 disyllabic native nouns in our database, 41 nouns (26.8%) lack an onset in the first (weak) syllable, such as àʔ.gù ‘woman’, ì.gì ‘fox’, while only 10 nouns (6.5%) lack an onset in the second (strong) syllable, and such cases are limited to where the first syllable ends in a glottal stop, as in práʔ.a ‘kick’ or ^m bruʔ.u ⁿ ‘night’, àʔ.u ⁿ ‘iguana’.

2.3 Tlapanec morphology

All monomorphemic stems, whether nominal or verbal, are either monosyllabic or disyllabic in Tlapanec. Noun stems may be augmented with the possessor suffixes, while verb stems can take various prefixes and suffixes, as will be described below.

Tlapanec verbs obligatorily encode tense-aspect-mode (TAM) categories by a prefix, such as incompletive (na-/nu-), completive (ni-), potential (ma-/mu-), etc. Tlapanec verbs manifest semantically based alignment (Duncan 2017; Navarro Solano 2012; Suárez 1983; Wichmann 1996, 2005, 2009). Our focus in this article is the agentive verbs, which encode the agent, that is the sole agentive argument of an intransitive verb or the more agent-like argument of a transitive verb, with prefixes such as 2sg agentive tara-/ta- or a glottal stop in the 3sg/1pl.in agentive prefix that will be discussed in this article or stem alternation (suppletion; cf. Section 3.4). These verbs correspond to transitive verbs, ‘unergative’ verbs and suppletive verbs in Duncan (2017).

3.1 Passive giʔ- ∼ gita-

Transitive verbs in Tlapanec can defocus their agent with the impersonal or passive prefix giʔ- ∼ gita- (cf. Suárez 1983: 211–214).^[3] The allomorphy of this prefix is conditioned by the number of stem syllables. In general, the former allomorph is selected when the stem is monosyllabic, while the latter is selected when the stem is disyllabic, regardless of the vowel length of the final syllable. The glottal stop in the first allomorph is further conditioned by the presence or absence of a glottal stop in the stem, or if the stem begins with a consonant cluster or not. The following examples illustrate this.

With disyllabic stems, the allomorph for the passive prefix is invariably gita-, as in Table 2. In this case, no further allomorphy is found. The corresponding active forms are shown in the right column to justify the stem forms.

With monosyllabic stems, the passive prefix has the monosyllabic allomorph, giʔ- or geʔ-, as in Table 3. The glottal stop of this allomorph is deleted in certain environments, and for the reasons to be discussed in Section 3.4, we contend that this glottal stop is present underlying even when it is not realized. First, when a stem is monosyllabic but does not contain a glottal stop or begins with a consonant cluster, the allomorph for the passive is giʔ- or geʔ-, with a glottal stop; note that this alternation treats s ^j in ‘make dance’ as a singleton. That the glottal stop belongs to the prefix and not to the stem is evident from the active form with the 1sg agent, which lacks a glottal stop.^[4] In the final two verbs, the prefix vowel is the mid e, rather than i (this is also observed in the completive prefix); the conditioning factor for this alternation in the vowel height is yet unknown.^[5]

The allomorph without the glottal stop is found when the stem lexically contains a glottal stop, as in ‘throw away’ in Table 4, or begins with a consonant cluster as in the rest of the table. Again, the corresponding active forms are shown in the right column to justify the stem forms. We hypothesize that this alternation is phonological, motivated by the avoidance of consecutive syllables with a glottal stop, and the avoidance of a glottal stop followed by a consonant cluster; we will come back to this issue in Section 3.4 and such constraints will be formulated in Section 4.1.^[6]

The only exception to the generalization so far is the verb -ʔgiʔ ‘set’. Even though this verb is monosyllabic, the passive prefix takes the allomorph gita-, which is generally attached to the disyllabic stems.^[7]

Having achieved the descriptive generalizations, we now come back to the issue of the alternation between the disyllabic and the monosyllabic allomorphs of the passive prefix. Here, what matters is the number of syllables in the stem, and not the number of the syllables that follow the passive prefix, including the enclitics or other compounded morphemes. This is illustrated by the following example, with ^h maa, which may be related to a morpheme ‘towards body’ (Carrasco Zúñiga 2006: 295), attached to the stem -ʦí ‘play’.^[8] Here, the passive prefix is followed by two syllables, but still selects the monosyllabic allomorph giʔ-, since the stem is monosyllabic.

Employing the subcategorization frames following Paster (2006: 12), the passive prefix allomorphy can be represented as follows. Following Inkelas (1991: Ch. 5), we assume that subcategorization frames can be sensitive to prosodic units, that is feet in this case. The frame in (9a) amounts to saying that the allomorph gita- subcategorizes for disyllabic stems, while that in (9b) stipulates that the allomorph gi/eʔ- subcategorize for monosyllabic stems. We saw above that the allomorph without a glottal stop occurs with stems containing a glottal stop or those beginning with a consonant cluster; since this is due to phonology; there is no need for an independent subcategorization frame. For the monosyllabic allomorph, the allomorphs with the vowel e are collapsed with those with the vowel i; recall that the conditioning factor of this alternation is yet unknown.

3.2 2sg agentive ta- ∼ tara-

The agentive verbs in Tlapanec encode the 2sg agent with the prefix ta- ∼ tara-.^[9] This allomorphy is conditioned again by the number of the syllables as well as the stem initial consonant. The alternation here thus resembles that of the passive prefix, but the distribution of these allomorphs is the opposite: the allomorph ta- is selected with disyllabic stems while the allomorph tara- is selected with the monosyllabic stems.

The alternation of the prefix is observed only when the stem begins with a voiceless consonant (t, ʦ, s or ʃ in our database); when the stem begins with a voiced consonant (d, b, m, n, y or w in our database), the allomorph is always ta-, regardless of the number of the syllables of the stem, as in Table 5.^[10] We will first look at the stems beginning with a voiceless consonant. When the stem is disyllabic, the 2sg prefix is ta-. The 3sg forms, without the 2sg agentive prefix, are given in the right column to justify the stem forms.

On the other hand, when the stem is monosyllabic, the 2sg prefix is tara-, as in Table 6. Again, the 3sg forms are given in the ^[11]right column to justify that the syllable ra does not belong to the stem.

When the stem begins with a voiced consonant, the 2sg prefix is always ta-, regardless of the number of the stem syllables. Table 7 contains monosyllabic stems beginning with a voiced consonant that take the allomorph ta-:

In the following examples in Table 8, the stems with an initial voiced consonant are disyllabic, and the 2sg prefix is ta-, as expected:

Finally, with the stems beginning with a glottal stop (all monosyllabic), both allomorphs are found; here, the conditioning factor for the selection of the allomorphs is yet unknown, although it may be the case that the feature [voice] of the consonant after the glottal stop determines the allomorphy. First, the following is a stem which occurs with the allomorph tara-:

On the other hand, the following stems occur with the allomorph ta-. Among these, ‘sit (him) down’ is also exceptional with respect to the passive allomorphy as we saw in (7) -ʔgíʔ ‘set’ in the preceding section, where it was observed that even though this verb is monosyllabic, the passive prefix takes the allomorph gita-, which is generally attached to the disyllabic stems. In (12), the final syllable has a long vowel and lacks the final glottal stop; this is because the form has the 3sg patient suffix -ii.

Again, employing the subcategorization frames, the allomorphy of the 2sg agentive prefix can be represented as follows. The allomorph tara- subcategorizes for monosyllabic stems beginning with a voiceless consonant (13a). Otherwise, including cases where the stem begins with a voiced consonant or when the stem is disyllabic, the allomorph ta- is employed, which is the ‘elsewhere’ allomorph (13b).

3.3 2sg agentive ta- ∼ Ø of suppletive verbs

In the preceding section, we saw that the 2sg agentive prefix has two allomorphs, tara- and ta-, the subcategorization of which is conditioned by the foot structure, as well as whether the stem initial consonant is voiced or not. This generalization holds for the regular verbs. On the other hand, Tlapanec has a couple dozen agentive suppletive verbs, as was mentioned briefly in Section 2.3, where the stems supplete according to the agent person categories. For suppletive verbs, the 2sg agentive prefix shows a different type of allomorphy.

For suppletive verbs, the allomorphy of the 2sg agentive prefix is between ta- and Ø-, rather than between ta- and tara-, conditioned by the number of stem syllables, as can be observed in Table 9. First, the 2sg agentive prefix has no exponent when the suppletive stem is monosyllabic. The following are the exhaustive list of monosyllabic suppletive verbs. For comparison, the 1sg and 3sg forms are also provided, in order to justify that these verbs do undergo stem suppletion according to the agent person categories:

Table 10, which follows, contains some of the suppletive disyllabic stems. Here, the 2sg forms always require the 2sg agentive prefix ta-, as in the case of the regular verbs discussed in the preceding section.

Incorporating the data from suppletive verbs, the subcategorization frames for the 2sg agentive prefix will be as follows. (14a) and (14c) are repeated from (13) above. The only addition is (14b), which states that the Ø- allomorph subcategorizes for suppletive monosyllabic stems.^[12]

3.4 Glottal stop of the 3sg and 1pl.in agentive prefixes

The fourth allomorphy which is sensitive to the stem size in Tlapanec is the alternation of the 3sg and 1pl.in agentive prefixes, which is between a glottal stop and zero. The conditioning factor of the presence or absence of this glottal stop is, again, the number of stem syllables (Cline 2013: 115–116). These prefixes are realized with a glottal stop when the stem is monosyllabic and does not contain a glottal stop or begin with a consonant cluster; otherwise, that is, when the stem is disyllabic, or monosyllabic and contains a glottal stop or begins with a consonant cluster, the glottal stop is not found, as in the case of the passive prefix (Section 3.1).

First, we will look at cases with disyllabic stems; in such cases, the glottal stop of this prefix is never realized, as can be observed in Table 11. In the following, we will show only the 3sg forms to illustrate the case; the same holds true for the 1pl.in forms. The 1sg forms are shown in the right column to justify the stem forms; tonal differences in the stems are due to Low Tone Spreading from the 1sg prefix nì- (Uchihara and Tiburcio Cano 2020).

When the stem is monosyllabic, whether a glottal stop is found depends on whether the stem already contains a glottal stop or begins with a consonant cluster, as in the case of the passive prefix discussed in Section 3.1. First, the following examples in Table 12 show stems without a glottal stop or initial consonant clusters; in such cases, the glottal stop of the prefix is realized. The 1sg forms in the right-hand column justify that the glottal stop does not belong to the stem. Recall that /s^j/ is treated as a singleton for this alternation.

The cases above contrast with verb stems which begin with a glottal stop, which is always realized throughout the paradigm:

When a monosyllabic stem contains a glottal stop or begins with a consonant cluster, the prefix glottal stop does not surface, as in Table 13:

We posit that the glottal stop of this prefix is underlying even when it is not realized; that is, when the stem is disyllabic, or is monosyllabic and contains a glottal stop or begins with a consonant cluster. This is because an underlying glottal stop blocks High Tone Spreading, as superficial glottal stop does (cf. Cline 2013: 27). Thus, the allomorphy of the 3sg/1pl.in agentive prefixes, of the presence and absence of the glottal stop, is phonological and not morphological, unlike the allomorphy we have discussed above. Thus, there is no morphological (suppletive) allomorphy, hence no need for subcategorization frames.

3.5 Summary

In this section we looked at three prefixes, the allomorphy of which is conditioned by the size of the stem. Such alternations have been represented with the subcategorization frames, which are repeated below; the 3sg agentive prefix allomorphy was argued to be due to a phonological process, and thus there is no subcategorization frame.

4.1 Passive allomorphy

First, the passive allomorphy discussed in Section 3.1 can be analyzed as follows under a P » M approach. Recall that the alternation was between the allomorphs gita- and giʔ-, the former occurring with disyllabic stems, while the latter with monosyllabic stems; with the latter, the glottal stop is deleted before a stem-initial consonant cluster or when the stem contains a glottal stop.

If *(tV.ˈσ) is the only constraint at work, we would expect that the allomorph giʔ- would always be selected, or that the t undergoes lenition to d or further to r, but this is not what we find; what we observe is that when t would occur in such a position, the passive takes the form giʔ-, rather than gita-. This is because the alternation of gita- and giʔ- is morphological, rather than phonological; there is no productive phonological process which deletes the entire syllable ta when it occurs in the weak position of the foot.^[15] That is, this alternation is a case of phonologically conditioned morphological alternation (Inkelas 2014: Ch. 9; Kager 1996; Nevins 2011; Paster 2006, 2015, among others). Further, we may assume that gita- is the ‘preferred’ (or default) allomorph, rather than giʔ-; if giʔ- were the preferred allomorph, the passive prefix would always be giʔ-, since this allomorph would never violate the constraint *(tV.ˈσ). In order to capture this insight, we adopt the constraint Priority, proposed by Mascaró (2007):^[16]

The phonological constraint *(tV.ˈσ) and the morphological constraint Priority are then ordered in the ranking P(honology) » M(orphology) (cf. Kager 1996; McCarthy and Prince 1993), namely *(tV.ˈσ) » Priority. The tableau in (22) shows that the allomorph giʔ- is selected with a monosyllabic stem, since it only violates the lower-ranked Priority.

(22)

On the other hand, (23) shows that the allomorph gita- is selected with a disyllabic stem, since candidate (a) does not violate either constraint:

(23)

Other common constraints to account for foot structure are also active in Tlapanec, such as Rightmost (Prince and Smolensky 2004: 46), which requires that the head foot is rightmost in prosodic word; Align(Wd, R; Ft, R) (McCarthy and Prince 1993) which is violated when no foot is present at the right edge of the word; FootForm = I, which requires that the feet have final prominence (Kager 1999: 172; Prince and Smolensky 2004: 63); GrWd = PrWd, which requires that grammatical words have prosody (Kager 1999: 152). These constraints are undominated; in the rest of the article these constraints are assumed not to be violated. On the other hand, Parse(σ), which requires that the syllables be parsed in feet, is dominated by other constraints. This is illustrated by (24) and (25). In (24), candidate (a) incurs violation of *(tV.ˈσ), and thus suboptimal. Candidate (d) incurs violation of Rightmost by having a non-final head foot, in addition to the violation of *(tV.ˈσ). Candidates (c) and (e) avoid violation of *(tV.ˈσ) as well as of Priority, but incur violation of undominated Align(Wd, R; Ft, R) and FootForm = I, respectively. Finally, candidate (f) vacuously satisfies *(tV.ˈσ) and Priority by not parsing syllables into feet, but violates undominated GrWd = PrWd by not having accent, in addition to the multiple violations of Parse(σ). The same holds true for (25) with a disyllabic stem.

(24)

(25)

When the stem contains a t in the weak branch of the foot, this t does not undergo any modification. This can be due to the affix-stem asymmetry (Beckman 1998; Bybee 2005; Urbanczyk 2011), or Faith _root (McCarthy and Prince 1994), which requires the root material (including a glottal stop) not be altered nor deleted.^[17] This constraint dominates *(tV.ˈσ) and Priority introduced above. The tableau in (26) illustrates how t in the weak position of the stem is preserved. Candidates (b) and (c) violate the undominated constraint Faith _root and thus are suboptimal, while candidate (d) is suboptimal to candidate (a) due to violation of Priority. The optimal candidate (a) only violates *(tV.ˈσ).

(26)

We saw in Section 3.1 that when a monosyllabic stem contains a glottal stop or begins with a consonant cluster, the passive allomorph is realized without a glottal stop. This can be accounted for by the following constraints. First, a self-conjoined constraint *[+constricted glottis]² _f (Marlett and Stemberger 1983; McCarthy 2008: 218) prohibits a sequence of glottal stops within a foot:

Next, a constraint against a glottal stop followed by a consonant cluster is also active:

These constraints are dominated by Faith _root; there are disyllabic stems which contain two glottalized vowels in consecutive syllables, or a sequence of a glottal stop followed by a consonant cluster. On the other hand, these constraints do not interact with *(tV.ˈσ) since the allomorph with a glottal stop does not contain a t, and thus the ranking between them is indecisive at this point. *[+constricted glottis]² _f and *ʔCC dominate Max _[+c.g] , the last of which prohibits deletion of a glottal stop, as we will see below.^[18]

The following tableau illustrates the selection of the allomorph gi-, without a glottal stop, when the stem contains a glottal stop. Candidate (a) violates *(tV.ˈσ), while candidate (b) violates *[+constricted glottis]² _f. These constraints are ranked higher than Priority which the optimal candidate (c) violates, since it selects the dispreferred allomorph giʔ- of passive:

(29)

The following tableau shows that the monosyllabic allomorph without a glottal stop, gi-, is selected when the stem begins with a consonant cluster. Candidate (a) violates *(tV.ˈσ) and candidate (b) violates *ʔCC, both of which are ranked higher than Priority, which the optimal candidate (c) violates.

(30)

When a monosyllabic stem does not have a glottal stop or begin with a consonant cluster, the allomorph with a glottal stop, giʔ-, is selected. This is illustrated in (31); here, candidate (a) violates *(tV.ˈσ) and thus is suboptimal, while candidate (c) violates Max _[+c.g] in addition to Priority and thus is suboptimal:

(31)

When a disyllabic stem contains two syllables with a glottal stop, both of these glottal stops are realized, due to Faith _root. This is illustrated in (32). Here, candidate (b) is suboptimal due to this constraint since one of the glottal stops in the input is not realized. Candidates (c) and (d) incur violation of Priority by having the dispreferred allomorph of the prefix, giʔ-:

(32)

Likewise, a stem can contain a glottal stop followed by consonant cluster, again due to Faith _root. The following tableau illustrates this. Here, candidate (b) is suboptimal for this reason.

(33)

4.2 2sg agentive ta- ∼ tara-

Equally, the 2sg agentive allomorphy described in Section 3.2 can now be accounted for as follows under the P » M model: as in the case of the passive allomorphy, 2sg agentive allomorphy is due to the constraint against t in the weak-branch of the foot, *(tV.ˈσ). Recall that 2sg agentive prefix has two allomorphs, ta- ∼ tara-, the first of which occurs with monosyllabic stems while the latter with disyllabic stems when the stem does not begin with a voiced consonant. When the stem begins with a voiced consonant, the allomorph is always ta-, regardless of the stem size. Again, we employ the constraint Priority to stipulate that the allomorph ta- is the preferred allomorph compared to tara-; for the allomorphy in question either preference works, but taking into account the alternation in the next subsection with the suppletive verbs, we conclude that ta- is the preferred allomorph out of the two. The phonological constraint *(tV.ˈσ) and the morphological constraint Priority are ranked as P » M, namely *(tV.ˈσ) » Priority.

The following tableaux illustrate this. In (34), with a monosyllabic root, candidate (b) violates *(tV.ˈσ) and thus is suboptimal, while in (35), with a disyllabic root, candidate (a) violates both *(tV.ˈσ) and Priority since it selects the dispreferred allomorph of the 2sg agentive prefix, tara-:

(34)

(35)

We showed above that tara- is not observed when the stem begins with a voiced consonant. Under a P » M model, this is due to *rVDV, which bans the sequence of a syllable beginning with r followed by another syllable beginning with a voiced consonant,^[19] which dominates *(tV.ˈσ). This is illustrated in the following tableaux. First, in (36), with a monosyllabic stem beginning with a voiced consonant, candidate (a) is suboptimal due to this *rVDV. The tableau in (37) shows a disyllabic stem beginning with a voiced consonant where the other syllable-initial consonant of the stem is also voiced. Here, candidate (c) is suboptimal since it violates the undominated Faith _root by altering the stem initial consonant, while candidate (a) incurs violation of *rVDV.

(36)

(37)

In a disyllabic stem the sequence rVDV is observed, as in -rígu ‘lower’ (out of the 423 disyllabic verb stems, eight stems (1.9%) have such a structure; cf. Table 3 and 5 below); this is again due to Faith _root introduced in the preceding subsection, which is undominated. When allomorphs which would avoid the sequence rVDV are not available, as in the case of other aspect-mood prefixes which begin with r, no segment is deleted despite the violation of *rVDV_. To prevent deletion of segments from such prefixes, we introduce Max _[-c.g] which prohibits deletion of segments that are not a glottal stop, which dominates *rVDV. Tableaux (38), with a rVDV sequence in the root, and (39), with a progressive prefix ra- attached to the stem beginning with a voiced stop, illustrate this. In both cases, the candidates with the rVDV sequence are selected since other candidates violate the undominated Faith _root or Max _[-c.g] .

(38)

(39)

4.3 2sg agentive ta- ∼ Ø of suppletive verbs

In Section 3.3, we saw that suppletive verbs are different from regular verbs in that disyllabic suppletive verbs take the allomorph ta- as in regular verbs, but monosyllabic suppletive verbs take zero allomorph, rather than tara- as in regular verbs. We can interpret this fact in the following way under the P » M analysis: in addition to the constraints discussed in the preceding subsections, namely *(tV.ˈσ) and Priority, the constraint against homophony (Baermann 2011) comes into play. That is, with suppletive verbs, when *(tV.ˈσ) cannot be obeyed, the 2sg agentive prefix has the allomorph zero, rather than tara- like in regular verbs, since for suppletive verbs the information about the agent person category is already encoded in the stem. In fact, homophony does not result even without the agent prefix for suppletive verbs, since the 2sg stem is always distinct from 3sg; the 3sg form would otherwise be homophonous without the 2sg agentive prefix for regular verbs (unless the glottal stop is realized; Section 3.4).^[20]

This constraint against homophony can be postulated as Anti-ident (Crosswhite 1999), which is based on the Ident constraints of Correspondence theory (McCarthy and Prince 1995). Anti-ident states that for two forms, S₁ and S₂, there must be some segment α which is a member of S₁ such that α is not identical to its correspondent in S₂:^[21]

With this constraint at hand, we can now modify the Priority constraint and the hierarchy for the 2sg agent forms. First, the Priority constraint for the 2sg agentive prefix will now be {ta- › Ø- › tara-}, instead of {ta- › tara-}, as was proposed in Section 4.2, to incorporate the zero allomorph of the 2sg agentive prefix with suppletive verbs. Secondly, the raking of these constraints will be as follows: Anti-ident » *(tV.ˈσ) » Priority. We will first discuss the cases of regular, non-suppletive verbs and then of suppletive verbs.

First, for regular monosyllabic verbs, such as -daʔ ‘throw away’, illustrated in (41), candidate (a), with the least preferred allomorph tara- of the 2sg agentive prefix, is suboptimal because it also violates *rVDV. On the other hand, candidate (c) violates Anti-ident since it results in homophony with the 3sg (as well as 1pl.in, 3pl) form nidaʔ. The optimal candidate (b) only violates *(tV.ˈσ).

(41)

A regular disyllabic verb is exemplified in (42) with the verb -ská ^h ma ‘show’; candidate (a) violates *(tV.ˈσ) and is thus suboptimal, while candidate (c) violates Anti-ident since it results in homophony with the 3sg form, niská ^h ma.

(42)

For suppletive monosyllabic verbs, such as -k ^h u/-ʦo/-p ^h u ‘bite’ (43), the absence of the 2sg agentive prefix in niʦo does not result in homophony with the 3sg form which is niʔk ^h u, and thus does not violate Anti-ident. Candidate (a) is suboptimal since it has the least preferred allomorph tara- of the 2sg agentive prefix. Candidate (b), which selects the most preferred allomorph, is ruled out by *(tV.ˈσ).

(43)

Suppletive disyllabic verbs are exemplified by the Tableau (44) with the verb – ríʔmáʔ/-gamaʔ ‘go up’. Candidate (a) incurs violation of *rVDV twice due to the sequence of two consecutive syllables with r, followed by a syllable with m. Candidate (c), with the zero allomorph of the 2sg agentive prefix, is suboptimal even though it does not result in homophony with the 3sg form (nigamaʔ), since it has the second preferred allomorph of the 2sg agentive prefix, Ø-. As a result candidate (b) is optimal.

(44)

4.4 Glottal stop of the 3sg and 1pl.in agentive prefixes

Finally, we saw in Section 3.4 that the glottal stop is deleted with disyllabic stems (and when it precedes a consonant-cluster or when the stem already contains a glottal stop, regardless of the stem size). Under the P » M model, the generalization is that the glottal stop can only be realized when the stem is monosyllabic (as well as other phonological factors, such as the presence of a lexical glottal stop in the stem and if the stem begins with a cluster or not). We interpret this descriptive generalization as follows: within a prosodic word, a glottal stop can only occur within the final, head foot (cf. Carrasco Zúñiga 2012: 20).^[22] This may be related to the crosslinguistic tendency that glottal stop is realized as close to a strong syllable as possible (González 2003: 107–109). In Huehuetepec Tlapanec, a glottal stop can occur in the only syllable of monosyllabic stems or the two syllables of the disyllabic stems. However, polymorphemic verb forms never have a glottal stop outside of this domain. One piece of independent evidence for this constraint is the fact that there is no disyllabic verb stem which begins with a glottal stop (*-ʔ(C)VCV), while there are monosyllabic stems beginning with a glottal stop, such as -ʔgaà ‘boil’, -ʔngà ‘tolerate’. If there were any disyllabic stems beginning with a glottal stop, this glottal stop would always appear before the head foot, since a post-vocalic glottal stop is always syllabified with the preceding vowel, even when there is no following consonant (Tiburcio Cano 2017: 48). This generalization, that a glottal stop can only occur in the head foot, can be formulated as the following Licensing constraint (cf. Steriade 1997):

This constraint dominates the faithfulness constraint Max _[+c.g] against deletion of a glottal stop: Lic-[+c.g] /HdFt » Max _[+c.g] . Other relevant constraints are *[+c.g]² _f and *ʔCC, which were already introduced in Section 4.1. First, (46) shows the 3sg prefix with a disyllabic stem; here, candidate (b) is suboptimal since the glottal stop is found outside of the final head foot, even though candidate (a) violates Max _[+c.g] by deleting the glottal stop of the prefix.

(46)

On the other hand, (47) illustrates a case with a monosyllabic stem without a glottal stop or a stem-initial consonant cluster (note that palatalized segments such as s ^j do not count as clusters for this process). Here, candidate (b) violates Max _[+c.g] since it deletes a glottal stop from the prefix.

(47)

When the stem is monosyllabic but contains a glottal stop, as illustrated in (48), candidate (a) is suboptimal due to *[+c.g]² _f, which prohibits a glottal stop in adjacent syllables. The same holds true for stems beginning with a consonant cluster due to the constraint *ʔCC (49):

(48)

(49)

4.5 Summary

In this section, we have provided an alternative, P » M analysis for the four cases of allomorphy that are based on the foot structure in Tlapanec presented in Section 3, under the subcategorization approach. Two of the morphemes that show foot-based allomorphy, the passive gita- ∼ giʔ- (Section 4.1) and the 2sg agentive ta- ∼ Ø ∼ tara- (Section 4.2, Section 4.3) are argued to be motivated by the same constraint against t occurring in the weak branch of the foot, (*tV.ˈσ). On the other hand, the presence or absence of a glottal stop in the 3sg/1pl.in agentive prefix is motivated by the constraint against a glottal stop outside of the head foot, Lic-[+c.g] /HdFt.

The following Figure 1 is a Hasse diagram summarizing the constraint interaction that accounts for the Tlapanec prefix allomorphy. The consistency of the ranking of these constraints has been checked with the OTSoft 2.5 (Hayes et al. 2013), and the Hasse diagram was drawn using the same program.

Figure 1:

Hasse diagram of constraint interactions.

5.1 Lack of a strong phonological motivation for the proposed constraints

The advantage of the P » M approach in general is its ability to state the markedness-driven motivation explicitly in terms of constraints, which conspire to trigger various phonological and morphological alternations to avoid such a marked structure. This advantage is the most obvious in the case of the deletion of a glottal stop in the 3sg agentive prefix with a disyllabic stem, which is due to a general phonological constraint in the language, namely the prohibition of a glottal stop outside of the head foot within the prosodic word. The same is true with the deletion of a glottal stop before a consonant cluster or with a stem which already contains a glottal stop that we observed with the passive and 3sg agentive prefixes. Such constraints also have typological and phonetic motivations: avoidance of a glottal stop in consecutive syllables is cross-linguistically common as was mentioned, and the avoidance of a glottal stop before a consonant cluster is possibly due to a constraint against complex syllable margins (*Complex; Kager 1999; 97).

However, not all the proposed constraints are typologically common, nor are synchronically motivated language internally. In the current case, the constraint *(tV.ˈσ) is considered to conspire to trigger allomorphic alternations in the passive and the 2sg agentive prefixes, with different ‘repair’ strategies: deletion of the syllable in the passive, and reduplication in the case of 2sg agentive. However, this constraint lacks generality and independent motivation in Tlapanec (Section 5.1.1). The same can be said for the constraint *rVDV introduced in Section 4.2 to account for the fact that the allomorph tara- of the 2sg agentive prefix is not allowed before stems beginning with a voiced consonant (Section 5.1.2).

5.2 Powerful theoretical add-ons

The second problem with the P » M approach proposed in Section 4 is that the analysis makes use of a number of non-standard mechanisms, making the analysis somewhat of a patchwork of powerful theoretical add-ons including the following.

First, a self-conjoined constraint, *[+c.g]² _f, was proposed in Section 4.1 to account for the dispreference of a glottal stop in the consecutive syllables. Kager (1999: 400) lists some conceptual problems with local conjunction, which also apply to self-conjoined constraints. First, minimal violation by itself should suffice to exclude the worst-of-the-worst, rather than an extra mechanism of constraint conjunction; accepting it means that the theory has two means to rule out excessive violation of faithfulness: (a) minimal violation of basic constraints, and (b) local conjunction. The second issue with local conjunction is that it undermines strict domination, which dictates that violation of higher-ranked constraints cannot be compensated for by satisfaction of lower-ranked constraints. Under self-conjunction, a constraint A is ranked too low to force the violation of B, but can nevertheless dominate B by joining forces [A]². Finally, allowing local conjunction would increase possible constraints in Gen: can any constraints be combined or self-conjoined?

The second issue with the P » M approach presented in Section 4 is that various constraints have been proposed to account for the prefix allomorphy that are not purely phonological: Priority, which dictates the preferred allomorph; and Anti-Ident against homophony, which is a transderivational (paradigmatic) constraint. In general, the null hypothesis is that faithfulness constraints are blind to morphologically structure, and thus allowing reference to morphological structure renders the machinery of OT too powerful (Kager 1999: 78).

5.3 ‘Too-many repairs’ problem

Generally, the main issue that P » M approach is faced with is that there are too many repair strategies to resolve the marked structure (cf. Yu 2007: Section 2.5.3). For instance, in Section 4, it was argued that the allomorphic alternations of the passive and the 2sg agent prefixes are due to the constraint *(tV.ˈσ), and that the allomorphs are selected so that such a structure is avoided. However, there are other strategies to avoid such a structure; one alternative solution would be to alter the morpheme order to avoid a marked structure. To avoid other such strategies, we would need more undominated constraints than were shown in Section 4, and the number of constraints that would be needed would be quite large.

5.4 The need of subcategorization frames even under a P » M analysis

In Section 4, we saw that the subcategorization frames are necessary anyway to account for some cases where the allomorph selection is not motivated by any markedness constraints, but rather lexically specified. For instance, we saw in Section 4.2 that the allomorphy of the 2sg agentive prefix is unpredictable for glottal-initial stems; there is no phonological motivation that triggers this allomorph selection, and thus the allomorph selection needs to be stated in subcategorization frames. The alternation of the vowels i ∼ e in the passive prefix (Section 3.1) would also require subcategorization frames for the same reason. Since subcategorization is already necessary even under a P » M analysis, a more parsimonious analysis would be to employ subcategorization for all of the allomorphy, as was proposed in Section 3.