3.1 Phonetics of Contrastive Phonation in Zapotec Languages

Zapotec languages are “laryngeally complex” (Silverman 1997), meaning they exhibit both contrastive phonation and lexical tone. Phonation contrasts may make use of a range of voice quality configurations, including glottal closures, creaky voice, breathiness, and so on. In the current paper, we use the term “phonation” to refer to the phonological phenomenon that relies on these voicing differences to create contrast, and we use the term “voice quality” to refer to the phonetic properties of voicing. Many Zapotec languages have been described as having a three-way vowel phonation contrast, such as a contrast among checked vowels (V^ʔ), rearticulated vowels (V^ʔV), and modal vowels (V) (Pérez Báez 2015; Beam de Azcona 2016; Crowhurst, Kelly, & Teodocio 2016). The glottalization in these checked and rearticulated vowels has most often been analyzed as the realization of a contrastive phonation type and not a consonant segment, as codas are often prohibited or highly restricted in Zapotec languages, and glottal stops never appear as onsets. As demonstrated in Table 1, contrastive phonation has phonologically diverse realizations across Zapotec varieties and can also have quite phonetically diverse realizations within a single variety (Arellanes Arellanes 2010; Esposito 2010; Crowhurst, Kelly, & Teodocio 2016).

Taken together, the studies represented in Table 1 suggest that the three-way glottalization contrast on vowels in Zapotec languages has a range of phonetic realizations depending on context and individual speaker and that the factors conditioning the different realizations are also diverse and not always categorical. The MacZ phonation contrast that is the focus of the current paper (and is described in the next section) is most similar to that of San Melchor Betaza Zapotec (Crowhurst, Kelly, & Teodocio 2016). One goal of the current study is thus to investigate whether two varieties with the same phonological characterization of phonation also mirror each other in their phonetic realizations of the contrast. If so, we expect to see that in citation form, word-final modal and rearticulated vowels terminate in breathiness, as found for Betaza Zapotec. We also expect word-final checked vowels in citation form in MacZ to vary between creakiness and a full glottal closure. For phrase-medial checked vowels, we expect to see fewer full glottal closures but more creakiness.

3.2 Phonological Characteristics of Macuiltianguis Zapotec

The focus of this paper is the phonation contrast in MacZ, exemplified in (1), where it can be seen that this contrast is largely independent from the lexical tone contrast that is described in more detail below. MacZ has five vowel qualities /a/, /e/, /i/, /o/, and /u/, and each of these can exhibit a contrast between modal, checked, and rearticulated phonation in open syllables.

As noted in the previous section, most descriptions of Zapotec languages consider this type of glottalization to be contrastive phonation, i.e., a feature of vowels rather than an independent consonant segment (see e.g., Chávez Peón 2010: 212–214). We follow such an analysis here, primarily because /ʔ/ surfaces in places few consonants appear, and conversely, does not surface in places we might expect consonants to appear.²

A more precarious question has to do with the status of so-called rearticulated vowels (V^ʔV) and whether they are best analyzed as single vowels with a glottal interruption, as in Table 1 and example (1), or as disyllabic sequences of a checked vowel followed by a modal vowel (V^ʔ.V). The single vowel analysis can be easily motivated for Zapotec languages in which the post-tonic vowel of disyllabic roots has historically been deleted resulting in almost exclusively monosyllabic roots (e.g., Chávez Peón 2010). In such cases, it would not make sense to propose that (C)V^ʔ.V is the only type of disyllabic root allowed. MacZ, however, retains post-tonic vowels and allows plenty of disyllabic roots, so this is not a reason to count out (C)V^ʔ.V as a possible root type. There are also some examples of words of the type (C)V₁^ʔ.V₂, such as ruʔa ‘mouth,’ and niʔa ‘foot,’ which cannot be analyzed as rearticulated vowels due to the change in vowel quality. If these are sequences of a checked vowel followed by a modal vowel, then V^ʔV sequences in which the vowel quality does not change could be analyzed the same way.

Further, V^ʔV is not allowed everywhere that modal (V) and checked (V^ʔ) vowels are allowed. Modal and checked vowels can appear in both stressed and unstressed syllables, while V^ʔV, if analyzed as a single vowel, would be restricted to stressed syllables in root-final position, as in belaʔa ‘precipice, cliff’.³ On the other hand, if V^ʔV is analyzed as a sequence of a checked vowel followed by a modal vowel, this would seem to open the door for many more trisyllabic roots in a language which historically has primarily allowed monosyllabic and disyllabic roots. However, such cases can often be shown to have been multimorphemic historically. For example, belaʔa ‘precipice, cliff’, could include a reflection of the proto-Zapotec prefix *pe-, which is generally analyzed as a type of animacy marker but also often found on plants and other nouns of the natural (and supernatural) world (e.g., Beam de Azcona 2016). Unlike in some other Zapotec languages, tone does not does not help disambiguate between a monosyllabic and disyllabic analysis of V^ʔV sequences in MacZ because under current analyses, the same sets of tones that are allowed across disyllabic roots are allowed on monosyllabic roots (Foreman 2006; Riestenberg 2017).

If vowels of the type V^ʔV do constitute single vowels with a glottal interruption, then they form part of a three-way contrast among modal (V), checked (V^ʔ) and rearticulated vowels (V^ʔV) in MacZ, as has been proposed for other Zapotec languages (e.g. Pérez Báez 2015 for Isthmus Zapotec; Crowhurst, Kelly, & Teodocio 2016 for Betaza Zapotec). Under such an analysis, the property that allows V^ʔV to create contrast is that the glottalization feature is anchored autosegmentally to the center portion of the vowel rather than the final portion of the vowel as for checked vowels (Arellanes Arellanes 2014; López Nicolas 2014).⁴ In this sense, the contrast has to do with timing rather than with the phonetic cues to glottalization, the latter being the central concern of this study. For this reason, and also due to their ambiguous status, we include words that terminate in V^ʔV in the current study but focus primarily on the contrast between checked and modal vowels. Still, we take advantage of the opportunity to explore the phonetic evidence related to the phonological status of rearticulated vowels and revisit this issue in our discussion.

Along with contrastive phonation, MacZ has a number of lexical tone contrasts, examples of which are given in Tables 2 and 3. Monosyllabic word roots with modal phonation exhibit a contrast among low, mid,⁵ falling, rising, and dipping tones (Table 2); dashes here represent unattested tone-phonation combinations. Monosyllabic word roots with checked phonation show a slightly restricted set of tone contrasts among low, mid,⁶ rising, and dipping; falling tones are not allowed on checked vowels. Word roots with rearticulated vowels can exhibit mid, high, falling, and rising tones. Across disyllabic roots, there are at least five contrastive tone patterns, regardless of whether the final vowel is modal or checked: low.low, mid.mid, high.high, high.low, and mid.high (Table 3). Though not particularly common, a handful of tonal minimal pairs can be found in the language. For instance, the word ìyyà with low tones means ‘flower’ while the word íyyá with high tones means ‘rock.’ The fact that such examples are somewhat rare indicates that tone may have low functional load in the language compared with phonation and other phonological contrasts.⁷

It is important to reiterate that checked vowels do not constitute any particular tone category in MacZ, as modal and checked vowels allow a similar set of contrastive tones. However, because this language has both contrastive phonation and contrastive tone, phonetic enhancement of one of these laryngeal features may have phonological consequences for the other. For instance, as mentioned above, Esposito (2010) found that breathy and creaky vowels in Santa Ana del Valle Zapotec always carry a falling tone. This is independent of a contrast between high and rising tones found on modal vowels. Falling tones, however, do not play a role in the tone contrast in this language. This suggests that enhancing a phonation contrast with a pitch cue could in turn obscure the distinctness of cues to lexical tone. In our discussion, we consider whether the existence of both tone and phonation in the language plays a role in the types of laryngeal cues available for phonetic enhancement of the phonation contrast.

5.1 Modal and Checked Vowels

Tables 5 and 6 show the number of tokens of modal and checked vowels in each speech context and for each speaker that exhibited each voice quality type.

The counts of tokens exhibiting the different voice quality types are largely similar across the speakers, except that Speaker 2 produced glottal closures for checked vowels in both speech contexts at a higher rate than Speaker 1, and with a higher rate of audible glottal releases in the citation form context. Speaker 2 also produced more modal vowels in phrase-medial context with breathy voicing than Speaker 1, and Speaker 1 produced more checked vowels in phrase-medial context with creaky voicing than Speaker 2. While the overall pattern for the two speakers is similar, we include Speaker as a random effect in our analyses below to account for this slight inter-speaker variation.

Note that nearly all vowel tokens, regardless of whether they are phonologically checked or modal, display some duration of modal voicing, typically the beginning and middle of the vowel. The different types of non-modal voicing typically occur in the final portion of the vowel. Therefore, the count data in Tables 5 and 6 do not on their own fully represent the voice quality differences across different phonation types. In order to provide more informative data, we also analyzed the differences in proportional durations of these voice quality types across vowels in the two speech contexts. Mean proportional durations were calculated for each type of voice quality interval (measured as described in Section 4 above) for modal and checked phonation vowels in both citation form and in phrase-medial contexts. These mean proportional durations are presented in Table 7 and visualized as boxplots of proportional duration in Figure 4 with the data from the two speakers collapsed.

Figure 4

Proportional duration of each voice quality type by phonological vowel type and elicitation context.

Phonologically modal vowels in both citation form and medial contexts are on average produced with modal voicing for at least 75% of the duration of the vowel; in citation form contexts, the modal portion is slightly shorter and the vowel tends to be followed by a relatively long period of breathiness. Checked vowels, on the other hand, tend to have a much shorter mean duration of modal voicing. In citation form, on average 41% of the total duration of a checked vowel consists of a glottal closure; in medial position this closure lasts for an average of 36% of the vowel. Phonologically checked vowels also tend to have periods of creaky voicing in both speech contexts.

5.1.2 Statistical analysis of F0 (modal vs. checked)

Table 13 shows the number of tokens of each phonological vowel type and utterance context that were produced with a change in F0, no change, or other F0 trajectory. Modal vowels in citation form tended to be produced with a rapid change or fluctuation in F0 trajectory in the final portion of the vowel, whereas checked vowels in the same context were unlikely to exhibit this change. Vowels in medial position, regardless of phonological vowel type, were also unlikely to exhibit a change in F0 trajectory. It was the checked vowels that were more likely to be coded as “other”; this is likely due to the fact that the onset of the glottalization associated with checked vowels can lead to aperiodicity in the signal and therefore unreliable pitch tracks. We excluded the few “other” tokens from the statistical analysis so that this unreliable data would not interfere with the main analysis of F0 behavior. The same data are visualized in Figure 5 with the “other” tokens omitted.

Table 13

Number of tokens with each F0 trajectory by phonological phonation type and utterance context.

CONTEXT	VOWEL TYPE	F0 TRAJECTORY
		CHANGE	NO CHANGE	OTHER
citation	modal	118	44	1
	checked	9	65	13
medial	modal	3	160	1
	checked	3	73	12

Figure 5

Counts of F0 trajectory types by phonological vowel type and elicitation context.

A mixed-effects logistic regression model was fit using the glmer function in the lme4 R package (Bates et al., 2015) to predict F0 trajectory. Table 14 presents the results of the regression model predicting “change” or “no change” in F0 for phonologically modal and checked vowels in the two utterance contexts. Modal was set as the reference level for phonological phonation type, and Citation Form was set as the reference level for context. Speaker was included as a random effect.

Table 14

Logistic regression model: F0 trajectory.

FIXED EFFECTS	ESTIMATE	STD. ERROR	T-VALUE	P-VALUE
(Intercept)	1.0515	0.4385	2.3982	0.0165
Type
Checked	–3.1036	0.4158	–7.4641	<0.001 ***
Context
Phrase-Medial	–5.1777	0.6246	–8.2889	<0.001 ***
Type * Context
Checked: Phrase-Medial	3.8973	0.9307	4.1873	<0.001 ***

The results in Table 14 show a significant main effect of phonological vowel type, such that checked vowels were significantly less likely to exhibit a change in F0 than modal vowels (p < 0.001). There was also a significant main effect of utterance context (p < 0.001); vowels produced utterance-medially were significantly less likely to exhibit a change in F0 than vowels produced in citation form. The model also revealed a statistically significant interaction between phonological phonation type and utterance context. A pairwise comparison showed that checked vowels in either utterance context, as well as phonologically modal vowels produced utterance-medially, were significantly less likely to be produced with a change in F0 than phonologically modal vowels produced in citation form (p < 0.001). In other words, changes in F0 are statistically most likely to occur in modal vowels produced in citation form than in any other vowel type in any other context.

5.2 Modal, Checked, and Rearticulated (V1 & V2) Vowels

Having shown the acoustic patterning of phonologically modal and checked vowels in this data set, we turn now to the acoustics of the two vowel portions that comprise V^ʔV vowel sequences in MacZ. We do so by conducting the same analyses as in the previous section, but this time examining all four vowel types (modal, checked, V1, and V2), paying special attention to the patterning of V1 and V2 vowels. As explained in section 3.2, the phonological status of the V^ʔV sequences is ambiguous, with some phonotactic and prosodic evidence in favor of an analysis of rearticulated (V^ʔV) vowels as a unitary, monosyllabic segment (and thus rearticulation as a third type of phonation, contrasting with modal and checked vowels), and other evidence in favor of analyzing V^ʔV as a sequence of a checked vowel followed by a modal vowel. To investigate whether there is phonetic support for the latter analysis, we analyze the acoustics of both portions of V^ʔV here in comparison with the analyses of modal and checked vowels presented above. Similarities between checked vowels and the first portions of rearticulated vowels and between modal vowels and the second portions of rearticulated vowels would provide some support for an analysis in which rearticulated vowels are in fact a sequence of a checked vowel and a modal vowel.

5.2.1 Statistical analyses of voice quality (modal, checked, and rearticulated vowels)

Tables 16 and 17 show the number of tokens for each vowel type and context that are produced with any duration of each of the voice qualities. For ease of comparison, modal and V2 vowels are shown in dark gray while checked vowels and V1 are shown in light gray; the color coding throughout this section is to emphasize the potential similarities between vowels in these two pairs. The proportion of tokens produced with each voice quality measure is similar between the two speakers across voice quality and utterance contexts. In the remainder of this section, we analyze the differences in proportional durations and include Speaker as a random effect to account for inter-speaker variation.

Table 16

Number of tokens (proportion of tokens) with each quality measure, Speaker 1.

CONTEXT	VOWEL TYPE	VOICE QUALITY MEASURES
		MODALITY	CREAKINESS	BREATHINESS	CLOSURE	RELEASE
citation	modal	84/84 (100%)	0/84 (0%)	56/84 (67%)	0/84 (0%)	0/84 (0%)
	checked	44/47 (94%)	15/47 (32%)	0/47 (0%)	33/47 (70%)	29/47 (62%)
	V1	20/23 (87%)	15/23 (65%)	3/23 (13%)	17/23 (74%)	0/23 (0%)
	V2	23/23 (100%)	0/23 (0%)	18/23 (78%)	0/23 (0%)	0/23 (0%)
medial	modal	90/91 (99%)	0/91 (0%)	2/91 (2%)	0/91 (0%)	0/91 (0%)
	checked	51/51 (100%)	24/51 (47%)	2/51 (4%)	36/51 (71%)	0/51 (0%)
	V1	23/23 (100%)	20/23 (87%)	1/23 (4%)	12/23 (52%)	0/23 (0%)
	V2	23/23 (100%)	2/23 (9%)	1/23 (4%)	0/23 (0%)	0/23 (0%)

Table 17

Number of tokens with each quality measure, Speaker 2.

CONTEXT	VOWEL TYPE	VOICE QUALITY MEASURES
		MODALITY	CREAKINESS	BREATHINESS	CLOSURE	RELEASE
citation	modal	84/87 (97%)	0/87 (0%)	58/87 (67%)	0/87 (0%)	0/87 (0%)
	checked	46/46 (100%)	17/46 (37%)	0/46 (0%)	43/46 (93%)	45/46 (98%)
	V1	18/20 (90%)	15/20 (75%)	1/20 (5%)	17/20 (85%)	0/20 (0%)
	V2	20/20 (100%)	1/20 (5%)	14/20 (70%)	0/20 (0%)	0/20 (0%)
medial	modal	84/94 (89%)	0/94 (0%)	15/94 (16%)	1/94 (1%)	0/94 (0%)
	checked	51/51 (100%)	11/51 (22%)	0/51 (0%)	49/51 (96%)	0/51 (0%)
	V1	24/25 (96%)	20/25 (80%)	1/25 (4%)	11/25 (44%)	0/25 (0%)
	V2	22/25 (88%)	0/25 (0%)	8/25 (32%)	0/25 (0%)	0/25 (0%)

Table 18 shows the proportional duration of each voice quality measure for all vowel types in citation form and in utterance-medial context, comparing data from the V1 and V2 portions of rearticulated vowels to the data previously presented in Table 7. The same data are visualized as boxplots in Figure 6.

Table 18

Mean proportional duration of each voice quality measure for vowels of each phonological type and produced in each context.

CONTEXT	VOWEL TYPE	VOICE QUALITY MEASURES
		MODALITY	CREAKINESS	BREATHINESS	CLOSURE	RELEASE
citation	modal	0.75	0	0.25	0	0
	checked	0.39	0.05	0	0.41	0.12
	V1	0.40	0.19	0.05	0.36	0
	V2	0.72	0.003	0.28	0	0
medial	modal	0.92	0	0.07	0.004	0
	checked	0.56	0.08	0.006	0.36	0
	V1	0.48	0.31	0.009	0.20	0
	V2	0.84	0.01	0.12	0	0

Figure 6

Proportional duration of each voice quality type by phonological vowel type and elicitation context.

V1 portions of rearticulated vowels in both production contexts show an average of 40–50% proportional duration of modal voicing, some period of creakiness, and a relatively long glottal closure. They are produced similarly to checked vowels medial position, in that they do not display the glottal releases typical of citation form checked vowels. It is to be expected that V1 vowels would resemble medial position checked vowels more than citation form checked vowels, as the citation form checked vowels are in final position and the V1 vowels are not; they are always followed by the V2 portion of the rearticulated vowel. One difference between V1 and checked vowels is that V1 tends to be produced with a greater proportional duration of creakiness, regardless of production context.

On the other hand, V2 vowels tend to behave like modal vowels in both contexts. In citation form, V2s are produced with modal voicing for an average of 75% of their duration, and then with breathy voicing for the remainder of the duration, closely mirroring the measurements for citation form modal vowels. In the phrase-medial context, both phonologically modal and V2 vowels are produced with even more modal voicing, 84% for V2 vowels and 92% for modal vowels on average, and a slightly shorter duration of breathy voicing.

As in section 5.1.1, statistical analyses were conducted using the mean proportional durations for different voice quality types. Individual mixed-effects linear regression models were fit using the lmer function in the lme4 R package (Bates et al., 2015), each predicting the mean proportional duration of one of the voice quality measures. The results of these models are discussed here in turn.

5.2.1.1 Statistical analysis of modal voicing (modal, checked, and rearticulated vowels)

Table 19 presents the results of the regression model predicting the proportional duration of modal voicing for phonologically modal, checked, V1 and V2 vowels in the two production contexts. Modal was set as the reference level for phonological phonation type, and Citation Form was set as the reference level for context. Speaker was included as a random effect.

Table 19

Mixed-effects linear regression model: proportional duration of modal voicing.

FIXED EFFECTS	ESTIMATE	STD. ERROR	T-VALUE	P-VALUE
(Intercept)	0.7473	0.0434	17.2348	0.0167 *
Type
Checked	–0.3571	0.0296	–12.0545	<0.001 ***
V1	–0.3493	0.0392	–8.9039	<0.001 ***
V2	–0.0298	0.0392	–0.7584	0.4485
Context
Phrase-Medial	0.1753	0.0244	7.1858	<0.001 ***
Type * Context
Checked: Phrase-Medial	–0.0062	0.0410	–0.1516	0.8795
V1:Phrase-Medial	–0.0904	0.0541	–1.6715	0.0951
V2:Phrase-Medial	–0.04770	0.0541	–0.8818	0.3782

The results in Table 19 show that whereas checked and V1 vowels are significantly different from modal vowels with respect to proportional duration of modal voicing (p < 0.001), there is no significant difference between modal vowels and V2 vowels. There was a significant main effect of context (p < 0.001), such that medial vowels were produced with a significantly longer proportional duration of modal voicing than vowels produced in citation form. A pairwise comparison reveals that the only pairs of vowels that did not show a significant difference in proportional duration of modality were those that compared either modal vowels and V2 vowels or checked vowels and V1 vowels. In other words, the statistics confirm here that with respect to proportional duration of modal voicing, V1 vowels are similar to checked vowels and V2 vowels are similar to phonologically modal vowels.

5.2.1.2 Statistical analysis of creaky voicing (modal, checked, and rearticulated vowels)

Table 20 presents the results of the regression model predicting the proportional duration of creaky voicing for phonologically modal, checked, V1 and V2 vowels in the two production contexts. Modal was set as the reference level for phonological phonation type, and Citation Form was set as the reference level for context. Speaker and Word were included as random effects.

Table 20

Mixed-effects linear regression model: proportional duration of creaky voicing.

FIXED EFFECTS	ESTIMATE	STD. ERROR	T-VALUE	P-VALUE
(Intercept)	0.0001	0.0117	0.0101	0.099301
Type
Checked	0.0536	0.0126	4.2585	<0.0001 ***
V1	0.1877	0.0168	11.1891	<0.001 ***
V2	0.0039	0.0168	0.2353	0.8141
Context
Phrase-Medial	<0.001	0.0097	–0.0010	0.9992
Type * Context
Checked: Phrase-Medial	0.02367	0.0164	1.4454	0.14881
V1: Phrase-Medial	0.1213	0.0216	5.6155	<0.001 ***
V2: Phrase-Medial	0.0107	0.0216	0.4944	0.6212

The results in Table 20 show that while both checked and V1 vowels differ from modal vowels with respect to their proportional duration of creaky voicing (p < 0.001), there is no statistical difference here between modal vowels and V2 vowels. Unlike with modal voicing, there is no main effect of context on proportional duration of creaky voicing. However, there was a significant interaction between vowel type and context. Pairwise comparisons show no statistically significant difference between modal and V2 vowels in either context. However, the differences in proportional duration between checked and V1 vowels are significantly different in both contexts (p < 0.001). Therefore, while modal and V2 vowels are produced with a similar proportional duration of creaky voicing in both contexts, the difference in proportional duration of creaky voicing between checked vowels and V1 vowels is significant across contexts.

5.2.1.3 Statistical analysis of breathy voicing (modal, checked, and rearticulated vowels)

Table 21 presents the results of the regression model predicting the proportional duration of breathy voicing for phonologically modal, checked, V1 and V2 vowels in the two production contexts. Modal was set as the reference level for phonological phonation type, and Citation Form was set as the reference level for context. Speaker and Word were included as random effects.

Table 21

Mixed-effects linear regression model: proportional duration of breathy voicing.

FIXED EFFECTS	ESTIMATE	STD. ERROR	T-VALUE	P-VALUE
(Intercept)	0.2460	0.0259	9.5143	0.01060 *
Type
Checked	–0.2457	0.0259	–9.4806	<0.001 ***
V1	–0.2017	0.0348	–5.7978	<0.001 ***
V2	0.0290	0.0348	0.8323	0.4059
Context
Phrase-Medial	–0.1828	0.0187	–9.7556	<0.001 ***
Type * Context
Checked: Phrase-Medial	0.1879	0.0315	5.9598	<0.001 ***
V1: Phrase-Medial	0.1450	0.0416	3.4873	<0.001 ***
V2: Phrase-Medial	0.0281	0.0416	0.6767	0.4988

As with previous voice quality measures, the model in Table 21 shows significant differences between modal vowels and both checked and V1 vowels (p < 0.001), but no significant difference between modal and V2 vowels (p = 0.4058). Here, again, there is a significant main effect of context (p < 0.001), such that medial vowels are significantly likely to be produced with less breathy voicing than those in citation form. Pairwise comparisons of the interaction between vowel types show no significant difference between modal vowels and V2 vowels in either utterance context. Similarly, there was no significant difference in proportional duration of breathy voicing between V1 vowels and checked vowels in either context. Therefore, results from this model further confirm that V2 vowels are produced similarly to modal vowels and V1 vowels are most similar to checked vowels.

5.2.1.4 Statistical analysis of duration of glottal closure (modal, checked, and rearticulated vowels)

Table 22 presents the results of the regression model predicting the proportional duration of glottal closure for phonologically modal, checked, V1 and V2 vowels in the two production contexts. Modal was set as the reference level for phonological phonation type, and Citation Form was set as the reference level for context. Speaker and Word were included as random effects.

Table 22

Mixed-effects linear regression model: proportional duration of glottal closure.

FIXED EFFECTS	ESTIMATE	STD. ERROR	T-VALUE	P-VALUE
(Intercept)	–0.0004	0.0258	–0.0158	0.9895
Type
Checked	0.4080	0.0169	24.0922	<0.001 ***
V2	0.3654	0.0225	16.2295	<0.001 ***
V2	0.0034	0.0225	0.1524	0.8789
Context
Phrase-Medial	0.0020	0.0135	0.1491	0.8815
Type * Context
Checked: Phrase-Medial	–0.0522	0.0227	–2.2976	0.0219 *
V1: Phrase-Medial	–0.1653	0.0300	–5.5144	<0.001 ***
V2: Phrase-Medial	–0.0043	0.0300	–0.1418	0.8873

Similarly to previous models, the results from this model show that there is a significant difference in proportional duration of the glottal closure between modal vowels and both checked and V1 vowels (p < 0.001), but no significant difference between modal and V2 vowels (p = 0.8789). While the main effect of context was not significant (p = 0.8815), the interaction between vowel type and context did achieve significance (p < 0.001). Pairwise comparisons show no significant difference between modal vowels and V2 vowels in either context; as supported by the previous models, it seems that these two vowel types behave relatively similarly in both contexts. While there was no significant difference between V1 vowels and checked vowels in citation form, the two vowels did differ significantly with respect to their proportional duration of the glottal closure in medial position.

5.2.1.5 Statistical analysis of duration of glottal release (modal, checked, and rearticulated vowels)

Table 23 presents the results of the regression model predicting the proportional duration of glottal release for phonologically modal, checked, V1 and V2 vowels in the two production contexts. Modal was set as the reference level for phonological phonation type, and Citation Form was set as the reference level for context. Speaker was included as a random effect.

Table 23

Mixed-effects linear regression model: proportional duration of glottal release.

FIXED EFFECTS	ESTIMATE	STD. ERROR	T-VALUE	P-VALUE
(Intercept)	–0.0001	0.0073	–0.0166	0.9892
Type
Checked	0.1174	0.0035	33.3302	<0.001 ***
V1	0.0006	0.0047	0.1291	0.8973
V2	0.0006	0.0047	0.1291	0.8973
Context
Medial	<0.001	0.0029	0.0032	0.9975
Type * Context
Checked: Medial	–0.1173	0.0049	–24.0557	<0.001 ***
V1: Medial	–0.0007	0.0064	–0.1209	0.9038
V2: Medial	–0.0007	0.0064	–0.1209	0.9038

Unlike with previous voice quality measures, the results in Table 23 show a significant main effect of phonological vowel type only between modal and checked vowels. There was no significant main effect of context (p = 0.9974), but the interaction between phonological vowel type and context did achieve significance (p < 0.001). Pairwise comparisons here showed significant differences only between checked vowels in citation form and all other vowel/context combinations. In other words, checked vowels in citation from had a significantly longer proportional duration of a glottal release than any other vowel produced in any other context.

5.2.1.6 Summary of statistical analyses of voice quality

To summarize the voice quality results across all four phonological vowel types, there is strong statistical evidence that modal vowels and V2 portions of rearticulated vowels are produced similarly with respect to all voice quality measures in all contexts. While V1 vowels and checked vowels also show many such similarities, they differ with respect to some voice quality measures in medial form, specifically the proportional duration of a glottal closure and release burst. These differences are likely due to the fact that the checked vowels in this data set are all word-final, but V1 vowels are inherently non-final in all cases, as they are always followed by the corresponding V2 portion of the rearticulated vowel.

5.2.2 Statistical analysis of F0 (modal, checked, and rearticulated vowels)

Table 24 shows the number of tokens in each F0 trajectory category by phonological vowel types and elicitation context. The data for modal and checked vowels is the same as previously reported in Table 13; data from the V1 and V2 portions of phonologically rearticulated vowels is added here for comparison.

Table 24

Number of tokens with each F0 trajectory by phonological vowel types and elicitation context.

			F0 TRAJECTORY
CONTEXT	VOWEL TYPE	CHANGE	NO CHANGE	OTHER
citation	modal		118/163(73%)	44/163(27%)	1/163(0.6%)
	checked		9/87(10%)	65/87(75%)	13/87(15%)
	V1		13/43(30%)	28/43(65%)	2/43(5%)
	V2		24/34(71%)	9/34(26%)	1/34(3%)
medial	modal		3/164(2%)	160/164(98%)	1/164(0.6%)
	checked		3/88(3%)	73/88(82%)	12/88(14%)
	V1		16/48(33%)	28/48(58%)	4/48(8%)
	V2		2/35(6%)	32/35(91%)	1/35(3%)

Table 24 shows the number of tokens for each vowel type and elicitation context that exhibited each type of F0 trajectory. Though the overall number of tokens coded as “other” is small, the checked vowels had more of these tokens than any of the other vowels. As reported above, we assume that this pattern is due to the fact that the onset of the glottalization required to create a checked vowel (that is, one that ends with a glottal closure) sometimes interferes with the periodicity required for pitch tracking. The same data are visualized in Figure 7 with the “other” tokens omitted.

Figure 7

Counts of F0 trajectory types by phonological vowel type and elicitation context.

Figure 7 shows that modal vowels in citation form had more than twice as many tokens with a change in F0 trajectory than it did tokens without one. Though there were fewer tokens of rearticulated vowels, the V2 portions of these vowels in citation form show a similar pattern; more than twice as many tokens with F0 changes than without. For checked and V1 vowels in all contexts, and for modal and V2 vowels in medial contexts, a majority of the tokens were produced with no change in the F0 trajectory.

A mixed-effects logistic regression model was fit using the glmer function in the lme4 R package (Bates et al., 2015) to predict F0 trajectory. We excluded the few “other” tokens from the statistical analysis so that this unreliable data would not interfere with the main analysis of F0 behavior. Table 25 presents the results of the regression model predicting “change” or “no change” in F0 for phonologically modal and checked vowels in the two utterance contexts. Modal was set as the reference level for phonological phonation type, and Citation Form was set as the reference level for context. Speaker was included as a random effect.

Table 25

Logistic regression model: F0 trajectory.

FIXED EFFECTS	ESTIMATE	STD. ERROR	T-VALUE	P-VALUE
(Intercept)	1.0777	0.5023	2.1453	0.0319 *
Type
Checked	–3.1643	0.4171	–7.5861	<0.001 ***
V1	–1.9729	0.4036	–4.8879	<0.001 ***
V2	–0.0180	0.4467	–0.0402	0.9679
Context
Medial	–5.2556	0.6230	–8.4357	<0.001 ***
Type * Context
Checke: Medial	3.9601	0.9315	4.2512	<0.001 ***
V1: Medial	5.5009	0.7882	6.9790	<0.001 ***
V2: Medial	1.2062	1.0401	1.1597	0.2462

The results show a significant main effect of vowel type (p < 0.001), such that checked vowels and V1 vowels were significantly less likely to exhibit a change in F0 than modal vowels; the difference between V2 and modal vowels was not significant (p = 0.9680). There was also a main effect of context; vowels produced in utterance-medial contexts were significantly less likely to be produced with a change in F0 trajectory than vowels that were produced in citation form.

Pairwise comparisons of the interaction between vowel type and context show no significant difference between modal vowels and V2 vowels in either context, supporting the notion that modal and V2 vowels are produced similarly with respect to F0 trajectory. Similarly, there was no significant difference between checked vowels and V1 vowels in citation form. However, in medial contexts, the difference between checked vowels and V1 vowels was significantly different (p = 0.0013), with V1 vowels more likely to show a change in F0 trajectory than checked vowels.

5.3 Overall Summary of Results

The phonetic characteristics of the MacZ phonation contrast presented in this section show that checked vowels are frequently produced with a glottal closure, while modal vowels are frequently produced without such a closure. In the citation form context, Speaker 1 produced 70% of the checked vowel tokens with a glottal closure, and Speaker 2 produced 98% of the checked vowel tokens with a glottal closure. In the phrase-medial context, Speaker 1 produced 75% of the checked vowel tokens with a glottal closure, and Speaker 2 produced 96% of the checked vowel tokens with a glottal closure. Modal vowels were almost never produced with a glottal closure for either speaker in either context. A glottal closure, or lack thereof, was thus the most consistent cue to the phonation contrast across the data examined, which is in line with the way the contrast between modal and checked vowels is generally described for Zapotec languages (see section 3.2).

However, the realization of additional phonetic cues (beyond the existence or non-existence of a glottal closure) differed depending on the speech context (citation form words in isolation versus words produced within a carrier phrase). Table 26 summarizes the results presented in this section with regards to the voice quality characteristics of phonologically modal and checked vowels as well as the V1 and V2 portions of rearticulated vowels in both the citation form and phrase-medial speech contexts.

For the citation form data, it is clear that phonologically modal vowels are likely to surface as modal followed by a period of breathiness, as was found for Betaza Zapotec by Crowhurst, Kelly, and Teodocio (2016). We additionally found that modal vowels tend to exhibit a fluctuation in F0 that is independent of final breathiness. Phonologically checked vowels exhibit phonetic glottalization, usually in the form of a full glottal closure with an audible and visible release burst. Modal vowels exhibited a significantly greater proportional duration of modal voicing than creaky vowels regardless of production context and a significantly greater proportional duration of breathy voicing than creaky vowels in citation form. Checked vowels exhibited a significantly greater proportional duration of glottal closure and burst than modal vowels in citation form, and a significantly greater proportional duration of creaky voicing than modal vowels in the phrase-medial context. This finding also mirrors that of Crowhurst, Kelly, and Teodocio (2016) for Betaza Zapotec, who found fewer full glottal closures and more creakiness in phrase-medial positions.

Overall, the phrase-medial data differs from the citation form data primarily in the fact that we see fewer phonetic cues to the contrast. Very few of the modal vowels exhibited breathiness or F0 fluctuations as they did in the citation form data. While phonologically checked vowels tend to surface in citation form contexts as modal throughout the majority of the vowel with a glottal closure in the final portion followed by an audible burst, we see no such bursts in the phrase-medial data.

The statistical analyses of phonetic cues to V1 and V2 vowels largely confirm the prediction that V1 vowels will show phonetic cues similar to those of checked vowels, and that that V2 vowels will show phonetic cues similar to those of modal vowels. This prediction, however, does not always hold for the V1 vowels, especially in citation form contexts. Phonologically checked vowels have a shorter period of creaky voicing than V1 vowels, and have a longer glottal closure and release on average than V1 vowels. We assume that this difference is due the fact that whereas V1 vowels are inherently non-word-final, as they are always followed by a corresponding V2, checked vowels here are always word-final and, in the citation form contexts, utterance-final.

6.1 Defining the Phonation Contrast

Phonetically-driven approaches to phonology differ in terms of how they characterize the realization of redundant phonetic cues associated with a given phonological contrast like those examined in this paper. As discussed above in our review of literature on phonetic enhancement (Section 2), Hall’s (2011) contrast and enhancement theory proposed that active contrastive features, as identified through a hierarchy of ordered phonological features (Dresher 2009; see Figure 1 in Section 2), are eligible for enhancement via any phonetic configuration that contributes to the perceptual distinctiveness of that feature. Boersma’s Functional Phonology (1998) argued against characterizing this process as one of phonetic enhancement, instead proposing that the set of phonetic cues that contribute to a contrast should be thought of as the gestural implementation of “Max” and “Min” Optimality-theoretic constraints on distinctive features. Boersma (1998) therefore rejects the notion of phonetic cues enhancing abstract phonological features (such as [back]), and instead proposes that all phonetic implementations correspond to perceptual features, which in turn correspond to a given acoustic-perceptual continuum (such as F2).

Both of these approaches assume that phonological contrasts rely on features whose elements should be maximally perceptually distinct. For Hall (2011), this is a binary [+/–] specification of traditional abstract phonological features. Under such an approach, we could assume that the phonation contrast in MacZ makes use of a [glottalized] feature, such as the laringización feature proposed by Arellanes Arellanes (2009), such that modal vowels are specified as [–glottalized] while checked vowels are specified as [+glottalized]. For Hall (2011), the glottalization distinction could be enhanced with a range of phonetic configurations, ranging from directly to indirectly related to the core acoustic and articulatory targets of the feature, as discussed in Section 2.

Boersma (1998) on the other hand, argues against the use of such binary features (p. 355), and prefers to understand features as corresponding to a continuous perceptual correlate (see also Boersma 2009). For example, the traditional [back] feature for vowels corresponds to the continuum of F2 values and the traditional [high] feature corresponds to the continuum of F1 values. Languages make use of these formant spaces in different ways through language-specific gestural implementations of these perceptual features. In the case of phonation contrasts, such an approach may be informed by existing proposals regarding voice quality continua, such as the proposal of Gordon and Ladefoged (2001), which associates breathy voice with a less constricted glottis (greater open quotient) and creaky voice with a more constricted glottis (smaller open quotient). Of course, for Boersma, the continuum should associate with a perceptual rather than articulatory feature. The open quotient gesture corresponds acoustically to the difference between the first two harmonics (H1 minus H2¹²) of the spectrum, with breathier sounds showing a greater difference between H1 and H2 than creakier sounds (e.g., Kuang 2017). This type of continuum is represented in Figure 8.

Figure 8

A simple articulatory and perceptual voice quality continuum.

Taking a Functional Phonology (Boersma, 1998) approach to analyzing the MacZ data, and drawing an analogy with how Boersma explains lip rounding of back vowels as part of the gestural implementation of MaxF2, we might wish to say that checked vowels are a gestural implementation of a “Min” constraint on the perceptual feature H1-H2, while modal vowels are the gestural implementation of a “Max” constraint on the same feature. Surface realizations of the former need not always be a complete glottal closure, and surface realizations of the latter need not always be voiceless; this variation can be captured by the interaction of the Min and Max constraints on H1-H2 with other markedness and faithfulness constraints. For example, there could be another constraint, Max-SNR, which prefers modal vowels that optimize signal-to-noise ratio over breathy or creaky voice, both of which can reduce recoverability for other important contrasts such as tone and vowel quality. A high ranking for such a constraint would result in winning candidates having modal voicing, despite the fact that modal voicing is in the middle of the continuum in Figure 8 and therefore not preferred by either the Min or Max H1-H2 constraints.

Having established how these two approaches might characterize the MacZ phonation contrast that is central to this paper, we move in the next few sections to discussing how they explain or do not explain the phonetic realizations revealed through our analysis of the MacZ data, addressing the advantages and disadvantages of each approach.

6.2 Theoretical Accounts of the Citation form Data

The phonetic characteristics of the MacZ phonation contrast for words in citation form are largely consistent with a contrast and enhancement approach (Hall 2011) in which each contrastive feature specification, in this case [+/–glottalized], is produced with additional phonetic cues that make the contrast sufficiently perceptible. Under Hall’s (2011) contrast and enhancement approach, redundant phonetic cues may enhance the auditory impression of the [+/–glottalized] feature in multiple ways and along different acoustic dimensions. Some enhancements work to push a perceptual cue to the contrastive feature further along a given acoustic continuum, while others involve separate acoustic effects that increase the overall salience of the distinction between contrastive elements. Such an approach is helpful for explaining the breathiness and F0 cues present on the majority of the phonologically modal vowels in citation form. Most of the modal vowels for both speakers have a breathy voice cue in the final portion of the vowel or throughout the entire duration of the vowel, and over half of the tokens show a non-steady F0 trajectory.

The breathiness cue in this case could be interpreted under Hall’s (2011) contrast and enhancement approach as a type of enhancement in which “a feature can be enhanced by the amplification of its articulatory and acoustic/auditory correlates” (p. 20). Analogous to producing a contrastively [–back] vowel in the front of the vocal tract rather than the center, breathy voice pushes the H1-H2 difference further to the left along the voice quality continuum presented in Figure 8. A Functional Phonology approach (Boersma 1998) would account for the breathiness cue in a similar way. Analogous to the way lip rounding on back vowels emerges from Boersma’s MaxF2 constraint, breathy voicing on modal vowels in MacZ could be seen to emerge from a MaxH1-H2 constraint that pushes the gestural implementation of modal vowels in citation form further to the left of the continuum presented in Figure 8.

However, the F0 fluctuations cannot easily be understood as pushing the H1-H2 feature further along its perceptual continuum in the same way, as there is not a direct acoustic relationship between these F0 fluctuations and the H1-H2 difference. We therefore understand the F0 fluctuation to be an instance of one of the more indirect relationships Hall (2011) describes, in which “a feature with a particular acoustic/auditory correlate can be enhanced by a separate acoustic/auditory effect that increases the relative salience of that correlate” (p. 20; emphasis added).

This indirect relationship has to do with the fact that the [–glottalized] feature in our data is also enhanced by breathiness, as discussed above. It has been shown in previous work that the switch from modal to breathy voicing can result in F0 perturbations (Garellek & Keating 2011). We therefore argue that MacZ speakers produce changes in F0 to increase the relative salience of the breathy correlate to the [–glottalization] feature, one of the types of phonetic enhancement that Hall’s (2011) contrast and enhancement approach accounts for.¹³ Importantly, given that our data show that the breathy voicing and F0 fluctuations do not always co-occur, our claim is that the F0 fluctuation is being used to optionally further enhance the phonation contrast, and not that it surfaces as an automatic mechanical byproduct of the breathy voicing enhancement described above. Functional Phonology (Boersma 1998) offers no clear mechanism to account for phonetic cues like these F0 fluctuations, which are not an obvious gestural implementation of MaxH1-H2 (or any unidimensional perceptual continuum) and instead add salience to the overall phonation contrast by drawing on an additional acoustic dimension.

The presence of creaky voicing and glottal bursts on checked vowels in the citation form data is also straightforwardly captured by Hall’s (2011) contrast and enhancement approach. Another of Hall’s (2011) possible phonetic enhancements occurs when “a feature with a particular articulatory correlate can be enhanced by the amplification of a natural mechanical by-product of that gesture” (p. 20). Creaky voicing and glottal bursts are natural mechanical by-products of the glottal closure gesture (e.g., Chong & Garellek 2018), and creaky voicing in particular has been found to be a phonetic correlate of the [+glottalized] feature across Zapotec languages (e.g., Arellanes Arellanes 2010; Chávez Peón 2010; Esposito, 2010; Crowhurst, Kelly, & Teodocio 2016). Therefore this is a predictable enhancement of the [+glottalized] feature of checked vowels according to Hall’s (2011) approach.

Here again, however, it is not clear how Functional Phonology (Boersma 1998) accounts for these additional cues. If we assume that checked vowels are a gestural implementation of the constraint MinH1-H2, the creaky voicing cue in fact moves us the wrong way along the continuum in Figure 8, as creaky voicing actually creates a greater H1-H2 difference than a glottal closure. In addition, the glottal bursts which are present on so many of the checked vowels in citation form do not seem to have a place along the continuum in Figure 8 whatsoever, as they have no direct relationship to the perceptual H1-H2 difference. The glottal closure that is at the right end of this continuum should be the most faithful implementation of a constraint like MinH1-H2, and indeed, a glottal closure is reliably produced in nearly all of the checked tokens in our data. However, the additional creaky voicing and burst cues are not easily accounted for following Boersma’s (1998) approach.

In sum, our results suggest that the checked/modal phonation contrast in MacZ exhibits phonetic enhancement following the predictions put forth by the contrast and enhancement theory (Hall 2011). A contrast and enhancement approach predicts that phonetic enhancement may occur along multiple dimensions to increase the salience of the overall contrast, which is precisely what we see in our data. This occurs in the citation form data via breathiness and F0 fluctuations for [–glottalized] and glottal bursts and/or creakiness for [+glottalized]. Contrast and enhancement neatly accounts for these particular phonetic characteristics, because it explicitly predicts how phonetic enhancements will be related to the phonological feature active in the contrast. Phonetic enhancement of the [+/– glottalized] feature in MacZ does not just take advantage of one unidimensional percept to glottalization, the way that lip rounding enhances the continuum of the vowel backness (F2) percept. Rather, cues may be a natural mechanical byproduct of an associated gesture, or they may involve an auditory effect that increases the salience of a separate auditory cue.

It is not clear how Functional Phonology (Boersma 1998) accounts for additional phonetic cues beyond those associated with a single perceptual continuum that underlies the main contrast. A constraint like MinH1-H2 only predicts gestural implementations that have a direct effect on the H1-H2 difference. Boersma requires that contrasts be explained using distinctive perceptual features (like F2), rather than abstract features with articulatory labels (like [back]), so we could consider other perceptual correlates of glottalization such as low periodicity and low overall spectral tilt (e.g., Garellek 2015). Still, it is not clear how to combine these into a single faithfulness constraint that explains cues such as F0 fluctuations and glottal bursts. It seems that these are only captured by appealing to a more abstract phonological feature, such as [glottal], and allowing for more indirect enhancements to this feature.

6.3 Theoretical Accounts of the Phrase-Medial Data

The most consistently occurring cue to checked phonation in MacZ, the glottal closure, is realized only on the final portion of the vowel after the tonal cue has been produced. This word-final placement of the cue could make the contrast particularly difficult to perceive in a phrase-medial context. According to the contrast and enhancement theory, this lack of perceptibility should trigger phonetic enhancements. However, the phonetic enhancements seen in the citation form data (breathiness, F0 fluctuations, creaky voicing, and glottal bursts) are not all present in the phrase-medial data. While the contrast and enhancement theory (Hall, 2011) neatly accounts for the enhancements of the overall contrast between checked and modal vowels, it does not straightforwardly explain why we see more phonetic enhancement in our citation form data than in the phrase-medial data. Although Hall (2011) is explicit that “phonetic enhancement is variable across languages, speakers and contexts” (p. 19), he does not provide mechanisms to predict or account for this variability.

Functional Phonology (Boersma 1998), on the other hand, explicitly accounts for such optionality across different production contexts by allowing for a stochastic element in constraint evaluation (Chapter 15: 329–346). As Boersma (1998) explains:

The account of optionality presented here naturally encapsulates pragmatics-based reranking. For instance, if you want to speak more clearly, you may raise all your faithfulness constraints by, say, 5 along the continuous ranking scale. In this way, an 80%–20% preference for place assimilation will turn into a 18%–82% preference against. Depending on whether the faithfulness constraint is ranked above or below its rival, slight variation may turn into obligations or the reverse. If the ranking difference is large to begin with, however, nothing happens; so we see that discrete properties of surface rerankability are compatible with, and may well follow from, a general continuous rerankability of all constraints (p. 346).

To account for the variability in our results across production contexts, we can appeal to variable rankings of interacting faithfulness constraints as Boersma (1998) suggests. However, as we argued above, constraints like MinH1-H2 that correspond to a single perceptual continuum as Boersma suggests are not in line with the range of phonetic cues found in MacZ. Rather, our constraints must be something like Max[glottal] and Min[glottal], faithfulness constraints that are gradiently satisfied by candidates with any subset of phonetic cues to the abstract [glottal] feature. Constraints of this more abstract nature better capture the range of phonetic cues shown in our data (breathy voicing, F0 fluctuations, creaky voicing, glottal bursts). We propose that the pressure to maximize perceptual distinctiveness is stronger in the citation form context than in the phrase-medial context. On the other hand, we suggest that the pressure to minimize articulatory effort (e.g., Lindblom & Maddieson 1988; Boersma 1998; Flemming 2002; 2004) and to maintain the perceptibility of of other contrasts, such as vowel quality and tone, are stronger in phrase-medial contexts than in citation form contexts, where hyperarticulation is common. In a Functional Phonology approach, which draws its basis from OT (Prince and Smolensky 1993), this interplay is represented through the interaction of faithfulness constraints, which represent the pressure to maximize perceptual distinctiveness, and markedness constraints, which represent the pressure to minimize articulatory effort.

We present an analysis of this type, adopting a maximum entropy (MaxEnt) approach (Goldwater and Johnson 2003; Wilson 2006; Hayes et al. 2009), a more recent probabilistic version of OT similar to Boersma’s (1998) Gradual Learning Algorithm. In a MaxEnt model, constraints are weighted for their relative importance rather than strictly ranked with respect to each other. Constraint violations are multiplied by the weight of that constraint, and the winning candidate is that which incurs the lowest weighted sum of violations. Unlike the Gradual Learning Algorithm, MaxEnt requires only one parameter to set, and is more generalizable to processes of learning in other domains (Goldwater and Johnson, 2003). For these reasons, we use MaxEnt as a tool to present an analysis in which the context-dependent phonetic enhancement patterns present in the data emerge from the relative weighting of perceptual faithfulness and markedness constraints. However, we do not claim that other stochastic grammar modeling methods are necessarily less suited to the data we present.

Tables 27 and 28 show how Max[glottal] and Min[glottal] constraints interact with other markedness and faithfulness constraints to account for the MacZ data. Table 27 captures the citation form data, in which checked vowels are produced as creaky with a final closure and burst, and phonologically modal vowels are produced with modal voicing throughout and a fluctuation in F0.¹⁴ Here, the markedness constraint *Effort, which militates against candidates that require articulatory effort, has a weight of 0; the winning candidate may be articulatory effortful without incurring any violations. The Max[glottal] and Min[glottal] constraints interact with Max-SNR, a markedness constraint that prefers candidates produced with modal voicing, as these increase the perceptibility of contrasts related to vowel quality and contrastive tone. Though candidate b. as the surface form of an input checked vowel incurs a violation of the relatively high-weighted Max-SNR constraint, it receives fewer total violation points than candidate a., which also violates Max[glottal], and c., which incurs no violations of Max-SNR but violates Max[glottal] twice. When the input vowel is phonologically modal, candidates that are produced with non-modal incur a violation of Max-SNR. Candidate d., which does not violate Max-SNR, incurs more violations of Min[glottal] than the optimal candidate, as it is produced neither with breathy voicing, which farthest to the left of the glottal continuum, nor the F0 fluctuation which we argue phonetically enhances the [–glottal] feature specification.

The differences in phonetic cues to the glottalization contrast across contexts is modeled here as a simple reweighting of the active constraints. Unlike in citation form (Table 27), *Effort in the modeling of phrase-medial data (Table 28) has a relatively high weighting, preventing articulatorily effortful candidates from surfacing. This high weighting captures the generalization that in medial form, articulatorily difficult phonetic enhancements do not surface, and modal voicing is more prominent. On the other hand, the weight of Min[glottal] in this context is 0, far outweighed here by Max-SNR. As a result of this relative weighting, optimal candidates are those that are produced with modal voicing, and therefore that allow for perceptibility of vowel quality and tone contrasts, at the expense of minimizing or maximizing the [glottal] percept.

We would emphasize, though, that theories which explain surface representations primarily through tradeoffs between maximizing perceptual distinctiveness and minimizing articulatory effort (e.g., Boersma; 1998) do not on their own offer the same explanatory power that a contrast and enhancement approach (Hall; 2011) offers for the data presented here. In a system that highly prioritizes the pressure to maximize perceptual distinctiveness, as we see for our citation form data, contrast and enhancement helps to predict what types of phonetic correlates we would expect to see (articulatory and perceptual cues directly and indirectly associated with enhancing the glottalization contrast) and not expect to see (random ways of adding in phonetic distinctiveness). We therefore see the principle of minimizing articulatory effort as working in tandem with the contrast and enhancement approach. Because the type of context predicts the extent to which articulatory ease may be prioritized by the speaker, it therefore also both constrains and predicts the types of phonetic enhancements of phonological contrasts that are likely to surface.

6.4 F0 as a Cue to Contrastive Phonation in a Laryngeally Complex Language

Though the F0 fluctuation as a phonetic enhancement to a [glottal] contrast can be easily captured by theories such as Hall’s (2011), we note here that it is somewhat surprising in the context of MacZ’s phonological system. As shown in Section 3.1, MacZ exhibits both contrastive tone and contrastive phonation. It would be reasonable to imagine that the presence of two laryngeal contrasts in a given system would limit the ways in which each of these contrasts is enhanced. In other words, the existence of contrastive tone in this language could be expected to eliminate the possibility of an F0 enhancement of the phonation contrast. However, our data show that the presence of linguistic tone in this system does not preclude F0 from being employed as a phonetic enhancement to the phonation contrast. Rather, the trajectory of F0 is a cue that enhances the phonation contrast while, apparently, not obscuring the perceptibility of the tone contrast, also cued by F0. We assume that this is made possible by the timing of laryngeal contrasts in MacZ, discussed above, in which contrastive tone is conveyed in the first part of the vowel and phonation contrasts only in the final portion. Though we leave a discussion of how temporal configurations fit into abstract phonological contrasts to future work, we note here that the F0 cue surfaces as a phonetic enhancement of the phonation contrast only after the tone gesture is complete.

The F0 cues to tone and phonation, respectively, are also perceptually distinguished by the nature of their pitch trajectories. The F0 cues to phonation appear to lack specific pitch targets and manifest primarily as a deviation from the pitch trajectory. The lack of an apparent pitch target, combined with the relatively short duration of the F0 fluctuations observed here, distinguish the cues to tone from the cues to phonation. We assume that this distinction makes for a situation in which F0 is able to simultaneously provide the primary cue to tonal contrasts and a phonetic enhancement of the phonation contrast. As a result, our data provide an instance of one laryngeal cue, F0, simultaneously contributing to two laryngeal contrasts, tone and phonation.

6.5 The Phonological Status of Rearticulated Vowels

The focus of this discussion has been the phonetic enhancement of the phonation contrast between checked (V^ʔ) and modal (V) vowels in MacZ. A secondary goal of this paper was to consider phonetic evidence related to the phonological status of rearticulated vowels, and whether they are best analyzed as single vowels with a glottal interruption (V^ʔV) or as disyllabic sequences of a checked vowel followed by a modal vowel (V^ʔ.V). A summary of the phonotactic arguments for and against each analysis was presented in Section 3.2. The current study presented an opportunity to add phonetic evidence to our understanding of the status of rearticulated vowels by comparing their acoustic characteristics to those of checked and modal vowels.

Under existing analyses of other Zapotec languages, rearticulated vowels are considered single vowels with a glottal interruption (V^ʔV) that differ from checked vowels primarily in that their [+glottalized] feature is anchored to the center portion of the vowel rather than the final portion of the vowel (Arellanes Arellanes 2014; López Nicolas 2014). Under such an analysis, the vowel portion following the glottalization is simply an echo or continuation of the vowel preceding the glottalization, and it does not have the phonological status of a modal vowel. Therefore if this is the correct analysis, we might expect to see acoustic cues to glottalization similar to those of checked vowels near the central portion of the vowel rather than at the end, but we would not necessarily expect phonetic enhancements like those of modal vowels to occur in the final portion of the vowel.

In Section 3.2, we suggested another possible interpretation of these vowels based on the phonotactic characteristics of MacZ. What have been assumed to be individual vowels with a glottal closure produced in the middle (V^ʔV), based largely on the syllable structures of other Zapotec varieties, could in MacZ be better analyzed as a sequence of two vowels with the same quality, the first checked and the second phonologically modal (V^ʔ.V). If this is the case, we would expect to see phonetic enhancements similar to those of checked vowels in the V1 portion of the vowel sequence and phonetic enhancements similar to those of modal vowels in the V2 portion of the sequence.

Our data showed that the V2 vowels showed the same phonetic enhancements as modal vowels (F0 fluctuations, breathiness) and patterned like modal vowels according to context (citation form versus phrase-medial). The V1 vowels showed phonetic enhancements similar to those of phrase-medial checked vowels, but with less likelihood of a full glottal closure and greater likelihood of creakiness. As discussed above, we posit that these differences may be due to the fact that these vowels necessarily have different positions in the word in our data, with our checked vowels always appearing word-finally and our V1 vowels necessarily appearing word-medially before the corresponding V2. In other words, the phonetic differences between V1 and checked vowels in our data do not preclude an analysis in which V1 vowels are in fact medial checked vowels. Though further evidence is needed to make a more conclusive claim about the phonological status of V^ʔV sequences, our data lends phonetic support to the idea that initial portions of these sequences could in fact be checked vowels and the final portions of these sequences could in fact be modal vowels, thereby supporting an analysis of purported rearticulated vowels as disyllabic vowel sequences.