Sibilant harmony in Santiago Tz’utujil (Mayan)

We analyze sibilant harmony in the Santiago Atitlán dialect of Tz’utujil (Mayan), a phenomenon that was briefly described by Dayley (1985). Novel data show that the obligatory harmony process (i) is asymmetrical (triggered only by [+ant] sibilants), (ii) progressive, and (iii) applies long-distance. Furthermore, we argue that the process is not stem-controlled. In contextualizing the phenomenon within the typology of sibilant harmony (Hansson 2010), we conclude that it is unique. Finally, we suggest that Santiago Tz’utujil sibilant harmony has been stable diachronically because the target segment /ʃ/ is always in the stressed syllable, thus being salient in the input during acquisition.

In section 4, we show that STz' sibilant harmony is typologically unique. Our STz' data come from in-person elicitations from 2017-2019 with four consultants in Guatemala, as well as online video elicitations from 2020-2021. Whenever the syntactic environment is helpful for understanding the data, we provide the target word in its sentential context.

Empirical description.
In this section, we expand on Dayley's (1985) description of sibilant harmony in STz' and arrive at the process described by the SPE-style rule in (7).
(7) Sibilant harmony in STz' ʃ → [+ant] / [+ant, +strid] … _ We use a descriptive rule format for ease of exposition, making no specific commitment as to the representation of harmony processes in the grammar in general (e.g., via agreement-bycorrespondence in OT - Rose & Walker 2004; see Rose 2011 for discussion). The sibilant harmony process in (7) has several properties that we will illustrate and discuss in turn in sections 3.1-3.4 and 4.1:  [+ANT, +STRID]. Dayley (1985) shows that /s/ is a trigger for harmony in STz' (2). Novel data in (10)  In (10a-c) above, harmony is triggered by a segment in the root. However, the trigger need not be a root segment -the /s/ in the causative suffix /-saa/ is also a trigger, as shown in (12) There are three candidate targets in the prefixal field for the assimilation process: (i) the completive aspect morpheme /ʃ-/, (ii) the /ʃ/ segment in the potential aspect morpheme /ʃt-/, and (iii) the /ʃ/ segment in the B2P agreement morpheme /iʃ-/. As shown below (and throughout the paper), the /ʃ/ in these prefixes never surfaces as [+ant]. We cannot assess the behavior of the other [-ant] sibilants /t ͡ ʃ/ and /t ͡ ʃ '/ within the prefixal field, since no STz' prefix contains these segments: We conclude that sibilant harmony in STz' is exclusively progressive. 5

THE HARMONY IS LONG-DISTANCE. All examples assessed so far have shown that [-ant] sibilants assimilate to a [+ant]
sibilant that is part of an immediately preceding morpheme. In this section, we show that the process is long-distance: the trigger and the target need not be in adjacent morphemes. This fact, and our discussion in 3.4 below, argue against an analysis that would take the STz' phenomenon to involve a phonologically-conditioned morphological alternation. 6 The crucial examples involve positional roots that are derived into transitive verbal stems. These stems can be passivized with /-Vʃ/, whose /ʃ/ segment is a target. Positional roots are a distinct category of roots in Mayan languages that are distinguished morphosyntactically and semantically from other root classes, like nominal and verbal roots (Henderson 2019;Polian 2017). Take for example the positional root /χer/, which surfaces in stems that are derived via suffixes that combine exclusively with positional roots: We conclude that sibilant harmony in STz' applies long-distance. This is consistent with the observation that sibilant harmony rarely applies across a compound boundary (see Hansson 2020). 6 See Paster (2014) for the criteria that distinguish whether an alternation arises due to phonological processes or from morphological subcategorization. 7 Two of our four consultants use an additional transitive suffix in this example type: In all examples shown so far, the target has been the /ʃ/ segment in the passive morpheme /-Vʃ/. However, this is not the only morpheme in which the harmony process materializes. The /ʃ/ segment in the nominalizer /-ʃik/ is also a target for the harmony process. This suffix attaches to a transitive stem and the resulting nominalization can be the complement of an auxiliary verb that is used in the progressive construction (Levin, Lyskawa & Ranero 2020), as illustrated below: A3S-PROG A3P-√buy-NMLZ P-many chicken 'He is buying many chickens.' A /ʃ/ segment in the nominalizer assimilates to a [+ant] sibilant in the root: An alternative interpretation of the data just described would posit that /-ʃik/ is decomposable into two affixes -the passive /-Vʃ/ and a nominalizer /-ik/. 8 In other words, this approach would require that nominalizations be derived from passivized stems. For example, Imanishi (2014) proposes this analysis for nominalizations in closely-related Kaqchikel (cf. Can Pixabaj 2015: 114-120 for K'iche'). If this analysis were correct for STz' as well, then the result of the harmony process would only materialize on the /ʃ/ segment of the passive /-Vʃ/. However, this morphological decomposition is not justified for STz'. First, the STz' nominalizations never surface with the vowel that is part of the passive morpheme. Second, some Kaqchikel nominalizations come in two versions, with a different voice morpheme in each (passive versus antipassive). Observe that the nominalization in (22a) contains a passive morpheme, whereas the nominalization in (22b) has an antipassive: quickly 'I want someone to quickly look for me.' Voice exponents are thus in complementary distribution in Kaqchikel nominalizations. In contrast, we have been unable to replicate similar data in STz' and examples like (22a-b) are not discussed by Dayley (1985) or García Ixmatá (1997;. Thus, there is no evidence from voice alternations in Tz'utujil that the sibilant segment in the nominalizer is an exponent of voice. Finally, the nominalizer suffix surfaces without a sibilant segment altogether in some nominalizations: 9 (23) Nominalization without sibilant a. /ʃ-Ø-in-t ͡ saq k'ij wej/ → [ʃint ͡ saq k'ij ̥ βej ̥ ] 'I made many tortillas.' COM-B3S-A1S-√make many tortilla b. /nuu-mχuon ruu-t ͡ saq-ik wej/ → [numχuon rt ͡ saqik βej ̥ ] / *[rt ͡ saqʃik] *[rt ͡ saqsik] A1S-PROG A3S-√make-NMLZ tortillas 'I am making tortillas.' The same can be observed with transitive roots that are passivized via the /-Vʃ/ suffix in general (see example (10c)): We thus could not posit that the sibilant is present underlyingly and undergoes deletion in a specific environment. As a result, we propose that the nominalizer has two lexically conditioned allomorphs: the allomorph /-ik/ combines with roots like /t ͡ saq/ 'make (tortillas)' (23) and /t ͡ s'aχ/ 'smash' (24), whereas /-ʃik/ combines with roots like /t ͡ s'uχ/ 'mock' (20b) and /loq'/ 'buy' (21c) (pace Levin, Lyskawa & Ranero 2020: 341).
To summarize, sibilant harmony in STz' is not morpheme specific, since it targets (i) the /ʃ/ segment in the passive /-Vʃ/ and (ii) the /ʃ/ segment in one of the allomorphs of the nominalizer, namely /-ʃik/.

Typological contextualization.
In this section, we lay out (and modify slightly) Hansson (2010)'s typologies of sibilant harmony systems and observe that the STz' pattern is an outlier. In section 4.1, we argue that the STz' phenomenon is an instance of progressive sibilant harmony that cannot be reduced to stem control, which is otherwise unattested. In section 4.2, we show that only one other language (Tlachichilco Tepehua) has been described as exhibiting harmony where only [+ant] sibilants are triggers. The combination of having sibilant harmony that is (i) progressive and (ii) triggered only by [+ant] sibilants makes the STz' phenomenon unique. We end by suggesting that sibilant harmony in STz' has been diachronically stable because the target is always in the stressed syllable. (26) Alternative interpretation of STz' sibilant harmony STz' sibilant harmony is stem-controlled, not progressive.
There are empirical and theoretical reasons to entertain (26). Baković (2000) assesses vowel harmony systems cross-linguistically and argues that stem-control gives rise to directionality in harmony systems, but directionality is epiphenomenal. 10 In a nutshell, progressive harmony arises in languages that employ suffixes because there is a trigger in the stem, which is to the left of a target in the suffixal field. Conversely, regressive harmony arises in languages that employ prefixes because there is a trigger in the stem, which is to the right of a target in the prefixal field. Baković argues that apparent cases of left-to-right (progressive) or right-to-left (regressive) harmony can be reduced to a single mechanism of stem-control, as opposed to two mechanisms (regressive and progressive spread), thus simplifying the grammar. Consider as an illustration sibilant harmony in Koyra (Omotic), which appears to be progressive (much like in STz' The process in (27) can be analyzed as stem-controlled, as opposed to progressive, given that the trigger is always part of the stem of affixation. Since this is a suffixing context, stem-controlled harmony gives rise to a "progressive" pattern epiphenomenally. Hansson (2010: 145) notes that seven languages exhibit what appears to be progressive sibilant harmony: Aari, Koyra, Benchnon, Zayse (Omotic); Rumsen (Costanoan); Izere (Bantu); and perhaps Wanka Quechua (Quechuan). However, he observes that sibilant harmony in all these languages can be analyzed as a stem-controlled process, raising the question of whether true leftto-right sibilant harmony exists (see Archangeli & Pulleyblank 2007: 367 for related discussion). Demi (Omotic; Seyoum 2008) and Sidaama (Cushitic; Kawachi 2007) should also be added to this list, since they are suffixing languages in which the result of sibilant harmony materializes to the right of the stem of affixation.
With this background in mind, two questions arise that serve to evaluate (26) as an analysis of STz' sibilant harmony. First, do we have definitive empirical reasons to reject the encoding of directionality altogether (left-to-right and right-to-left) for the representation of sibilant harmony? Second, can STz' sibilant harmony specifically be analyzed as stem-controlled? The answer to both questions is no.
First, while Baković (2000) shows that most vowel harmony processes can be reduced to stem-control (though see Rose & Walker 2011), directionality must be invoked to account for several cases of sibilant harmony. Samala (also called Ineseño; Chumashan) illustrates this most clearly. 11 In this language, the rightmost sibilant in a word determines the [± ant] feature of all sibilants that precede it, whether they be segments in the stem or prefixal field. In the following examples, the stem is underlined, /s-/ indicates 3 rd person subject, /-it/ indicates 1 st person object, /-us/ indicates 3 rd person object, and /-waʃ/ is a past marker: (28) Samala sibilant harmony is regressive (Applegate 1972  Since directionality must be invoked for sibilant harmony like Samala's -in this case, a right-toleft spread mechanism -this undermines the case for disallowing a left-to-right mechanism a priori. We thus reject a theoretical prior for adopting the alternative in (26) as an analysis of STz' harmony.
Moving to the second question -whether a stem-control analysis holds water at all specifically for STz' sibilant harmony -consider what Hansson (2010: 144) observes about stemcontrolled harmony in general: "the clearest cases of stem-control are those where harmony affects prefixes and suffixes alike, yielding bidirectional harmony 'outwards' from the root." Akan (Niger-Congo, Kwa) vowel harmony illustrates this. Note that the [± ATR] value of the affix vowels below is determined by the stem, which is underlined: With this in mind, the data we discussed in section 3 pose a significant challenge to a stem-control analysis of STz' sibilant harmony. Recall that much like Akan, STz' employs both prefixes and suffixes. As we showed, however, STz' is unlike Akan in that the relevant segments in the prefixal field never alternate. This is unexpected for a stem-controlled harmony process. In order for (26) to be descriptively adequate, then, one would need to stipulate that the domain of application for sibilant harmony in STz' excludes three morphemes in the prefixal field: (i) the completive aspect /ʃ-/, (ii) the potential aspect /ʃt-/, and (iii) the B2P agreement /iʃ-/. At this juncture, one could point out the empirical observation that some sibilant harmony systems do seem to be limited to specific domains -i.e., the root and derivational affixes in Totonacan languages (Hansson 2010: 326). This appears only to be an empirical tendency, however, and given the small number of languages exhibiting sibilant harmony (see (31)), it is unclear whether this tendency reveals something deep about limitations on the representation of sibilant harmony. For instance, data that contrast with the Totonacan pattern are readily available. In Tseltal (Mayan), the effect of sibilant harmony materializes on the incompletive aspect and A3S agreement prefixes (30b-c): We conclude, then, that (26) fails to characterize sibilant harmony in STz'.
To summarize this section, we considered an alternative analysis of the STz' harmony process that would take it to be stem-controlled, rather than progressive (26). We argued against this analysis, since (i) there is no justification for rejecting the representation of directionality altogether in sibilant harmony and (ii) such an analysis must resort to stipulation. We maintain, then, that STz' instantiates an instance of truly progressive sibilant harmony that cannot be reduced to stem control. 12 4.2. A TYPOLOGY OF TRIGGER CLASSES. Sibilant harmony patterns may be divided into three types based on the class of segments that triggers the process (Hansson 2010: 356-367). 13 To Hansson's typology below, we add Slovenian (Jurgec 2011;see Hansson 2020), Dime (Seyoum 2008;see Hansson 2010: 50), and Sidaama (Kawachi 2007; see Rose 2011) as type (31b) languages, and move Tseltal from type (31b) to (31a). Finally, we add STz' to type ( In what we have termed type (31a), Hansson identifies the following: Navajo, Chiricahua Apache, Plains Apache, Tanana (Athabaskan), Barbareño, Samala, Ventureño (Chumashan), Southern Paiute (Uto-Aztecan), Nebaj Ixil (Mayan), and possibly Misantla Totonac (Totonac); we add Tseltal (Mayan). The second type (31b) seems to be the most common. Hansson identifies several languages (some with multiple dialects) that fit this type: Tsuut'ina, Dene-tha (Athabaskan), Wiyot (Algic), Aari, Koyra, Benchnon, Zayse (Omotic), Moroccan Arabic (Semitic), Berber (various dialects; Afro-Asiatic), Coptic (various dialects; Afro-Asiatic), Nkore-Kiga, Rwanda, Rundi, Shambaa and Izere (Bantu); we add Slovenian (Slavic), Dime (Omotic), and Sidaama (Cushitic). As is clear from (31), languages are unevenly distributed across the possible types, an empirical observation that has the potential to shed light on the source and representation of sibilant harmony processes. What is most interesting for our purposes is the dearth of languages in type (31c). Before our addition of STz' to type (31c), Hansson (2010) had identified a single language where only [+ant] sibilants are triggers: Tlachichilco Tepehua (Totonacan). 15 As shown below, sibilant harmony is regressive in Tlachichilco Tepehua, unlike in STz': 12 Other cases that might qualify are Navajo, which may have progressive harmony within the prefixal field (see Hansson 2010: 148-152 for discussion and qualifications) and Teralfene Flemish (Hansson 2010: 145), though the latter requires more investigation. We leave a formalization of the representation of truly progressive harmony for the future. Note that certain models of consonantal harmony are built to disallow genuine progressive systems; see Hansson (2010: 261-294) for discussion. 13 This typology excludes languages with sibilant co-occurrence restrictions in roots that may or may not evidence a synchronically active process. 14 Hansson (2010) notes that classifying some languages as (31b) instead of (31a) can be an analytical choice. For instance, two languages classified as (31b) by Hansson (Nkore-Kiga and Tseltal) actually exhibit a symmetrical system. However, Hansson claims that [+ant] triggers are more restricted than [-ant] ones. For example, he argues that Tseltal exhibits ʃ → s only if the target is sandwiched between two /s/ triggers (Kaufman 1971 (30)) lead us to classify Tseltal as a (31a) language. 15 Hansson (2010: 46) observes that Tz'utujil has sibilant harmony (citing Dayley 1985), but he does not classify the pattern into his typology. On page 389, he classifies Tz'utujil sibilant harmony as being a static morpheme structure constraint on roots, so Hansson must have been referencing syllable internal co-occurrence restrictions (Dayley 1985: 31), rather than the rule in (1); see section 5 below.
(32) Tlachichilco Tepehua sibilant harmony (Watters 1988  Hansson argues that the relative rarity of type (31c) languages is not accidental, linking it instead to the Palatal Bias -a robust empirical generalization related to speech errors in production: (33) Palatal Bias in relation to sibilants (Shattuck-Hufnagel & Klatt 1979) Speech errors involving substitutions like s → ʃ are much more common than ʃ → s In a nutshell, Hansson claims that whatever mechanism underlies the Palatal Bias (an empirical observation) also gives rise to sibilant harmony. We therefore expect more languages where sibilant harmony gives rise to s → ʃ only (31b), as opposed to ʃ → s only (31c).
In light of this proposal, Hansson must explain the existence of Tepehua harmony, which is ʃ → s (31c). He speculates that Tepehua harmony has an extraordinary diachronic source: it arose from the grammaticalization of a language-specific sound symbolism pattern. Totonacan languages are characterized by sibilant alternations within a stem, where a stem with a [-anterior] sibilant maps to a meaning indicating a greater intensity than a stem with a [+anterior] sibilant: (34) Sound symbolism in North Central Totonac of Apapantilla, Puebla (MacKay 1994) a.
[tʃutʃu] 'she sucks' Hansson proposes that Tlachichilco Tepehua speakers reanalyzed a sound symbolism pattern as a purely grammatical process. The pattern of ʃ → s harmony in Tepehua is therefore not a challenge to the claim that whatever underlies the Palatal Bias also gives rise to sibilant harmony. Instead, it is an outlier that can be explained independently (see Hansson 2010: 360-363). STz', however, is a type (31c) language for which we cannot invoke an extraordinary diachronic source -e.g., there is no sound symbolism mechanism that could explain the outlier nature of the STz' process. Nevertheless, let us point out a characteristic of STz' grammar that might contribute to the diachronic stability of its sibilant harmony: the /ʃ/ in which the harmony process materializes is always in the final syllable of the word, which bears stress: (35) STz' is stress-final (data repeated from (19)) /ʃ-in-sin-ɓaa-Vʃ-a/ → [ʃin.sin.ɓaa.sá] 'I was undressed.' We conjecture that this property of STz' grammar makes the ʃ → s alternation particularly salient for the learner. If we assume that whatever underlies the Palatal Bias would make the STz' pattern diachronically unstable, then this particular characteristic of STz' grammar -the fact that the target of harmony always occurs in the final syllable -has contributed to the pattern's stability over generations of transmission. We end by noting that adding a single new outlier to the typology in (31) does not conclusively cast doubt on Hansson (2010)'s proposal connecting the mechanism underlying the Palatal Bias to the skewed patterning of sibilant harmony cross-linguistically. However, it appears that more detailed work in the Mayan family could establish even more examples of the rare (31c) type. For example, DuBois (1981) briefly notes that Sakapultek, another language of the K'ichean branch, appears to exhibit the same sibilant harmony system as STz ': 16 (36) Sakapultek sibilant harmony (adapted from DuBois 1981: 137, 207) a. /tsili-ʃ/ → [tsilis] 'it was returned' √return-PASS b. /kuɁum-asa-ʃ/ → [kuɁmasas] 'it was moved' √move-CAUS-PASS c. /r-ts'unu-ʃ-iik/ → [r-ts'unu-s-iik] 'its being asked for' A3S-√ask.for-PASS-NMLZ At present, we do not have as detailed a description for Sakapultek as we do for STz', so we hesitate to conclude that a third language should be added to type (31c). DuBois (1981: 46) suggests that the same sibilant harmony process also exists in Poqom (K'ichean), though we have been unable to find corroboration in published sources (e.g., Santos Nicolás & Benito Pérez 1998;Malchic Nicolás, Mó Isém & Tul Rax 2000). Romelia Mó Isém and Carlos Humberto Cú Cab (p.c.) also point out to us that some dialects of Q'eqchi (K'ichean) have the same sibilant harmony process, though this has not been documented either.
To summarize, we added STz' to the typological distribution of sibilant harmony based on trigger type, showing that it is one of two outlier languages where only [+ant] sibilants are triggers. We laid out Hansson's (2010) proposal that linked the scarcity of type (31c) languages to the mechanism underlying the Palatal Bias and conjectured that the stress-pattern in STz' might explain how the sibilant harmony process in the language has been stable diachronically.

Conclusions and future work.
We showed that STz' exhibits a typologically unique process of sibilant harmony. The combination of properties that make it an outlier are the following: the process (i) is triggered solely by [+ant] sibilants, (ii) is progressive, and (iii) cannot be reduced to stem-control.