Conjunction in Colonial Valley Zapotec

Colonial Valley Zapotec (CVZ) refers to the language attested in a set of documents written in Oaxaca, Mexico, during the colonial period (1550–1810). In this paper I discuss the four primary conjunction strategies attested in CVZ documents: chela, huanee, =la, and asyndetic conjunction. These conjunction strategies are used interchangeably in CVZ throughout a wide time frame and geographic area. In some passages, a scribe alternates between different conjunction strategies to emphasize subgroupings within the construction. To analyze these examples, I introduce a theory of localized sensitivity, an expansion of the current typology of conjunction.


Introduction.
Colonial Valley Zapotec (CVZ), the language attested in a corpus of colonial-era (1550-1810) documents from Oaxaca, Mexico, has four primary strategies of conjunction ('and' coordination). This begs the question, how might they be used differently in context? Haspelmath (2004Haspelmath ( , 2007 has described how conjunction markers may be restricted to a particular syntactic category of conjunct (e.g., NPs) or a particular semantic type of conjunction. This is called SyntaCtiC or SeMantiC SenSitivity (see 2). However, the CVZ conjunction system does not fit into this typology since the four conjunction strategies do not show evidence of the sensitivity Haspelmath describes (4). Instead, there are complex constructions in which multiple conjunction strategies are used, wherein the strategies take on what I call loCalized SenSitivity (5). A conjunction system of this kind is undescribed in the literature, so this paper contributes to the typology (1) t/a/M-verB =SuBj oBj qui-xee =nij tobij missa Modern Zapotec languages are tonal and have complex vowel phonation contrasts, and it is reasonable to suppose that CVZ was similar, although these contrasts cannot be read from the CVZ orthography (see Smith Stark 2003:230).
The CVZ corpus is formed by two categories of alphabetic documents spanning from the middle of the sixteenth century up until the beginning of the nineteenth century: printed books written under the auspices of the Catholic Church and manuscripts written by native speakers. The printed resources include a grammar and dictionary written by Fray Juan de Cordova (1578aCordova ( , 1578b) and a bilingual Catholic doctrine (Feria 1567). These documents were clearly translated into CVZ from Spanish, but there is no information about the method of translation or the degree of fluency, in either language, of the translators.
The handwritten manuscripts are legal documents, such as testaments, land deeds, and bills of sale, written by native speakers of CVZ for local governmental purposes. These documents, although narrated from the point of view of a testator or landowner, were written by scribes who decided on the actual wording of the document and may have worked from templates. In this paper I refer to the "scribe" instead of the "speaker" when discussing the author of a document. In some cases, the documents were later translated into Spanish for use in court cases (e.g., to prove ownership of a piece of land); again, in the case of the contemporary Spanish translations, there is no information as to the fluency of the translator, although Heher et al. (2014) found that at least some translators were not native Spanish speakers. The orthography in both categories of documents is highly inconsistent. 4 Existing modern resources on CVZ include an index of Junta Colombina de México 1893 by Whitecotton and Whitecotton (1993), an online, searchable version of Cordova's (1578b) dictionary (Oudijk and Miceli 2015), a Unless otherwise noted, all examples are CVZ data; a full list of the CVZ sources referenced is given in 7. The glosses reflect my current understanding of CVZ morphosyntax, and all errors are my own. Glosses of data from other languages are based on the sources cited, but in some cases unnecessary detail has been suppressed. 3 As shown in 7, citations from CVZ sources generally follow the following scheme: Sourcepage;line. 4 See Smith Stark 2003 andBroadwell 2015a for discussion of CVZ orthography and phonology.
FieldWorks Language Explorer (FLEx) database (see Broadwell and Lillehaugen 2013), and a digital explorer of CVZ texts .
This paper uses data from sixteen documents spanning from 1567 to 1766, from eleven different towns. A full list of these documents is given in 7.
2. Conjunction typology. The focus of this paper is ConjunCtionthat is, conjunctive ('and') coordination, as opposed to disjunctive ('or') or adversative ('but') coordination. Conjunction constructions vary considerably across languages. In this section, I summarize conjunction typology as discussed by Haspelmath (2004Haspelmath ( , 2007Haspelmath ( , 2013. I use this typology as a basis for describing the CVZ conjunction constructions in 3 and 4. In conjunction constructions, two or more ConjunCtS are linked by a Con-junCtion MarKer; in English, this is the word and. These constructions are either aSyndetiC (lacking an overt marker) or SyndetiC (containing an overt marker, highlighted in boldface), as demonstrated in (2). Asyndetic conjunction is sometimes also called conjunction by juxtapostion. Syndetic conjunction constructions may be BiSyndetiC, with one conjunction marker for each conjunct, or MonoSyndetiC, with fewer conjunction markers than conjuncts. Since the syndesis of a conjunction construction is determined by the conjunction marker, the markers themselves are described as monosyndetic or bisyndetic. In this paper I use the term ConjunCtion StrategieS to refer to both overt conjunction markers and the process of asyndetic conjunction. Notice that in English, and may conjoin many different syntactic categories, including NPs (cats and dogs), VPs (ate and drank), and PPs (to the store and to the bank). Some languages disallow conjunction of a particular category. For example, Arabic disallows VP conjunction, forcing use of clause conjunction instead (Haspelmath 2007:22). Furthermore, some languages use different conjunction strategies for different syntactic categories. This is called SyntaCtiC SenSitivity; the conjunction marker is sensitive to the syntax of the conjuncts. For example Somali ([som]; Cushitic, Somalia) has three conjunction markers: iyo for NPs, oo for VPs, and -na for clauses.
(3a) Somali NP conjunction (iyo) rooti iyo khudrat bread and fruit 'bread and fruit' (Haspelmath 2007: ex. 60a) (3b) Somali VP conjunction (oo) Suuqa tag oo soo ibi rooti market go and andative buy bread 'Go to the market and buy bread!' (Haspelmath 2007: ex. 60b) (3c) Somali clause conjunction (-na) Carrur-tu waxay joogaan dugsi-ga waxay-na children-the 3pl.foC be school-the 3pl.foC-and bartaan Af-Soomaali learn language-Somali 'The children are in school, and they learn Somali' (Haspelmath 2007: ex. 60c) The most common syntactic sensitivity is a binary contrast between NP conjunction and VP/clause conjunction; this is attested in Dagbani ([dag]; Gur, Ghana). Haspelmath (2004:11) presents an implicational map for syntactic sensitivity, shown in (4). The "range" of a given conjunction strategy must cover a continuous segment of the sequence. For example, if a conjunction marker can conjoin both noun phrases and verb phrases, then it must also be able to conjoin adjectival phrases, assuming such conjunction constructions exist in the language.
(4) noun phrase -adjective phrase -verb phrase -clause In 4.2 I examine the range of syntactic categories attested in CVZ conjunction constructions and show that CVZ conjunction strategies do not have syntactic sensitivity.
There are several semantic subtypes of conjunction. For example, we can distinguish natural ConjunCtion, in which the conjuncts habitually go together as a unit, from aCCidental ConjunCtion (Haspelmath 2007:23). This may also be referred to as tight vs. looSe conjunction (Haspelmath 2004). The pair 'mother and father', for example, is often marked as natural conjunction. Languages may distinguish natural and accidental conjunction by using different conjunction strategies; this is called SeMantiC SenSitivity. For example, Erzya Mordvin ([myv]; Finno-Ugrian, Russia) uses asyndetic conjunction for natural conjunction and the marker di for accidental conjunction.
(5a) Erzya Mordvin natural conjunction (asyndetic) t'et'at avat father.pl mother.pl 'father and mother' (= 'parents') (Haspelmath 2007: ex. 62a) (5b) Erzya Mordvin accidental conjunction (di) t'ikšeń di sivel'eń grass and meat 'grass and meat' (Haspelmath 2007: ex. 62b) We can observe a weaker version of this phenomenon in written English, where the ampersand may be used to indicate a closer bond and set a pair of items off from a larger list, as in Add salt & pepper, cumin, and coriander. Notice that in this case the conjunction strategies are sensitive to the semantics of the construction-that is, to the relationship between the conjuncts. In other languages, different conjunction strategies are used depending on the semantic category of the conjuncts. For example, in the Asmat languages (Trans-New Guinea, West Papua), eněrím conjoins proper names, while other noun phrases are conjoined by the suffix -am. Takia ([tbc]; Oceanic, Papua New Guinea) uses separate conjunction strategies for human vs. nonhuman conjuncts (Haspelmath 2004:12).
Because there are many possible semantic categories of conjuncts, and because the distinction between natural and accidental conjunction is culturally specified, it is difficult to determine if a conjunction strategy is semantically sensitive given the absence of native speaker judgments. In 4.3 I discuss examples where different CVZ conjunction strategies are used in almost identical constructions and conclude that there is no evidence of semantic sensitivity in CVZ.

Overview of CVZ conjunctions
3.1. Previous literature. Broadwell's (2002) is the only previous linguistic study that deals specifically with CVZ conjunction constructions. Broadwell's analysis focuses on data from Feria 1567, a Catholic doctrine written in Spanish with a parallel Zapotec translation. He describes the use of some CVZ coordination markers, including the conjunction marker =la (see (11) in 3.2), with a comparison to coordination in modern Zapotec languages and a discussion of the historical development (see 3.3).
CVZ conjunction is also discussed in some of the colonial resources on the language. In his grammar of CVZ (1578a), the Dominican priest Fray Juan de Cordova mentions two conjunction markers: chela and =la (6). He describes =la as being postposed and gives an example in which =la is attached to the end of each conjunct; that is, he shows its use as bisyndetic. This example is glossed in (7). Cordova does not give an example using chela.
(6) Et, que, y atque, y mas, quoque, tambien. Para las dos primeras sirve chela vel la, postpuesto. vt. V.g. Llamame a pedro y a juan y alonso, coxeni pedro la juan la alonso la. Several conjunction markers are also listed in Whitecotton and Whitecotton's (1993) index of the Junta Colombina de México 1893. Chela is listed as meaning 'and', 'also', and 'consequently' (1993:45). Huanee is described as meaning "tambien, juntamente, y[gualmente]" 5 [also, together, equally] (1993:125). =la is listed with the definition "y (pospuesta)" [and (postposed)] (1993:143). 6 However, none of these descriptions of CVZ conjunction are very robust; indeed none even mention asyndetic conjunction, although it is attested throughout the corpus with some frequency. None of these sources address the question of why CVZ has so many conjunction strategies, nor do they describe how these strategies are used in context.
The current paper deals with the four conjunction strategies-chela, huanee, =la, and asyndetic conjunction-which account for the majority of conjunction constructions in CVZ documents. In 3.2 I introduce each conjunction strategy and describe its basic usage. In 3.3 I discuss conjunction in Modern Zapotec languages. In 4 I analyze the distribution of the conjunction strategies throughout the corpus with an eye to Haspelmath's (2007) typology. Finally, in 5 I discuss cases where multiple conjunction strategies are used within a single construction.

Basic constructions.
Chela is monosyndetic, as shown in (8), where the conjunction marker appears twice to conjoin three conjuncts. (Throughout the paper, I surround each conjunct with numbered brackets.) It is possible that chela is morphologically complex, containing the morpheme =la 'and', although che does not apparently carry meaning in CVZ.
The string chela appears in other morphologically complex words, such as lechela 'spouse'. Chela is also a verb root with a wide variety of meanings related to togetherness and joining: 'to be joined', 'to fit together', 'to set bones', 'to bind books', 'to thicken or harden' (e.g., rochela; Whitecotton and Whitecotton 1993:300).
Huanee has identical construction and usage to chela; as shown in (9) it is monosyndetic (two instances of the conjunction marker conjoin three conjuncts). A translation of huanee using these definitions might be 'very much with' or 'very much and', indicating that huanee might imply a closer semantic relationship. However, this meaning does not seem to be productive in CVZ; as I show in 4.3, huanee does not have semantic sensitivity. =la is almost always attested as bisyndetic, as shown in (11), where =la attaches to each of the conjuncts beni nabani and beni coti.  (12) and (13), where =la is used monosyndetically. These monosyndetic examples occur in different documents, Al697 and Co721, which were written in different towns and are separated by about 20 years. The Spanish translator of Co721 interpreted the coordination in (13) as disjunctive (meaning 'or'). The proclitic la= is indeed a disjunction marker in CVZ (Broadwell 2002), but the use should still be bisyndetic. Contemporary Spanish translations of CVZ documents often contain errors, so this cannot be used as definitive evidence.
It is possible that =la was used monosyndetically in some contexts. According to Haspelmath (2004:9), bisyndetic coordinators sometimes appear in monosyndetic constructions "when the coordinands are long, and especially when they are clauses." However, the apparently monosyndetic uses of =la are quite distinct-(12) shows conjunction of proper NPs, whereas (13) shows conjunction of PPs. Since =la is attested bisyndetically in a wide variety of circumstances, including clause conjunction (see 4), perhaps these examples are scribal errors. Further study may reveal otherwise as the CVZ corpus grows.
Asyndetic conjunction, although not described in colonial resources on CVZ, is attested throughout the corpus. In asyndetic conjunction constructions, conjuncts are conjoined by juxtaposition (there is no overt conjunction marker), as shown in (14). The data in this paper come from a corpus of 86 fully analyzed conjunction constructions involving chela (38 constructions), huanee (20), =la (15), and asyndetic conjunction (13).
In addition to these four primary conjunction types, a few other conjunction markers appear sporadically throughout the CVZ corpus. Chelañee is listed by Whitecotton and Whitecotton (1993:45) as meaning "asimismo; también; e ó y (conjunción)" [likewise; also; and (conjunction)], but it is attested only occasionally. Huachela is also listed (1993:121), but it does not appear at all in the corpus. The conjunction marker chela huanee appears twice in the corpus, for example: It is possible that this double-conjunction marker construction was meant to convey emphasis (i.e., 'both X and Y').
From these examples of rare conjunction markers we can infer that there is a set of conjunctive morphemes in CVZ that were used in various combinations. However, chela, huanee, and =la are by far the most common overt conjunction markers. Because the other markers are so rare, I do not address them in my general discussion.
The conjunction marker y, borrowed from Spanish, is also attested several times in the CVZ corpus. It is usually used in context with other Spanish words, for example in the (recited) declaration of 'Jesus, Mary, and Joseph', as in (16), or in Spanish numbers, as in (17). Y sometimes appears in conjunction constructions along with native CVZ conjunction strategies (see 5.2).

Modern Zapotec conjunction.
As mentioned, the exact relationship between CVZ and the modern Western Valley Zapotec languages is unclear. I refer the reader to figure 1 for the Zapotec family tree. Since there is some link, however, it is worthwhile to consider how the CVZ conjunction strategies may be related to those in modern Zapotec languages. Strangely, there is very little correspondence between CVZ conjunction markers and those documented for other Zapotec languages.
Conjunction constructions are not robustly described in most sources for Zapotec languages. However, what descriptions are available suggest that the diversity of conjunction strategies found in CVZ is not present in modern Central Zapotec. Most sources of modern Western Valley Zapotec languages cite a single conjunction marker beginning with /n/-for example, San Lucas Quiaviní Zapotec nah 'also, as well'. 7 (18) San Lucas Quiaviní Zapotec: NP conjunction with nah R-càa'z=a' g-ùuny=a' liaz=a' sti'lly re'nn ba'i, haB-want=1Sg pot-do=1Sg house=1Sg style this really [ 1 cosye'nn 1 ] nah [ 2 ba'nny 2 ] nah kitchen and bathroom and 7 Valley Zapotec languages also often use a conjunction/comitative marker borrowed from Spanish con, 'with', as in San Lucas Quiaviní Zapotec cëhnn (Munro and Lopez 1999). Forms like this do not show up in the CVZ data and must therefore have been borrowed more recently. haB-feed poSS-chickens 'My mother feeds the pigs and the chickens.' (Black 1994: ex. 348c) The relationship between modern Valley Zapotec conjunction markers and the three CVZ conjunction markers (chela, huanee, and =la) is not abundantly clear. It is possible that CVZ chela is cognate to San Lucas Quiaviní Zapotec chiru', which Munro and Lopez (1999)  In an informal context on Twitter, when asked how to say 'and' in their language, a speaker of Santa Catarina Albarradas Zapotec (Central, Albarradas) responded chuuru (@Ri-cardo01596181 2017a). In an unrelated tweet, the same speaker used churu to conjoin two NPs (@Ricardo01596181 2017b). A speaker of Teotitlán del Valle Zapotec (Central, Western Valley) responded that she had a similar word, chiru, in her language (@JanChvzSanti 2017).
[ 2 nyèe'c Dì:i'zh Tiilly que'ity r-gwè:=e'ng 2 ] even language Spanish neg haB-speak=3Sg.prox 'She's wearing a dress † and she doesn't even speak Spanish' (Munro and Lopez 1999) † Lit. 'she is standing equipped with a dress' Broadwell (2002) suggests CVZ =la may be related to the CVZ disjunctive la=, which he in turn compares to modern Valley Zapotec yes/no question particles (for example, San Dioniscio Ocotepec Zapotec lá=). The <nee> in CVZ huanee could be related to San Lucas Quiavini Zapotec nèe 'as well, too, also' (Munro and Lopez 1999). 9 Regardless, the diversity of conjunction strategies in CVZ does not appear in modern Valley Zapotec languages.
Northern Zapotec languages show more diversity. In addition to the /nV/ conjunction marker (na'), Butler (1980) lists two other conjunction markers in Yatzachi el Bajo Zapotec: len and nach. In the narratives in Butler's volume, however, nach is often glossed as 'entonces' [then] when applied to clauses, and len is often glossed as 'con' [with] when applied to NPs, so it is unclear if these are truly pure conjunction markers. Sonnenschein 2004 lists na' as the sole conjunction marker in San Bartolome Zoogocho Zapotec (also Northern) and treats lenh as a comitative marker. Long and Cruz list na', len, and nach as NP conjunction markers in San Bartolome Zoogocho Zapotec, but for clause conjunction they only list na', translating nach as 'entonces, y después' [then, and after] instead (1999:472-73). Although there is diversity of conjunction strategies in these languages, the markers len and nach do not appear related to the CVZ conjunction markers.

Geographic and temporal distribution.
Considering that CVZ has four conjunction strategies, it is important to ask how they are used in context-for example, are they a sign of dialectal variance, or are they restricted to different contexts? In this section, I show that none of the conjunction strategies are restricted to a particular temporal or geographic subset of the corpus. In 4.2 and 4.3, I show that none of the conjunction strategies carry syntactic or semantic sensitivity of the type described in 2.
The modern Valley Zapotec languages exist on a dialect continuum, with each pueblo speaking a distinct form of the language. Despite this modern diversity, CVZ is surprisingly consistent across many different towns and over the course of about 300 years. This homogeneity is also represented in the conjunction strategies. All four conjunction strategies are attested in individual documents from San Bartolome Coyotepec, San Sebastian Tectipac, and San Jéronimo Tlacochahuaya (Co721, Te626, and Tl675, respectively). Each CVZ manuscript was produced by a single scribe, and so these documents show that an individual CVZ scribe could make use of all the conjunction strategies. Additionally, chela, =la, and asyndetic conjunction are attested in a document from San Antonio Ocotlán (Oc686). This distribution covers most of the geographic range of CVZ documents (figure 2); thus the diversity of conjunction strategies does not appear to represent dialectal variation. Table 1 shows the distribution of the conjunction strategies over time between 1560 and 1740. After 1600, the distribution is even; the blank spaces can be explained by the variation in corpus size over time.
Huanee is the only conjunction marker not attested in the sixteenth century, and notably it is the only conjunction not attested in Feria's (1567) Doctrina (this text is ususual in using =la in most conjunction constructions). It is possible that huanee was not commonly used at the time. However the Doctrina was created under instruction from the church, so the exact source of the text is unknown, and it may not accurately represent the full range of CVZ grammar. Since native-speaker manuscripts from the sixteenth century are relatively rare, it might be coincidental that huanee is not attested in the current corpus for that time period. Haspelmath (2007) describes how conjunction strategies may be syntactically sensitive; that is, they are limited in which syntactic categories they may conjoin (see 2). In this section, I argue that the CVZ conjunction strategies do not carry this kind of sensitivity by showing the variety of grammatical categories that each conjunction strategy may conjoin. This evidence can be viewed two ways. Because any of the four strategies may conjoin, for example, NPs, the strategies are not distinguished by their ability to conjoin NPs (and so syntactic sensitivity is not at play). Similarly, because the conjunction chela, for example, may conjoin NPs, VPs, clauses, and relative clauses, it does not appear to be limited in the syntactic categories it can conjoin (and therefore is not syntactically sensitive).

Syntactic sensitivity.
As shown in (21) 10 In this paper I discuss clause juxtaposition as a method of conjunction in CVZ. Many Oto-Manguean languages also use clause juxtaposition to signal subordination (Thompson et al. 2007:241) and may distinguish this usage from asyndetic clause conjunction via prosody (see Palancar 2012 on Otomi). Although subordination through clause juxtaposition is also attested in CVZ, for each example of asyndetic clause conjunction analyzed for this paper, I have carefully considered the context to rule out other possible interpretations. In writing, some contexts are ambiguous, but it is reasonable to assume that where a conjunctive interpretation makes sense in context, it is also possible in CVZ.
Per the implicational map discussed in Haspelmath 2004 (see (4) in 2), conjunction strategies that conjoin both NPs and clauses are also likely to conjoin VPs. This conjecture is supported by the CVZ data since chela, =la, and asyndetic conjunction are all attested conjoining VPs, as shown in (23) Table 2 summarizes the distribution of the four conjunction strategies with regard to the syntactic category of their conjuncts. This is based on a set of 86 well-analyzed examples. None of the strategies appear to be restricted in terms of which syntactic categories they may conjoin, and likewise no syntactic category is exclusive to a single conjunction strategy. Thus none of the conjunction strategies carry syntactic sensitivity. 11 4.3. Semantic sensitivity. Haspelmath (2007) also describes semantic sensitivity, wherein conjunction strategies are sensitive to either the semantics of the construction (e.g., natural vs. accidental) or the semantics 11 It is still possible that certain strategies are preferred for particular syntactic category. In this sample, there is a clear preference to conjoin clauses with chela; choice of conjunction strategy is dependent on conjunct category with p < 0.05 (Pearson's Chi-squared test). However, because of the nature of this corpus, the conjunction examples have been gathered by a combination of chance encounter and searches for conjunction markers, skipping many underanalyzed portions of documents, so the sample should not be considered random. Regardless, this dataset is sufficient to show that the strategies do not carry syntactic sensitivity. of the conjuncts (e.g., animate vs. inanimate). Since these distinctions are culturally relative, it is difficult to prove absolutely that the CVZ conjunction strategies do not carry semantic sensitivity. However, in this section, I show that each strategy is used in a variety of semantic contexts, with substantial overlap.
The examples in (25) show chela and asyndetic conjunction attested in nearly identical constructions; in both cases, the strategies are conjoining the members of the Trinity. This minimal pair is evidence that these two conjunction strategies are not differentiated by the semantics of their construction.  All four conjunction strategies are also attested with conjuncts of similar types. Each of the four is used for a wide variety of conjuncts-for example, (animate) proper names, as shown in (27), and inanimate common nouns, as shown in (28) The examples in this section are summarized in table 3. Although complete paradigms cannot be established within the current CVZ corpus (and would be extremely difficult regardless without a native speaker), these data show that each conjunction strategy is used in a variety of contexts, and they do not appear to be differentiated by any semantic sensitivity.
To summarize, syntactic and semantic sensitivity do not explain the presence of multiple conjunction strategies in CVZ as they do in other languages since each strategy is used for a range of syntactic categories and in a variety of semantic contexts. Additionally, each conjunction strategy is attested in a wide range of locations, and they are all attested throughout most of the time period for CVZ documents. The four conjunction strategies seem to be interchangeable.
However, although CVZ conjunction strategies are not linked to any particular subtype of conjunction, scribes did use the various conjunction strategies to break a list into smaller units. In the next section I discuss conjunction constructions that make use of more than one conjunction marker and introduce the concept of localized semantic/syntactic sensitivity in CVZ.

Localized sensitivity.
In 4, I showed that none of the four CVZ conjunction strategies demonstrate syntactic or semantic sensitivity, and in general that there is no hard restriction on the use of a particular strategy. In each of these examples, only one conjunction strategy was used, although we have seen that multiple strategies may be used in a given document (see 4.1). In this section, I discuss examples in which the scribe uses multiple conjunction strategies within a single construction.
For example, in (29) the scribe uses both huanee and chela. Huanee conjoins the two NPs, 'two rods of beads' and 'two rings', while chela operates at the clause-level, conjoining 'I have two rods of beads . . .' and 'I have two yoke of oxen'. The structure of this conjunction construction may be informally represented by the tree shown in (30)  It appears that the change in conjunction strategy signals to the listener/ reader that the conjunctive relationship is going to change-that is, that there has been a shift to a different level of the structure. Since the shift is between noun phrase conjunction and clause conjunction, this appears to be the syntactic sensitivity described in 2. However, as I showed in 4.2, this relationship between chela and huanee is not predictable. Both chela and huanee are unrestricted in the syntactic categories they may conjoin, and each may conjoin both noun phrases and clauses (see (21) and (22)

in 4.2).
In (31), the scribe again uses both huanee and chela within a single conjunction construction. In this case, all of the conjuncts are NPs. However, semantically the conjuncts are broken into two groups: the testator's two children and his wife. Thus, the construction may be represented by the tree in (32), where 'Ambrosio and Jacinto' forms a separate entity. Again, the change in conjunction strategy signals a shift to a new part of the tree. To borrow from Haspelmath's (2004) terminology, I describe huanee as the tight conjunction marker (T) because it applies to the more closely related pair of siblings, and chela as the loose conjunction marker (L) because it conjoins the previous group of the siblings with the final conjunct, the wife. This is not, however, exactly the same as the tight/loose conjunction Haspelmath describes, which refers specifically to conjuncts that habitually go together, such as "needle and thread. In each of these constructions, it appears that the scribe used the different conjunction strategies to indicate a shift between different levels of conjunction. This phenomenon is similar to but distinct from the pattern of sensitivity described in Haspelmath 2007, wherein each conjunction strategy is set aside for use in a particular context.
Both constructions in (29) and (31) begin with using huanee and then switch to using chela. However, in the first case chela marks a switch in syntactic category of the conjuncts, whereas in the second, chela marks a change in the semantic relationship between the conjuncts. Since neither conjunction marker has semantic/syntactic sensitivity (see 4), there is no evidence that the listener/ reader knows what type of conjunct will come after chela, only that there will be a change. The sensitivity is not inherent to the conjunction strategies but appears to be activated only within a particular construction (e.g., chela is not tied to clause conjunction, it just happens to play that role in 29). For this reason, I call the phenomenon loCalized SenSitivity and say that a conjunction strategy is loCally SenSitized for a particular role within a construction.
This phenomenon can also be described in terms of scope, where switch in conjunction strategies defines the scope of the conjunction marker. While the operation of conjunction is associative (the grouping of the conjuncts does not affect the truth conditions), conjuncts are often analyzed as grouped into pairs, and a list of three conjuncts has ambiguous constituency. Using multiple conjunction strategies resolves this ambiguity, with a switch from one conjunction strategy to another marking a constituent boundary.
In 5.2, I present further examples of localized sensitivity and discuss the distribution of the different conjunction strategies across different contexts. I then discuss some limitations of localized sensitivity (5.3) before presenting my conclusions.

Distribution.
This use of multiple conjunction strategies, which I am calling localized sensitivity, is attested many times throughout the corpus in a wide variety of contexts. In this section, I provide a few illustrative examples and discuss the distribution of conjunction strategies in these constructions.
Example (33) contains a more complex construction than previously discussed, demonstrating how intricate conjunction constructions in CVZ can be. The construction is a list of possessions that the testator is giving away. Huanee is used as the loose conjunction marker (the "default" marker) throughout. Within the long list, the scribe uses chela huanee, =la, and chela to separate off pairs of items that should be interpreted as units (tight conjunction). For example, =la separates 'a (prayer) mat and altar', a pair of religious objects that go together, from the rest of the list.  -3;8-13) Most conjunction constructions in the corpus involve clauses and NPs, so conjunction strategies are often locally sensitized for these syntactic categories. However, localized sensitivity is used in many situations, and sometimes other syntactic categories are at play. In (34), the use of multiple conjunction markers distinguishes relative clause conjunction (performed with chela) from NP conjunction (=la).  -41r;7-9) Although the Spanish conjunction marker y is not fully productive in CVZ, and it is only attested in the context of the other Spanish words, a switch from y to a Zapotec conjunction marker could also signal a shift in the conjunction hierarchy. In (35), two witnesses are listed with a regidor, a type of local official. =la performs tight conjunction, conjoining the two witnesses, while e (a phonological variant of y) is used for loose conjunction.  The fact that the phenomenon of localized sensitivity could be applied to a borrowed conjunction marker indicates that the process was very productive and also reinforces how arbitrarily the different conjunction strategies were used in CVZ.
In total, I have analyzed 19 CVZ conjunction constructions that make use of more than one conjunction strategy; these constructions are summarized in table 4. 14 The phenomenon appears in ten different documents, from eight different towns. The table is sorted by a hierarchical ranking of types of conjunction based on scope, whereby, for example, clause conjunction is higher (more broad) than VP conjunction, and loose NP conjunction is higher than tight NP conjunction.
In this dataset, each conjunction strategy is used for multiple roles; for example, =la is used in various examples for VP, loose NP, and tight NP conjunction. The role of a particular conjunction strategy might change within a particular text (for example, chela is used for both clauses and NPs in Te626).
There are, however, some patterns in how different conjunction strategies are sensitized. Of the 20 examples, 15 use chela as the highest level (most broad) 14 Does not include the two examples discussed in 5.3. conjunction strategy, mostly at the clause level. 15 Notably, =la is never used in this position. The distribution for the second-highest level of conjunction strategy is more even: four use chela, eight use huanee, four use =la, and three use asyndetic conjunction. These patterns, however, do not allow a reader to predict what function a particular conjunction marker will play in a given construction. Within the current corpus, it is not possible to specifically define the boundaries of an instance of localized sensitivity, although the sensitivity assigned to a particular conjunction strategy may change within a single document. It is also unclear whether the sensitivity expires after a certain point or is retained and eventually replaced.

Counterexamples to the proposal.
Although multiple conjunction strategies are frequently used in long lists or constructions with multiple syntactic categories, this process is not obligatory. A single conjunction strategy may play multiple roles within a construction, for example in (36) In the examples in 5.2, it is intuitively clear what role a particular conjunction strategy is performing within the larger construction. However, I have found two constructions in which the utility of multiple conjunction is less clear. In (37), =la performs representative conjunction (R), conjoining various possible fibers for clothing. For the last part of the construction, 'and other types of clothes', which reinforces the representative nature of the conjunction, the scribe switches to chela. In this case there is an apparent semantic shift, but it is not very dramatic, and it cannot be easily visualized as a hierarchy, unlike other examples. (Al697-62r;8-12) † I believe this passage refers to, in order, the twelve Articles of the Faith (the Apostles' Creed), the Seven Sacraments, the Ten Commandments, and the Five Precepts of the Church.
The last conjunct, 'all the words of our Lord God', is similar to the last conjunct in (37) and seems to be hinting at representative conjunction. Since chela is used to conjoin the last conjunct, it might make sense to interpret huanee as separating off the Apostles' Creed and the Seven Sacraments as a pair and the Ten Commandments and the Five Precepts as a pair. In terms of modern Catholic theology, this is not a particularly salient grouping of these doctrines. It is unclear how a native speaker would interpret the roles of the various conjunction markers as they read this text. 16 6. Conclusion. Using data from a set of 16 CVZ documents, I have presented an overview of conjunction in Colonial Valley Zapotec. The four conjunction strategies attested in CVZ (chela, huanee, =la, and asyndetic conjunction) are used interchangeably throughout the corpus and do not carry semantic or syntactic sensitivity of the type described by Haspelmath (2004). However, the conjunction strategies may be locally sensitized by the scribe in a specific sentence to differentiate between two types of conjunction within that sentence. Based on the 1626 Testament of Juan López (Te626), a single conjunction strategy can be sensitized to multiple roles over the course of a long text. No patterns based on location or time are apparent, but they may appear as the corpus of analyzed data grows. A conjunction system of this type has not been previously described, so this analysis calls for an expansion of the current conjunction typology.
The exact relationship between CVZ and modern Valley Zapotec varieties remains unclear. As discussed in 3.3, only one conjunction marker (of the form /nV/) is documented for most Valley Zapotec languages; the complicated conjunction system found in the CVZ documents is not apparent. This might indicate that CVZ is not representative of the Zapotec language spoken during this time frame but is instead a separate, purely written language developed to best serve the nature of these religious and political documents.
However, the complexity of conjunction in CVZ still calls for further study into conjunction in modern Zapotec languages. Sources on Northern Zapotec languages list multiple conjunction markers. In Butler's grammar of Yatzachi el Bajo Zapotec, there is one example in which multiple conjunction markers are used in a single construction, shown in (38). The first four clauses are each of the form gwxi'e X, 'he bought X' and are conjoined by nach. The last clause is attached with the marker na', which implies that this is a representative (nonexhaustive) list. 'He bought corn, and he bought garlic, and he bought chiles anchos, and he bought onions, and who knows what else he bought.' † (Butler 1980:258) † Butler's translation: "Compró maíz y compró ajo, y (compró) chiles anchos y (compró) cebollas, y quién sabe qué más compró." This is notably similar in structure to the CVZ example in (37) in 5.3. Although no conclusions can be drawn from this single example, it is possible that Yatzachi el Bajo Zapotec has a system of localized sensitivity similar to that in CVZ, and further research into Northern Zapotec conjunction is warranted.
7. CVZ Sources. Data in this study came from a combination of (i) analysis derived from a FLEx database for CVZ documents (see Broadwell