2.1 Intransitive stems

Intransitive roots can be described as roots that appear without an overt v/Voice $^{0}$ suffix in intransitive stem forms, as in (7). Here and below, I do not parse out a null v/Voice $^{0}$ morpheme, but I assume that a ‘Ø’ v/Voice $^{0}$ head is present.

Intransitive stems are marked by the intransitive status suffix -i in perfective and imperfective aspects (7a–b), and the irrealis -ok in the prospective aspect (7c).Footnote ^[5] The status suffix -i is dropped when the stem is not phrase final (see Dayley Reference Dayley1981, Henderson Reference Henderson2012, Buenrostro Reference Buenrostro2013), unless its omission would result in a final complex coda (Mateo Pedro Reference Mateo Pedro2011), discussed below (see footnote 9). Compare (8a), in which -i is dropped, with the form in (8b), in which the subject appears preverbally and -i is not dropped.

The single argument of the intransitive stem is cross-referenced with an absolutive morpheme, known as ‘Set B’ in Mayanist literature. The Set B marker attaches to the stem-initial aspect marker, as shown in the forms in (7) above. As in other Mayan languages, there is no overt third person singular Set B marker.

Other types of roots may also appear in intransitive stem forms, but require the presence of overt derivational suffixes, to which we return below. The intransitive stem template is shown in (9). Intransitive stems share a common inflectional exterior: TAM, Set B (possibly null), and the status suffix -i (possibly dropped). √ ITV roots appear directly in this frame (9a), while other types of roots must be derived via overt morphology in order to appear in the intransitive stem (9b).

2.2 Transitive stems

Transitive roots may appear directly in transitive stem forms, as in (10).

The transitive root appears with the transitive status suffix -V’ in perfective, imperfective, and prospective aspects; the suffix is -a’ for roots with non-back vowels [a], [e], and [i], and harmonic with the root vowel for roots with [o] and [u] vowels. As with the intransitive suffix, the transitive status suffix is dropped when the stem is not phrase-final. Transitive stems appear with two person $/$ number-marking morphemes: objects are marked with Set B clitics (null for third person) and transitive subjects are cross-referenced with Set A (ergative) prefixes.

In Mayanist literature, a division is drawn between ‘root transitive’ stems, like the ones in (10), and ‘derived transitive’ stems (see, e.g., Coon Reference Coon2016). Root transitives are formed directly from transitive roots, while derived transitives are formed from other types of roots and appear with a special status suffix, -ej, shown in (11).

These derived transitives include both transitives derived by overt morphology, as in (11a), as well as a number of zero-derived forms, often denominals, as with the nominal root tz’ib’ ‘letters, writing’ in (11b). Note that the transitive suffix -V’ does not co-occur with -ej, and -ej is not dropped phrase-finally.

Root and derived transitive templates are given in (12).

2.3 Verbal stem summary

The TAM markers and ‘status suffixes’ seen to this point are summarized in the table in (13).Footnote ^[6]

The suffixes in (13) are listed together here for ease of reference, but note that they do not form a unified category, and it is not clear that they serve a specific derivational ‘function’. While -i and -V’ appear only in phrase-final position, -ej is never dropped. The intransitive suffix is replaced with -ok in irrealis contexts like the prospective ol, but the other two suffixes are not. Furthermore, while both -i and -ej appear on stem forms that have been derived, -V’ only appears immediately following transitive roots.

This is in keeping with the description of so-called ‘status suffixes’ in works such as Henderson (Reference Henderson2012). As Henderson notes, Mayan status suffixes do not alter the transitivity of a stem, but rather ‘reflect valency information already available from the lexical content of the predicate, or from a combination of lexical information and derivational morphology’ (Henderson Reference Henderson2012: 747). While the choice of status suffix depends on properties like transitivity, TAM, and clause type, this information is generally also represented elsewhere – for example in the stem-initial TAM marker. Following Clemens & Coon (Reference Clemens and Coonto appear), and consistent with the order of morphemes on the stem, I locate the status suffix in the head of a projection at the top edge of the verbal maximal projection, above VoiceP, in a projection labelled ‘ssP’. I tentatively assume that the choice of suffix may be determined by selection, but leave a full analysis of the role of status suffixes as a topic for future work. We return to verb stem structure below in Section 3.

2.4 Positional roots

Positionals form a distinct class of roots throughout the Mayan family, distinguishable by their special stem-forming morphology as well as their meaning (see, e.g., Haviland Reference Haviland1994, Henderson Reference Henderson2016, Reference Henderson2017). Semantically, they typically make reference to position, shape, aggregation, or surface quality. Henderson (Reference Henderson2016: 585) writes that positionals ‘lexicalize interval states (sitting, standing, lying down, etc.) as well as gradable properties (broken, fat, flexible, etc.)’. For Tsotsil, Haviland (Reference Haviland1994: 733) refers to an apparent ‘preoccupation with space, shape, and configuration’, reflected in the large class of positional roots. For Chuj, Hopkins (Reference Hopkins1967: 76) notes that positional roots may be ‘distinguished from other form classes by a number of derivational reduplication processes which occur with no other form class’.

Positionals are a class of roots, but there is no special class of positional surface stems (see, e.g., England Reference England1983, Haviland Reference Haviland1994, contra Evans & Levinson Reference Evans and Levinson2009). Positional roots in Chuj may be identified by their ability to form stative (aspectless or ‘non-verbal’) predicates through the addition of the suffix -an, often accompanied by a directional particle, as in (14).

Henderson (Reference Henderson2017) argues that positional roots in Kaqchikel denote measure functions and that they must always be derived into a relational expression before they may be inflected. Indeed, in order to form eventive transitive or intransitive predicates, positional roots in Chuj must always appear with one or more of a number of derivational suffixes, followed by the appropriate status suffix (-i for intransitives or -ej for transitives). In other words, positional roots follow the templates for derived intransitive and transitive stems in (9b) and (12b) above.Footnote ^[7]

While the tripartite division among intransitive, transitive, and positional roots above is a useful point of departure, in some cases, a given root may not belong clearly to one or another class, or may belong to multiple classes (Haviland Reference Haviland1994, Lois Reference Lois, Shklovsky, Mateo Pedro and Coon2011). This is particularly true in the case of transitive and positional roots, which appear to share certain derivational possibilities across the Mayan family (see Haviland Reference Haviland1994, Coon & Preminger Reference Coon, Preminger, Avelino, Coon and Norcliffe2009, and Henderson Reference Henderson2016, Reference Henderson2017), as well as in Chuj in particular (Hopkins Reference Hopkins1967). Nonetheless, canonical positional roots may be distinguished from transitive roots by the former’s inability to form transitive stems without the presence of overt derivational morphology. I focus on canonical cases here, setting aside the puzzle of what Haviland (Reference Haviland1994) terms ‘mixed categories’ for future work.

2.5 Summary

In addition to providing an introduction to roots and stem-formation in Chuj, this section underscores the importance of distinguishing between, for example, an intransitive root and an intransitive stem. As seen above, Chuj roots may be classed according to their formal derivational behavior, and, to some extent, their semantics (see (3) above). However, roots do not inflect directly for person $/$ number and temporal information. Rather, additional morphology is required to form stems, and a single root may enter into a variety of different stem forms through the addition of the ‘derivational’ (v/Voice $^{0}$ ) and ‘status’ suffixes seen above. Thus, while roots in Chuj clearly belong to root classes, these classes do not directly correspond to surface lexical categories. Below, I argue that a transitive root (√ TV), for example, requires exactly as much derivation to form a verbal surface stem as a nominal root (√ NOM) does. In the sections that follow, we look in more detail at morphology used to form verb stems in order to address the question of what information is contributed by the root and what is contributed by the functional heads.

3.1 Unergatives

Many agentive intransitives in Chuj are formed from nominal roots with the suffix -w.Footnote ^[8] Examples of denominal -w verbs and their corresponding nominal roots are shown in (17).

Note that verbs borrowed from Spanish – for example karrel from the Spanish infinitive form correr ‘to run’ – enter Chuj as nominals and must be derived into verb forms (also discussed in Haviland Reference Haviland1994 for Tsotsil and Coon Reference Coon2013 for Ch’ol). These borrowed forms suggest that the -w process is at least semi-productive, and not restricted to frozen lexical items (cf. Buenrostro Reference Buenrostro2013). However, the process is not entirely productive (i.e. many nominal roots may not appear with -w), and, as the forms above indicate, the meaning of the resulting verb stem is not necessarily predictable from the meaning of the nominal root. I take this to be compatible with the proposal that the v/Voice $^{0}$ head instantiated by -w merges directly with the root, formalized below, within the domain of special meaning (Arad Reference Arad2003).

Examples of underived nominal roots in nominal environments are shown in (18); the corresponding intransitive -w stems are in (19).Footnote ^[9]

Intransitive verb forms are also derived from positional roots (Section 2.4), as in the table in (20). The verbal stem form indicates some kind of movement in the position denoted by the positional root.

Recall that positional roots in Chuj may be identified by their ability to form stative predicates with -an. Examples of positional roots in stative -an stems are given in (21); the same roots in intransitive verb stems with -w are shown in (22).

These stems fit the pattern of intransitive stems in Chuj identified in Section 2.1 above: they have a single argument, marked with Set B $/$ absolutive (null for third person singular). They also appear with the intransitive status suffix -i in the perfective and imperfective aspects (22a), and with the irrealis -ok in the prospective (22b). I propose that  stems like the one in (23a) have the structure in (23b).

Specifically, the suffix -w occupies a bundled v/Voice $^{0}$ head (Pylkkänen Reference Pylkkänen2002, Harley Reference Harley, D’Alessandro, Franco and Gallego2017), which merges directly with the root, categorizing the stem as verbal and introducing the external argument in its specifier position (along the lines of Kratzer Reference Kratzer, Rooryck and Zaring1996). The ‘status suffix’ occupies the highest head in the verbal projection, labelled ssP (Clemens & Coon Reference Clemens and Coonto appear). The root undergoes successive head movement through v/Voice $^{0}$ to , resulting in basic verb-initial order (discussed in Section 3.2), and forming the verb stem; the order of morphemes is in line with the Mirror Principle (Baker Reference Baker1988). The TAM particle is located in Infl $^{0}$ (Aissen Reference Aissen1992), which cliticizes to the verb stem.

While both agentive intransitive -w and regular transitive v/Voice $^{0}$ heads introduce an external argument in their specifier position, agentive intransitive -w differs in a crucial respect: transitive v/Voice $^{0}$ assigns inherent ergative case to the argument in its specifier, while -w v/Voice $^{0}$ does not (see, e.g., Woolford Reference Woolford1997, Legate Reference Legate2008 and works cited there on inherent ergative).Footnote ^[10] Following Coon Reference Coon2017, I assume that Set A (ergative) agreement goes hand-in-hand with inherent ergative case assignment; the ergative prefix represents the spell-out of features of the external argument on the verb stem as the result of feature-sharing triggered by inherent case assignment.Footnote ^[11] As expected, ergative agreement is absent in agentive intransitive forms like (23a). Instead, in -w forms, the single absolutive argument must be licensed by finite Infl $^{0}$ , as shown in (24). Following Coon, Mateo Pedro & Preminger (Reference Coon, Mateo Pedro and Preminger2014) for Q’anjob’al, first and second person Set B morphemes like the one in (19b) are pronominal clitics – they are triggered by agreement with finite Infl $^{0}$ and attach to the TAM head.Footnote ^[12]

The derivation of an intransitive -w stem in (24) may be contrasted with the proposed derivation of a full transitive in (25); see Coon et al. (Reference Coon, Mateo Pedro and Preminger2014), Armstrong (Reference Armstrong2015), and Clemens & Coon (Reference Clemens and Coonto appear).

Note that in the -w stem above, Infl $^{0}$ licenses the absolutive external argument; in a transitive construction like (25), Infl $^{0}$ licenses the absolutive object (here, I set aside the question of whether the object raises above the subject, as in Coon et al. (Reference Coon, Mateo Pedro and Preminger2014), focusing just on which heads license which arguments). The proposal here is in keeping with the fact that absolutives are consistently licensed by Infl $^{0}$ , that agreement with Infl $^{0}$ results in the Set B clitics, and that both derived and underived intransitive verbs have absolutive (Set B) subjects. These structures will be relevant to the proposal for antipassives discussed in the following section.

3.2 Incorporation antipassives

Chuj has been described as having two types of antipassive (e.g. Maxwell Reference Maxwell1976, Dayley Reference Dayley1981, Buenrostro Reference Buenrostro2013): (i) an absolutive antipassive and (ii) an incorporation antipassive, -w.Footnote ^[13] Here, we focus on the latter, returning to the absolutive antipassive in Section 4. A transitive ${\sim}$ incorporation antipassive pair is shown in (26).

In the transitive in (26a), the subject is marked with the Set A (ergative) prefix and, in the absence of post-verbal material, we would find the transitive status suffix -a’ (cf. Tas ix-a-xik-a’? – ‘What did you chop?’). The object in the transitive in (26a) appears with a noun classifier, te’, the classifier for wood-based elements (Hopkins Reference Hopkins2012b, Royer Reference Royer2016). As described for Chuj and other Q’anjob’alan languages (e.g. Craig Reference Craig1986 for Popti’; Zavala Reference Zavala and Senft2000 for Akatek; Royer Reference Royer2016 for Chuj), noun classifiers appear either preceding nominals in referential contexts (27a), or alone functioning as referential pronouns (27b). Although they are not limited to definite contexts (Craig Reference Craig1986, Royer Reference Royer2016), I translate them with the English definite determiner for simplicity, and I assume that they occupy D $^{0}$ .

The transitive in (26a) above has two full arguments, and like other transitive stems, the subject is marked with a morpheme from the Set A (ergative) series (here ko-). Incorporation antipassive forms like the one in (26b), on the other hand, only ever appear with a single person $/$ number-marker on the verb. In (26b), the Set B (absolutive) morpheme -onh cross-references the subject. Although an apparent object appears, here k’atzitz ‘wood’, unlike in (26a), it does not appear with its classifier.

As others have noted (Maxwell Reference Maxwell1976, Dayley Reference Dayley1981, Buenrostro Reference Buenrostro2013), incorporation antipassive stem forms like the one in (26b) appear with a non-oblique post-nominal ‘object’, but there are restrictions. First, the nominal must be bare and non-referential, and it must appear immediately adjacent to the verb stem. The ‘object’ in an incorporation antipassive may not appear with numerals (28a), demonstratives (28b), or noun classifiers (28c).

The incorporated object may also not be possessed, nor may it be a pronoun, as shown by the ungrammaticality of the forms in (29).

However, as Maxwell (Reference Maxwell1976) describes, the object is not restricted to a bare nominal. Certain modifiers may appear with the objects in the incorporation antipassive construction, as in (30).Footnote ^[14]

Because only a limited number of modifiers may appear in this construction, Maxwell proposes that forms like niwak kaxlan ‘fat chicken’ in (30) are actually compounds. I suggest instead that modifiers that appear in these constructions are Chuj’s true adjectives and sit below D $^{0}$ , while others are relative clauses (‘the one that is X’); this analysis is compatible with diagnostics for distinguishing adjectives from relative clauses in Martínez Cruz (Reference Martínez Cruz2007). Forms like niwak kaxlan in (30) are thus bare NPs (or nPs) of type ${<}$ e,t ${>}$ . Either possibility is compatible with the analysis below.

The apparent object in an incorporation antipassive construction is ‘discourse opaque’ (see Farkas & de Swart Reference Farkas and de Swart2003). As shown by the sequence in (31), the incorporated NP may not be picked up by pronouns later in the discourse. Recall that noun classifiers serve a pronominal function in Q’anjob’alan languages; here, anh is the classifier used for nouns denoting plant-based elements (Craig Reference Craig1986, Hopkins Reference Hopkins2012b, Royer Reference Royer2016).

Following Maxwell (Reference Maxwell1976), we may conclude that these apparent objects in the incorporation antipassive construction are not ‘true nominal arguments’, from either a syntactic or a semantic point of view. I propose a syntax for these constructions in Section 3.2.1 and then turn to the semantics in Section 3.2.2.

3.3 Unacccusatives

Up to this point, we have seen the suffix -w appear on three of the four types of roots identified above: it appears on nominal and positional roots to form unergative predicates (Section 3.1), and it also appears on transitive roots to form incorporation antipassives (Section 3.2). Some examples are repeated in (45).

The fourth type of root discussed in Section 2.1 above is the class of intransitive √ ITV roots. I propose here that all √ ITV roots are unaccusative and that unergative surface stems are always derived. Although independent unaccusativity diagnostics do not to my knowledge exist, Chuj verbs that correspond to canonical unergatives in languages for which there are such diagnostics are overwhelmingly derived by some sort of suffix or series of suffixes: either the -w described above or one of several other consonantal suffixes (see Hopkins Reference Hopkins1967).

For example, cross-linguistically, unergatives often (though not always) include manner-of-motion verbs, verbs denoting volitional actions, as well as verbs of bodily function (e.g. Perlmutter Reference Perlmutter1978). In Chuj, such verbs are generally morphologically complex in form. Positional roots derive many manner-of-motion verbs, as seen in Section 3.1 above. Additional derived intransitives belonging to these categories are shown in (49).

While many of the putative unergative stems are transparently derived from other roots, for some the source is not transparent (e.g. taj is a nominal root meaning ‘pine’). The proposal that all unergatives in Chuj are derived is in keeping with proposals in which unergative verbs are derived from nominals roots via abstract verbal structure – even in languages that lack morphological evidence for this derivation (Hale & Keyser Reference Hale and Keyser1993, Reference Hale, Keyser, Alsina, Bresnan and Sells1997); see also the discussion of Ch’ol in Section 3.4.2 below and Davis (Reference Davis, Mendikoetxea and Uribe-Etxebarria1997) on St’át’imcets (Salish).Footnote ^[21]

The class of √ ITV roots, on the other hand, includes verbs of directed motion like b’at ‘go’, ek’ ‘pass’, em ‘descend’, and och ‘enter’, as well as change-of-state verbs like k’ib’ ‘grow’ and cham ‘die’ – forms that we would expect to pattern as unaccusatives if independent diagnostics were found (see also examples in (7) above). These roots are impossible with -w.Footnote ^[22]

Note that under the analysis presented here, the impossibility of -w on intransitive (i.e. unaccusative) roots is unsurprising: -w is a v/Voice $^{0}$ head that attaches to a root and merges an external argument; the internal argument in a -w stem is either absent entirely in the case of √ NOM and √ POS roots (Section 3.1) or incorporated in stems formed from √ TV roots (Section 3.2). Intransitive √ ITV roots, on the other hand, are unaccusative and combine only with an internal argument. Semantically, I propose that they are of type ${<}e,<s,t>>$ , like transitive roots, as in (3) above; unlike transitive roots, they may not combine with the transitive-forming v/Voice $^{0}$ head which merges an agent (I do not take a particular stance on the source of this distinction). These conflicting requirements then create a licensing problem. The -w v/Voice $^{0}$ head introduces an external argument but does not assign inherent ergative case. The introduced argument must be licensed by finite Infl $^{0}$ , leaving the internal argument of the √ ITV root unlicensed.

One imaginable possibility would be to incorporate the unaccusative subject, alleviating the licensing problem. In Chuj, unaccusatives are unable to incorporate their subjects (a noted point of variation among languages that allow transitive objects to incorporate, described in Mithun Reference Mithun1984). For example, in (48a), the intransitive root ul ‘dissolve’ appears in an intransitive stem with a full DP subject. The ungrammatical example in (48b) attempts to combine the intransitive (unaccusative) root with a bare nominal subject, and then to introduce an external argument via -w.

Although nothing internal to the analysis here would rule out a form like (48b), the absence of unaccusative subject incorporation is unsurprising from a typological perspective.

3.4 Interim summary and a Ch’ol interlude

4.1 Two kinds of passive

The reappearance of -w in the absolutive antipassive suffix -waj does not seem to be an accident. Indeed, although some works list -waj as the antipassive suffix, Hopkins (Reference Hopkins1967) decomposes it into -w plus a suffix -aj:

As the above quotation suggests, and as previewed in Section 2 above, the suffix -w is one of a number of consonantal suffixes that attach to roots in order to form stems: CVC-C (see the template in (6) from Section 2 above). The suffix -aj then attaches to some of these CVC-C bases to form an intransitive stem (which then further requires the presence of the intransitive status suffix -i).

Take for example the stem el-k’- ‘steal’. This stem appears to be derived from the intransitive root el ‘to leave’, followed by the derivational suffix -k’. This stem then forms a derived transitive with the suffix -ej, as in (57a). It also forms an intransitive (unaccusative) with the suffix -aj, which is then followed by the intransitive status suffix -i, shown in (57b). Buenrostro (Reference Buenrostro2013) lists -aj as the passive suffix for derived transitives; I gloss -aj as div for ‘derived intransitive’, and propose here that it has a broader use than simply passive.Footnote ^[24]

Although further work is needed on the large number of stem-forming suffixes described by Hopkins (Reference Hopkins1967), examples like (57b) provide a first clue that -waj should be decomposed into the general ‘agentive intransitive’ suffix -w from Section 3, plus the suffix -aj. Indeed, we find an analogous alternation between a consonantal suffix and a -C-aj suffix in Chuj passives, summarized in (58).

What I label ‘agentless passives’ in Chuj are formed with the suffix -j.Footnote ^[25] This suffix attaches to transitive roots like man ‘buy’ and choj ‘grind’ in (59) to form an intransitive (unaccusative) stem. As with other intransitive stems in Chuj, the single argument is Set B and the intransitive status suffix -i appears stem-finally.

The agentless passive forms in (59) stand in contrast to passives formed with -chaj, shown in (60). As shown in (64), -chaj attaches to transitive roots – here yam ‘catch’ and mol ‘gather’ – to form an intransitive unaccusative stem. Again, the single argument is marked Set B and the stem is suffixed with -i.

I propose that the difference between agentless passives with -j in (59) and regular passives with -chaj in (60) lies in the presence of an implicit agent. The contrast is comparable to the difference between anticausatives and passives in English (see, e.g., Roeper Reference Roeper1987, Baker, Johnson & Roberts Reference Baker, Johnson and Roberts1989, Levin & Rappaport-Hovav Reference Levin and Rappaport-Hovav1995). In English, for example, the former may appear with agentive by-phrases and agent-oriented adverbs (61a), while the latter may not (61b).

Contrasts such as the one in (61) have been used to argue that passives contain an implicit agent, while anticausatives do not (see Bhatt & Pancheva Reference Bhatt, Pancheva, Everaert and van Riemsdijk2006 and Williams Reference Williams2015 for overviews). I offer a brief sketch of the analyses for the two constructions in Chuj, beginning with the -chaj passive in Section 4.1.1 and turning to the -j agentless passive in Section 4.1.2.

4.2 Implicit arguments and the suffix -aj

Having examined two types of passives, we return to the suffix -aj. In both antipassives and passives, the presence of -aj correlates with the availability of oblique DPs that are associated with particular thematic roles (see the table in (58) above). In the non-aj constructions, on the other hand, the argument slot is either already filled (by the bare NP object in the case of the incorporation antipassive), or is not projected at all (in the agentless passive with -j). Minimal pairs are provided in (72) and (73) below.

While the absolutive antipassive with -waj permits the theme to be realized as a PP (72b), incorporation antipassives like (72a) prohibit oblique themes altogether, requiring instead that the theme be a bare NP.

In the preceding section, we found that agentless passives like the one in (73a) permit an oblique, but this DP is not associated with the notional agent. Instead, it may have any semantically plausible causal relation to the event, as a result of the meaning of the relational noun that introduces it. In a regular passive like (73b), the oblique must be understood as the agent.

As foreshadowed above, I propose here that the suffix -aj in both absolutive antipassives and regular -chaj passives correlates with the presence of an implicit argument – an argument that contributes to the semantic interpretation, but is never overtly realized. Passives like (73b) and (74) have an implicit external argument, while antipassives like (72b) and (75) have an implicit internal argument. In both passives and antipassives, I propose that -aj is an overt reflex of Existential Closure (Diesing Reference Diesing1992).

In the forms in (78) and (79), I represent -aj as a morpheme located between the v/Voice $^{0}$ head which introduces the external argument, and the intransitive status suffix which demarcates the edge of the verb stem. This is in keeping with the proposal that the root undergoes head movement through the functional spine, landing in the projection that hosts the status suffix. This location for existential closure is also compatible with existing semantic work. For example, in Diesing’s (Reference Diesing1992) system, existential closure happens at the edge of VP – which, at the time of her work, was assumed to host all of the participant arguments. Translating this into modern terms, we might expect existential closure to take place just above the projection hosting the external argument, here v/VoiceP. More recently, Chung & Ladusaw (Reference Chung and Ladusaw2004) proposed that ‘[p]redicates must have their participant arguments (semantically) saturated at the event level’ (Chung & Ladusaw Reference Chung and Ladusaw2004: 11). The event argument delimits the event level, and is the highest argument of a predicate, combining after all participant arguments have been saturated. Again, we would expect existential binding to happen just above v/VoiceP – exactly in line with the order of morphemes on the Chuj stems.

I assume that the implicit argument combines with the root via regular Functional Application, but is a variable that must be existentially bound higher in the derivation. Optionally, a PP may be adjoined higher and associated with the implicit argument. Compare the transitive construction in (76a) with the absolutive antipassive in (76b).

In both constructions, the transitive root tek’ ‘kick’ composes with an internal argument via Functional Application: in (76a) it is the DP nok’ pelota, while in (76b) it is the unexpressed implicit argument, represented as

As the contexts and translations of the sentences in (76) suggest, there is a semantic difference between a transitive construction and an antipassive in which the notional patient is expressed as a PP: in the transitive, the patient has a higher degree of ‘affectedness’ than in the corresponding antipassive (Hopper & Thompson Reference Hopper and Thompson1980). This is in keeping with existing literature on antipassives (see, e.g., Spreng Reference Spreng, Johns, Massam and Ndayiragije2006, Basilico Reference Basilico, Cuervo and Roberge2012, Polinsky Reference Polinsky2017). I set the details of this construction – as well as semantic properties of the implicit internal argument, see Bhatt & Pancheva (Reference Bhatt, Pancheva, Everaert and van Riemsdijk2006) and Williams (Reference Williams2015) for discussion and overviews – aside as a topic for future work.Footnote ^[28]

5.1 Summary

This detailed tour through the Chuj verb stem provides evidence for the division of labor in (77), repeated from (2) above. Specifically, while functional projections are responsible for introducing the external argument, as well as for clausal licensing and agreement generally, it is the root itself that determines whether or not it will semantically compose with an internal argument.

I argued that Chuj’s rich verbal morphology offers a window into the syntax of its argument structure. In Chuj, as in other Mayan languages, roots may be clearly classed based on their stem-formation possibilities. Following a range of work in Mayan linguistics, and discussed in Section 2, it is clear that root class does not correspond directly to surface lexical category: a √ TV is not a transitive verb, an √ ITV is not an intransitive verb, and √ POS does not correspond to any surface lexical category (Haviland Reference Haviland1994, Henderson Reference Henderson2017).

I proposed instead that root classes in Chuj may be at least partially distinguished based on their semantic types, summarized in (3) above. Specifically, √ TV and √ ITV roots are of type ${<}e,<s,t>>$ and thus combine directly with an internal argument DP to yield an event predicate, while √ NOM and √ POS roots – the latter discussed in depth in Henderson (Reference Henderson2017) – do not. As noted above, this is in line with the proposal by Davis (Reference Davis, Mendikoetxea and Uribe-Etxebarria1997) that all predicates are based on roots that combine with a single internal argument.Footnote ^[29] Surface verbs are formed by combining a root with one of a number of v/Voice $^{0}$ heads (often of the form -C in Chuj); we saw in Section 3 that √ TV, √ POS, and √ NOM roots may all form agentive intransitive stems with the v/Voice $^{0}$ head -w.

Here, we focused particularly on transitive roots, which may be identified by their unique ability to form transitive surface stems with no overt derivational suffix (Section 2.2). Options for transitive roots, along with their argument structural properties, are summarized in (78). For each v/Voice $^{0}$ head, I list the nature of the internal and external arguments, where present, along with the functional head responsible for licensing $/$ agreement. Crucially, note that while we find variation in three out of four columns, what is common to all constructions is the presence of some type of internal argument. As in (77), this is an inalterable property of the root itself.

In a full transitive construction as in (78 $/$ 79a), the null v/Voice $^{0}$ head introduces the external argument and assigns it ergative case. Infl $^{0}$ licenses the internal argument, the transitive object (see Coon et al. Reference Coon, Mateo Pedro and Preminger2014 for discussion). In both regular and agentless passives in (78 $/$ 79b–c), the internal argument is the only overtly present argument, and is licensed by finite Infl $^{0}$ . These two passive heads differ insofar as the v/Voice $^{0}$ head -ch introduces an implicit agent, while -j does not. The implicit agent need not be licensed, but must be existentially bound by the suffix -aj, obligatorily present in (79b). Finally, we found two types of antipassive with -w, as in (79d–e). In both, the -w head introduces an external argument but does not assign it ergative case. The external argument must then be licensed by Infl $^{0}$ . Full DP internal arguments are therefore impossible; the internal argument must be either a bare NP (79e) or an implicit argument that does not need case (79d). In the latter ‘absolutive antipassive’ construction, the suffix -aj is required on the stem and a PP adjunct is possible.

Again, there is no option in which the transitive root selects no complement. Transitive roots like jax ‘grind’ always semantically compose with an internal argument, whether as a full DP argument via Functional Application, as in the transitive and passive constructions from (79a–c), an incorporated NP via Restrict, as in (79e), or an implicit argument, detectable by the suffix -aj from (79d).

We focused specifically on the v/Voice $^{0}$ head realized by the suffix -w. This particular head is instructive because when it attaches to a transitive root an apparent conflict arises: there are not enough licensing heads in the derivation to permit a DP internal argument (Infl $^{0}$ must license the external argument DP), but the transitive root nonetheless must semantically compose with something. The fact that -w stems formed from nominal and positional roots do not combine with an internal argument to create event predicates provides evidence that this is a property of the root itself. This is further corroborated by the appearance of the suffix -aj in absolutive antipassives – the only stem to apparently lack an internal argument. Based on a comparison with two types of passive constructions, I proposed that this morpheme is an overt reflex of the existential binding of an implicit object.

5.2 Implications and avenues for future work

This work fits in line with a larger body of literature that takes roots to be not completely specified for their argument structure or ‘thematic grids’. Rather, argument structure is at least partially determined during the syntactic derivation (see, e.g., Halle & Marantz Reference Halle and Marantz1993, Arad Reference Arad2003, Borer Reference Borer2005, Lois & Vapnarsky Reference Lois, Vapnarsky, Lois and Vapnarsky2006, Lois Reference Lois, Shklovsky, Mateo Pedro and Coon2011, Harley Reference Harley, D’Alessandro, Franco and Gallego2017, among many others). However, in line with works such as Alexiadou et al. (Reference Alexiadou, Anagnostopoulou, Schäfer and Frascarelli2006) and Levinson (Reference Levinson2007, Reference Levinson2014), roots in Chuj clearly have more information than just loose semantic content (contra Borer Reference Borer2005, Acquaviva Reference Acquaviva2009, and others; see discussion in Levinson Reference Levinson2014). Minimally, roots belong to classes, which may be at least partially distinguished based on semantic type.

In Chuj, we saw that roots that belong to the class of transitives obligatorily semantically compose with an internal argument – the internal argument cannot be removed or suppressed in later stages of the derivation, although the manner in which it is realized depends on higher functional heads. This is in keeping with proposals such as that in Koontz-Garboden (Reference Koontz-Garboden2009) on the impossibility of removing arguments in later stages of the derivation, and Alexiadou et al. (Reference Alexiadou, Anagnostopoulou, Schäfer and Frascarelli2006), Wood (Reference Wood2015), and others who argue that voice alternations (like passive ${\sim}$ anticausative) are not derived one from the other, but rather all are derived from the root. The focus in much of this previous work is on external arguments, but Chuj provides evidence that the same holds true for internal arguments. This also lends support to proposals in which roots directly compose with internal arguments, as in Harley (Reference Harley2014).

Here, the focus is on roots rather than surface stems – two categories that are easily distinguishable in a language like Chuj, but more difficult to distinguish in English and other languages without rich verbal morphology. Although the implementations are different, the proposals here about differences among root classes find similarities in discussion of lexical categories in works such as Baker (Reference Baker2003). For Baker and others, verbs are lexical items that combine directly with an argument (see also Davis (Reference Davis, Mendikoetxea and Uribe-Etxebarria1997)), while nouns are those lexical items that introduce a referential index. This general approach is compatible with the proposals regarding √ TV,  √ ITV, and √ NOM roots discussed above.

The large classes of √ POS roots found across Mayan – and, as Henderson (Reference Henderson2017) points out, the correspondingly small classes of true adjectives – make it clear that different languages may make different choices about how the conceptual space is carved up into root classes, and even what these root classes may be. Nonetheless, it is an interesting topic for future work to consider what Mayan languages might tell us about root classes and categories in other languages. Levinson (Reference Levinson2014), for example, distinguishes at least three classes of verb roots in English: Root Creation Verbs (‘braid’, ‘pile’), Explicit Creation Verbs (‘build’, ‘bake’), and Change of State Verbs (‘open’, ‘freeze’). Although a more systematic study is needed, the first type never seem to belong to the √ TV class in Chuj (or Kaqchikel; Henderson, personal communication); instead, forms with equivalent meanings surface as derived transitive stems (Section 2.2), derived from nominal or positional roots. Among other differences between the latter two categories, Levinson shows that theme arguments are required of Change of State verbs, but optional for Explicit Creation Verbs. If we can take Chuj’s morphology to be instructive more generally, perhaps the theme in Explicit Creation Verbs is present implicitly – and if English had an -aj suffix, we would find it exactly here.