Left branch extraction, object shift, and freezing effects in Tumbalá Ch’ol

Ch’ol (Mayan) exhibits asymmetries in what is available for left branch extraction. While both numerals and interrogative possessors are able to extract from absolutive subject position, only numerals may extract from absolutive object position. To capture this asymmetry, I provide evidence that objects with overt possessors always undergo object shift, blocking left branch extraction. This follows from the Freezing Principle (Ross 1974; Wexler & Culicover 1977), or a ban on extraction from a moved constituent. Objects with numeral modifiers do not obligatorily undergo object shift and therefore may extract from the object. In addition to numerals and interrogative possessors, I present and analyze possibilities for other elements to extract out of various positions. I situate this work within Agree-based theories of extraction (e.g. Rackowski & Richards 2005; van Urk & Richards 2015; Branan 2018) and discuss this proposal’s theoretical implications. Unless otherwise noted, all data comes from the author’s fieldwork.


Introduction
Ch'ol (Mayan) exhibits asymmetries in what is available for left branch extraction. Interrogative possessors may extract from absolutive subjects in (1a), but not from absolutive objects, as in (1b). Numerals, on the other hand, may extract from absolutive subjects (2a) and absolutive objects (2b).

Nominal word order
In the nominal domain, possessors follow possessees, as in (8a). When the possessor is interrogative, it precedes the possessee, generating the possessor-possessee order in (8b).

Clausal word order
Like many Mayan languages, Ch'ol is a verb-initial language and has alternating VOS/VSO word order (Vázquez Álvarez 2002b;Coon 2010;Clemens & Coon 2018). This alternation is determined by structural properties of the object. VOS word order is associated with NP objects, and VSO with DP objects. That is, bare object NPs such as waj 'tortilla' in (10a) must appear next to the verb, generating the pragmatically neutral VOS word order. VSO word order is ungrammatical with NP objects, as shown by (10b). Other modifiers, such as numerals, may also appear with objects in VOS object position (discussed further in Section 3). ]. pfv a3-eat-tv nc-Rosa tortilla. Intended: 'Rosa ate a tortilla.' DP objects, or objects with material in or above D, must appear in VSO object position. 1 In Tumbalá Ch'ol, VSO word order with a full DP object was judged felicitous in a context in which the speaker is looking at a dead deer and says (11a). DP objects are judged to be ungrammatical in VOS object position, as in (11b) The preverbal position is associated with interrogative, focused, and topicalized constituents. As Ch'ol is an obligatory wh-fronting language, interrogative words such as majki 'who' and chuki 'what' must appear preverbally in (12a) and (13a). In situ wh-words are ungrammatical, as per (12b) and (13b). 2 (12) a. Chuki i ta' i-k'ux-u t i aj-Rosa? what pfv a3-eat-tv nc-Rosa 'What did Rosa eat?' b. *Ta' i-k'ux-u chuki aj-Rosa? pfv a3-eat-tv what nc-Rosa Intended: 'What did Rosa eat?' (13) a. Majki i ta' i-k'ux-u waj t i ? who pfv a3-eat-tv tortilla 'Who ate a tortilla?' b. *Ta' i-k'ux-u waj majki? pfv a3-eat-tv tortilla who Intended: 'Who ate a tortilla?' Focused-marked constituents in Ch'ol may also appear in the clause-initial position (Vázquez Álvarez 2011), as shown in (14), though in situ focus has also been shown to be possible in Ch'ol (Clemens et al. 2017). (14) a. Aj-Rosa i ta' i-juch'-u sa' t i . nc-Rosa pfv a3-grind-tv masa 'Rosa ground masa.' b. Sa' i ta' i-juch'-u t i aj-Rosa. masa pfv a3-grind-tv nc-Rosa 'Rosa ground masa.' In sum, the pragmatically neutral order for transitive verbs with two overt arguments is VOS. As presented above, there are restrictions on VOS objects, summarized in Table 1. NPs are grammatical in VOS object position, but not in VSO. DPs, on the other hand, are banned from appearing in VOS object position. Next, I present Coon's (2010) analysis for deriving the restrictions on objects as well as provide new data that expands on the object restrictions in Table 1.

Deriving word order in Ch'ol
The word order restriction is summarized in (15).
(15) Nominals with material in or above D 0 are banned from VOS object position in Ch'ol. (Coon 2010: 361, ex. 23) Any word order analysis that derives VSO via object shift is compatible with the current proposal (i.e. Aissen (1992); Coon (2010)). Here, I will adopt Coon's (2010) analysis of (15), where objects with material in or above D shift to a position above the vP. Evidence for object shift comes from the placement of adjuncts and the semantic interpretation of shifted and unshifted objects. I additionally provide new data demonstrating that (15) holds for nominals with overt possessors in object position. The possibilities for object shift will be important in capturing the left branch extraction asymmetries discussed in Section 4.

A predicate-fronting analysis of VOS/VSO (Coon 2010)
According to Coon (2010), VOS word order follows from the two factors in (16). I refer the reader to Coon (2010) for details motivating these factors.
(16) A predicate-fronting proposal for verb-initial word order a. Strong agreement features on T requiring the verb to move overtly to T; b. A general absence of head movement. (Coon 2010: 355) VOS word order is derived, as in (17), where the vP, containing the verb and object NP, fronts to Spec,TP to check strong agreement features on T.
To derive VSO word order with full DP objects, the object first shifts out of the vP to a projection Coon (2010) labels AbsP, following Massam (2000). Then the vP fronts to Spec,TP, as in (18), generating VSO word order.
Importantly, VSO word order is derived, in part, via object shift. Additional support for object shift comes from the placement of adjuncts, to which I turn next. 3

Adjunct placement as a diagnostic for object shift
The placement of adjuncts has been used as a test in Germanic languages for object shift (i.e. Holmberg (1986);Diesing (1992;) and can provide evidence that the object has moved. As previously discussed in Coon (2010), temporal adverbs and prepositional phrases can intervene between the DP object and the verb, generating V-XP-(S)-O order, as in (19a). An adjunct may not intervene between an NP object and verb, as in (19b). 4 The adverb may only appear after the NP object, as in (19c). 5 3 Coon (2010: 365 footnote 11) remains agnostic to the structural location of the aspect marker, suggesting that one possibility is that it heads an Asp(ect)P above vP and it is AspP, rather than vP, which fronts. 4 While temporal adverbs and PPs display similar distributional behavior with respect to NP and DP objects, other adverbials attach directly to the VP, such as the prefix cha'-in (i).
(i) Tyi i-cha'-boñ-o otyoty jiñi wiñik. pfv a3-again-paint-tv house det man 'The man painted the house again.' (Coon 2010: 373) As suggested by Coon (2010: 373), I take the adverb in (i) to be a VP-level adverb. Evidence for this comes from the fact that (i) means that the house has been painted before, but not necessarily by the same man, suggesting the adverb does not have scope over the subject. These VP modifiers can have scope over the object and thus seem to attach lower in the verbal domain than the adverbs and PPs discussed above. Directionals are another set of elements that seem like adverbs in Ch'ol and may appear between a bare noun object and verb. However, as Coon (2013) notes, directionals are different from other adverbials, as they must appear next to the verb, without anything else intervening. Directionals have been reported to be a closed class of items in a number of Mayan languages (England 1976: Mam;Haviland 1993;Aissen 1994: Tsotsil;Mateo Toledo 2004: Q'anjob'al). Mateo Toledo (2004) argues that directionals in Q'anjob'al are clitics to the verb. Aissen (2009) proposes that the directionals in Tsotsil are serial verb constructions. For these reasons, I do not use directionals as a diagnostic for object shift. 5 No difference was exhibited with PP adjuncts for the word order facts in (19). The adverb ak'bi 'yesterday' may be replaced with a PP adjunct such as tyi kotytoty 'in my house' and the word order facts remain the same.
(19) Data based on Coon (2010: ex (40) and (41) According to Coon's (2010) analysis, these adverbs and PPs attach as adjuncts to VoiceP. As the NP object remains within the vP, nothing may intervene between it and the verb, and the ungrammaticality of (19b) is predicted. The vP moves over the null subject and adjunct, generating the grammatical V-O-XP-(S) word order of (19c), given in ( To derive (19a), repeated in (21a), the DP object shifts to a position above vP, below the subject and adjunct. The vP then fronts to Spec,TP, generating the order V-XP-(S)-O, modeled in (21b).
In sum, the placement of adjuncts with respect to the object provides evidence for whether object shift has taken place. That is, if an adjunct can intervene between the verb and object, the object has shifted. Next, I introduce new data on objects with overt possessors, which pattern with the DP objects just discussed.

Overt possessors in object position trigger object shift
Data from alternating VOS/VSO word order and the placement of adverbs provides evidence that the restriction in (15) also holds for nominals with overt possessors in object position. As seen with objects with determiners, VOS word order is prohibited for objects with overt possessors, as in (22a). Rather, objects with overt possessors appear in VSO object position in (22b). As in (19a) for DP objects, an adjunct may intervene between the verb and an object with an overt possessor as in (23). 6 (23) Ta' j-k'ux-u ak'bi [ DP i-waj aj-Eve ]. pfv a1-eat-tv yesterday a3-tortilla nc-Eve 'I ate Eve's tortilla yesterday.' For these reasons, I analyze objects with overt possessors as DPs, which consequently trigger object shift. This will be important for capturing the ban on interrogative possessor extraction from object position.

Objects with numeral modifiers can shift
Numerals can appear in VOS object position, as shown in (24a). New data in (24b) demonstrates that numerals may also occur in VSO object position, indicative of object shift.
]. pfv a3-eat-tv nc-Rosa two-clf tortilla 'Rosa ate two tortillas.' As may be expected, shifted and unshifted objects with numerals give rise to different semantic interpretations. Evidence for this comes from their scopal relations with adverbs. In (25), the PP adjunct may appear before or after the object cha'tyikil wiñik 'two men'. While (25a) is felicitous in a context in which the speaker sees two different men every day, (25b) is felicitous in a context in which the speaker sees two specific men every day.
(25) a. Ju-jum-p'ej k'iñ mi j-k'el [ cha'-tyikil wiñik ] tyi bij. one-red-clf day imf a1-see two-clf man prep way 'Every day, I see two men on the road.' one-red-clf day imf a1-see prep way two-clf man 'Every day, I see two (specific) men on the road.' 2 > ∀ These interpretational differences are predicted by Diesing's (1992) Mapping Hypothesis, where unshifted objects such as in (25a) are existentially closed at the VP level, as they are within the domain of existential closure, (i.e. within the VP). Therefore, the object in (25a) takes low scope with respect to the adverb. Presuppositional or specific objects as in (25b) shift out of the domain of existential closure (i.e. outside of VP), indicated by the PP that intervenes between the verb and the object. In (25b), the object scopes over the adverb and the interpretation is that the speaker sees two specific men every day on the road. The interpretations in (25) constitute further semantic evidence for object shift, compatible with Diesing's Mapping Hypothesis. The optional shifting of objects with numerals will be relevant for capturing the left branch extraction asymmetries. Before summarizing the Ch'ol object shift facts, I briefly consider reasons why object shift occurs in Ch'ol. Motivations for object shift have been explored in a number of languages (Holmberg 1986;Diesing & Jelinek 1995;Diesing 1996;Vikner 2005). For Ch'ol, I suggest that overt definiteness is incompatible with an existential interpretation (see also discussion in Diesing & Jelinek (1995: 130) for German). Overt material associated with definiteness includes determiners, demonstratives, and overt possessors. Bare nouns, on the other hand, remain in the VP and may be existentially closed there. 7

Summary of object restrictions
An updated summary of object restrictions is given in Table 2. While bare objects must appear in VOS object position, objects with determiners or overt possessors must appear in VSO object position-indicative of object shift. Objects with numerals may appear in VOS or VSO object position. These object shift facts lay the foundation for capturing the asymmetries exhibited with respect to left branch extraction.

Left branch extraction data
In this section I lay out the empirical facts for the left branch extraction, or subextraction, of a number of elements in the Tumbalá dialect of Ch'ol. Left branch extraction is when an element (located in the left branch of a larger NP or DP) is moved via A-bar movement. In the Mayan literature, interrogative possessor extraction has been noted before for the Tila dialect of Ch'ol (Coon 2009) and the closely related language Tsotsil (Aissen 1979;1987;. This study adds new data on the possibilities for extracting a variety of elements from different positions in the Tumbalá dialect of Ch'ol.

Interrogative possessor extraction
Interrogative possessors may extract from absolutive subjects, as in (26a), where the interrogative possessor majki appears before the verb. Pied-piping is also an available option, as shown by (26b), where the constituent containing the interrogative possessor and its possessee fronts to the preverbal position. Semantically, left branch extraction is associated with focus on the extracted constituent. In (26b), a speaker may be asking simply whose cow fell, but in (26a), the speaker may not have properly heard the name of the cow-owner and is asking for the interlocutor to repeat it.
Extraction of an interrogative possessor is also possible out of ergative subjects in (27a), though speaker variation is attested. Pied-piping is an option available to all speakers, as in (27b). 9 8 The traces represent where the interrogative possessor originates (i.e. postnominally). Interrogative possessors undergo additional movement within the DP (not shown by the traces) to generate the correct surface word order, in which the interrogative possessor is prenominal. 9 In the Tila dialect of Ch'ol, extraction from ergative subjects is ungrammatical (Coon 2009 who a3-sister pfv a3-see-tv-b2 'Whose sister saw you?' Extraction of an interrogative possessor from the subject of an unergative verb is grammatical for some speakers (and slightly better than (27a)), as shown in (28a). Pied-piping, again, is an available option for all speakers, as in (28b).
Extraction of an interrogative possessor from an absolutive object, on the other hand, is ungrammatical, as in (29a). 10 In this case, pied-piping is the only available option, as in (29b). 11 10 With the applicative suffix -be, possessor extraction from object position is possible, as in (i). Following Coon & Henderson (2010), (i) is an external possessive construction, indicated by the applicative marker, and not an instance of left branch extraction.
(i) Majki i ta' a-k'el-be i-chich t i ? who pfv a2-see-appl a3-sister 'Whose sister did you see?' 11 As noted by Coon (2009) for the Tila dialect of Ch'ol, Ch'ol allows recursive possessive constructions. Additional data confirms that the Tumbalá dialect has recursive possessive constructions and that left branch extraction patterns similarly. Extraction is banned from absolutive objects (i), but possible from absolutive subjects (ii).
who a3-sister pfv a2-see-tv 'Whose sister did you see?' Paralleling the pattern for objects of transitive verbs, extraction of an interrogative possessor out of the direct object or indirect object of a ditransitive verb is also ungrammatical, as in (30) and (31). 12 When occurring in ditransitive objects, interrogative possessors must pied-pipe their possessees.
who pfv a2-sell-appl a3-car nc-Rosa Intended: 'Whose car did you sell to Rosa?' Only possible meaning: 'Who did you sell Rosa's car to?' ? who pfv a2-sell-appl car a3-sister Intended: 'Whose sister did you sell a car to?' b. *Majki i ta' a-choñ-be [ i-chich t i ] karu ? who pfv a2-sell-appl a3-sister car Intended: 'Whose sister did you sell a car to?' Finally, interrogative possessors inside PPs must pied-pipe the entire PP, as in (32a). They may not extract from PPs as per (32b).
While extraction of multiple possessors from ergative subjects is possible for some speakers in (iiia), the speakers that accept (iiia) judge extraction of two possessors ungrammatical in (iiib). I leave an analysis on this difference for future work.
who a3-plate a3-dog pfv a3-hit-tv-b2 'Whose dog's plate fell on you?' 12 I provide two possible word orders for the ditransitive constructions, as multiple overt postverbal nominals are rare in Mayan languages. It seems that when the direct object is bare, the pragmatically neutral ordering of postverbal objects is V-DO-IO.
who a3-house pfv arrive-iv-b2 prep Intended: 'Whose house did you go to?' I follow previous work, which argues that that PPs in Ch'ol are adjuncts (Coon 2013;Coon & Preminger 2011), and therefore adjunct islands. This claim is further supported with the data above, where extraction of the PP object in (33) and extraction of the interrogative possessor in (32b) are banned. Additional support for the islandhood of PPs comes the fact that PPs are never selected as complements in Ch'ol. For instance, picture-of-NPs are expressed with possessive phrases-'a picture of Rosa' in Ch'ol would be 'Rosa's picture'.

Numeral extraction
Numerals may extract from absolutive subject position in (34a) and absolutive object position in (34b). 13 (34) a. Cha'-kojty i ta' yajl-i [ t i wakax ]. two-clf pfv fall-iv cow 'Two cows fell.' Absolutive subject b. Cha'-kojty i ta' i-k'el-e [ t i wakax ] aj-Rosa. two-clf pfv a3-see-tv cow nc-Rosa 'Rosa saw two cows.' Absolutive object When only the numeral extracts, the numeral alone is focused (rather than the whole nominal argument). For instance, (34a) may be a response to someone who mistakenly said that only one cow fell. 14 As shown for interrogative possessors, extraction out of ergative subjects receives mixed judgements, as in (35).
two-clf pfv a3-eat-tv-b1 cat 'Two cats bit me.' Ergative subject Numeral extraction is possible from the direct object of a ditransitive verb, as in (36a), but degraded from the indirect object (36b). The data in (36)  Further data reveals a clear difference between extraction from the direct object and extraction from the indirect object in Ch'ol. In (36b), the classifier used with 'two' is the human classifier (-tyikil). When both objects take the same classifier (-kojty), as in (37) with chili peppers and cats, the extracted numeral unambiguously modifies the direct object 'chili peppers'.
(37) Cha'-kojty ta' k-äk'-e ich mis. two-clf pfv a1-give-appl chili cat. 'I gave two chili peppers to the cat.' NOT: 'I gave chili peppers to two cats.' The example in (37) provides evidence that extraction from the direct object is easier than extraction from the indirect object. I return to why this may be the case in Section 5.7. Finally, it is not possible to extract numerals from PPs, as shown by the ungrammatical (38). I take (38) as further evidence for the islandhood of PPs in Ch'ol.

Demonstratives and non-interrogative possessors cannot extract
For completeness, I present data on other elements, which may not extract from any position. Demonstratives such as ixä 'that' may not extract from any argument position, as shown by the data in (39) Ditransitive IO I return to these ungrammatical instances of extraction in Section 5.5.

Summary
The left branch extraction facts from this section are summarized in Table 3. 15 Crucially, extraction of interrogative possessors is banned from object position of transitive and ditransitive verbs. Numerals, on the other hand, can extract out of the object position. Extraction from subjects of transitive verbs and indirect objects is degraded. Demonstratives and non-interrogative possessors may never extract out.

Shifted objects are islands to extraction
I argue that object shift, discussed in Section 3, freezes the object, creating an island. Elements within that object are subsequently prohibited from extracting out. This follows from the Freezing Principle in (41), which bans moving something out of an already moved element (i.e. (Ross (1974); Wexler & Culicover (1977); see also discussion in Corver (2006)).
(41) The Freezing Principle (Wexler & Culicover 1977: 17) Moved constituents are islands to extraction. 15 As a reviewer points out, elements which can extract are those that are prenominal. However, not all prenominal elements can extract: demonstratives and determiners cannot extract out, for example. While the Freezing Principle in (41) may be too strong 16 -indeed, not all moved constituents are islands for extraction (see, e.g. Konietzko et al. 2018)-object shift has been shown to create islands in a number of languages, including English (Chomsky 1973;Lasnik 2001), German (Diesing 1992), and Ukrainian (Mykhaylyk 2010). Here I argue that Ch'ol, too, bans extraction from shifted objects. Recall that nominals with overt possessors are banned from VOS object position and must always shift under Coon's (2010) analysis. Like non-interrogative possessors, interrogative possessors also trigger object shift before the interrogative phrase fronts to the preverbal position. It then follows that for interrogative possessors in object position, the object first undergoes object shift, creating an island as per the principle in (41). This accounts for the ungrammaticality of interrogative possessor extraction from object position. 17 Given the principle in (41), extraction from, not extraction of, the shifted object is banned. As noted above, the interrogative possessor can pied-pipe the possessee, which is predicted on this analysis. As Coon (2010: 368) notes, the entire object constituent may extract from the specifier position of AbsP, and indeed there is no ban on the movement of object constituents in Ch'ol (see example (12a) above).

Shifted numerals and adjectives may not extract
Further support for the current proposal comes from shifted objects with numeral and adjective modifiers. While objects with overt possessors always shift, numerals and adjectives may appear in VOS or VSO object position. It is therefore possible to test whether extraction of the numeral is grammatical when the object shifts. Extraction of a numeral from the object position is only possible in VOS word order, as shown by (42a). Extraction of a numeral from a shifted object, indicated by VSO word order, is ungrammatical, as in (42b).
(42) a. Cha'-kojty i ta' i-k'el-e [ t i wakax] aj-Rosa. two-clf pfv a3-see-tv cow nc-Rosa 'Rosa saw two cows.' b. *Cha'-kojty ta' i-k'el-e aj-Rosa [t i wakax]. two-clf pfv a3-see-tv nc-Rosa cow Intended: 'Rosa saw two cows.' Yet more evidence that object shift blocks extraction comes from the extraction of adjectives from object position. Under the current proposal, we would expect that extraction of an adjective from a VSO object would be ungrammatical. Indeed, adjectival extraction from the object is only possible from VOS objects in (43a), not VSO, as in (43b). 18 16 See also discussion on the Revised Extraction Constraint from Diesing (1992: 128). 17 Left branch extraction takes place after the vP fronts to Spec,TP. Under the predicate-fronting analysis adopted here, the vP moves with NP objects to Spec,TP. I argue that this movement does not induce an island effect. The NP is still within its licensing domain (i.e. within the domain of v, which Agrees with the object and assigns it case (Coon 2017)). I propose that left branch extraction of the interrogative possessor is not blocked in these cases, as the object has not moved away from its licenser, v. This builds on recent Agree-based accounts of extraction and unlocking, further discussed in Section 6. 18 Adjectival extraction is not possible from both ergative and absolutive subjects. When adjectives appear discontinuous from nouns in these positions, they are interpreted as secondary predicates or depictives, as shown in (i).
(i) a. *Pek' ta' i-k'el-e-yoñ wiñik. short pfv a3-see-tv-b1 While it is impossible to test whether unshifted objects with overt possessors allow extraction of an interrogative possessor, as they always shift, data from shifted objects with numerals and adjectives can provide support for the bleeding effects of object shift.

Moved absolutive subjects block left branch extraction
Absolutive subjects, like objects, in Ch'ol originate as complements to the verb. Unlike objects, absolutive subjects with overt possessors do not always move (see also Coon 2010: 362). Nevertheless, if they do move, indicated by the placement of adjuncts, extraction from the subject is ungrammatical. In (44), the subject has an overt possessor, and the PP adjunct may appear between the verb and subject, or after the verb and subject.
white pfv fly-iv dir nc-bird Intended: 'The white bird flew away.' Absolutive subject Speaker comment: A translation of this would be 'the bird flew away white'. However, speakers comment that this does not make sense.
See Vázquez Álvarez (2002a) and Martínez Cruz (2007) for further discussion on secondary predication, depictives, and adjectives in Ch'ol. At the moment, I do not know why adjectives may only extract from object position. I note this as an issue for future research, but highlight that the pattern in (43) further supports this paper's main claim that if the object shifts, the adjective may not extract. 19 Speakers have similar judgements for (44) and (45) if the PP is replaced with the temporal adverb ak'bi, as in (i) and (ii).

Shifted ditransitive objects block left branch extraction
Recall that extraction of interrogative possessors is ungrammatical from objects of ditransitives, but the extraction of numerals is grammatical for direct objects of ditransitive verbs. The argument for ditransitive objects proceeds in a similar fashion: when the object shifts, extraction is blocked from the object. I again use adjunct placement as a diagnostic for object shift.
While interrogative words must always front to the preverbal position, it is possible to test object shift with objects that have overt possessors. New data from direct objects with interrogative possessors reveals that they, too, undergo object shift, as in (46). In (46), the adverb ak'bi 'yesterday' comes between the verb and direct object ikaru ajRosa 'Rosa's car', indicating that the direct object has shifted. Following the line of argument laid out here, it is correctly predicted that extraction of interrogative possessors from objects of ditransitive verbs is ungrammatical: overt possessors trigger object shift with ditransitive objects as well, bleeding extraction from object position.

(46)
Ta' k-choñ-be-yety ak'bi [ i-karu aj-Rosa ]. pref a1-sell-appl-b2 yesterday a3-car nc-Rosa 'I sold you Rosa's car yesterday.' Numerals provide further evidence that object shift bleeds extraction: when an object with a numeral has shifted, the numeral cannot extract out. When an adverb intervenes between a ditransitive object and verb in (47a), extraction of the numeral is blocked. When the verb and object are adjacent, however, extraction may proceed as expected in (47b).

Ruling out demonstrative and non-possessor extraction
Demonstratives and possessors may never extract from any position. I suggest that demonstratives project in Ch'ol 20 and that there is a ban on the A-bar extraction of the demonstrative head (see discussion on movement in, e.g., Travis 1984). 21 Numerals, adjectives, and interrogative possessors, on the other hand, all form their own constituents and therefore have the possibility of extracting out. While this suggestion admittedly raises some questions about the nature of demonstratives and other determiner-like elements in Ch'ol, I leave these issues open for future research.
To rule out possessor extraction, I propose that this is due to the inability of (non-interrogative) possessors to move to the edge of DP. Under a phase-based theory of movement (Chomsky 2000;2008), in order to extract out of a phase, constituents must move to the edge of that phase. Adopting the proposal that DPs are phases (Chomsky 2008), a 20 See, e.g., Nicolae (2015) for an analysis of weak demonstratives in Romanian as heads. 21 The determiner jiñ(i) may appear discontinuous to the noun (Vázquez Álvarez 2011: 337), but as discussed by Vázquez Álvarez (2011) and Little & Wiegand (2018), it also acts as a focus marker. In Ch'ol and many Mayan languages, the preverbal position is the focus position. For the present paper, I follow Vázquez Álvarez (2011) and Little & Wiegand (2018) in assuming that jiñ(i), when discontinuous from nouns, is the spellout of focus features in the preverbal focus position, rather than an extracted constituent.
possessor must be able to reach the edge of the DP phase to extract out. 22 To illustrate, take the possessive construction in (48a) and Coon's (2010) analysis of (48a) in (48b). In (48b), the possessee nP fronts to a position between DP and PossP, which Coon (2010) labels as IP, paralleling vP fronting in the clausal domain. This generates the correct possessee-possessor word order.
Interrogative possessors always appear before their possessees, as in (49a). This involves additional movement of the interrogative possessor to Spec,DP, as shown in (49b). As the interrogative possessor is at the edge of the DP phase, it has the possibility to extract out.  (49), other features, such as focus, do not. It may be expected that if a possessor fronts within the DP, it gives rise to a focus interpretation, paralleling focus movement in the clausal domain. However, as shown by (50), even when prosodically stressed, a non-interrogative possessor may never occur before its possessee. In Ch'ol, the possessor in (50) is unable to move to the edge of the DP phase, evidenced by its inability to appear prenominally. As it cannot move to the edge of the DP, it cannot extract out. Interrogative possessors, on the other hand, do move to the edge of DP, driven by wh-features, as in (48), and therefore have the ability to extract out.
(50) *aj-Rosa i i-wakax t i nc-Rosa a3-cow Intended: 'Rosa's cow' The differences in wh-and focus-driven movement within the nominal domain have parallels in the verbal domain. While wh-features always trigger movement within the verbal and nominal domains, focus features do not: in situ focus is possible in Ch'ol (Clemens et al. 2017).

Nonperfectives
As previewed in Section 2.1, under Coon's (2013) analysis of split ergativity in Ch'ol, nonperfective forms are possessed nominals. The nonperfective aspect marker is a syntactic predicate that takes a nominalized clause as its argument. For example, the progressive marker woli in (51a) is analyzed as taking the DP i-k'ux waj aj-Rosa as its complement.
To derive VOS word order, the nP fronts to a position within the DP, as shown in (51b), paralleling vP fronting to Spec,TP in the clausal domain in (17). I propose that the objects in (52) undergo object shift in nonperfective aspect as well (see also Coon 2010: ex. (30)). Though nonperfective forms are not derived in Coon's (2010) analysis of word order, I propose that DP objects must shift to a position above nP, labeled again as AbsP. After the object shifts, the nP moves to Spec,IP, as in (53b). The object majki i-chich 'whose sister' is in its shifted position in Spec,AbsP, shown in (53b), so extraction of the interrogative possessor majki is blocked. Pied-piping of the whole DP is the only available option, shown in (53b) for (53a). 23 ]. prog a3-eat det tortilla nc-Juan(=encl) 'Juan is eating the tortilla.' For cases like (i), I suggest that the subject is in a right topic because it is marked with the enclitic =i, an intonational enclitic associated with topicalized constituents. Indeed, even with perfective verbs, VOS word order with full DP objects is possible when the subject has this enclitic, as in (ii).
a3-grind-tv det masa nc-Rosa=encl 'Rosa ground the masa.' A detailed looked into the prosody and distribution of the subjects in (i) and (ii) is needed to conclusively determine if they are right topics. Nevertheless, I note these potential instances of VOS word order with DP objects and suggest that the subjects are right topics, as has been argued before for similar data in Tsotsil (Aissen 2016; and other Mayan languages (Can Pixabaj 2004;Curiel 2007). The main claim remains the same: full DP objects shift in both perfective and nonperfective aspects, blocking left branch extraction.

Extraction from specifiers
Extraction from an ergative argument or indirect object is subject to variation. I suggest that this is due to a language-specific dispreference on extracting from specifiers in Ch'ol. Unlike absolutive subjects and objects, both ergative subjects and indirect objects are introduced in a specifier position: ergative subjects in Spec,VoiceP and indirect objects in Spec,ApplP (as per standard analyses of ditransitive objects). Under the current proposal, ergative subjects are not prima facie islands for extraction, given the Freezing Principle in (41). In fact, there is no evidence that subjects of transitive verbs move from their base position in Spec,VoiceP (Coon 2010;2013;Clemens & Coon 2018). Nevertheless, variability in judgements is unsurprising, given crosslinguistic facts on the variability of extraction out of subjects (i.e. Chomsky 2008;Polinsky et al. 2013, inter alia). Similarly, indirect objects are introduced in specifier position. It is difficult, or impossible, as in (37), for an extracted numeral to modify an indirect object. Indirect objects therefore provide additional evidence that extraction from specifiers is more difficult in Ch'ol than from complement position. Indeed, Müller (2018: 538) notes that a ban on extraction from specifiers can be a language-specific restriction. In the context of this paper, I conclude that extraction from specifiers is dispreferred and the leave the specific reasons for this variability as an open question for future work.

Another freezing effect
Above, I argued that shifted objects induce freezing effects, subsequently blocking left branch extraction. It may then be expected that movement operations other than object shift may induce freezing effects in Ch'ol. This is borne out in relative clause structures, as in (54). (54) *Majki i ta' i-k'el-e-yety [ i-chich t i ] k [ ta'=bä i-cha'l-e soñ t k tyi who pfv a3-see-tv a3-sister pfv=rel a3-do-tv dance prep k'iñijel ]? party Intended: 'Whose sister that danced at the party saw you?' Adopting a raising analysis of relative clauses (e.g. Bhatt 2002), the head majki ichich 'whose sister' in (54) originates in the relative clause 'that danced at the party' and then moves to the matrix clause. 24 It is thus not surprising, based on the analysis proposed here, that the extraction of the interrogative possessor in the matrix clause is not possible for any of the speakers consulted: the relative clause head has been moved and becomes an island for extraction. Freezing effects occur with other moved constituents in Ch'olfurther support for the current proposal.

Summary
Taking VSO word order and the distribution of adjuncts as indications of object shift, we see that the left branch extraction asymmetries follow naturally from the restrictions on VOS objects. When objects shift, they become frozen. For objects with overt interrogative possessors, object shift always occurs, which means that extraction from the object is always blocked. For objects with numerals, extraction of the numeral is only blocked if 24 A reviewer notes that for Bhatt (2002), the raised head in English is an NP not a DP. As detailed in Section 3, nominals with overt possessors pattern with DPs in Ch'ol. While a detailed look into the structure of relative clauses in Ch'ol is outside the scope of the present paper, I suggest that there may be variation across languages as to what raises. For example, Erlewine & Gould (2016) argue for a DP head-raising analysis of Japanese relative clauses. The point of (54) is simply that freezing effects seem to be exhibited elsewhere in the language. the object has moved. I suggested that demonstratives may not extract from any position due to a ban on the A'-movement of heads. I further argued that non-interrogative possessors may not extract out, as they cannot move close enough to the edge of the DP to do so. Nonperfective aspects exhibit the same left branch extraction asymmetries. I argued that object shift, too, occurs within nominalized complements of nonperfective aspects. I also suggested that extraction from specifiers is dispreferred in Ch'ol, explaining why extraction from subjects and indirect objects is more difficult or impossible. I ended this section with another instance of a freezing effect in Ch'ol.

Implications for theories of extraction
The proposal detailed in this article has implications for recent Agree-based approaches to extraction. Under those analyses, Agree, in the sense of Chomsky (2000;, with a nominal unlocks that nominal for extraction (i.e. Rackowski & Richards 2005;van Urk & Richards 2015;Branan 2018). Agreement morphemes in this context are taken to be the result of an Agree relation between a probe with unvalued φ features and a goal (an NP or DP) with valued φ features. This is modeled in (55). The probe on X in (55)  Agree A reviewer notes that the proposal laid out here may be in conflict with Rackowski & Richards's (2005) proposal of extraction. Using evidence from Tagalog, Rackowski & Richards (2005) propose that shifted nominals are indicative of Agree and only then is extraction licensed. In other words, for Rackowski & Richards (2005), object shift feeds extraction, whereas for the present claim, object shift bleeds extraction. Nevertheless, I argue that the proposal here is compatible with Rackowski & Richards (2005) if the ordering of object shift and Agree is taken into account. Though Rackowski & Richards (2005) concentrate on extraction patterns in Tagalog and not left branch extraction, I extend their proposal to left branch extraction. To illustrate with an example from Tagalog, take (56), where the possessor 'buffalo' has extracted out. For Rackowski & Richards (2005), 'the buffalo's horn' moves to Spec,vP indicative that v has Agreed with the object. 25 The verb Agrees with the whole nominal 'the buffalo's horn', indicated by the agreement affix in (the object voice marker) on the verb. Agree consequently unlocks the DP, and the possessor may then extract out. In (56), Agree takes place with shifted nominals to license extraction.

(56)
Tagalog (Kroeger 1991: 31) (my bolding added) Ang=kalabaw, p<in>utul ng=magsaka ang=sungay. nom=buffalo perf-cut-ov gen=farmer nom=horn. 'The buffalo, the farmer cut off the (i.e. its) horn.' In Ch'ol, on the other hand, Agree takes place with unshifted objects. Coon (2013; for Ch'ol and Preminger (2014) for other Mayan languages argue that an Agree relation is established between v and the object in the complement of V. Though these works do not explicitly discuss object shift and its relation to Agree, in the context of this paper, this means that Agree takes place before object shift. Shifting the object after Agree is what bleeds left branch extraction. 26 Put another way, as long as the Agree configuration is maintained, the nominal is unlocked and extraction is possible. For Ch'ol, this means that as long as the object remains within vP, where v can license it, left branch extraction is possible. This Agree-based proposal of extraction also can explain why objects within fronted vPs are not islands for extraction. As the object is still within the domain of its licenser, v, the Agree configuration is maintained, allowing extraction. Once the object moves from the vP via object shift, I suggest that the unlocking effects of Agree are reversed so left branch extraction is no longer permitted. On the surface, there seems to be a clash with Rackowski & Richards (2005). However, if the ordering of Agree and object shift is taken into account, the present proposal and that of Rackowski & Richards's (2005) are not at odds. For Ch'ol, Agree takes place before object shift. For Tagalog, object shift is indicative of Agree. I therefore propose that the ordering of Agree and object shift can parametrically vary across languages. Given this variation on Agree and movement operations in Tagalog and Ch'ol, it is predicted, and even expected, that there exist different patterns with respect to left branch extraction crosslinguistically. 27

Conclusion
This paper contributes new empirical data on left branch extraction in a less studied language. I argued that restrictions on VOS objects can capture asymmetries exhibited in left branch extraction. I adopted Coon's (2010) proposal for word order, in which VSO word order is derived from VOS word order, in part, via object shift. Overt possessors always trigger object shift, which is why interrogative possessors may never extract from object position. Numerals, on the other hand, do not obligatorily trigger object shift and may extract from unshifted objects, but not from shifted ones. This follows from the Freezing Principle (Ross 1974;Wexler & Culicover 1977), or a ban on extracting from a moved constituent.
I ended with some of the current proposal's implications for theories of extraction. While, at first blush, there seems to be a clash with Rackowski & Richards's (2005) analysis of Agree and object shift in Tagalog, I argue that the current proposal is not at odds with theirs. If the ordering of Agree and object shift is taken into account, variation with respect to extraction is expected. 26 As argued for by Coon (2017), absolutive subjects and objects have different licensers. Absolutive objects are licensed by v, while absolutive subjects are licensed by INFL/T. For the present paper, both v and INFL/T license the absolutive argument before object shift, or movement of the absolutive subject. Adopting phase theory (Chomsky 2000;2008), CPs and transitive vPs are phases. Intransitive vPs are not phases and for this reason, INFL/T probes into the complement of the verb to license the absolutive subject, in turn licensing extraction. 27 For more on the relative order of operations as a point of cross-linguistic variation see Obata et al. (2015).
I thank a reviewer for pointing this reference out to me.