2.1 Basic phonology

This section describes the phonological inventory of SMP Mixtec, with all generalizations drawn from Eischens & Hedding (2025). Table 2 lists the phonemic consonants. One notable feature is the contrast between plain and prenasalized stops and affricates, a common contrast in Mixtec languages (Marlett 1992; Iverson & Salmons 1996).

SMP Mixtec has five phonemic oral vowels and three phonemic nasal vowels (Table 3). Important for our discussion in §4 is the fact that the language has oral mid vowels ([e] and [o]), but not nasal mid vowels ([ẽ] and [õ]). Restrictions on the distribution of mid vowels are common across Mixtec languages (Pike 1947: 168–169).

Finally, in SMP Mixtec, vowels only contrast for nasality after the plain (non-prenasalized) stops, affricates, and fricatives. Everywhere else, their orality/nasality is predictable. We discuss these distributional patterns in depth in §4.

There are at least five contrastive tones in SMP Mixtec (Peters 2018). Three are level tones: High [V́], Mid [V], and Low [V̀]. These contrast with contour tones, which include a Low-to-High rise [V̌] and at least one falling tone [V̂]. The mora is the tone-bearing unit, meaning that any mora may bear any one of the language’s five contrastive tones. Contours of three tones or more cannot be hosted on a single mora – these are only found on bimoraic long vowels.

SMP Mixtec also contrasts modal, laryngealized, and breathy phonation on the penultimate mora of stems (though see Peters 2018 for a consonantal analysis of [ʔ] and [h]). We concur with Macaulay & Salmons (1995) that these are best analyzed as suprasegmental features of the morpheme, though this analysis is not crucial for the main points of this paper. These contrasts are discussed at length in §3.

SMP Mixtec has a series of pronominal clitics which typically follow their hosts. In §2.2 we discuss vocalic enclitics of the form /=V/. In §4.4 we argue that these enclitics are parsed into the same prosodic word as their hosts, and form a bimoraic foot with root material. In §3.1.2 we argue that larger, CV enclitics are instead external to the prosodic words of their hosts.²

2.2 Vocalic enclitics

SMP Mixtec features a number of pronominal enclitics that overwrite the final vowel of the stem to which they attach (Ostrove 2018; Peters & Mendoza 2020; Mendoza 2020; Hedding 2022). Because they feature prominently in §4, we spell out their relevant morphophonological behavior in detail here. (Larger, CV clitic pronouns are discussed in §3.1.2.)

Vocalic enclitic pronouns /=V/, which are used primarily as subject markers, markers of possession, and the objects of prepositions, are listed in Table 4 below:

We classify these morphemes as clitics because they are non-selective with respect to their hosts: they may directly follow, and ‘lean’ on verbs, adverbs, nouns (in possession), and prepositions (e.g. Macaulay 1987a; b; 2005).

Similar vocalic enclitics are found in other varieties, where they interact in various ways with stem-final vowels, including coalescence (Ixtayutla, Penner 2019: 217), diphthong and glide formation (Alcozauca, Uchihara & Mendoza Ruiz 2022: 626–627), the loss of a stem-final vowel (Alcozauca, Uchihara & Mendoza Ruiz 2022: 626–627; Huajuapan, Pike & Cowan 1967: 12–13), and in some cases, unrepaired vowel hiatus (Huajuapan, Pike & Cowan 1967: 13). Of these strategies, SMP Mixtec displays the complete loss of stem-final vowels (5) and the conversion of stem-final vowels into consonant off-glides (6).

Whether the stem-final vowel is lost or becomes an off-glide depends on the identity of the stem-final vowel and the identity of the preceding consonant. For example, stem-final /i/ is lost after [ʃ], as in (5), but retained as an off-glide after [s], as in (6). Whether the stem-final vowel is lost or becomes an off-glide is also subject to interspeaker variation for some words. Even with these restrictions, it is the case that each enclitic is capable of completely overwriting stem-final vowels.

Vocalic enclitics also undergo tonal variation when overwriting stem-final vowels (Peters & Mendoza 2020). When a high-toned enclitic attaches to a low-final stem, the result is a low-to-high rise (7). If a low-toned enclitic combines with a high-tone-final stem (8) or a rise-final stem (9), then the result is a falling tone. Sometimes, contour tones are produced through the combination of an enclitic with a mid-tone-final stem, though this is lexically-determined (Peters & Mendoza 2020, cf. (5)–(6)).

These enclitics also interact with the nasality of stem-final vowels, as summarized in Table 5. These patterns are the focus of §4, so we defer detailed discussion till that point.

Similar patterns of nasality have been described for vocalic enclitics in other Mixtec languages, such as Huajuapan Mixtec (Pike & Cowan 1967), Ixtayutla Mixtec (Penner 2019), Yoloxochitl Mixtec (DiCanio et al. 2020), and San Martín Duraznos Mixtec (Auderset et al. 2024). In these varieties, it is common for the 2sg enclitic to be invariably nasal (Pike & Cowan 1967; North & Shields 1977; Gerfen 1996; Becerra Roldán 2019; Penner 2019; DiCanio et al. 2020; Auderset et al. 2024), for the 1pl.incl to be invariably oral (Pike & Cowan 1967; DiCanio et al. 2020; Auderset et al. 2024), and for the 1sg enclitic to preserve the nasality/orality of the base (Pike & Cowan 1967; Becerra Roldán 2019; Penner 2019; Auderset et al. 2024) (just as in Table 5). In other Mixtec languages, vocalic enclitics show uniform patterns of nasality. For example, all vocalic enclitics in Mixtepec Mixtec take on the nasality/orality of the final vowel of the root to which they attach (Belmar Viernes 2024).

3.1 The distribution of laryngeals

Across Mixtec languages, [ʔ] has a restricted distribution, which is often defined in terms of the couplet. For example, Josserand (1983: 228) writes that “[ʔ] occurs in couplet-medial position”, and Macaulay & Salmons (1995: 54) claim that “[ʔ] is…a feature of the couplet”. Because the distribution of [ʔ] is analyzed in terms of the couplet, this section outlines the behavior of laryngeals (both [ʔ] and [h]) in SMP Mixtec, arguing that their distributional characteristics are best understood in terms of a bimoraic foot instead of a morphological root. The argumentation broadly follows the lines of Penner (2019), who convincingly argues that the distribution of [ʔ] in Ixtayutla Mixtec is best defined in terms of a bimoraic foot.

Our argumentation relies on a non-isomorphism between the morphological root and the bimoraic foot in long roots. Given that most roots in SMP Mixtec are bimoraic, the root and bimoraic foot are usually coextensive. However, there are also a number of roots in SMP Mixtec which are trimoraic or larger. Many of these roots contain prefixes that are no longer productive, or are fossilized compounds. Examples of such constructions can be seen in (10)–(13).³

The semantic contribution of most fossilized prefixes in SMP Mixtec, such as those in (10)–(11), is no longer clear. For example, the [^͡tsì-] in (10) likely descends from a classifer used for animals and round things, since similar prenominal classifiers are productive in other varieties (e.g. de León Pasquel 1988: 142 on Coatzoquitengo Mixtec). In SMP Mixtec, [^͡tsì-] is used in many animal names, but also with some plants (e.g. [^͡tsìkʷàʰǎ], ‘orange’) and adjectives (e.g. [^͡tsìkàβà], ‘crooked’). Additionally, it appears in some words in which it is not clearly a prefix but is instead integrated into the bimoraic root, such as in [^͡tsìnã̀] ‘dog’ and [^͡tsĩ̀ĩ́] ‘mouse’. Finally, it has a phonologically distinct form in some words. For example, it is [^͡tʃì-] in [^͡tʃìkʷii] ‘fox’. The [nã-] in (11) may derive from a ‘repetitive’ marker described as unproductive in other Mixtec varieties (e.g. Alexander 1988: 245–246; Zylstra 1991: 103–104). It is similarly unproductive in SMP Mixtec, where repetition of an action is signaled by the post-verbal adverb [tùʰkù] ‘again’.

The forms in (12) and (13) may derive historically from compounds, though they are now lexicalized. For example, the [ka-] in (12) may be a compound of [kaʰsa] ‘will do’ and [^͡tʃiɲũ] ‘work (n)’, though such CV truncation in compounds is not synchronically productive. Finally, the [ta-] in (13) is of uncertain origin, though the full form likely derived from a compound containing the word [ⁿtaˀǎ] ‘hand’. Because of the non-productivity and opaque meaning of such prefixes and compounds, a number of scholars maintain that at least some words that are trimoraic or larger, like those in (10)–(13), are synchronically mono-morphemic, with the antepenultimate syllable now analyzed as a part of the root (e.g. Penner 2019; DiCanio et al. 2020).

Roots of this shape allow for the morphological and prosodic analyses of the Mixtec couplet to be tested against each other, because each analysis makes different predictions (Table 6). If the couplet corresponds to the morphological root, then generalizations that hold of the couplet should hold of both bimoraic and trimoraic roots. However, if the couplet corresponds to the bimoraic foot, then there should be an asymmetry between bimoraic and trimoraic roots, given that the foot may only include two morae. For reasons which will become clear in this section, we follow Penner (2019: 211) in analyzing the foot as aligned to the right edge of the prosodic word. Under this assumption, there is a non-isomorphism between the root (enclosed in vertical bars) and the foot (enclosed in parentheses) in trimoraic roots, as seen in (14) vs. (15). This non-isomorphism leads to the predictions in Table 6.

3.2 Rising tones

Tonal phonotactics, and tonal melodies in particular, are frequently described for Mixtec languages in terms of the couplet. In fact, Pike’s (1948) original coining of the term tonemic couplet was in the service of describing tonal melodies in a Mixtec language. Since tonal distributions are often analyzed in terms of the couplet, we describe one aspect of tonal phonotactics that points to the foot as an active unit in SMP Mixtec phonology. In particular, we show that underlying rising tones may only surface foot-internally.⁵ That is, rising tones may be found on either mora of the bimoraic foot, but not on material that precedes or follows the foot.

Many words in SMP Mixtec contain monomoraic rising tones. These are most common on the final mora of the word (38), but there are many bimoraic words with an initial rising tone (39). For readability, we include the tonal melodies in parentheses next to the IPA transcriptions in this section.

Rising tones are phonologically distinct from low (40), mid (41), and high (42) tones, with which they contrast.

Despite their ubiquity, rising tones have a restricted distribution in SMP Mixtec. Specifically, underlying rising tones are only ever found on the penultimate or final mora of a prosodic word. That is, there are no underlying rising tones on fossilized prefixes, and neither are there rising tones on pronominal enclitics.

This pattern means that there are no trimoraic or quadrimoraic roots in which a rising tone occurs on the antepenult or preantepenult. This is in spite of there being many trimoraic and larger roots in which rising tones occur within the right-aligned foot, as shown in (43).

Additionally, CV clitic pronouns that occur to the right of the root do not host rising tones or falling tones – they only host level tones. This is in spite of the fact that some clitic pronouns have independent counterparts with contour melodies. For example, (46a) shows a independent pronoun with a H-L melody, whose dependent form has only a H tone (46b) instead of an HL contour tone, though falling tones are allowed root-internally (Peters 2018; Eischens & Hedding 2025). This tonal distinction between independent and dependent pronoun pairs suggests a pressure against contour tones on dependent pronouns. This can be interpreted either as contour tones being restricted to the foot, or as contour tones being restricted to the domain of the prosodic word, which we have argued CV clitics fall outside of.

So, in a phrase like (47), underlying rising tones may only occur in the bimoraic foot, which is in parentheses and excludes the CV clitic pronoun.

As was the case for laryngeals, the restricted distribution of rising tones is straightforwardly accounted for by a generalization in terms of a right-aligned, bimoraic foot _Rt| …_Ft(μμ) |: underlying rising tones only occur foot-internally, not foot-externally. Under the assumption that at least some words that are trimoraic or larger constitute a single root, the absence of any underlying rising tones on the antepenult or preantepenult in such words makes clear that the root is not an especially useful unit for describing the distribution of underlying rising tones.

Additionally, it is worth pointing out that the distribution of laryngeals was potentially analyzable either as a property of the foot or of stressed morae. The distribution of rising tones cannot be accounted for by making reference to stressed morae — underlying rising tones are licensed on both morae in the foot, not just the stressed one (whether feet are taken to be trochaic or iambic in SMP Mixtec).

3.3 Summary

In this section, we have described two phonotactic patterns in SMP Mixtec that have been analyzed as properties of the couplet in other Mixtec languages. These are the distribution of laryngeals and rising tones.

Using the logic laid out in Penner (2019), we have argued that both of these phonological patterns point to the existence of a bimoraic foot aligned to the right edge of a prosodic word. That is, when considering roots that are trimoraic or larger, it is clear that the final two morae have different phonological properties from preceding material. Since all of this phonological material is contained within the same root, the generalizations in question cannot be accounted for by appealing to morphological structure. Instead, both generalizations are straightforwardly accounted for by assuming the existence of a bimoraic foot that is right-aligned to the prosodic word.

In this sense, these results align with the arguments in Penner (2019) that the bimoraic foot is representationally equivalent to the cross-Mixtec couplet. Since the couplet has been used to demarcate the distribution of laryngeals and tonal melodies in Mixtec languages (Macaulay & Salmons 1995; Daly & Hyman 2007), and since the foot can be used in the same way to account for the same generalizations, then an analysis of the couplet as coextensive with a bimoraic foot is appropriate.

However, this is not the whole story: patterns of co-occurrence between oral and nasal segments have also been analyzed as a property of the Mixtec couplet, but these patterns are not compatible with a solely prosodic analysis. Instead, any account of the data refer directly to morphological structure, at least in SMP Mixtec. We turn to these facts in the next section.

4.1 Representative nasal air pressure data

In this section, we illustrate patterns of vowel and consonant nasality with nasal air pressure data from two speakers of SMP Mixtec (see Gerfen 1996 and Herrera Zendejas 2014 for previous nasal airflow studies on Mixtec languages). Oral and nasal air pressure recordings were made with Glottal Enterprises (GE) PT-2E pressure transducers, mounted on an oro-nasal mask. The GE oro-nasal mask has separate oral and nasal chambers, which allows oral and nasal air pressure to be recorded more-or-less independently. The oral pressure transducer was mounted directly on the mask, and the nasal pressure transducer was mounted on a GE DRTH-1 handle connected to the mask.

The air pressure transducers were connected to a GE MS-110 pressure transducer unit with a GE BFC-2 cable. For both speakers, the air pressure recordings were made at 11,025 Hz using Audacity. The amplitude of the nasal and oral air pressure channels were normalized to [0,1] for each speaker separately. All data processing, including normalization, was carried out with custom scripts in Praat and R (R Core Team 2013; Boersma & Weenink 2020).

Our measure of vowel nasality is ‘nasalance’. Nasalance is calculated by dividing the amplitude of pressure in the nasal chamber by the sum of the amplitude of pressure in the oral and nasal chambers (A_nasal / (A_nasal + A_oral)). Nasalance thus ranges from 0 to 1, is lower in oral sounds, and is higher in nasal sounds.⁶

Nasalance was calculated by low-pass filtering the absolute value of the normalized oral and nasal air pressure signals from [0, 280]Hz, smoothing the filtered signals with an 11ms averaging window centered on each measurement point, then taking the ratio A_n / (A_n + A_o) at each point. Nasalance was set to N/A for any point with nasal air pressure at or below 1% of the maximum for each speaker.

In plots, we only show nasalance for vowels, nasal consonants, and [ʔ] and [h], which may be contextually nasalized in SMP Mixtec. A representative token of the amplitudes of individual chambers, as well as nasalance, is given in Figure 1. The first row shows the normalized amplitude of air pressure from the oral cavity, the second shows the normalized amplitude of air pressure from the nasal cavity, and the third shows low-pass filtered nasalance. Note that the increased nasal air pressure in the second syllable in Figure 1 corresponds to increased nasalance on nasal [ĩ] vs. oral [i] in the first syllable. For reasons of space, we only present low-pass filtered nasalance in subsequent figures, omitting the original oral and nasal pressure channels. In all nasalance plots, numbers below segmental transcriptions indicate duration in ms.⁷

Figure 1: Oral air pressure, nasal air pressure, and low-pass filtered nasalance during chìxin [^͡tʃìʰʃĩ] ‘stomach/estómago’ (Speaker RDC).

We collected data from two participants, each of whom produced target words from a wordlist embedded in the carrier phrase in (48).

Speaker NGC produced 105 items, with 4 repetitions each (420 total productions). Speaker RDC produced a 70-item subset of the wordlist produced by NGC, with 2 repetitions each (140 total productions). Summary plots like Figure 2 also include tokens from a separate dataset in which speaker RDC produced 84 morphologically-simplex items, with 3 repetitions each (252 total productions). This extra dataset was included to increase the number of tokens used to calculate baseline nasalance levels for oral and nasal vowels. Full wordlists are included in the supplementary materials, along with nasalance plots for each recorded token.

Nasalance levels may vary by vowel height. In particular, nasal [ã] has markedly lower nasalance in our data than nasal [ĩ] or [ũ]. This is clear in Figure 2, which shows a summary of nasalance across vowel qualities for contrastively oral /V/ and nasal /Ṽ/. The weaker nasalance for nasal /ã/ compared to /ĩ ũ/ may be a result of a wider aperture in the oral cavity for low vowels, leading to increased oral airflow, relative to nasal airflow (e.g. Young et al. 2001).

Figure 2: Mean nasalance across vowel qualities for contrastively oral /V/ (left panel) and nasal /Ṽ/ (right panel). Measurements taken from root-internal vowels which were not adjacent to any nasal consonant. Measurements reflect middle 40–75% of each vowel (steps 7–13 of 17 total). Circles indicate means, bars ±1 standard deviation. Dashed red lines and surrounding bands indicate mean nasalance for contrastively oral /V/ and nasal /Ṽ/ across all vowel qualities, and ±1 standard deviation. Numbers at bottom of plot indicate observations per condition.

4.2 The phonotactics of nasality

A contrast between oral and nasal vowels is a consistent characteristic of Mixtec languages (Rueda Chaves 2021). As is common in languages with such a contrast, oral and nasal vowels are not contrastive in every phonological context. For example, vowels following nasal consonants are usually obligatorily nasal (Pike & Cowan 1967; Zylstra 1980; Marlett 1992; Gerfen 1996; León Vásquez 2017), as shown for SMP Mixtec in (49).

Phonotactic restrictions on nasality have often been cast in terms of the couplet in the Mixtec literature. There are many examples of this, and we list a few here: Pike & Cowan (1967: 5) write that “If the first of two vowels is nasal in a monomorphemic couplet, the second vowel is usually nasal”. Hunter & Pike (1969: 30) say, “Allophonic nasalization of vowels is best described in relation to the couplet”. North & Shields (1977: 28) state that “Nasal vowels not preceding and following /m n ñ/ are restricted in their distribution in the couplet”. Zylstra (1980: 21) notes that “Nasalization of vowels extends through couplets of the form CVV and CVʔV where the vowels are identical”. Gerfen (1999: 260) says of nasalization that “a rule will associate [nasal] to the rightmost vowel of a couplet, while another [rule] will subsequently spread [nasal] to the left”. Paster & Beam de Azcona (2004: 67) write that, in the Yucunany dialect of Mixtepec Mixtec, “In couplets, vowel nasalization usually occurs in both syllables or neither”. Rueda Chaves (2019: xv) treats nasalization as a characteristic feature of couplet boundaries (original: “se abordan la nasalización y la glotalización como rasgos característicos de los lindes del couplet”). For Penner (2019: 265), “The domain of nasal phonotactics in [Ixtayutla Mixtec] is the couplet”. Finally, after defining the couplet as a bimoraic minimal word, Becerra Roldán (2023: 75) writes, “Nasal morphemes have a floating [+nasal] feature that associates to the right boundary of the minimal word and spreads leftward to sonorants” (original: “Los [morfemas nasales] poseen un rasgo flotante [+nasal] que se asocia al linde derecho de la palabra mínima y se propaga a la izquierda a segmentos resonantes”).

As these quotes show, patterns of nasalization are intimately tied to the notion of the couplet. Because of this, nasalization is another appropriate domain for testing the two main hypotheses about the nature of the couplet.

We argue in this section that restrictions on nasal and oral segments are, contrary to the results in §3, best analyzed in terms of the morphological root, and not the foot. As we show, restrictions on nasality hold exceptionlessly within roots, but not necessarily within feet, pointing to the root as a correspondent of the Mixtec couplet.

4.3 Restrictions in mono-morphemic words

This section describes the phonotactic restrictions that hold on sequences of nasal and non-nasal segments in mono-morphemic roots, focusing on three exceptionless restrictions, which are common across Mixtec languages (Marlett 1992; Rueda Chaves 2021).⁸

The first restriction is that vowels in mono-morphemic roots are obligatorily oral when following approximants like [β], [l], and [j], as shown in (50)–(51) and Figure 3, where nasalance values are within the range for oral vowels established in Figure 2.

Figure 3: [jiβà] (50) (Speaker RDC).

Additionally, vowels in roots are obligatorily oral when following prenasalized consonants, as shown in (52)–(55), and illustrated for (52) in Figure 4.

Figure 4: [kòⁿtò] (52) (Speaker NGC).

The second restriction, noted at the beginning of this section, is that vowels following nasal consonants are obligatorily nasal in roots.⁹ This is exemplified in (56)–(57) and Figure 5 (recall that nasal [ã] has lower nasalance than other nasal vowels; Figure 2).

Figure 5: [ɲãnĩ] (56) (Speaker NGC).

The third and final restriction discussed here concerns adjacent vocalic morae. Specifically, sequences of adjacent vowels must match in nasality or orality within roots (58). The same is also true of vowels separated by [ʔ] or [h], as exemplified in (59)–(60). This is illustrated using air pressure data for (58) in Figure 6, and for (60) in Figures 7, 8. Note especially the nasalization of [h] in Figure 8, which suggests the spread of nasality between the two vowels.¹⁰

Figure 6: [kʷáà] (58a) vs. [kʷã́ã̀] (58b) (Speaker NGC).

Figure 7: [^͡tsi̥kʷàhǎ] (60a) (Speaker NGC).

Figure 8: [kʷã́ʰã̌] (60b) (Speaker NGC).

These three generalizations, which are listed together in Table 7, hold exceptionlessly within monomorphemic forms.

Having defined these restrictions, we next turn to a construction in which the morphological and prosodic views of the couplet make differing predictions.

4.4 Vocalic enclitics

The morphological and prosodic views of the Mixtec couplet make different predictions regarding the behavior of a particular construction in SMP Mixtec in which there is a non-isomorphism between the morphological root and the bimoraic foot. The key forms involve bimoraic roots followed by the enclitic pronouns described in §2, which completely or partially overwrite the preceding vowel (61).

When vocalic enclitics like [=ì] overwrite the final vowel of the root to which they attach, they take over the timing slot of the root-final vowel and are therefore foot-internal, as we will argue shortly. In these cases, the foot and morpheme boundaries are non-isomorphic, as visualized in the following pair of examples. (62) shows the bimoraic root [ⁿtaˀǎ] (‘hand’), in which the root and foot boundaries are aligned, and (63) shows the morphologically-complex foot [ⁿtaˀ=î] (‘my hand’), in which the morpheme and foot boundaries are misaligned.¹¹

Given this non-isomorphism between root boundaries and foot boundaries, morphological and prosodic analyses of the Mixtec couplet make differing predictions. Since (62) involves a single root and (63) involves a morphologically-complex structure, the ‘couplet = root’ approach allows for the two words to behave differently with respect to patterns that take the couplet as their domain. However, since (62) and (63) each constitute a single foot, the ‘couplet = foot’ approach predicts no difference between them with respect to characteristics of the couplet.

As we show in detail below, restrictions on the co-occurrence of oral and nasal segments within roots in SMP Mixtec may be violated in morphologically complex structures. These violations occur across morpheme boundaries that are foot-internal (63). This means that morphological structure must be used to define the domains in which nasal phonotactics hold—prosodic structure alone is not sufficient.

4.5 Violations of nasal phonotactics across morpheme boundaries

We turn now to cases in which phonotactic restrictions on nasal and oral segments—repeated in Table 8—are violated across the boundary between a root and vocalic enclitic. These patterns point to the root, not the foot, as the domain within which constraints on nasality are defined.

Recall that vocalic enclitic pronouns in SMP Mixtec behave differently with respect to nasality (Table 9). One is invariably nasal, another is invariably oral, and the others alternate in their nasality. In this section, we focus on enclitics with invariable nasality/orality; the following section considers alternations in enclitic nasality.

We illustrate patterns of enclitic nasality with nasalance traces from individual examples/tokens, as well as with summary nasalance plots. In summary nasalance plots, we include the first three enclitics in Table 9 (1sg [=ì], 1pl.incl [=é], and 2sg [=ṹ]), but not 3sg.n [=à]. The reason for excluding 3sg.n [=à] is that its nasal/oral alternation is more variable than the other three enclitics, making it less straightforward to interpret in summary plots; we discuss this variability in §4.6.

The pronouns that are invariably oral or nasal often give rise to violations of the phonotactic restrictions in Table 8. For example, restriction 1 states that prenasalized consonants and approximants are always followed by oral vowels. This can be violated in constructions involving the 2sg enclitic [=ṹ]. (69)–(70) show the nasal enclitic [=ṹ] following approximants. The air pressure patterns in Figure 9 show high and rising nasalance on the word-final vowel.

Figure 9: [jiβ=ũ̌] (69b) (Speaker NGC) (compare with Figure 3 [jiβa] (69a)).

This pattern is consistent within and between the two speakers recorded, as shown in Figure 10. Specifically, when following approximants, nasalance for the 2sg [=ũ] is clearly higher than for other enclitics, as well as unmodified, root-final vowels.

Figure 10: Mean nasalance for enclitic vowels and root-final vowels after approximants /β j l/. Apparent violation of nasal phonotactics highlighted in grey. See Figure 2 for additional details.

In the same way, the 2sg enclitic [=ṹ] can follow prenasalized consonants, as shown in (71)–(72) and Figure 11.¹²

Figure 11: [kòⁿt=ũ̌] (72b) (Speaker NGC) (compare with Figure 4 [kòⁿtò] (71)).

The 2sg [=ṹ] is consistently nasal after pre-nasalized stops and affricates across the items tested, as shown in Figure 12. Nasalance for other enclitics and unmodified, root-final vowels is low in this environment. It is clear that restriction 1 can be violated in morphologically-complex constructions involving inherently nasal vocalic enclitics.

Figure 12: Mean nasalance for enclitic vowels and root-final vowels after prenasalized stops and affricates /ⁿt ⁿ^͡ts ⁿ^͡tsʲ ⁿ^͡tʃ/. Apparent violation of nasal phonotactics highlighted in grey. See Figure 2 for additional details.

Restriction 2 can also be violated in morphologically-complex environments. This restriction states that vowels following nasal consonants are obligatorily nasal. However, this generalization can be violated in constructions containing the 1pl.incl enclitic [=é].¹³ This is exemplified in (73)–(74). The air pressure example for speaker NGC in Figure 13 shows final [=é] with lower nasalance than for preceding nasal [ã] — the nasal vowel which generally has the lowest nasalance overall in our data (Figure 2).¹⁴

While there is clearly some coarticulatory nasality on [=é], due to the influence of the preceding nasal [ɲ], nasalance declines significantly over the course of the vowel. This pattern of descending nasalance is not characteristic of phonologically nasal vowels in our data (as is clear from plots provided here). We conclude that [=é] remains phonologically oral in this environment.

Figure 13: [ɲãɲ=ě] (73b) (Speaker NGC) (compare with Figure 5 [ɲãɲĩ] (73a)).

The low nasalance of 1pl.incl [=é] after a nasal consonant is apparent in the summary plot in Figure 14. Though nasalance for [=é] is somewhat higher than the oral baseline, this is likely due to coarticulation with preceding nasal consonants; its nasalance is far lower than that of the other enclitics, and root-final vowels, which are all clearly nasalized.

Figure 14: Mean nasalance for enclitic vowels and root-final vowels after nasal /m n ɲ/. Apparent violation of nasal phonotactics highlighted in grey. See Figure 2 for additional details.

Restriction 3, which requires that adjacent vowels match in nasality, can also be violated in morphologically-complex constructions. In fact, it can be violated in both directions. First, if an inherently nasal enclitic overwrites the second of two adjacent oral vowels, the result is an oral-nasal sequence (75)–(76). A nasalance plot for (75b) is given in Figure 15.

Figure 15: [ʃà=ũ̌] (75b) (Speaker RDC).

Second, if the inherently oral 1pl.incl [=é] overwrites the second of two nasal vowels, the result is a nasal-oral sequence (77). This is illustrated in Figure 16.

Figure 16: [^͡tsʲã́=ě] (77b) (Speaker NGC).

The nasality of the 2sg [=ṹ] in hiatus with a preceding oral vowel, and the orality of the 1pl.incl [=é] in hiatus with a preceding nasal vowel, are shown in a summary plot in Figure 17.

Figure 17: Mean nasalance for enclitic vowels immediately following stem-final vowels, after vowel overwriting in /CV₁V₁ = V₂/ → [CV₁ = V₂]. Apparent violation of nasal phonotactics highlighted in grey. See Figure 2 for additional details.

We have seen that restrictions on the co-occurrence of nasal and oral segments hold robustly root-internally, but are violated across morpheme boundaries. Importantly, these restrictions do not necessarily hold foot-internally, as shown schematically in (78)–(80). If vocalic enclitic pronouns are foot-internal, as we claim, segment sequences which violate nasal-oral co-occurrence restrictions are contained within the same foot.

Enclitic pronouns are foot-internal, but external to the domain of nasal phonotactics (being outside of the root, morphologically). It follows that the foot does not define the domain of phonotactic restrictions on nasality.

These patterns provide evidence that the couplet in SMP Mixtec is not exclusively coextensive with the foot, contrary to the patterns discussed in §3. Instead, nasal phonotactics—argued to hold over the couplet in many Mixtec languages (§4.2)—coincide with the morphological root in SMP Mixtec.

The data discussed in this section have involved enclitic pronouns whose nasality/orality is invariant. This leaves out two enclitics, namely 1sg [=ì] and 3sg.n [=à], which alternate in nasality. In the following section, we briefly describe the behavior of these enclitics, arguing that they are fully consistent with our claim that the root, and not the foot, is the domain of nasal co-occurrence restrictions in SMP Mixtec.

4.6 Alternations across morpheme boundaries

As noted in §2, the 1sg and 3sg.n enclitics alternate in nasality depending on the stem they attach to. In this section, we first discuss their shared behavior, and then we highlight cases in which the 3sg.n enclitic [=à] behaves variably. In most cases, however, the nasality of both of these enclitics can be predicted from the nasality of the stem-final vowel that they overwrite.

The nasality of the 1sg enclitic [=ì] is completely predictable based on the nasality of the stem-final vowel, and the same is usually true of the 3sg.n enclitic [=à]. Specifically, they surface as oral when overwriting an oral vowel (81), and as nasal when overwriting a nasal vowel (82), illustrated in Figures 18, 19.

Figure 18: [kúʰk=ì] (81b) vs. [sòʰk=ĩ̂] (82b) (Speaker NGC).

Figure 19: [kúʰkʲ=à] (81c) vs. [sòʰk=ã̂] (82c) (Speaker NGC).

This generalization is illustrated for the 1sg enclitic [=ì] in the summary plot in Figure 20, which shows the nasalance of vocalic enclitic pronouns when they overwrite contrastive oral and nasal root-final vowels. Alternating 1sg [=ì] has low nasalance when overwriting an oral vowel, and high nasalance when overwriting a nasal vowel. Note also the invariant nasalance of 1pl.incl [=é] and 2sg [=ṹ]; again, 3sg.n [=à] is not included in the plot due to the patterns of variability we discuss shortly.

Figure 20: Mean nasalance for stem-final and enclitic vowels after voiceless obstruents, where nasality is contrastive. Alternations based on nasality of the stem-final vowel highlighted in grey. See Figure 2 for additional details.

The same patterns hold when the nasality of the final vowel of the root is predictable: 1sg [=ì] is oral after approximants and prenasalized consonants, and the same is usually true of 3sg.n [=à] (83)–(84). Though representative nasalance traces are excluded for reasons of space, the low nasalance of 1sg [=ì] in these contexts can be seen in the earlier summary plots in Figures 10 and 12.

When following a nasal consonant, both enclitics always surface as nasal. This is exemplified in (85) and illustrated for 1sg [=ì] in the summary plot shown earlier in Figure 14.

Finally, when overwriting the second of two consecutive vowels, 1sg [=ì] takes on the nasality of the preceding vowel (86a), (87a), as seen earlier in Figure 17. The 3sg.n enclitic usually does, as well (86b), (87b).

4.6.1 Variability in 3sg.n [=à]

The 3sg.n [=à] enclitic displays some variability that merits discussion. While it often preserves the nasality or orality of the stem-final vowel, it sometimes surfaces as nasal when overwriting an oral vowel, as seen in (88) and Figures 21, 22. In these examples, there is no coarticulatory source for the increased nasalance on the 3sg.n enclitic.

1. (88)
1. a.

1. Xàà à
1. [ʃà=ã̀]
2. chin=3sg.n
1. ‘Their chin/su barbilla’

1.  
1. b.

1. Yiva à
1. [jiβ=ã̀]
2. plant=3sg.n
1. ‘Their plant/su hierba’

Figure 21: [ʃà=ã̀] (88a) (Speaker NGC).

Figure 22: [jiβ=ã̀] (88b) (Speaker RDC).

Almost all of the cases of this exceptional nasality in our dataset occur when the enclitic overwrites a low vowel /a/. However, the nasality of 3sg.n [=à] is not entirely predictable from the quality of the vowel it overwrites, and may be subject to interspeaker variation. For example, speaker NGC produced all repetitions of /ⁿ^͡tʃiʰʃǐ =à/ ‘their corncob/su elote’ in (89) with a final oral vowel (Figure 23), while speaker RDC produced it with a final nasal vowel (Figure 24).

1. (89)

1. Nchixǐ à
1. [n^͡tʃiʰʃ=â] ∼ [n^͡tʃiʰʃ=ã̂]
2. corncob=3sg.n
1. ‘Their corncob/su elote’

Figure 23: [ⁿ^͡tʃiʰʃ=â] (89) (Speaker NGC).

Figure 24: [ⁿ^͡tʃiʰʃ=ã̂] (89) (Speaker RDC).

It appears that the 3sg.n enclitic [=à] does alternate in nasality, but less predictably than the 1sg enclitic [=ì]. Because of this, we have excluded it from summary plots throughout, and we leave a precise analysis of its alternation for future work. However, we note that it, too, can trigger violations of the nasal phonotactic constraints described in §4.2. For example, it occurs as nasal in hiatus with an oral vowel (Figure 21) and after an approximant (Figure 22), providing another case where nasal phonotactic constraints are violated across morpheme boundaries.

4.7 Summary

In this section, we have shown that phonotactic restrictions on the co-occurrence of nasal and oral segments hold exceptionlessly within the root, while strings that violate these restrictions are always separated by a morpheme boundary. Considered in isolation, this result is consistent with the morphological definition of the couplet: the domain in which nasal phonotactic rules are upheld is the root.

Purely prosodic analyses of nasal phonotactics, based on foot structure, do not make the right predictions. Sequences of segments which violate nasal phonotactic restrictions are not, in general, separated by foot boundaries. Instead, the entire string is contained in the same foot.

In sum, the patterns described here constitute evidence that the morphological root is a crucial domain in defining the behavior of nasality. This conclusion is different from §3, where the bimoraic foot and not the morphological root was necessary to describe the domain in which laryngeals and underlying rising tones may surface.

These two results are not necessarily at odds with each other: in SMP Mixtec, some phonological patterns are defined in terms of the foot, and some are defined in terms of morphological structure. However, from the perspective that the Mixtec ‘couplet’ stands in a one-to-one correspondence relation with a single grammatical unit, these results are problematic. The correct conclusion appears to be that properties attributed to the couplet are not all properties of a single category. Instead, the Mixtec ‘couplet’ is a just shorthand term for the set of patterns which occur in canonical roots: morphologically-simplex, bimoraic words (Carroll 2015: 56). We return to this theme in §5, after a consideration of alternative analyses of nasal phonotactics which attempt to maintain a purely prosodic characterization of those patterns.

4.8 Alternative analyses