4.1. Sound inventory of Zongozotla Totonac

Our analysis of the overall sound inventory of Zongozotla Totonac reveals a system that is quite similar to those described for other Totonac languages (Beck, 2024; Davletshin, 2019), particularly those belonging to the Highland group (Aschmann, 1946; Román Lobato, 2008; Troiani, 2007): as shown in Table 2, consonants contrast six places of articulation, voicing is not contrastive in obstruents (but stops are usually phonetically voiced word-internally after nasals), and affricates and fricatives contrast in terms of central versus lateral airflow.

As in many Totonacan languages (Herrera Zendejas, 2023; McFarland, 2009; Román Lobato, 2008), /q/ can variably be realized as a stop, affricate [q^χ], fricative [χ, ʁ], or as a pre-fricated affricate [^χq^χ]. To our current knowledge, this variation is not predictable. Glottal stops occur phonetically before word-initial vowels, which themselves may be modal or glottalized (a point to which we return when describing vowels), but otherwise are found almost exclusively word-finally. This is a crucial point for our analysis in Section 4.4. In contrast to the cross-linguistic tendency for glottal “stops” to be mostly (creaky-)voiced, even utterance-finally (Garellek et al., 2023), in ZoT phrase-final glottal stops are almost categorically realized as “canonical” glottal stops, i.e., with a period of sustained voiceless closure followed by a release. Glottal /h/ is realized as [x] or (before /i/) [ç] in onset position, but we nonetheless consider it to be underlyingly glottal due to its phonological behavior, which we describe in Section 4.3.

All sonorants tend to be gradiently breathy-voiced or devoiced when they occur phrase-finally. For example, /aw, aj, an/ can each surface as [aw̤~aw̥, aj̤~aç, an̤~an̥] at ends of phrases. The lateral approximant /l/ only occurs prevocalically, and so this devoicing does not affect the contrast between /l/ and /ɬ/. This tendency for phrase-final breath (Duarte-Borquez et al., 2024) is an important fact to which we will return in Section 5.2.1.

There are three contrastive vowel qualities, each of which contrasts further in terms of length and phonation type (see also Table 3). Vowels /i, u/, irrespective of their length or phonation type, lower to [e, o] categorically adjacent to a uvular, even when a nasal intervenes, and are orthographically marked with e, o, as in (1); before /h, ʔ/, high-vowel lowering is common but more variable in occurrence (and rarely marked orthographically), and is often gradiently realized. For example, /ih/ can surface as [ih], [eh], or [iᵉh]. In Spanish loans like /peɾo/ ‘but,’ mid vowels occur non-adjacent to any uvular or glottal and thus could be considered phonemic as well (but they do not contrast in terms of length or phonation). Vowel /a/ also tends to be produced further back as [ɑ] adjacent to uvulars, as found across the Totonacan family (e.g., in Mecapalapa Tepehua; Herrera Zendejas 2023). In all contexts, short /a/ is often realized as [ə].

Of particular relevance for this paper is the limited distribution of certain vowels. In ZoT long vowels do not appear word-finally. This is in contrast to neighboring varieties from Zapotitlán and Ozelonacaxtla, where word-final long vowels are described. Cognates with word-final long vowels in neighboring varieties appear in ZoT with a final consonant, especially /j, h/. For example, /t͡ʃaqiːj/ chaqé:y ‘lo lava, s/he washes it’ from the example in (1) is /t͡ʃiqiː/ in Ozelonacaxtla Totonac (Román Lobato, 2008, p. 34).

Modal and glottalized vowels are subject to other phonological processes and have restricted distributions that differ from the patterns reported for related varieties: (1) As in other varieties of Totonac, all vowels at the beginning of a word sound glottalized; for closely-related Zapotitlán Totonac, Aschmann (1946, 1983) analyzes immediately word-initial vowels as underlyingly glottalized (“laryngealized”). This is largely due to phonetic reasons, namely that word-initial vowels are irregular in voicing and often preceded by a glottal stop. We will return to this point in Section 4.3; (2) Unlike in neighboring varieties, in ZoT glottalized vowels also do not occur underlyingly in word-final position; and (3) modal vowels devoice phrase-finally. These facts will be crucial for our analysis of the relationship between glottal stops and glottalized vowels in Section 4.4, and for our broader discussion in Section 5.

As with other varieties of Totonac, ZoT has surface-contrastive placement of lexical stress (which is also dependent to some degree on the morphology; MacKay, 1994; McFarland, 2009), but length and phonation contrasts are found on both unstressed and stressed syllables. In our broad phonemic transcriptions then, we avoid marking stress unless its placement is non-default (e.g., if stress appears fixed within a given morph of a polymorphemic word), but narrow allophonic transcriptions will always be marked for lexical stress.²

4.2. The phonetics of modal and glottalized vowels

In this section, we describe the phonetic characteristics of the phonation contrast on vowels in Zongozotla Totonac. For a language with a vocalic phonation contrast, ZoT stands out for two reasons: (1) the contrast has a very low functional load, and (2) the contrast appears to be very weak, with phonologically glottalized vowels almost never showing perceptually strong glottalization.

Although many words have glottalized vowels, we were able to find only a few minimal or even near-minimal pairs between modal and glottalized vowels in ZoT. Our phonetic analysis will involve only nine (near-)minimal pairs because so far those are the only pairs we have managed to find. Our impressions listening to the recordings suggest that the contrast is more strongly realized on long vowels than on short vowels, but in general the contrast is quite subtle. This is illustrated in Figure 2, which shows two representative (near-)minimal pairs differing in terms of vowel phonation: /kiɬniʔ/ versus /kḭɬniʔ/ and /paskaːt/ versus /kaka̰ːt/. For both pairs the glottalized vowel looks remarkably similar to the modal one, with regular voicing throughout. But the glottalized vowels in both /kḭɬniʔ/ and /kaka̰ːt/ have more energy in the higher frequencies than their modal counterparts, and the long glottalized vowel for /kaka̰ːt/ shows a more pronounced drop in voicing amplitude toward its end than does its modal counterpart, /paskaːt/.

Figure 2

Minimal and near-minimal pair in terms of phonation, with the relevant vowels enclosed in red rectangles. Words with glottalized vowels appear on right. The first pair (top row) was uttered by female speaker APG 426; the second pair (bottom row) by male speaker IPN 389.

Based on our informal audiovisual investigation, we expected that glottalized vowels will be differentiated from modal ones primarily in terms of their spectral tilt and voicing intensity, though we also expected there to be an interaction with vowel length. To control as much as possible for other sources of acoustic variation, we restrict this analysis to minimal or near-minimal pairs, controlling for placement of lexical stress and the manner of articulation of surrounding consonants, as both factors may influence voice quality (Garellek, 2022). Our list of nine word pairs is shown in Table 4.³

As mentioned earlier, words were read three times in isolation, followed by once in the carrier. Across all speakers, a total of 557 tokens were analyzed. The target vowels were segmented based on the onset and offset of F1 and F2 excited by voicing. The acoustic measures were calculated using VoiceSauce (Shue et al., 2011), which outputs a value of each measure every millisecond. We targeted the acoustic measures shown in Table 5 (we refer the reader to Garellek, 2022 for an overview). These were selected because they capture different components of glottalized (i.e., more constricted) phonation. We avoided adding additional measures that were highly correlated with those in Table 5.

Residual H1* is a relatively new measure that assesses the amplitude of H1 (corrected for formant frequencies and bandwidths) but also factors out the RMS energy. It is calculated by modeling H1* as a function of RMS energy, and then subtracting the product of the modeled RMS energy’s beta value and the particular token’s RMS energy from that token’s H1* level; see Chai & Garellek (2022) for more details. The more constricted the phonation, the lower the value of Residual H1*. Chai and Garellek found that this measure was better correlated with glottal constriction and less susceptible to error than the better-known spectral tilt measure, H1*–H2*.

We selected HNR < 500 Hz (harmonics-to-noise ratio below 500 Hz) as a noise measure, because it focuses on noise specifically in the lowest frequencies and thus captures F0 irregularity well. We also included subharmonics-to-harmonics ratio, because constricted voice qualities can be characterized by more prominent subharmonics in the spectrum.

Other measures included in the model were root-mean-squared (RMS) energy (the overall energy in the signal from both noise and signal), which can be lower due to increased constriction; strength of excitation (SoE), a measure of voicing intensity specifically (also expected to be lower due to increased constriction); and F0 (fundamental frequency), which is usually lower in constricted voice qualities, but can be higher for certain types of tense voice (Gordon & Ladefoged, 2001; Keating et al., 2015). VoiceSauce’s default settings were used to calculate these parameters. The F0 was calculated using STRAIGHT (Kawahara et al., 1998) and the formants via Snack (Sjölander, 2004).

F0 values greater than 3 standard deviations from each speaker’s mean were considered outliers and removed. Within each vowel category, we calculated the Mahalanobis distance on the F1–F2 space between each individual token relative to the mean of the category; the larger the Mahalanobis distance is, the more deviant the token is from the center of the category and the more likely there is a tracking error for that vowel. We also considered tokens with a Mahalanobis distance larger than 6 as outliers. We then removed all timepoints with F0 and formant outliers, and averaged each measure over the entire vowel. The outlier removal led to the exclusion of 14 out of the 557 tokens. Duration was log-transformed due to its being log-normally distributed.

To determine which acoustic measures, when taken together, predict glottalized versus modal vowels, we fitted a Bayesian logistic regression model using brms (Bürkner, 2018) in R. We predicted vowel phonation (dummy coded with “modal” as baseline) by the scaled mean acoustic measures and their interaction with vowel length and with random intercepts by speaker and word pair. The model was specified with weakly informative priors as Normal(0,1.5) in log-odds space for both the intercept and the fixed effects, and was fitted to draw 5000 samples in each of four Markov chains, with a burn-in period of 1000 iterations per chain.⁴

We find compelling evidence for the following parameters as predictors of the phonation contrast: SoE (β = –0.94, 95% CrI = [–1.33,–0.59]), Residual H1* (β = –0.83, 95% CrI = [–1.21,–0.45]), RMS energy (β = –0.67, 95% CrI = [0.34, 1.02]), duration (β = 1.01, 95% CrI = [0.60, 1.44]), and the interaction between RMS energy and vowel length (β = –0.57, 95% CrI = [–0.91,–0.25]). Compared to modal vowels, glottalized vowels are longer and have lower SoE and Residual H1*, suggesting weaker voicing due to greater glottal constriction. Glottalized vowels also have more RMS energy overall, but this effect is largely driven by long vowels (Figure 3).

Figure 3

Credible acoustic predictors of the phonation contrast. Violin plots show the distribution and points show each observation. The larger point indicates the mean value.

Therefore, glottalized vowels differ from their modal counterparts in ways that are expected for creaky versus modal contrasts. However, the differences in mean are very small and there is a large overlap in distributions. According to our perceptual impressions, vowel glottalization is also markedly weaker than expected, given the first and third authors’ extensive experience with various kinds of glottalized or creaky vowels in other languages such as Hmong, Mazatec, Mixtec, !Xóõ, and Zapotec (DiCanio et al., 2015; Keating et al., 2023). Therefore, we can conclude that the phonation contrast in ZoT is not only marginal (in terms of the rarity of minimal pairs), but also that it is only weakly realized phonetically.

To explore further the phonation contrast and its phonetic realization, in Figure 4 we plot time course data for the four acoustic measures (aside from duration) that emerged as credible predictors. (The top facet showing RMS energy is further subdivided by vowel length, given the interaction that emerged in the regression model.) For SoE and Residual H1*, the distinction between modal versus glottalized vowels is realized throughout most of the vowel’s duration. For RMS Energy, the increase in overall energy associated with glottalized long vowels is also found throughout the vowel.

Figure 4

Time course plots for the three acoustic measures (excluding duration) that emerged as predictors of the contrast. Shaded areas indicate Loess smoothing about the mean value.

Next we explore the variation across word pairs. Figure 8 in the Appendix shows, for each acoustic measure, the difference between the modal versus glottalized vowel in each word pair. We can see that some measures are more robust across word pairs: strength of excitation is higher for modal vowels across all pairs except for xkut-xku’ta and paqxa-pa’qxa, which show no difference along this measure; duration is lower for modal vowels across all pairs of words except for xkut-xku’ta (no difference) and paqlh-pa’qlh (effect in the opposite direction). The other acoustic measures show either more pairs that overlap with 0, or more pairs whose differences are in the opposite direction from the main effect for that measure. There is also no single pair of words for which a main effect (in the expected direction) is not found for at least one acoustic measure. This confirms that the phonation contrast is indeed found for all pairs, even if it is weak in its phonetic realization.

Finally, we explore cross-speaker variation. Figure 9 in the Appendix shows the difference between the modal versus glottalized vowels, here averaged across all word pairs, for each acoustic measure. We see that some speakers more reliably produce a difference in phonation than others. For example, RLC 505 produces the contrast mainly in terms of Residual H1*. Of all measures, SoE shows the more robust pattern across speakers. This is similar to what was found across word pairs.

In sum, the acoustic differentiation between modal and glottalized vowels is very weak, and the main acoustic measures that emerged as predictors of this contrast (vowel duration, RMS energy, Residual H1*, and SoE) vary both across speakers and word pairs in terms of whether and to what extent they reliably differentiate the phonation types.

4.3. The distribution of glottal stops

In this section, we provide an analysis of the distribution of glottal stops in Zongozotla Totonac. Based on their limited word-internal and phrasal distributions, we will argue that glottal stops should be considered marginally contrastive sounds in the language.

Evidence for the phonemic status of glottal stop comes from word-final position, where /ʔ/, /h/, and ∅ contrast, as in (2).

Although words ending in an underlying glottal consonant appear to be common in the lexicon, the glottal consonants (as well as final vowels) only surface as in (2) when they occur phrase-finally. Elsewhere, word-final /ʔ/ and /h/ delete and word-final vowels retain their voicing, as shown in Figure 5. This is presumably due to a constraint, common across Totonac languages, against C#C sequences across phrase-medial word boundaries (Román Lobato, 2008; McFarland, 2009, §2.6.5.1; Garcia-Vega, 2022, §2.3.1; Quintana Godoy, 2022). When a word ends in a consonant, the constraint is satisfied via deletion (of the glottal consonant) or /-i/ epenthesis (if the first word ends in a non-glottal consonant, as in /paskaːt/ paská:t ‘mujer, woman’). Examples of these alternations are seen in (3).

Thus glottal stops, though contrastive word-finally, rarely surface within a phrase. This is common across Totonac languages, and has led some researchers to posit that phrase-final glottal stops are either inserted phrasal markers (Ozelonacaxtla Totonac; Román Lobato, 2008), or that there is a change in progress whereby lexical glottal stops are being reanalyzed as postlexical, i.e., phrasal, markers (Filomeno Mata Totonac; McFarland, 2009). We will return to this point in Section 5.

Figure 5

Comparisons between /ʔ, h, a/ in phrase-final (left) versus phrase-medial (right) positions. All tokens were produced by speaker APG 426. Note the absence of glottal consonants phrase-medially.

In contrast to /h/, glottal stops cannot appear in coda position in the middle of the word. Indeed, morphological derivation that should in theory lead to a word-medial coda /ʔ/ instead results in deletion of the glottal stop. For instance, ZoT has a first-person plural possessive nominal construction ki(n)-X-kán ‘our X.’ When the nominal root ends in /-h/, that glottal is retained even though it occurs word-medially; in contrast, a root ending in /ʔ/ loses its glottal in comparable derived forms, as in (4).

In sum, underlying word-final glottal consonants do not surface except when the word is phrase-final. Glottal fricatives have a less restricted distribution, as they can appear in word-medial codas (as well as in onset position, where they surface as [x] or [ç]; see Section 4.1).

Now we turn to glottal stops that appear in non-final position within the word. Vowel-initial words are often preglottalized; that is, they begin with a glottal stop. Figure 6 shows a token of [ʔuˈkʃiɬḁ] ukxilha ‘lo ve, s/he sees it/him,’ where the word-initial vowel is preceded by a glottal stop and some irregular voicing at the beginning of the vowel.

Figure 6

Vowel-initial [ʔuˈkʃiɬḁ] ukxilha ‘lo ve, s/he sees it/him,’ with a preceding glottal stop. Token uttered by speaker RLC 505.

Immediately word-initial vowels (i.e., those not preceded by an onset consonant) also do not appear to contrast on the surface in terms of phonation. These facts have led some researchers (e.g., Aschmann (1946) for Zapotitlán Totonac) to treat word-initial vowels in Totonac as underlyingly glottalized rather than modal. In that case, the glottal stop would be part of the phonetic implementation of a glottalized vowel in word-initial position. Thus if true for ZoT, the word illustrated in Figure 6 could be transcribed phonemically as /ṵkʃiɬa/ instead of as /ukʃiɬa/. But contemporary understanding of the sources of irregular voicing (Garellek, 2022) suggests that irregular voicing on the initial vowel should not be used to motivate its underlying status as glottalized. Glottal stops usually lead to creaky voice on an adjacent vowel (Garellek, 2014; Garellek et al., 2023), so word-initial vowels in ZoT (as well as in Zapotitlán Totonac) can “appear” glottalized by virtue of an inserted glottal stop that is motivated by phrasal position rather than by lexically-specified phonation.

We also note that a third alternative exists, namely that all syllables must have an onset, such that so-called “word-initial” vowels are in fact preceded by a phonemic glottal stop (e.g., /ʔukʃiɬa/). But we believe such word-initial glottal stops are best analyzed as postlexical glottalization of initial vowels, rather than as phonemic /ʔ/, because they occur phrase-initially, where domain-initial strengthening in many languages leads to glottalization (Cho, 2016; Garellek, 2014, 2022). If the glottal stop were underlying, we would need some explanation for why it almost invariably does not surface in phrase-medial position. For example, when a vowel-final word like /ʃkṵta/ occurs in our carrier sentence before vowel-initial /ijmah/, hiatus is never repaired: [ˈʃkṵta ˈijma-n ta̰t͡ʃiˈwḭːn]; see also the lack of glottalization in hiatus for the phrase-medial tokens in Figure 5. Moreover, if word-initial glottal stops were underlying, we would need to explain why intervocalic glottal stops are so rare, especially given that word-final /ʔ/ is always realized strongly as a “canonical” glottal stop. But if instead we assume that the glottal stop is a result of prosodic strengthening of a word-initial vowel, that would make ZoT similar to the many languages with variable word-initial glottalization.

Ultimately, we assume that the phonation contrast does exist underlyingly in word-initial position, and that (verbal) morphophonological alternations provide evidence of the contrast. The form [ʔuˈkʃiɬḁ] ukxilha ‘lo ve, s/he sees it/him/her’ alternates with first-person [kuˈkʃiɬḁ] kukxilha ‘lo/la veo, I see it/him/her,’ where the verbal root begins with a modal vowel. In contrast, the verb root for ‘go’ is /a̰n/, with a glottalized vowel. We know this because that vowel is predictably glottalized, even word-medially; e.g., [hka̰n] jka’n ‘voy, I go.’ Therefore, we posit that /ukʃiɬa/ begins with a modal vowel underlyingly, and undergoes word-initial glottalization. Unfortunately, we did not elicit forms to test whether the word-initial phonation contrast is maintained or neutralized on the surface after word-initial glottalization occurs, but this should be tested in future work.

In word-medial position, glottal stops are extremely rare and variable across speakers. To our knowledge, these can occur only in hiatus between two glottalized vowels, as in (5). The comitative prefix is /ta̰ː-/, which we know contains a glottalized vowel because the glottalization predictably surfaces with this prefix, as in [ta̰ːtɬaˈwaj] ta:’tlawáy ‘lo hace con él, s/he does it with him.’ The verb root for ‘go’ is /a̰n/, with a glottalized vowel that is predictably glottalized. When /a̰n/ is preceded by the comitative prefix, the verb can surface as [ta̰ːˈʔa̰n] ta:’án ‘se va con él, s/he leaves with him,’ with a glottal stop between the two glottalized vowels, or as [ta̰ːˈa̰n], without the glottal stop. In our data set, instances of phonetic glottal stop [ʔ] in this word are speaker-specific: half of the eight speakers consistently say [ta̰ːˈʔa̰n], whereas the other half consistently say [ta̰ːˈa̰n], as illustrated in Figure 7. These groups do not appear to be structured according to age or gender. Therefore, we analyze word-medial glottal stops as a speaker-dependent form of prosodic strengthening of glottalized vowels under hiatus.

In sum, in this section we showed that, while some glottal stops can be considered contrastive in Zongozotla Totonac, the distribution of those phonemic glottal stops is limited. They contrast only in coda position; onset glottal stops, which are found before word-initial vowels and between glottalized vowels at a morphological juncture, are analyzed as derived from prosodic strengthening of a word-initial vowel or from hiatus involving two glottalized vowels. Thus in terms of their lexical phonotactics, glottal stops appear to have a limited distribution similar to that found in other unrelated languages, including Chinese varieties with coda /ʔ/ (Chai, 2022) as well as Western Muskogean (Ulrich, 1993).

Figure 7

Cross-speaker variation in the realization of /ta̰ːˈa̰n/ ta:’án ‘se va con él, s/he leaves with him’ without an intervocalic glottal stop (left) versus with a glottal stop (right). The token on the left is from speaker SLC 388, whereas the one of the right is from RLC 505.

But the distribution of phonemic glottal stops is especially limited, because they cannot appear in word-medial coda position; they are only found word-finally, and morphophonological alternations that involve suffixation to a root ending in a glottal stop results in the loss of this sound. Therefore, there must be an active constraint in the language against word-nonfinal (phonemic) glottal stops. And postlexical phrasing also matters, as ZoT further restricts surface glottal stops to phrase-final position.

4.4. The phonological relationship between glottalized vowels and glottal stops

Is there a phonological relationship between glottalized vowels and glottal stops in Zongozotla Totonac? While we have been treating them as distinctive sounds in the language, in fact [V̰] and [Vʔ] sequences are in complementary distribution: glottalized vowels generally do not occur word-finally, whereas [Vʔ] sequences only occur word-finally (and recall that the [ʔ] only surfaces at the end of a phrase). Instances of derived word-final glottalized vowels provide the sole exception to this distribution: if a word ends in /V̰h/, the /h/ is not realized phrase-medially (recall (3)). Thus /skuhma̰h/ skujma’h ‘está trabajando, s/he is working’ is realized as [ˈskuhma̰] in phrase-medial position—without the word-final /h/ but retaining a glottalized vowel that appears word-finally in the derived phrase-medial form. In the same environment, a word ending in a glottal stop would be realized with a word-final modal vowel. A summary of these distributions by phrasal position is shown in (6). Note that glottalized vowels are never followed by a coda glottal stop, and in word-final position long vowels are not found in open syllables or in those closed by a glottal consonant.

The complementary distribution of [V̰] versus [Vʔ] sequences at the very least suggests that [Vʔ] sequences are allophonic realizations of underlying word-final /V̰/. Nonetheless, we argue that they should be treated as contrastive sounds. One reason is that glottalized vowels and glottal stops are perceptually quite distinct, in a way that would render them, at the very least, “quasi-phonemes” (Kiparsky, 2014). Another reason is that, as we have argued in Section 4.2, the contrast between modal versus glottalized vowels is marginal, with the latter being only weakly differentiated from the former. Yet the realization of glottal stops is phonetically robust, so much so that they are typically realized as canonical voiceless glottal stops.

But morphophonological alternations provide probably the strongest evidence that [V̰] and [Vʔ] represent distinctive entities in synchronic ZoT. Recall in (4) that, in the first-person plural possessive ki(n)-X-kán construction, if the nominal root ends in an /h/, the fricative is retained; in contrast, root-final /ʔ/ disappears, regardless of whether the root-final glottal stop is in a stressed (as in [t͡ʃiˈt͡ʃiʔ] ‘perro, dog’) or unstressed syllable (as in [ˈpa̰ʃniʔ] ‘cochino, pig’), as shown in (7). If [Vʔ] were the word-final allophone of underlying /V̰/, then we would expect it to surface as [V̰] when it appears in this construction. Instead, the root-final /ʔ/ disappears and the preceding vowel remains modal: /kin-t͡ʃit͡ʃiʔ-ˈkan/ ‘nuestro perro, our dog’ surfaces as [kint͡ʃit͡ʃiˈkan], not as *[kint͡ʃit͡ʃḭˈkan, kint͡ʃit͡ʃiʔˈkan]. This cannot be explained by a constraint against glottalized vowels in this construction: with /liːwa̰j/ li:wa’y ‘carne, meat,’ the glottalized vowel remains after suffixation.

Therefore, the phonological relationship between glottalized vowels and glottal stops can also be considered what Hall (2013) calls “intermediate.” On the one hand, they are in complementary distribution, suggesting an allophonic relationship. On the other hand, glottalized vowels are phonetically very distinct from glottal stops, and the sounds’ divergent behavior in alternations provides evidence that /ʔ/ should be considered synchronically contrastive with glottalized vowels.

5.1. Implications of the phonetically-weak phonation contrast

Glottalization plays an outsize role in both diachronic and synchronic phonological analyses across the Totonacan family. In terms of historical analyses, MacKay & Trechsel (2018, p. 55) state: “The key question that must be addressed in any reconstruction of [Proto-Totonac-Tepehua] concerns the status of laryngealized vowels and glottalized consonants in the protolanguage.” That is because scholars disagree on whether the protolanguage should be hypothesized as having glottalized vowels (Arana Osnaya, 1953; Brown et al., 2011; Levy, 1987) or (different sets of) glottalized consonants, including clusters with glottal stops (Davletshin, 2019; MacKay & Trechsel, 2018).

But the analysis of glottalization in present-day Totonac and Tepehua languages is also contentious. As McFarland (2009, p. 13) writes in her dissertation on Filomeno Mata Totonac: “The writers of published grammars of all varieties [of Totonac-Tepehua languages] have struggled with the analysis of glottal features, which are variously accounted for as the phoneme /ʔ/, ejection on stops, laryngealization of vowels, and/or part of the vocalic nuclei, (either Vʔ, ʔV or ʔVʔ).” Part of the reason for the analytic struggles is that, across Totonac varieties, the historical contrast between modal versus glottalized vowels shows a continuum between full maintenance and complete loss, with many varieties having lost the contrast (in favor of modal vowels) in all, or nearly all, contexts. For instance, in his dictionary of Zapotitlán Totonac (which is part of the Highland group of Totonac languages), Aschmann (1983, p. 141) writes: “En varios pueblos se utiliza [una contracción o un cierre en la laringe] sólo al final de la palabra u oración.” (“In several towns [laryngeal constriction or closure] is used only at the end of the word or utterance.”) Further, the phonetic differentiation between modal versus glottalized vowels can be very weak. We have shown this to be true in ZoT, where the glottalized vowels can be described phonetically as at most “slightly constricted,” that is to say “tense.” The same has been argued for Olintla Totonac, though with more limited phonetic analysis (Tino Antonio, 2020). Huehuetla Totonac, another Highland variety, is reported to have only modal vowels; it further allows vowel-final words to variably end in a glottal stop (Troiani, 2007), which suggests that the presence of glottal stops is unrelated to the lexical phonology.

Why would some varieties of Totonac, including ZoT, have a weak phonation contrast, and why do others appear to have no contrast in phonation at all? We suggest that the relatively small number of contrasts (Davletshin, 2019) contribute to the phonetic weakening and eventual loss of the phonation contrast. As we discussed in Section 4.2, the functional load of the phonation contrast is low, both in ZoT and across the family. When there is competition in the lexicon, such as when a word has at least one minimal pair competitor, we can expect a tendency toward hyperarticulation of the phonetic attributes that helps differentiate competitors (Nelson & Wedel, 2017; Wedel et al., 2018). Conversely, a relative dearth of competition is associated with phonetic hypoarticulation. Moreover, a higher functional load of a contrast can inhibit diachronic loss of that contrast (Wedel et al., 2013). Therefore, diachronic phonation weakening and loss across Highland Totonac varieties can have had their origins in functional considerations of the contrast, namely the lack of competition between lexical items differing in terms of phonation type. This may also help explain why glottalized vowels are phonetically so weakly differentiated from modal ones in ZoT. Are glottalized vowels realized as only weakly non-modal (1) because of the lack of competition in the lexicon, (2) because the contrast is undergoing diachronic loss, (3) for both these interrelated reasons, or (4) for independent reasons? We don’t know the answer, but we suspect an ongoing change toward phonation loss (possibly due to its low functional load) is at play. If phonation is indeed being lost in ZoT via gradual weakening of the phonetic characteristics of glottalization, then this would likely be a case of “merger by approximation” (Labov, 1994, p. 321), involving the gradual coalescence of two categories. Mergers by approximation can sometimes result in a single category bearing the same mean value as one of the original categories. This seems to be at work here: glottalized and modal phonation appear to be coalescing, with the remaining category being phonetically modal.

The phonetically weak realization of the phonation contrast also has important implications for comparative studies of Totonac glottalization. If modal versus glottalized vowels are only weakly differentiated in some varieties, then glottalized vowels are likely to go underdescribed. In fact, this appears to be true for previous work on ZoT. In his field notes transcribing words across several Highland Totonac varieties, McQuown (1976) transcribed ZoT words with word-final glottal stops, but did not transcribe non-final vowels as glottalized, even where we do. Thus for the word ‘árbol, tree’ in ZoT, McQuown transcribed kíwiˀ, i.e. /kiwiʔ/. In contrast, our own orthographic rendition and phonemic transcription are ki’wi’ /kḭwiʔ/, indicating a word-medial glottalized vowel. It is not the case that McQuown heard but neglected to transcribe non-final glottalization; in his transcription of Zapotitlán Totonac, McQuown transcribes the same word with glottalization on both vowels (cf. Aschmann (1983)’s dictionary spelling as qui’hui’, i.e., /kḭwḭ/). Thus McQuown did hear and transcribe non-final glottalized vowels in Zapotitlán Totonac, but in ZoT he probably did not hear non-final glottalized vowels, and thus did not transcribe them.

This has important implications for comparative and diachronic work on Totonac. For example, McQuown (1990) variably transcribed word-medial glottalized vowels in Coatepec Totonac, another Highland variety: for instance, ‘árbol, tree’ is listed as both /kḭwiʔ/ (p. 493) and as /kiwiʔ/ (p. 44), and word-medial glottalized vowels do not appear to be transcribed in other texts by McQuown (Levy, 2020). Were glottalized vowels variably transcribed in Coatepec because, as in ZoT, they were too weak to be detected? This is an important question to answer for comparative work, because Coatepec is cited in comparative studies as an example of a Totonac variety that lacks the phonation contrast (Davletshin, 2018; MacKay & Trechsel, 2018). In sum, impressionistic transcriptions for Coatepec Totonac and other varieties of Totonac should be revisited using detailed field research and phonetic analysis like the kind we have employed here. We hope to do this in the near future, using the elicitation list and strategy that we employed for Zongozotla. While it is likely that some varieties have indeed lost the phonation contrast, it seems equally likely that others retain the contrast in the way that Zongozotla does—with a very weak phonetic implementation of the glottalization. Extensive careful and comparative elicitation and analysis will be needed to resolve this question.

5.2. Explaining the complementary distribution between glottalized vowels and glottal stops

Another problem for synchronic analyses concerns word-final glottal stops, especially in varieties such as Zongozotla Totonac, where glottal stops only surface phrase-finally. Recent hypotheses of Proto-Tepehua-Totonac (Brown et al., 2011; Davletshin, 2018; MacKay & Trechsel, 2018) reconstruct the proto language, if tentatively, as having glottal stops in word-final position. This has implications for how we analyze the complementary distribution in ZoT between glottalized vowels and glottal stops, which we argue below can either be inherited or innovated.

In ZoT, word-final glottal stops are always realized as “canonical” glottal stops, i.e., with mostly voiceless constriction (see panel (a) of Figure 5). But in perhaps most languages, glottal stops are realized with voiced glottalization throughout the constriction (Garellek et al., 2023). This means that the distinction between creaky vowels [V̰] and vowel-glottal stop sequences [Vʔ] should not be posited mainly in terms of some kind of assumed “typical” phonetic realization, because in fact creaky vowels and modal vowels followed by glottal stops typically look and sound the same. For this reason, Garellek et al. (2023) argue that the choice of analyzing glottalization as [V̰] versus [Vʔ] should be based primarily on its phonological patterning rather than on its phonetic properties. Indeed, in Section 4.4 we argued that the strongest evidence for treating final [ʔ] as underlying /ʔ/ (rather than as a phrase-final realization of /V̰/) is due to the glottal stop’s phonological behavior, not its phonetic realization.

Why might the phonetic similarities between glottalized vowels and glottal stops matter for Totonac more broadly? First, the fact that [V̰] and [Vʔ] can be realized in the same way provides a potential explanation for the complementary distribution found in ZoT between glottalized vowels (which do not occur word-finally) and glottal stops (which mainly occur word-finally). That is, it is possible to posit a change such as *V̰# > Vʔ#, as language users reinterpreted a word-final glottalized vowel as a modal one followed by a glottal stop. Indeed, Davletshin (2018) has already posited this as a sound change for some Totonac varieties.

Second, the fact that /V̰/ and /Vʔ/ can be phonetically identical provides another reason to reexamine previous analyses of glottalization in Totonac. For instance, where Zapotitlán Totonac has word-final glottalized short vowels /V̰/, ZoT only has /Vʔ/; thus, ‘árbol, tree’ is /kḭwḭ/ in Zapotitlán Totonac but /kḭwiʔ/ in ZoT. (Interestingly, there are very few⁵ words with word-final /Vː/ or /V̰ː/ in Aschmann (1983)’s dictionary, and their cognates in ZoT lack a final long vowel. For example, the dictionary entry lāntlā’ (/laːntɬa̰ː/) ‘cómo, how,’ with a final long glottalized vowel, corresponds to ZoT /laːntah/.) In this paper, we argue that there is phonological evidence for positing word-final /ʔ/ instead of /V̰/ in ZoT. Is there comparable phonological evidence in Zapotitlán Totonac for word-final /V̰/? We don’t yet know, but the answer would bear on crucial questions related to the reconstruction of Proto-Tepehua-Totonac. Clearly then, the study of Totonac sounds could benefit from more, and deeper, phonological and phonetic analyses.