Nominalizations and the structure of progressives in Chuj Mayan

This paper investigates the structure of progressives and nominalizations in Chuj, an understudied Mayan language of Guatemala. Like many other Mayan languages, Chuj shows aspect-based split ergativity: the otherwise ergative head-marking pattern in the language disappears in the progressive aspect. In other Mayan languages—for example Ch’ol (Coon 2010; 2013) and Yucatec (Bricker 1981)—the appearance of a non-ergative pattern in the progressive has been attributed to nominalization. In Chuj, however, there is no clear morphological reflex of nominalization, as is found in other languages in the family. Using data from negation, particle placement, and agreement, we argue that Chuj progressives nonetheless involve an aspectual matrix predicate and a nominalized embedded verb. This provides a clear structural explanation for the split pattern. Finally, we distinguish different types of nominalizations in Chuj: those which are nominalized directly from a root, and those which are nominalized above verbal projections.


Introduction
In this paper we investigate the structure of progressive clauses in Chuj, an understudied language from the Q'anjob'alan branch of the Mayan family, spoken in the department of Huehuetenango in Guatemala. We provide evidence that progressive clauses involve an aspectual predicate and an embedded nominalized clause, and that this nominalization is the source of split ergativity in the language. Following previous work on other Mayan languages (Larsen & Norman 1979;Bricker 1981;Mateo Pedro 2009;Coon 2010), as well as cross-linguistically (Laka 2006;to appear;Coon 2013), split ergativity in Chuj may thus be seen as structurally based. 1 This analysis contrasts with approaches in which splits are taken to involve different case/agreement features relativized to particular aspectual values (see e.g. Anand & Nevins 2006;Ura 2006;AnderBois & Armstrong 2016), as well as with proposals in which the split is taken to be purely morphological (e.g. Legate 2014 for nominal splits).
This result is important because the nominal nature of Chuj progressive markers is not at all obvious from their surface form. This contrasts with Mayan languages like Ch'ol, in which there is morphological evidence of nominalization in progressive stem forms (Coon 2013). In fact, while Chuj does have stem forms that are very clearly nominalized (-el forms, described below), these stems are impossible in progressive constructions. We provide evidence first that there is nevertheless a structural difference between progressive and non-progressive constructions, and second, that nominalizations in Chuj may be small (formed directly from a root, as in the -el forms) or large (nominalized above the projection of verbal material). This connects to existing literature arguing for different sizes of nominalizations (see e.g. Abney 1987;Grimshaw 1990;Harley & Noyer 1998;Borsley & Kornfilt 2000;Alexiadou 2001;Embick 2010). Only the latter type-the larger nominalizations containing verbal material-may appear in the progressive construction; we connect this to their ability to project arguments (as in Grimshaw's 1990 Event Nominals).
The remainder of this paper is organized as follows: section 2 provides background on split ergativity in the Mayan family, along with details of Chuj's split pattern. As little descriptive material exists in Chuj-and virtually none in English-section 3 gives a brief overview of the major characteristics of the language. The core evidence for our proposal is presented in sections 4 and 5. In section 4 we argue that the progressive aspect markers are predicates, and in section 5 we demonstrate that the verbs which they embed are nominalized. In section 6 we present a formal account of the progressive nominalizations, comparing them with other nominals in the language and with nominalizations crosslinguistically. We conclude in section 7.

Aspect-based split ergativity in Chuj
Like many other Mayan languages, grammatical relations in Chuj are head-marked on the predicate via two sets of person markers (e.g. Larsen & Norman 1979;Grinevald & Peake 2012;Coon 2016a;Aissen to appear). In Chuj, the head-marking pattern shows an aspectual split (see e.g. Buenrostro 2013). Clauses in non-progressive aspects-along with aspectless nonverbal predicates, discussed below-exhibit an ergative-absolutive alignment. Transitive subjects are co-indexed with an ergative prefix, known as "Set A" in Mayan linguistics. In (1), this is the 1st person plural ko-. Transitive objects and intransitive subjects are cross-referenced with a set of absolutive clitics, such as -ach in (1), called "Set B" (conventions for using "=" vs. hyphen for parsing out Set B are described below). 2 (1) a. Ix-ach-ko-chel-a'. pfv-b2-a1p-hug-tv 'We hugged you.' b. Ix-ach-b'ey-i. pfv-b2-walk-itv 'You walked.' In the progressive aspect, as in (2), we find the split: both transitive and intransitive subjects are now cross-referenced with the Set A prefix series. Set B is impossible on progressive intransitives, as in (3).

(6)
Ch'ol perfective a. Tyi i-jats'-ä-yoñ. pfv a3-hit-tv-b1 'She hit me.' b. Tyi majl-i-yoñ. pfv go-itv-b1 'I went.' prog a3-go-nml 'She's going. ' We argue that this type of proposal is also correct for Chuj progressives; the structure we propose below is represented in (8). The proposal that the progressive aspect marker is a predicate which embeds the contentful lexical material in some type of complement is not new (see the works cited above), and is discussed explicitly for Chuj in Buenrostro (2007;. Specifically, Buenrostro proposes that lan (and its variants, discussed below) is a predicate derived from "a positional root which takes as its argument the clause containing the principal meaning" (Buenrostro 2013: 123). As discussed and observed below, other apparently nonfinite complement clauses in Chuj also show this type of split in person marking. The questions that remain are (i) what is the status of the complement clause? and (ii) what triggers the split in person marking? In this paper we argue that the embedded complement clause is a nominalization, and that the split in person marking is directly connected to the nominalized status of the embedded predicate. Crucially, across most of the Mayan family, Set A markers cross-reference not only transitive subjects, but also possessors. Compare the 1st person plural Set A markers in the progressives in (8) with the possessive from in (9). Under a nominalization analysis, the notional subjects of progressive forms like those in (8) are grammatical possessors, as indicated by the suggested literal translations above.
(9) ko-nun a1p-mother 'our mother' In Ch'ol, we find clear initial support for the nominalization hypothesis in the morphological form of the complement clause predicates, shown in (10) below. While an intransitive verbal stem appears with the verbal "status suffix" -i (discussed in section 3), as in (10a), intransitives in the progressive aspect appear instead with the suffix -el, as in (10b). Suffixes of the form -Vl appear on nominals throughout Ch'ol and other Mayan languages (see e.g. Bricker 1981), discussed in more detail in Coon (2010; and in section 6 below. (10) Ch'ol intransitives a. Tyi wäy-i-yety. pfv sleep-itv-b2 'You slept.' b. Choñkol [ a-wäy-el ]. prog a2-sleep-nml 'You're sleeping.' Unlike in Ch'ol, Chuj shows no morphological difference between stems appearing in perfective and progressive aspects, aside from the change in person marking. Compare the Ch'ol intransitives above with the Chuj intransitives in (11).
a2-itv 'You're sleeping.' Buenrostro (2007: 255) explicitly rejects a nominalization analysis of embedded clauses, suggesting instead that these embedded clauses are verbal. Though she does not formalize the nature of the alignment split, one could stipulate that functional projections in matrix versus embedded intransitive clauses have different case/agreement features, but are structurally and categorically identical. Below we argue that despite the morphological similarities, the stem forms in (11a) and (11b) are nonetheless structurally and categorically different from one another: the former is a verbal stem, while the latter is a nominalization. This analysis is not only empirically supported by the Chuj facts, but also allows for a non-stipulative account of the "extended ergative" person marking system in embedded forms. First, we turn to general background on Chuj.

Chuj background
As in other Mayan languages, predicates in Chuj can be divided into two types: verbal predicates and nonverbal predicates (see Grinevald & Peake 2012;Coon 2016a). The former generally denote events and require a TAM marker, as in examples seen so far above. Nonverbal predicates, like those in (12), generally denote states and never appear with a TAM marker; adjectival and nominal predicates fall into the latter type and do not appear with an overt copula, as shown by the examples in (12a)-(12b). There exist several transitive nonverbal predicates, like 'know' in (12c); as in verbal transitives, the subject is marked Set A and the object is marked Set B. In these examples, temporal distinctions may be made through the addition of adverbial material, or inferred from context. TAM-less nonverbal predicate constructions will be relevant to the discussion of the progressive below. A template for a Chuj transitive verbal predicate is given in (13); these components are discussed briefly below. For general descriptions of Chuj, see Hopkins (1967);Maxwell (1976) ;Domingo Pascual (2007); Buenrostro (2013).
As in other Mayan languages, core nominal arguments are cross-referenced by Set A and Set B markers on the predicate. In discourse-neutral contexts nominal arguments appear post-verbally, but may also appear in preverbal topic/focus positions (see England 1991;Aissen 1992 on Mayan generally, and Bielig 2015 on Chuj). Both VOS and VSO orders are possible, and further work is needed to determine the factors governing this distribution in Chuj (see England 1991 andClemens & on postverbal argument order in Mayan). As described for other Q'anjob'alan languages (e.g. Craig 1986 for Popti' and Zavala 2000 on Akatek), Chuj possesses a series of nominal classifiers which appear either preceding nominals in referential contexts (14a), or alone as pronouns (14b) (Buenrostro et al. 1989;Hopkins 2012b The initial h-of these morphemes is not pronounced as such, but is an orthographic convention used to indicate the absence of an epenthesized word-initial glottal stop (initial glottal stop is inserted before other vowel-initial forms in Chuj, discussed for example  (16), and discussed in more detail in Buenrostro (2007) and Carolan (2015). The past perfective marker ix may be dropped, as in Q'anjob'al (Mateo Toledo 2011); in Chuj, the alternation between ix and Ø appears to indicate a difference in recent versus more distant past, discussed in Carolan 2015. 3 We assume following Aissen (1992) and other work on Mayan languages that the aspectual particles tz, ix/Ø, and ol occupy finite Infl 0 ; the status of lan is discussed in detail below. c. Ol-ach-in-chel-a'. prosp-b2-a1-hug-tv 'I will hug you.' Stem-final status suffixes are found throughout Mayan languages, and vary with transitivity and clause type. In both imperfective and perfective stem forms, intransitives appear with the status suffix -i and transitives appear with the transitive suffix -V', here -a'. 4 These status suffixes are typically dropped when not in phrase final position, as in many other Mayan languages (see e.g. Henderson 2012). In the prospective forms in (17c) and (18c), the pattern is similar except that intransitives appear with the irrealis suffix -ok, discussed further below. Contrasts between imperfective, perfective, and prospective constructions in (17) and (18) on the one hand, and progressives in (19) and (20), on the other, will be the focus of the remaining sections.

(19)
Lan ha-b'ey-i. prog a2-walk-itv 'You're walking.' Lan hach=in-chel-an-i. prog b2=a1-hug-sub-itv 'I'm hugging you.' Note that unlike in the perfective, imperfective, and prospective aspects, the stem in the progressive is written orthographically as a separate word from the aspect marker (see e.g. Buenrostro 2004;Domingo Pascual 2007), an intuition shared by speakers we have consulted. As previewed in section 2, intransitives appear with Set A marking crossreferencing their subjects, rather than the Set B marking found on intransitives elsewhere in the language. Finally, note that in the transitive form the verb stem appears with a suffix, -an, glossed following Buenrostro (2004) as sub for "subordinate clause", and then the intransitive suffix -i. We offer an account for this morphology in section 6 below.
A note on the status and placement of the Set A (ergative, possessive) and Set B (absolutive) morphemes is in order, and will be relevant to the discussion below. As noted above, Set A morphemes are prefixal; following Coon (to appear) we assume that Set A agreement is the spell-out of a low, in situ agreement relationship between the head that introduces the external argument (i.e. Voice 0 or possessessive n 0 ) and the external argument itself (see also Wiltschko 2006 on ergative agreement in Halkomelem Salish). We propose, following Coon, Mateo Pedro & Preminger (2014) on Q'anjob'al, that Set B morphemes are the result of clitic doubling which arises due to an Agree relationship between the head responsible for licensing the absolutive argument, and the argument itself ("the associate" of the clitic double; see e.g. Anagnostopoulou 2003;Harizanov 2014). In verbal predicates like those in (17)-(18), this head is the TAM marker, located in Infl 0 (discussed further in section 6 below). Infl 0 licenses the absolutive argument and triggers cliticization of the Set B morpheme, which attaches to the TAM marker. The structure of a verbal predicate, like the transitive in (18b) above, repeated in (21), is shown in (22).

Coon & Carolan
A note on the status and placement of the Set A (ergative, possessive) and Set B (absolutive) morphemes is in order, and will be relevant to the discussion below. As noted above, Set A morphemes are prefixal; following Coon (to appear) we assume that Set A agreement is the spell-out of a low, in situ agreement relationship between the head that introduces the external argument (i.e. Voice 0 or possessessive n 0 ) and the external argument itself (see also Wiltschko 2006 on ergative agreement in Halkomelem Salish). We propose, following Coon, Mateo Pedro & Preminger (2014) on Q'anjob'al, that Set B morphemes are the result of clitic doubling which arises due to an Agree relationship between the head responsible for licensing the absolutive argument, and the argument itself ("the associate" of the clitic double; see e.g. Anagnostopoulou 2003;Harizanov 2014). In verbal predicates like those in (17)-(18), this head is the TAM marker, located in Infl 0 (discussed further in section 6 below). Infl 0 licenses the absolutive argument and triggers cliticization of the Set B morpheme, which attaches to the TAM marker. The structure of a verbal predicate, like the transitive in (18b) above, repeated in (21), is shown in (22).
PFV-B2-A1-hug-TV 'I hugged you.' The verb root undergoes head movement through Voice 0 to the status suffix, forming the verb stem. We represent v 0 /Voice 0 as a single head, though nothing critical hinges on whether there is a null v 0 or the two are "bundled" (see Harley to appear). Following Clemens & Coon (2016), we label the head hosting the status The verb root undergoes head movement through Voice 0 to the status suffix, forming the verb stem. We represent v 0 /Voice 0 as a single head, though nothing critical hinges on whether there is a null v 0 or the two are "bundled" (see Harley to appear). Following Clemens & Coon (2016), we label the head hosting the status suffix simply ss 0 ; what is important here is that it is a head at the edge of the verbal projection. Set A agreement is the result of a local relationship between the Voice 0 head and the in situ external argument, spelled out as a prefix on the stem: in-chel-a' (see Coon to appear for Ch'ol). The TAM head enters into an Agree relationship with the object, triggering the Set B clitic double on the TAM particle ix. In non-progressive aspects, we follow others in writing the complex as a single phonological word and do not represent clitic boundaries internal to these stems. This is consistent with the prosodic behavior of Set B morphemes and is supported by speaker intuitions as well as preliminary phonetic data (Cora Lesure, p.c.). We assume following previous work that Set B markers in Chuj are syntactic clitics, but that prosodic effects may affect their phonological realization. 5 We offer a detailed proposal for the structure of progressives, including the attachment of Set A and Set B morphemes, below. First, in section 4 we show that the progressive lan behaves as a nonverbal predicate, while the other aspect markers ix/Ø, tz, and ol do not. As noted above, this part of our analysis accords with other work on Chuj, for example discussion in Buenrostro (2007;. Next, in section 5 we provide evidence for our more controversial proposal that the forms embedded under the progressive marker are nominalizations.

Progressives as predicates
This section investigates the behavior of the progressive aspect marker lan in Chuj. We show that it behaves like other nonverbal predicates in Chuj with respect to negation (section 4.1) and particle placement (section 4.2). Finally, in section 4.3, we compare lan to other embedding verbs in the language and discuss a likely origin for the marker lan as originating from a positional predicate; see also Pascual (2007) on Q'anjob'al.

Negation
In non-progressive aspects, negation in Chuj is expressed by a preverbal particle, here man, and a particle laj which follows the verb stem, as shown by the transitive and intransitive prospective sentences in (23). Note that -ok-which normally appears on prospective intransitives, even in non-final position-is absent from the negated intransitive in (23b); compare (17c) above.
(23) a. Man ol-ach-in-chel laj. neg neg 'I will not hug you.' b. Man ol-in-b'ey laj. neg prosp-b1-walk neg 'I will not walk.' In the imperfective aspect the negative marker man and the imperfective morpheme tz combine to form max, and in the perfective aspect man and the perfective ix combine to form maj, shown in (24). Despite this difference, laj consistently follows the verb stem in these forms, as in (23).
honh=y-il laj winh ewi. neg.pfv b1p=a3-see neg clf.masc yesterday 'He didn't see us yesterday.' 5 Buenrostro (2013) represents both TAM and Set B morphemes as phonological clitics; for example, a stem like (21) is written as ix=ach=in-chel-a'. Based on preliminary phonetic data (Cora Lesure, p.c.), this appears to be borne out in the progressive aspect. However in non-progressive aspects, given the general inability of certain prosodically-governed material to appear within the stem (discussed in section 4.2), we provisionally adopt the proposal in Bennett, Harizanov & Henderson (2015) for Kaqchikel in which a prosodic "smothering" operation turns the stem into a single inseparable word. The phonology of the progressive aspect is discussed further in section 4.
Negation in the progressive aspect follows a different pattern, shown in (25). First, the progressive morpheme lan appears with the irrealis marker -ok, seen in (25) below. Second, note the difference in placement of the particle laj: in (23) laj follows the verb stem, while in (25) it follows the progressive marker.
(25) a. Man lan-ok laj hach=in-chel-an-i. neg prog-irr neg b2=a1-hug-sub-itv 'I'm not hugging you.' b. Man lan-ok laj hin-b'ey-i. neg prog-irr neg a1-walk-itv 'I'm not walking.' Compare the lan forms in (25) with the negated stative predicate k'ayb 'um 'teacher' in (26). The progressive morpheme lan patterns with other nonverbal predicates in taking the suffix -ok; the Pred-ok stem in both is followed by the particle laj.

(26)
Man hin=k'ayb'um-ok laj. neg b1=teacher-irr neg 'I'm not a teacher.' As two anonymous reviewers point out, the distribution of irrealis -ok and negative laj not, in and of lead to the conclusion that lan is a predicate. Under one possible analysis, laj is an enclitic that attaches to the first prosodic word in a certain domain, regardless of its lexical category; since the progressive lan is an independent word, we would predict the difference between (23)/(24) and (25) on purely phonological grounds. Indeed, one reviewer notes that the distribution of the cognate particle in K'ichee', ta(j), discussed in Henderson (2012), is prosodically governed and is not restricted to attaching to predicates. Similarly, -ok appears on elements of various categories-e.g. numerals, directionals-in irrealis contexts. While further work is needed to fully understand the nature of both -ok and laj in Chuj, their distribution here both reinforces the claims that (i) progressive lan behaves differently from the other aspect markers, and (ii) that it patterns with other nonverbal predicates in Chuj. Craig (1977: 93) reports similar facts for related Popti' (formerly Jakaltek), also from the Q'anjob'alan branch. A Popti' negated progressive is shown in (27a), and can be compared with the negated intransitive stative predicate in (27b). Craig notes that "[t]he fact that lanhan is a higher predicate and a stative verb is indicated by the negative construction." 6 (27) Popti' (Craig 1977: 94) a. Mat lanhan-oj ha-wayi. neg prog-irr a2-sleep 'You are not sleeping.' b. Mat sonlom-oj hach. neg marimba.player-irr b2 'You are not a marimba player.' progressives in Chuj Mayan Art. 22, page 11 of 35

Particles
The particles =xo 'already' and =to 'still' provide additional evidence for a structural difference between the progressive and non-progressive markers. Specifically, just as with negation above, the progressive morpheme behaves as a nonverbal predicate with respect to the distribution of =to and =xo. Examples (28) and (29) below show the contrast between verbal and nonverbal predicates. In NVPs like (28), =to and =xo attach directly to the predicate.

(28)
Nonverbal predicates a. Tzalajnak=to ix unin. happy=still clf.fem child 'The girl is still happy.' b. K'ayb'um=in=xo. teacher=b1=already 'I am already a teacher.' In the verbal predicates in (29), =to and =xo appear sentence-initially attached to an added morpheme to. The morpheme to is also a complementizer used to embed finite clauses, and we suggest that it is inserted here in order to host the clitic. 7 Specifically, one might analyze forms like (29) along the lines of English do-support. 8 Compare the behavior of (North American) English negative =n't, which may attach to auxiliary verbs (30a)-(30b) and copular be (30c), but requires do-insertion in other contexts (30d)-(30e): (30) a. Mary isn't going b. Mary hasn't gone c. Mary isn't a teacher d. Mary doesn't have the book e. *Mary hasn't the book.
Assume that like English =n't, Chuj =to and =xo must cliticize to a certain head position in the derivation. In some cases, this position may be filled by movement, in which 7 Buenrostro (2013: 121) reports variation in the placement of =to and =xo in some verbal predicates. In addition to forms with an initial to, like those in (29), she provides forms like (i), in which the clitic follows the prospective ol.
(i) Ol=to=in-mujlaj-ok. prosp=still=b1-work-irr 'I will still work.' More work is needed to determine if this is a more general point of variation. It is worth noting that the only forms like (i) provided by Buenrostro involve the prospective ol (not perfective ix or perfective tz) the prospective shows independent differences, such as the appearance of the irrealis -ok in intransitives (see (17) above). We leave this as a topic for future work, noting that under the prosodic smothering account discussed in footnote 5, this type of variation may be attributed to features of the aspect marker. 8 We thank an anonymous reviewer for fuller discussion of this idea. case the particle attaches to the moved head: the NVP in (28). In other cases, movement is blocked and the complementizer to is inserted, as in (29).
We now turn to progressives. Again in parallel with the behavior of nonverbal predicates, to is not inserted in the progressive aspect. Instead, the particles =xo and =to attach directly to the progressive marker, as shown in (31) and (32).
Note that unlike with the negative particle laj in section 4.1 above, it is not obvious how a prosodic analysis would capture the fact that =to/=xo attach to the progressive lan in (31) and (32), but require the insertion of a C 0 head in the case of verbal predicates in (29). However, a consistent generalization emerges under the proposal that the progressive lan simply is a nonverbal predicate. NVPs-perhaps due to an absence of added functional structure found on verbal predicates-may move to the position to which =to and =xo attach (by assumption, C 0 ). We suggest that the added functional structure (i.e. status suffixes the aspectual) found on verbal predicates blocks this movement and the dummy element to is inserted in order to host the clitics.
Setting aside the details of the verbal and nonverbal predicates, the important point here is again that the progressive marker lan behaves exactly like other nonverbal predicates in its ability to combine directly with =to and=xo. As with the facts in the previous section, this difference in behavior does not in and of itself point to a categorical difference between lan and other aspect markers. However, the fact that lan not only behaves differently from the other aspectual morphemes, but also consistently behaves like nonverbal predicates, lends support to this analysis. Adopting the proposal of Buenrostro (2007;, we propose that the progressive morpheme simply is a nonverbal predicate, on par with other NVPs, shown in (33).
In section 5 we argue that the structural similarity extends to the complements in (33), in brackets, which we propose are both nominal. First we turn to the origin of the progressive marker. progressives in Chuj Mayan Art. 22, page 13 of 35

Progressives and positional predicates
Finally, we compare the progressive marker lan with other embedding verbs. Note that the stem forms embedded under the progressive aspect marker in (34a) and (35a) are identical to stem forms which appear under elements that are clearly matrix predicates, like yamoch 'begin' in (34b) and (35b).
But what kind of a predicate is lan? First note that lan is the most common progressive marker we have encountered in our work on Chuj, Domingo Pascual (2007: 155) also lists wan, Hopkins (2012a) has both wan and wal, Buenrostro (2013: 149) gives lan, wan, and lanhan, and Buenrostro (2004)  c. Leman=to y-ak'-an lesal winh. prog=still a3-give-sub pray clf.masc 'He is still praying.' While we have yet to determine whether there is any semantic difference among these markers, we suggest that the fact that there are several is consistent with the progresive being expressed as a lexical predicate-not as a functional aspectual particle (in contrast with tz, ix, and ol). 9 Buenrostro (2013: 123) proposes that the progressive comes from a positional root (see also Pascual 2007 on Q'anjob'al). Positionals form a distinct class of roots in Mayan languages, identifiable by their semantics (usually referring to physical features or spatial configuration), as well as by the distinct morphology required to form stem forms (see e.g. Haviland 1994;Henderson 2016). In Chuj, positional roots form stative predicates with the addition of the suffix -an (Hopkins 1967). The progressive lan may be a reduced version of the positional lanh; lanh can appear with the positional suffix -an, forming a progressives in Chuj Mayan Art. 22, page 14 of 35 stative positional predicate meaning something like 'extended (over some space)', as in (37a). Compare this with the positional root linh 'standing' in (37b).

(37)
Chuj (Domingo Pascual 2007: 190) a. Lanh-an ek' kamix sat lu'um. extended-pos dir shirt on ground 'The shirt is lying (extended, carelessly) on the ground.' b. Linh-an ek' nok' chej. standing-dir clf.animal horse 'The horse is standing.' Cross-linguistically, it is unsurprising to find a locative basis for progressive expressions. Just as the shirt in (37a) is extended over space, the event in the progressive is extended over time; see Bybee, Perkins & Pagliuca (1994) for cross-linguistic similarities between locative and progressive constructions, and discussion for Mayan in Coon (2013).
Just as the predicate lanhan in (37a) combines with a nominal complement-kamix 'shirt'-we argue that the reduced progressive form lan is also a nonverbal predicate, and that the stem form it embeds is a nominalized verb. In the following section, we review evidence in favor of treating the complement of lan as a nominalization.

Complements as nominalizations
In this section we provide evidence in favor of treating the stem form embedded under the progressive as a nominalization. Like other nominals, these stems may: serve as subjects (section 5.1) and trigger agreement (section 5.2). In section 6 we provide further details on the structure of these forms and the typology of nominalization. While progressive stems differ from other nominal forms in Chuj with respect to certain diagnostics (e.g. they may not appear with demonstratives), we show that these differences are expected from a cross-linguistic point of view.
Recall that while the predicative nature of lan has been proposed in previous work on Chuj (see also Mateo Toledo 2003; 2013 on Q'anjob'al), the proposal that its complement is nominal is new. The nominalization analysis advocated for here provides an explanation for the split in person marking in embedded clauses: Set A is found in both transitive and intransitive subjects because these are grammatical possessors (see e.g. Larsen & Norman 1979). This work not only contributes to our understanding of the typology of nominalization, but also provides a natural account of the person split in Chuj progressives and embedded clauses more generally.

As arguments
The progressive stem forms we are examining are shown again in brackets in (38) and (39). In (38) we observe the main facts that need to be explained: (i) the intransitive stem in (38a) appears with Set A (ergative/possessive) marking co-indexing the thematic subject, and (ii) the transitive stem in (38b) appears with a suffix -an, followed by what appears to be the intransitive suffix -i. Recall that the suffix -i drops when it is not in phrase final position, as shown by the forms in (39)  Like other nominals, the bracketed stem forms in (38) and (39) can appear as subjects. The nonverbal predicate in (40) has a regular possessed noun as its subject. Recall that as in other Mayan languages, there is no overt equative copula in Chuj.
Man te wach'-ok laj [ NP ko-kape ]. neg very good-irr neg a1-coffee 'Our coffee isn't very good.' Progressive stem forms may also appear as subjects, as in (41); compare the progressive stems in (39)  As with nominalizations in English, not all nominalizations are appropriate as subjects to all predicates (see section 6.3 below), but with the right context, these are acceptable. Another set of examples is discussed by Buenrostro (2004: 256), shown in (44). Just as a possessed noun can appear as the subject of an aspectual predicate like lajw in (43)

Triggering agreement
In the forms above, the bracketed nominalization serves as the intransitive subject of either a nonverbal predicate (41) Under the analysis presented here, the intransitive predicate lajwi 'finish' in (44) "shows" null, whether it is a regular nominal, as in (44a), or a nominalized clause as in (44b). Analogously, if the progressive aspect marker lan is the predicate, and its complement is a nominal argument, we do not expect to find any overt reflex of this relationship. Compare the progressive form in (46a) with the nonverbal predicate in (46b).
clf.masc man 'The man is happy.' While the absence of morphology in the forms above is consistent with our account, it is hard to draw conclusions from missing morphology. If these nominalizations are true arguments, we expect them to also govern the appearance of overt 3rd person Set A morphology. This prediction is borne out. Compare the more complex perfective and progressive forms in (47) below. The appearance of 3rd person Set A morphology in the example in (47b) may initially come as a surprise: it is absent from the perfective form in (47a), and note that the thematic subject is 1st person plural.
a. Ix-numx-i ko-mixnaj-i. pfv-stop-itv a1p-bathe-itv 'We stopped bathing.' b. Lan s-numx-i ko-mixnaj-i. prog a3-stop-itv a1p-bathe-itv 'We're stopping bathing.' Our analysis of these two sentences is shown in (48). 10 We represent a null morpheme in (45c) and in forms below for clarity, but this is not intended to represent an analysis in favor of a null morpheme as opposed to the absence of any morpheme at all. Rather, we include it to demonstrate that the absence of an overt morpheme is expected for 3rd person absolutive arguments. progressives in Chuj Mayan In (47a)/(48a), the intransitive matrix verb numxi takes the possessed nominal komixnaji 'our bathing' as its single absolutive argument. Since (like all other nominalized verbs) the argument is 3rd person singular, we see no overt agreement morphology. The progressive in (47)/(48) is more complex. As usual under this analysis, lan-unlike ix-must take a nominal complement. Here, it is a complex possessive construction snumxi komixnaji, literally, 'our bathing's stopping'. We propose that komixnaji 'our bathing' is the grammatical possessor of the nominal numxi 'stopping'. Note that the nominal form of 'stop' in (48b) is homophonous with the verbal form in (48a); we return to this below.
Like other possessors, komixnaji follows the possessum and triggers Set A marking on it: here 3rd person s-. Compare the complex possessive construction in (48b), repeated in (49a), with the uncontroversial complex possessive form in (49b).
[ NP ko-nun ] i ] a3-house a1p-mother 'Our mother's house.' To summarize, while there is no overt morphological evidence for the nominalization of stem forms embedded under the progressive aspect morpheme (i.e. no dedicated nominalizing morpheme), we argue that the distributional facts here lend support to an analysis of these forms as nominal. In particular, our account provides a natural explanation for the appearance of 3rd person Set A agreement on complex progressive constructions like (47b).
Nonetheless, the progressive forms do not pass all nominal tests in the language. Furthermore, as one reviewer points out, non-nominal clauses might also be expected to serve as subjects and to trigger 3rd person agreement. In section 6 we propose a structure of the progressive stem forms as nominalizations that include (possibly complex) internal verbal structure. Following much previous work on nominalization, we suggest that differences in behavior between the progressive stem forms and other derived nominals fall out from differences in the size of the nominalizations. We argue that the appearance of Set A (possessive) person marking on subjects is best understood under this account.

The structure of nominalizations
In this section we provide an analysis of the internal structure of progressive stem forms. In section 6.1 we review a concern raised by Buenrostro (2007) about transitive nominalizations, and suggest based on previous work on nominalizations that this is not a problem for Chuj. Next, in section 6.2, we discuss formal details of our account of the nominalization of progressive stems. We argue that progressive stem forms are nominalized above a verbal projection and that the Set A agreement cross-references the grammatical possessor. The possessor controls a phonetically null argument in the subject's base thematic position-that is, in progressive stem forms, arguments are projected internal to the nominalization. Finally, we compare the embedded progressive forms with other smaller nominal forms in 6.3.

Transitives and intransitives
In this section, we present a specific proposal for the structure of the nominalized intransitive and transitive forms. First, it is worth noting that Buenrostro (2007) has discussed these constructions, and concluded that the forms embedded under lan are not nominal: "One of the most frequent explanations for [the progressive] complement clause consists in saying that these are nominalized verbs. The explanation is based in the idea that the ergative marker of the intransitive verb is interpreted as possessive. However, when we see [transitive examples] this hypothesis is not sustainable, since the transitive verb stem has both of its person markers" (Buenrostro 2007: 255).
Chuj perfective and progressive forms are presented in (50) and (51) for comparison. Note that Buenrostro's concern is with forms like the embedded transitive in (51a), which appears with a Set A marker co-indexing the subject, and morpheme co-indexing the object.
(50) a. Ix-ach-ko-chel-a'. pfv-b2-a1p-hug-tv 'We hugged you.' b. Ix-onh-b'ey-i. pfv-b1p-walk-itv 'We walked.' The assumption underlying Buenrostro's concern is that nominalizations must be intransitive. Indeed, based on comparison with many other Mayan languages, this is a reasonable expectation. In languages like Mam (England 2013) and Q'eqchi' (Berinstein 1985), nonfinite embedded forms must first be detransitivized-i.e. passivized or antipassivized-in order to undergo embedding. In the Q'eqchi' form in (52a), for example, the embedded verb is an incorporation antipassive: an antipassive morpheme appears on the transitive root, and the object must be bare and nonreferential (see Massam 2001 on pseudo noun incorporation). In the nominalized form in (52b), only the notional object is expressed on the embedded verb, cross-referenced with Set A marking (see also Imanishi 2014 for this type of construction in Kaqchikel).

(52)
Q'eqchi' (Berinstein 1985: (2014) offer a formal account for why embedded transitives are impossible in some, but not all, Mayan languages, which we review briefly in section 6.2 below. Relevant here is first, that there is no a priori problem with a nominalization which appears with a thematic object/patient (i.e. is transitive), and second, that the presence or absence of an internal argument may give us clues to the internal structure of the nominalization. Grimshaw (1990), for example, discusses two types of nominalizations: Result Nominals (RNs) and Event Nominals (ENs; also known as Process Nominals). The former may denote a variety of things related to the verb, and have no internal argument structure. The latter have some verbal properties: they denote complex events and obligatorily project arguments. We will argue that the progressive stems in Chuj belong to the latter type.
Grimshaw points out that many nominals in English are ambiguous between Result Nominals and Event Nominals, and that this ambiguity gives rise to the appearance of optionality of the complement (i.e. the thematic object, the patient in (53)). For example, the English verb examine in (53a) requires a complement, while the nominal examination in (53b) apparently does not.
However, as Grimshaw points out, this optionality disappears when the form is disambiguated. For the ambiguous nominal expression in (54), the modifier frequent forces a complex Event Nominal reading. While a complement need not occur under a Result Nominal reading, omitting the complement when the EN reading is forced is ungrammatical, as shown by the contrast between (54a) and (54b).
(54) English nominals (Grimshaw 1990: 50) a. *The frequent expression is desirable. b. The frequent expression of one's feelings is desirable.
The distinction between simple RNs and complex ENs also has an effect on the interpretation of possessors. In (55a), Grimshaw (1990: 48) notes that the possessor, John, may be understood as the "possessor, author, or taker of the exam". However, in (55b)-when the noun behaves as a theta-role-assigning EN-the possessor must be interpreted as a thematic agent.
(55) English possessors (Grimshaw 1990: 48) a. John's examination was long. b. John's examination of the patients took a long time. We argue that Chuj transitives like (51a) above, repeated here in (56), are comparable to English forms like the one in (55b)-namely, they are complex Event Nominals.
(56) Lan [ NP hach=ko-chel-an-i ]. prog b2=a1p-hug-sub-itv 'We're hugging you.' (lit. ∼ 'Our hugging of you is happening.') Of course, the semantic ability to take a complement may be separated from the syntactic licensing mechanisms required (see e.g. Grimshaw 1979). While an English verb like examine in (53a) above may appear with a nominal complement, the nominal form examination requires the addition of the preposition of in (53b). Following work by Ordóñez (1995) for Popti ' and Coon et al. (2014) for Q'anjob'al, we argue that the Chuj suffix -an-which appears in embedded transitive forms like (56)-serves a similar function to that of English of: it is required in order to license the transitive object in a nominalization. In Chuj, we propose below that transitive objects are normally licensed by finite Infl 0 (vs. English v 0 ), giving us the desired result that all transitive nominalizations will require -an. Following Chomsky (2000; (and much preceding work), we assume that nominals must be assigned abstract case in order to be licensed in the derivation, and that case is assigned by functional heads, for example v 0 and finite Infl 0 . A range of work on ergative languages seeks to understand the licensing mechanisms found in these systems (see e.g.  Legate (2008) refers to as absolutive=nominative (abs=nom) languages-languages in which absolutive arguments are licensed via agreement with T 0 / Infl 0 (see also Bok-Bennema 1991;Campana 1992;Murasugi 1992, and others). Ergative is assigned inherently by transitive to the external argument in its thematic position (Woolford 1997). In a transitive clause, finite Infl 0 must license the object across the inherently-licensed subject. This type of system-previewed in (22) above-is schematized in (57) for transitives and (58) for intransitives. 11

(57) Transitive
Nominalization and the structure of progressives 27 As Legate notes, in an ABS=NOM language absolutive arguments are predicted to be uniformly absent in nonfinite embedded clauses, which lack finite Infl 0 . 12 Though Coon et al. (2014) do not discuss Chuj in any detail, Chuj fits the pattern of its close relative Q'anjob'al, which is argued to be an ABS=NOM language, like Mam and Q'eqchi'. 13 Building on work on cognate suffixes in Popti' by Ordóñez (1995), and in Q'anjob'al by Coon et al. (2014), we argue that the suffix -an in Chuj nonfinite embedded transitives is the realization of a functional head capable of licensing the 11 An unaccusative intransitive is represented in (58). We assume that unergative subjects in Chuj would be licensed in the same way (see Coon 2016b).
(58) Intransitive ructure of progressives 27 an ABS=NOM language absolutive arguments are predicted t in nonfinite embedded clauses, which lack finite Infl 0 . 12 014) do not discuss Chuj in any detail, Chuj fits the pattern njob'al, which is argued to be an ABS=NOM language, like on cognate suffixes in Popti' by Ordóñez (1995), and in t al. (2014), we argue that the suffix -an in Chuj nonfinite the realization of a functional head capable of licensing the other words, -an may be thought of as an accusative case r as it licenses transitive objects, or, in Ordóñez' terms, as contrasts with an absolutive=default (ABS=DEF) system, in which abl default, collapsing two different underlying mechanisms of syntactic ts are licensed by v 0 (i.e. accusative), while intransitive subjects are liative). Ch'ol is an example of the latter type of language: absolutive mbedded transitives-since they are licensed by v 0 -but impossible in e Coon et al. (2014) for discussion. H-ABS" language and the extraction of transitive subjects generally re-Focus construction. As shown in (i), the suffix -an doubles as the Agent gnate suffix -on in Q'anjob'al: ixim CLF wa'il? tortilla rtilla?' on ergative extraction restrictions (e.g. Campana 1992; Bittner & Hale gue that the extraction of transitive subjects in Q'anjob'al is the result of : transitive objects must move above the subject in order to be licensed tive subject in its base position. The Agent Focus construction fixes an additional case-assigner. We do not discuss this further here in the that this account provides a unified explanation of the appearance of -an s and Agent Focus environments. As Legate notes, in an abs=nom language absolutive arguments are predicted to be uniformly absent in nonfinite embedded clauses, which lack finite Infl 0 . 12 Though Coon et al. (2014) do not discuss Chuj in any detail, Chuj fits the pattern of its close relative Q'anjob'al, which is argued to be an abs=nom language, like Mam and Q'eqchi'. 13 Building on work on cognate suffixes in Popti' by Ordóñez (1995), and in Q'anjob'al by Coon et al. (2014), we argue that the suffix -an in Chuj nonfinite embedded transitives is the realization of a functional head capable of licensing the internal argument. In other words, -an may be thought of as an accusative case assigning head, insofar as it licenses transitive objects, or, in Ordóñez' terms, as an incorporated (case-assigning) preposition. What is crucial for us here is that an analysis of -an as a case-assigner both (i) explains its presence in nonfinite embedded transitives (and its absence in intransitives); and (ii) explains the ability for an absolutive argument to appear internal to a nominalization. Specifically, we propose that the nominal stem forms embedded under the progressive aspect marker are nominalized above the projection containing -an, explaining the possibility for full transitives to appear in nominalizations.
For concreteness we assume, following Coon et al. (2014), that the suffix -an occupies a special kind of Voice 0 head, and that this head assigns case to the transitive object and merges the thematic subject in its specifier position, as shown in (60). Unlike regular transitive Voice 0 , the head occupied by -an does not assign inherent ergative case to the subject position (compare (22)), which is occupied by PRO.
(i) Mach ix-man-an ixim wa'il? who pfv-buy-af clf tortilla 'Who bought the tortilla?' Building on other work on ergative extraction restrictions (e.g. Campana 1992;Bittner & Hale 1996), Coon et al. (2014) argue that the extraction of transitive subjects in Q'anjob'al is the result of a case assignment problem: transitive objects must move above the subject in order to be licensed by Infl 0 , trapping the ergative subject in its base position. The Agent Focus construction fixes this problem by bringing in an additional case-assigner. We do not discuss this further here in the intererst of space, but note that this account provides a unified explanation of the appearance of -an in both embedded transitives and Agent Focus environments. b2=a1p-hug-sub-itv 'We're hugging you.' (lit. ∼ 'Our hugging you is happening.') (60) Step 1: The embedded verb stem an occupies a special kind of Voice 0 head, and that this head assigns case to the transitive object and merges the thematic subject in its specifier position, as shown in (60). Unlike regular transitive Voice 0 , the head occupied by -an does not assign inherent ergative case to the subject position (compare (22)), which is occupied by PRO. 'We're hugging you.' (lit. ∼ 'Our hugging you is happening.') (60) Step 1: The embedded verb stem We adopt the proposal in Coon et al. (2014) that the presence of the intransitive status suffix in these forms is connected to the fact that the subject position is occupied by controlled PRO. As discussed in section 3, the verb root-here chel 'hug'-undergoes head-movement through Voice 0 up to the head hosting the status suffix, resulting in the stem chel-an-i, in line with Baker's Mirror Principle (Baker 1988).
Nominalization occurs above the verbal projection, as shown in (61). A possessor is merged in Spec,nP, controlling the null embedded subject. We assume that just as Set A marking in the verbal domain reflects an agreement relation between We adopt the proposal in Coon et al. (2014) that the presence of the intransitive status suffix in these forms is connected to the fact that the subject position is occupied by controlled PRO. As discussed in section 3, the verb root-here chel 'hug'-undergoes head-movement through Voice 0 up to the head hosting the status suffix, resulting in the stem chel-an-i, in line with Baker's Mirror Principle (Baker 1988).
Nominalization occurs above the verbal projection, as shown in (61). A possessor is merged in Spec,nP, controlling the null embedded subject. We assume that just as Set A marking in the verbal domain reflects an agreement relation between Voice 0 and the external argument, Set A marking in the nominal domain reflects an isomorphic relationship between possessive n 0 and the possessor (see Coon to appear). In (61), the root again undergoes successive cyclic head movement, this time landing at the top of the nominal projection. (61) Step 2: Nominalization Nominalization and the structure of progressives 29 Voice 0 and the external argument, Set A marking in the nominal domain reflects an isomorphic relationship between possessive n 0 and the possessor (see Coon to appear). In (61), the root again undergoes successive cyclic head movement, this time landing at the top of the nominal projection. (61) Step 2: Nominalization As previewed in section 3, we connect the position of the Set B clitic to the head responsible for licensing the absolutive argument. In verbal predicates, Infl 0 licenses absolutive arguments and the Set B clitic double attaches to the TAM particle, as in (21)-(22) above. When Infl 0 is missing, -an is inserted to license the As previewed in section 3, we connect the position of the Set B clitic to the head responsible for licensing the absolutive argument. In verbal predicates, Infl 0 licenses absolutive arguments and the Set B clitic double attaches to the TAM particle, as in (21)-(22) above. When Infl 0 is missing, -an is inserted to license the transitive object and Set B cliticizes to the embedded stem (i.e. the minimal word containing the head which triggered the doubled clitic).
The structure for intransitives is somewhat simpler, as there is no need for an additional licensing mechanism. The single argument of the intransitive is a controlled PRO, which is controlled by the possessor merged in Spec,nP above the verbal projection headed by the status suffix. A structure for the sentence in (62) is shown in (63).
a1p-walk-itv 'We're walking.' (lit. ∼ 'Our walking is happening.') (63) 30 Coon & Carolan The proposed structures for transitive and intransitive progressive stems account for the core facts we sought to explain. First, both transitive and intransitive subjects are cross-referenced by Set A marking. Under our approach, this is because both subjects are controlled PRO; the overt argument is a co-indexed grammatical possessor. Second, the presence of a special suffix -an in embedded transitives is connected to the fact that transitive objects are normally licensed by finite Infl 0 , which is absent in embedded environments. In languages such as Mam and Q'eqchi', transitives are simply impossible in nonfinite embedded contexts (see (52) above). Languages of the Q'anjob'alan branch, however, have a special licensing morpheme inserted in exactly those environments in which transitive objects require case. The appearance of the intransitive status suffix -i is connected to the fact that the embedded subject does not receive ergative case; see Ordóñez (1995) and Coon et al. (2014) for extensive discussion of these constructions.
A final issue remains: in both transitive and intransitive progressive stems in (61) and (63), we represent a null nominalizing head, n 0 . Recall that Chuj contrasts with some other split-ergative Mayan languages in having no clear morphological evidence of nominalization. Compare, for example, the Chuj progressive intransitive in (64)  'You're sleeping.' (lit. ∼ 'Your sleeping is happening.') The proposed structures for transitive and intransitive progressive stems account for the core facts we sought to explain. First, both transitive and intransitive subjects are crossreferenced by Set A marking. Under our approach, this is because both subjects are controlled PRO; the overt argument is a co-indexed grammatical possessor. Second, the presence of a special suffix -an in embedded transitives is connected to the fact that transitive objects are normally licensed by finite Infl 0 , which is absent in embedded environments. In languages such as Mam and Q'eqchi', transitives are simply impossible in nonfinite embedded contexts (see (52) above). Languages of the Q'anjob'alan branch, however, have a special licensing morpheme inserted in exactly those environments in which transitive objects require case. The appearance of the intransitive status suffix -i is connected to the fact that the embedded subject does not receive ergative case; see Ordóñez (1995) and Coon et al. (2014) for extensive discussion of these constructions. A final issue remains: in both transitive and intransitive progressive stems in (61)  We propose that both languages share the underlying structure in (63). However, while Chuj has a null n 0 and an overt verbal status suffix, Ch'ol exhibits the reverse. The fact that we do not find two overt morphemes is perhaps unsurprising, given (i) the tendency for status suffixes to delete in non-phrase-final position in many Mayan languages, and (ii) the fact that vowel hiatus is frequently resolved by syncope (Bennett 2016). One possibility is that the choice of whether to realize the verbal status suffix -i (Chuj) or the nominalizing head -el (Ch'ol) is simply a morphological accident.
A plausible alternative is that the suffix -i found in Chuj progressive stems is in fact a n 0 head, which is accidentally homophonous with the intransitive status suffix. In his discussion of Q'anjob'al, Mateo Pedro (2009) notes formal similarity between the Q'anjob'al progressive stem suffix -i and the nominalizer -ik in Kichean-branch languages. Here we do not take a stance on whether -i in progressives is the itv marker, or an accidentally homophonous nominalizer, but simply note that either possibility is compatible with the analysis presented above.

Types of nominalization: -i vs. -el
In section 5 above we reviewed evidence for the nominal nature of the stem forms analyzed in sections 6.1-6.2. Specifically, these stems may appear as sentential subjects and may trigger 3rd person singular Set A agreement in certain constructions. However, the progressive stem forms do not pass all distributional tests for nouns in the language, and indeed we find certain deverbal stems which are more transparently nominal than the progressive stems above. Compare the forms in (66) below, discussed in Buenrostro (2007).

(66)
Chuj (Buenrostro 2007: 262) a. Ix-in-b'at [ wa'-el ]. pfv-b1-go eat-nml 'I went to eat.' b. Ol-ach-b'at [ mol-oj kape ]. prosp-b2-go gather-nml coffee 'You will gather coffee.' Here we focus on the intransitive forms like (66a), which provide a more direct comparison with intransitive -i forms discussed above, though we return briefly to -oj forms like (66b) below. As noted above, suffixes of the form -Vl are found on nominals across Mayan (see e.g. Bricker 1981). If our analysis above is on the right track, we then have two types of nominalized intransitive stems in Chuj: We propose that the differences in behavior between these two groups of nominals fall out from independently observed differences in types of nominals cross-linguistically.
Specifically, the -el nominals belong to Grimshaw's class of Result Nominals (RNs), while the -i nominals found in progressive stems correspond to Grimshaw's Event Nominals (ENs), introduced in section 6.1 above (see also Abney 1987;Harley & Noyer 1998;Borsley & Kornfilt 2000;Alexiadou 2001; Embick 2010 on different types of nominalizations). 14 First, note that both -el and -i nominals may appear as subjects for example of the nonverbal predicate in (68); we return to the differences in translations below.
(68) a. Man te wach-ok-laj [ ko-munlaj-el ]. neg very good-irr-neg a1p-work-nml 'Our work isn't very good.' b. Man te wach-ok-laj [ ko-munlaj-i ]. neg very good-irr-neg a1p-work-itv 'Our working isn't very good.' Furthermore, with certain aspectual verbs, such as yamoch in (69) pfv-a2-begin a2-work-itv 'You began to work.' However, as shown in (70), the -el forms may also appear with the indefinite jun, the demonstrative particle tik, and as shown in (70a), may be fronted to preverbal focus position. 15 The corresponding -i nominal in (70b) is impossible with any combination of these markers. A second point of difference directly relevant to the proposal in this paper is that -el forms are banned or strongly dispreferred in progressive environments, as shown in (71a). Progressives instead require the -i forms discussed in section 6.1, shown in (71b). These properties are summarized in the table in (72). While the -el and -i forms pattern similarly with respect to the diagnostics in rows (a) and (b), the diagnostics in rows (c) and (d) make -el forms look "more nominal" and -i forms look "more verbal". We tie this contrast to independent work on nominalization, arguing that only -i nominals contain verbal structure and thus correspond to Grimshaw's Event Nominals.
Structurally, we propose that the differences here fall out naturally from an analysis in which -el forms are nominalized lower in the structure, as in (73), while -i forms are nominalized above the verbal projection, as in (74) and discussed in section 6.2.
(73) 34 Coon & Carolan Structurally, we propose that the differences here fall out naturally from an analysis in which -el forms are nominalized lower in the structure, as in (73), while -i forms are nominalized above the verbal projection, as in (74) and discussed in section 6.2.
Crucially, only the -i forms project an argument structure, here a thematic subject (PRO); the -el forms do not. Either form may then appear with a possessor, but in the more complex -i nominalization, the possessor controls a null PRO in thematic subject position. This explains the fact-independently discussed for English in (55) above-that while the possessor in (73) may bear any contextually appropriate relation to the work, the possessor in (74) must be understood as the person or entity carrying out the work. 16 16 A reviewer asks about the possibility of arbitrary PRO readings in the absence of an overt possessor to bind PRO. Initial evidence suggests that these may indeed possible; compare the examples in (i).
ropose that the differences here fall out naturally from an analms are nominalized lower in the structure, as in (73), while -i ed above the verbal projection, as in (74) and discussed in sec- munlaj work e -i forms project an argument structure, here a thematic subrms do not. Either form may then appear with a possessor, but -i nominalization, the possessor controls a null PRO in then. This explains the fact-independently discussed for English while the possessor in (73) may bear any contextually approwork, the possessor in (74)  Crucially, only the -i forms project an argument structure, here a thematic subject (PRO); the -el forms do not. Either form may then appear with a possessor, but in the more complex -i nominalization, the possessor controls a null PRO in thematic subject position. This explains the fact-independently discussed for English in (55) above-that while For example, while a bare non-referential object may appear as the complement as in (66b), repeated in (77a) below, a full referential DP te' kape 'the coffee' is impossible, as in (77b). 19 This is again consistent with the proposal from section 6.2 above that the morpheme -an is required in embedded transitives to license the direct object.
(77) a. Ol-ach-b'at [ mol-oj kape ]. prosp-b2-go gather-nml coffee 'You will gather coffee.' b. *Ol-ach-b'at [ mol-oj te' kape ]. prosp-b2-go gather-nml coffee intended: 'You will gather the coffee.' Returning to the properties in (72), the fact that both -i and -el nominals may serve as a subject-see row (a) in (72)-is consistent with the fact that both forms are ultimately nominal in nature. Turning to row (b) in (72), we note that aspectual verbs like 'begin' are frequently optional restructuring verbs (Wurmbrand 1998;. Crucially, restructuring verbs are verbs which combine with smaller embedded elements, generally taken to lack an embedded (e.g. PRO) subject. If this is on the right track, yamoch could be considered an optional restructuring verb in Chuj. In (69a) it combines with a small complement (restructuring), while in (69b) it combines with a larger complement (non-restructuring).
The restriction of D 0 -level elements to appearing only with smaller -el forms in row (c) of (72) above has some cross-linguistic precedent. Compare the ungrammaticality of determiners and demonstratives with English poss-ing gerunds (see Lees 1963;Borsley & Kornfilt 2000). Under the analysis presented here, the Chuj -el forms are analogous to English nominals like criticism, which do not project verbal functional material (Grimshaw's RNs). The -i forms are comparable to English gerund forms, like the one in (78b). While nothing in the present analysis predicts the fact that one form should appear with D 0 elements and the other should not (see Grimshaw 1990 andHarley &Noyer 1998 for some discussion), we suggest that our analysis gains support from this crosslinguistic parallel.
(78) a. We discussed this/that/the criticism of the book. b. *We discussed this/that/the criticizing the book.
Finally, we suggest that the difference in argument structure of -i and -el nominals is related to the contrast in their ability to appear under the progressive lan, shown again in (79). The progressive lan is an intransitive predicate which takes only a single argument: the possessed nominalized clause. We suggest that the impossibility of -el nominals under lan may be related to a preference for thematic roles to be assigned by predicates. In (79a), the null subject is projected and assigned a thematic role inside the nominalized clause, controlled by the higher possessor; in (79b) the participant is associated only with the possessor position, literally equivalent to 'Your work is happening'. Though this latter form is at best highly marked (not unlike its literal English translation), note that lan may appear with small nominal complements-a plain noun k'in in (80a) and an RN munlajel in (80b)-with the addition of the directional em 'down' in (80).
(80) a. Lan em ha-k'in. prog dir.down a2-day/party 'Your party is happening.' b. Lan em munlaj-el. prog dir.down work-nml 'Work is happening.' The examples in (80) suggest that there is no syntactic restriction preventing lan from combining with "smaller" nominals. We leave the semantic contribution of the directional as a topic for future work, noting for now parallels with existential-type constructions cross-linguistically, for example in English: A party is going *(down) or There is a lot of work ?(in the alley); see also Bybee et al. (1994) on cross-linguistic relations between progressive and locative constructions, noted above.

Conclusions
In this paper, we examined progressive constructions in Chuj, an understudied Mayan language of Guatemala. Building on work on other Mayan languages, we suggested that the appearance of a split in person marking in the progressive aspect is due to differences in structure between progressive and non-progressive aspects. Namely, the progressive aspect marker behaves as an intransitive predicate, taking a nominalized verb form as its complement.
First, in section 4, we showed that the progressive aspect marker behaves unlike perfective (ix/Ø), imperfective (tz), and prospective (ol) aspects in a number of respects, and instead patterns with the Chuj class of nonverbal predicates in terms of the appearance of negation and aspectual particles. Next, in section 5 we provided evidence that the complement to the progressive maker shares certain properties with other nominals in the language. In particular, it may saturate the argument position of a predicate and may trigger overt 3rd person singular agreement on a head.
In section 6 we proposed a structural account of these forms, comparing them with different types of nominalizations cross-linguistically, and within Chuj. Specifically, we argued that these progressive stems are nominalized above the vP layer and project argument structure, corresponding to Grimshaw's (1990) Event Nominals. Following work on Ch'ol by Coon (2010;, we proposed that subjects are null PROs, controlled by possessors generated in the nominalized nP layer. Taken as a whole, the analysis of these forms both accounts for the presence of Set A marking on embedded intransitives, as well as the appearance of the suffix -an in embedded transitives (required to case-license the embedded object in the absence of finite Infl 0 ).
Finally we examined a different "more noun-like" type of nominal stem form. Though more work is needed to fully understand the range of differences between intransitive -i and -el nominals, the proposal that -i-nominalizations occur above vP and contain a thematic PRO (ENs), while -el-nominalizations are smaller (RNs), accounts for a range of facts and makes clear connections between nominalization in Chuj and nominalization in other languages.
While the core analysis of Chuj's progressive aspect builds on existing work in Mayan, the result that Chuj progressive stems are nominalized is not trivial, since-unlike other languages for which such studies have been conducted-there is no overt morphological difference between intransitive verbal stems and intransitive nominal stems in Chuj. We thus hope to have shown how the careful investigation of distribution of forms can result in differences which are not apparent on the surface.
Future work is needed to understand the semantics of these lan constructions, including the dispreference for Result Nominals under lan. Another avenue for further investigation is the range of contexts in which the nominalized "progressive" stem forms appear. In related Q'anjob'al, Mateo Toledo (2003), developed in Mateo Toledo (2013), identifies other environments in which these stem forms-which he labels "nonfinite forms"-are used, including in adverbial clauses and depictive secondary predicates (see also Pascual 2010 andMateo Toledo 2012). Preliminary evidence suggests that these uses can be found in Chuj as well, as shown by the sentences in (81) (2013) on Q'anjob'al, we suggest that all of these involve embedding. Under the analysis laid out above, there is nothing progressive about the stem forms themselves; rather, the progressive interpretation comes entirely from the embedding predicate lan. Under this analysis, the forms in (81) involve stative predicates (i.e. te jelan and te lajan); see also Henderson & Coon (2016) on adverbial embedding in Kaqchikel. Unlike previous work, we maintain that all of these embedded forms are nominalized. The different sizes of nominalization discussed above provide an explanation for the fact that these forms pass some, but not all, nominal diagnostics (discussed in Mateo Toledo 2013 for Q'anjob'al), and this nominalization provides a straightforward account of the appearance of the shift in person marking in all of these constructions.