2.1. A note on data sources and elicitation methodology

Kaqchikel is a Mayan language of central Guatemala from the K’ichean-Mamean (Eastern) branch. It is described as ‘vulnerable’ (Moseley Reference Moseley2010), as it is spoken by approximately 410,000 speakers, most of whom are bilingual in Kaqchikel and Spanish (Eberhard, Simons & Fennig Reference Eberhard, Simons and Fennig2022) and vary with respect to language dominance. The degree of inter-speaker variation is generally high, even among people living in the same area and with similar socioeconomic backgrounds (Patal Majzul, García Matzar & Espantzay Serech Reference Patal Majzul, Matzar and Serech2000).

Lyskawa & Ranero (Reference Lyskawa and Ranero2022) highlight the importance of studying individual grammars in their examination of microvariation in verbal agreement patterns among speakers of Santiago Tz’utujil, a language closely related to Kaqchikel. Their work cautions against broad generalizations across speakers, particularly perilous in communities with widespread unbalanced bilingualism. Following their approach, we conducted in-depth examinations of individual speakers’ grammatical knowledge over an extended period.

The data presented in this study, unless otherwise noted, come from three native Kaqchikel speakers between 30 and 45 years of age. All consultants are bilingual in Kaqchikel and Spanish, reside in Patzún (Chimaltenango department, Guatemala), and have comparable educational backgrounds. They use Kaqchikel actively in their daily lives. To ensure reliability, we tested each construction multiple times with each speaker using different lexicalizations. Additionally, we compared our findings with existing descriptive literature, including grammars and dictionaries, documenting any discrepancies.

2.2. Basics of Kaqchikel (morpho)syntax

Kaqchikel is a head-marking, morphologically, and syntactically ergative language, with common subject and object pro-drop (1). Grammars state that the unmarked word order in Kaqchikel is VOS (verb-object-subject); however, speakers of the Patzún variety prefer the SVO (subject-verb-object) order (García Matzar & Rodríguez Guaján Reference García, Pedro and Guaján1997, Patal Majzul et al. Reference Patal Majzul, Matzar and Serech2000, Clemens Reference Clemens and Kenstowicz2013). There are several derivational pathways to verb-initial orders (Clemens & Polinsky Reference Clemens, Polinsky, Everaert and van Riemsdijk2017, Clemens & Coon Reference Clemens and Coon2018), but these are not essential for our discussion.

Kaqchikel nominal phrases bear no case-marking, and only a handful of nouns can be marked for plurality. The rich verbal morphology makes up for the lack of nominal marking. In finite clauses, both the subject and the direct object are cross-referenced with agreement prefixes on the verb, as in (1), with the standard order of morphemes being tense/aspect-absolutive(-ergative)-root(-caus-pass/ap/tv). Table 1 shows the agreement prefixes; using the standard notation adopted in Mayanist literature, we gloss ergative as (Set) A and absolutive as (Set) B. Tense is not expressed separately, and a single Tense-Aspect prefix is used instead, usually glossed as either completive/incompletive (García Matzar & Rodríguez Guaján Reference García, Pedro and Guaján1997; Patal Majzul et al. Reference Patal Majzul, Matzar and Serech2000; Patal Majzul Reference Patal Majtzul2007; Henderson Reference Henderson2012; Heaton, Deen & O’Grady Reference Heaton, Deen and O’Grady2016, among others) or perfective/imperfective (Imanishi Reference Imanishi2014); we use the former notation.

Table 1. Person–number markers in Kaqchikel

Kaqchikel predicates form two groups: non-verbal (prohibiting Tense-Aspect marking) and verbal (requiring Tense-Aspect marking). The former group includes stative positionals (Pye Reference Pye, Shklovsky, Pedro and Coon2011, Armstrong Reference Armstrong2017), adjectival, and nominal predicates. Verbal predicates include simple intransitives, simple (underived) transitives, derived transitives with the suffix -j, and derived intransitives (antipassives, passives).

In this paper, we focus on the verbal-predicate type, illustrated in (2). Intransitive verbs can be divided into unergatives and unaccusatives. This division is supported by difference in agreement. In a nutshell, third-person plural agreement with an external argument is obligatory, while such agreement with an internal argument is optional (Burukina Reference Burukina2021); consider (2). In that regard, Kaqchikel is similar to Santiago Tz’utujil, a closely related K’ichean language (Levin, Lyskawa & Ranero Reference Levin, Lyskawa and Ranero2020; Lyskawa & Ranero Reference Lyskawa and Ranero2022). It is worth noting that the unergative/unaccusative distinction is not uniformly represented across Mayan languages. For instance, in Ch’ol (Western branch), all intransitive verbs are unaccusative (Coon Reference Coon2013), and covert transitives (see Hale & Keyser Reference Hale, Keyser, Hale and Keyser1993) correspond to what can be considered simple unergatives in other languages.

Mayan languages are divided into two core types with respect to absolutive case-assignment: low-absolutive and high-absolutive languages (Tada Reference Tada1993; Coon, Mateo Pedro & Preminger Reference Coon, Pedro and Preminger2014). In low-absolutive languages, absolutive is licensed in the verb phrase and is therefore available in non-finite forms, while in high-absolutive languages, the finite inflectional head is responsible for absolutive licensing; hence, absolutives do not occur in non-finite structures. This difference in licensing correlates with the order of morphological exponents in the VP. Kaqchikel instantiates the high-absolutive type. Infl uniformly assigns absolutive case downward, either to the sole DP argument or, in active transitive clauses, to the internal DP argument. Ergative is assigned lower in the structure, by a transitive Voice/v (see Aldridge Reference Aldridge2004 and Legate Reference Legate2008 for this approach beyond Mayan).

There is no unanimous perspective on the precise composition of the clausal spine in Mayan transitive clauses, in particular concerning the range of projections in the vP. Under a single vP approach, the VP is dominated by a single verbal projection vP and the external argument is base-generated in spec,vP, where it receives ERG from v in a spec-head configuration (see Coon Reference Coon2017 for a detailed demonstration of this approach).

A differentiated VoiceP-vP approach assumes that the VP is dominated by two separate projections – VoiceP and vP – and that the external argument is base-generated in the lower projection but is licensed by a higher functional head. Assuming that VoiceP and vP are both present, they can be represented as split or as bundled (see Pylkkänen Reference Pylkkänen2008 for the original proposal). Imanishi (Reference Imanishi2020), Burukina (Reference Burukina2021), and Ranero (Reference Ranero2021) adopt a split approach to Kaqchikel, while Coon, Baier & Levin (Reference Coon, Baier and Levin2021) propose that Mayan languages have a bundled VoiceP/vP head, which they represent as vP for simplicity.

3.1. Kaqchikel antipassives: Main types

Navigating the antipassive landscape in Mayan is challenging because of rampant differences between descriptive and theoretical vocabularies, between Spanish and English labels, and across individual descriptions. Our aim in this section is to bring some order to this descriptive chaos. In what follows, we rely mostly on the existing descriptive literature, including García Matzar & Rodríguez Guaján (Reference García, Pedro and Guaján1997); Patal Majzul et al. (Reference Patal Majzul, Matzar and Serech2000); and McKenna Brown, Maxwell & Little (Reference McKenna Brown, Maxwell and Little2006).

Kaqchikel has four constructions that traditional grammars describe as antipassive: absolutive, focus, oblique, and incorporating antipassive.

In the absolutive antipassive (Spanish antipasivo absoluto), the verb is detransitivized, and the object is not expressed; the sole remaining subject is cross-referenced by an absolutive prefix. This antipassive is marked with -on/un on underived transitive verbs and -n on derived transitive verbs, where that suffix replaces the derived transitive status suffix -j. Absolutive antipassivization is highly productive in Kaqchikel (Patal Majzul Reference Patal Majtzul2007).Footnote ² In what follows, we refer to this type as null antipassive (AP_null).

In the focus antipassive (Spanish antipasivo de enfoque), the object is expressed as an independent noun phrase; similar to the direct object in the active transitive, it appears to be a full DP as it can contain a determiner, a possessor, modifiers, etc. However, unlike regular transitives, focus-antipassive verbs only take an absolutive prefix; this prefix can index either subject or object, instantiating so-called omnivorous agreement: agreement with the argument bearing a higher person-feature value (Preminger Reference Preminger2014). The predicate in this construction is marked with -o/u for underived transitives, and with -n for derived transitives. As this short description shows, the traditional name of this construction is misleading, as it has as much in common with transitive clauses (see also Heaton Reference Heaton2017). As a reflection of this parallel with transitives, focus antipassive has been referred to as Agent Focus (AF) in the formal Mayanist literature, the term we use below.

A typical claim is that AF is restricted to contexts involving A-bar extraction of the Agent, although this restriction needs to be qualified in at least two respects. First, AF does not occur under topicalization; in Patzún Kaqchikel, the SVO order with a subject-topic is usually understood as non-derived (i.e., not involving A-bar movement), and it does not require special marking on the verb. Second, literature reports that AF can sometimes be used without agent fronting (Patal Majzul et al. Reference Patal Majzul, Matzar and Serech2000, Ajsivinac & Henderson Reference Ajsivinac and Henderson2010); however, we have not been able to observe such occurrences.

AF has received significant attention in theoretical literature; for discussion, see Clemens (Reference Clemens and Kenstowicz2013), Erlewine (Reference Erlewine and Kenstowicz2013, Reference Erlewine2016), Preminger (Reference Preminger2014), Henderson & Coon (Reference Henderson and Coon2018), and Ranero (Reference Ranero2021) on Kaqchikel; Coon et al. (Reference Coon, Pedro and Preminger2014) on Q’anjob’al; Aissen (Reference Aissen, Ostrove, Kramer and Sabbagh2017b) on Tzotzil; and Stiebels (Reference Stiebels2006) and Coon et al. (Reference Coon, Baier and Levin2021) on Mayan AF in general.

García Matzar & Rodríguez Guaján (Reference García, Pedro and Guaján1997: 376) use the term antipasivo de enfoque to refer to both AF constructions and constructions where the logical object is expressed by a PP with a relational noun ((i)chin in Kaqchikel). They appear to use this umbrella term because the verb in the two types of contexts is marked the same and because the logical subject in sentences with an (i)chin-object is often focus-fronted, similar to the logical subject in AF constructions. Likewise, Patal Majzul et al. (Reference Patal Majzul, Matzar and Serech2000) suggest that (i)chin-PPs are used when the agent is focus-fronted.Footnote ³ However, McKenna Brown et al. (Reference McKenna Brown, Maxwell and Little2006) mention no connection between focalization and antipassive verbs with an oblique object. To avoid the confusion between AF and the construction whose logical object appears as a PP headed by the preposition (i)chin, we will use the neutral descriptive term oblique antipassive (AP_obl) to refer to the latter (see more in Section 3.2).

Finally, some grammars distinguish incorporating antipassive (Spanish antipasivo de incorporación); e.g., García Matzar & Rodríguez Guaján (Reference García, Pedro and Guaján1997: 380). Its object is a bare noun, and only the agent is cross-referenced on the verb by an absolutive marker. The predicate is typically marked with -o/u for underived transitives, and with -n for derived transitives, similar to AF; however, McKenna Brown et al. (Reference McKenna Brown, Maxwell and Little2006: 181) provide examples where an underived transitive predicate with a bare object is marked -on, as in AP_null.

3.2. Patzún Kaqchikel data

The data obtained from our consultants confirm that they consistently draw a line between AP_null and AF. Both constructions are highly productive, and so far, we have not found a transitive predicate that would not be acceptable in either of them.

In AP_null (always marked -Vn in Patzún Kaqchikel), the object must be absent. No Agent extraction is required. In AF (marked -Vn on derived stems and -o/u with CVC roots), the object either is a full DP or remains unpronounced. In our data, the Agent in AF is always fronted. In what follows, we only provide examples with CVC roots, since in such cases AP_null and AF are morphologically distinct.

Our data support Heaton’s (Reference Heaton2018) observation that the Patzún dialect is one of the most conservative in morphologically distinguishing AP_null and AF (in other varieties, the -n/o distinction is neutralized).

We observed significantly more inter- and intra-speaker variation in the distribution of the oblique antipassive. None of our consultants produced spontaneous examples with an object introduced by a relational noun, either when speaking Kaqchikel all along or when asked to translate examples from Spanish. The main reason for that has to do with the independent use of the relational noun (i)chin to introduce a possessor. For example, in (6), the possessor either is cross-referenced by a Set A prefix on the possession or is embedded inside the (i)chin phrase, in which case the Set A marker is absent.

Data from Patal Majzul (Reference Patal Majtzul2007) and a Bible corpus (courtesy of Robert Henderson) indicate that (i)chin possessors are highly productive. Consequently, when presented with examples of detransitivized predicates with (i)chin objects, consultants rejected such sentences, insisting that (i)chin phrases denote entity ownership. Heaton (Reference Heaton2017) and Ranero (Reference Ranero2021) reported that their consultants did not find oblique AP problematic in the elicitation context and argued that the (i)chin AP should be grouped together with AF (see also the discussion of oblique antipassives in Section 3.1 above).

Similarly, only a couple of examples of incorporating antipassive were accepted by the native speaker consultants, both with the AP -on and the AF -o marking on the verb; we mention those in Section 4.3. For now, we will focus on the AP_null vs. AF distinction, which is productive in Patzún Kaqchikel.

4.1. Diagnostics of implicit arguments

4.1.2. Extra-linguistic reference and discourse reference

The unpronounced object in AF can receive a definite, specific, or deictic interpretation, as it must be identified with either an extra-linguistic referent or a referent previously mentioned in the discourse, similarly to unpronounced objects in active transitive clauses. Generally, there is a preference to interpret the null object in AF deictically, as shown in (7) and several subsequent examples. In contrast, the unpronounced object of AP_null cannot be understood as referring to a specific person/object and is interpreted existentially, which makes it comparable to simple non-specific indefinites (7c).

Unpronounced objects in AF can have a discourse referent, but no such discourse reference is possible in AP_null, as further shown by the contrast in (8). The null object can refer back to a shirt in AF (8a) but not in the antipassive (8b), which makes the continuation that the girl lost the shirt infelicitous in the latter case.

Next, consider differences between AP_null and AF with respect to follow-up questions. Recall that null objects in AF tend to be interpreted deictically ((7b) above). In an AF sentence, it is redundant to ask about the identity of a deictic object, as the object’s identity is already established. The purport of example (9) is that Pedro purchased an animal in plain sight, so no guesswork is involved.

In contrast, the antipassive in (10) describes the general action of shopping, and there is no indication of items that changed hands, which makes asking about such items felicitous.

Negation is another context where the contrast between the unpronounced objects in AF and AP becomes noticeable. The base active transitive example is given in (11a); here, even a missing object is still interpreted as definite and specific, which makes the contrastive continuation felicitous. A similar pattern is observed with AF in (11b). However, if the first sentence has the antipassive (11c), it is interpreted as being about weaving in general (‘she does not weave at all’), and the follow-up statement is perceived as a contradiction.

4.1.3. Reference to a disjunctive antecedent

Related to the discourse-reference pattern just discussed, certain unpronounced objects can pick a referent from a disjunction presented in the preceding discourse, matching the choice implied in the antecedent (Cyrino & Lopes Reference Cyrino and Lopes2016, Sakamoto Reference Sakamoto2016, Landau Reference Landau2018).

Consider the following example with two transitive predicates. An unpronounced object in the second clause is interpreted as referring to one of the stated options in the disjunction in the first clause:

In contrast, if the second clause is antipassive, no reference to the disjunction introduced in the first clause is possible. In (13), the reading that Señora Ixkat will weave the same thing as Señora Nikte is unavailable.

Checking the interpretation of the unpronounced object in AF in disjunctive contexts is impossible because AF is incompatible with chuqa’. We hypothesize that the reason for this empirical gap lies in pragmatics. Recall that AF in Kaqchikel serves to (contrastively) focus the agent; meanwhile, chuqa’ brings in the (contrastive) topic interpretation, and the two readings clash.

4.1.4. Paycheck pronouns

So far, discourse/extra-linguistic reference data showed that unpronounced objects of AF behave as regular anaphoric/deictic pronouns, similar to silent objects of active transitives. In addition, these objects allow a bound variable reading and can function as paycheck pronouns. Consider the context where two women, a grandmother and a young girl, each made an huipil, after which the grandmother sold her huipil, and the girl lost hers. This situation can be described as in (14).

The objects of the transitive verbs in (14b) can be null, as shown in (15). The strict reading is preferred (i), but a bound-variable reading (ii) is also available.

The AF construction with an unpronounced object is also felicitous (16); the context here has to be slightly modified, since AF requires that the Agent be focus-fronted.

In contrast, AP is incompatible with the context presented here. Using it implies that the grandma sold her huipil but the girl just lost something. This incompatibility follows from the observation that there is no corresponding object to relate to the one in the transitive construction.

Again, we see parallels between transitives and AF and another point of difference between AF and AP.

4.1.5. Depictive licensing

Both overt and null objects of active transitive and AF predicates alike can license depictives: secondary predicates describing the state or condition of a participant concomitant with the time of the main event. This supports the idea that such null objects are present in syntax and indicates that they are so-called strong unpronounced arguments – namely, null pronouns, pros – and not structurally smaller weak silent φPs or NPs (see Landau Reference Landau2010 for arguments that only strong implicit arguments can serve as subjects of predication, including secondary predicates). Consider (18), where the baseline sentence (18a) shows that a depictive can modify a direct object. Example (18b) with a transitive verb and a silent object is ambiguous because the depictive can be interpreted as modifying either the subject or the object. The same ambiguity is attested in AF (18c).

Unpronounced objects of AP_null, however, do not allow modification by a depictive predicate. Thus, the only reading available in (19) is the one where the depictive modifies the subject (Maria).

4.1.6. Scope effects

Indefinite objects of transitive verbs can take wide scope over VP-adverbs, as in the following example, where ‘one huipil’ scopes over ‘again’:

A transitive clause and an AF clause with an unpronounced object receive the same interpretation: the unpronounced object scopes over the adverbial expression.

In the antipassive construction, the interpretation is just the opposite, implying that Señora Nikte was again engaged in weaving and made different items each time:

4.1.7. Implicit-argument diagnostics inapplicable in Kaqchikel

The following two tests are often used to diagnose the syntactic presence of an unpronounced argument: control of an embedded PRO (as in English The doctor recommends <ec _i> [PRO _i to eat more vegetables]) and binding of anaphoric pronouns (The town council helps <ec _i> with self’s _i problems); see Rizzi (Reference Rizzi1986) and Bhatt & Pancheva (Reference Bhatt, Pancheva, Everaert and van Riemsdijk2006). However, these tests are untenable in Kaqchikel.

When it comes to control, there are no infinitives in Kaqchikel, and most of the predicates that would involve control in other languages embed a finite clause or an oblique nominal dependent. Kaqchikel equivalents of typical object control sentences in English or Spanish usually involve either a causative or a periphrastic construction; for instance, ‘permit’ = ‘give permission’ (the Permission Holder is encoded as an oblique Recipient).

As for binding, Kaqchikel reflexives and reciprocals (encoded as the base -i’ with a Set A marker that cross-references the antecedent) are restricted to the direct object position and cannot be used as possessors or be included in an oblique construction headed by a relational noun (Burukina Reference Burukina, Stockwell, O’Leary, Xu and Zhou2019). Because of this, it is impossible to construct an example where an unpronounced direct object could, in principle, c-command such a pronoun.

The diagnostics available in Kaqchikel (discourse reference, depictive modification, scope) are sufficient to point to robust differences between unpronounced objects in transitive and AF constructions on the one hand and AP constructions on the other. These differences are summarized in Table 2, which shows that unpronounced objects in active transitives and AF pattern together, whereas the antipassive is different.

Table 2. Properties of unpronounced objects in Kaqchikel

Having established systematic differences across the two kinds of unpronounced objects, we now turn to the syntactic status of these objects.

4.2. The status of unpronounced objects in antipassives: pros, φPs, NPs, or absent?

4.2.1. Small or absent?

The results presented above confirm that unpronounced objects in active transitive clauses and AF clauses are silent personal pronouns – pros (see Epstein Reference Epstein1984, Rizzi Reference Rizzi1986, Borer Reference Borer, Lapointe, Brentari and Farrell1998, among others, on unpronounced arguments as pros). At the same time, the status of the unpronounced object in AP is less clear. Two analytical options are available: (i) AP unpronounced objects are syntactically present deficient φPs (or even smaller NP/Ns; see below), which would be consistent with their indefinite/non-specific interpretation, and (ii) AP unpronounced objects are not projected at all. Based on several considerations, we argue for the latter approach.

First, a φP-analysis does not find empirical support in Kaqchikel. φPs are assumed to be ‘variables whose value is constrained by the value of its φ-set’ (Landau Reference Landau2010: 383). φPs can be used as deictic or anaphoric (Déchaine & Wiltschko Reference Déchaine and Wiltschko2002, Landau Reference Landau2010); however, AP unpronounced objects never receive a definite reading but are only interpreted existentially. Additionally, φPs are expected to exhibit some featural specification (gender, person, number; Šereikaitė Reference Šereikaite2022). This is not what is observed in Kaqchikel.

One might propose that AP unpronounced objects are structurally smaller than φPs and consist of a single silent nP/NP/N. Such an approach may find support in occasional examples of AP_null with a bare object reported in the literature, such as pon-on wäy ‘bake-AP tortillas’ or b’an-on xajab’ ‘make-AP sandals’ (McKenna Brown et al. Reference McKenna Brown, Maxwell and Little2006: 181). However, except for stable collocations (23), our consultants would not accept such examples (see Buenrostro Reference Buenrostro2013 on root-NP complement combinations receiving a special meaning).Footnote ⁷

We propose that in (23), a bare NP is merged in the complement of the VP, but it does not function as an argument; see Maxwell (Reference Maxwell1976) and recently Coon (Reference Coon2019) for a similar proposal on Chuj antipassives. Instead, the NP of the type <e,t> forms a complex predicate with the lexical verb. The requirement to form a complex predicate may explain why only some, mostly lexicalized, combinations AP_null + N are allowed.

4.2.2. Absent in syntax and/or in semantics?

Our proposal, whereby the internal argument in antipassives is not projected in the structure, naturally leads to the following question: What allows that object to be absent? Consider, for instance, the classical Theta Criterion, according to which a thematic role must normally be assigned to an argument present in syntax. The absence of an otherwise expected argument can be accommodated in several ways. We discuss them below, arguing ultimately that the best explanation is that AP clauses lack an internal argument entirely.

The object argument is part of the theta-grid. Assuming that the object argument is still required by the theta-grid of the predicate (i.e., there is a thematic role to be assigned), the following two accounts may be proposed. In the first scenario, the argument remains unsatisfied (in the sense of E. Williams Reference Williams1985), and the thematic role remains unlinked. A problem with this option is that it is not entirely clear how to derive the strictly existential reading of the unpronounced object and to ensure that the result is interpretable.

Alternatively, as proposed by some researchers, the argument variable is existentially bound. In this case, a number of analyses include the assumption that the existential closure is performed by some functional head in syntax; see Bruening (Reference Bruening2013) on the existential closure of external arguments in English passives and Coon (Reference Coon2019) on passives and antipassives in Chuj. To summarize, these options are as follows:

The object argument is not part of the theta-grid. One could also assume that the theme/patient argument is absent at both the syntactic and semantic level; in other words, it is not in the theta-grid of the predicate. This in turn can lead to two different scenarios. Unlike (24), the alternatives in (25) do not correspond to different mechanisms that can explain the absence of an argument within the same framework but rather represent two different approaches to the nature of thematic relations, which, following A. Williams (Reference Williams2015), can be identified as projectionist and separationist. Under the assumption that relations are projected by the lexical verb itself, we may be dealing with homonymous transitive/intransitive predicates or a single predicate that undergoes a valency-changing transformation already in the lexicon (see Reinhart & Siloni Reference Reinhart and Siloni2005). In other words, the theta grid of the AP predicate is inherently different from that of a transitive predicate, and the former does not have an argument linked to the Theme/Patient role (25a). Such an approach, however, is challenged by the high productivity and regularity of Kaqchikel AP; recall that practically every transitive verb can be antipassivized and thus would need to have a lexical twin.

An alternative is that the entailed relations are not necessarily realized by a corresponding constituent at both the syntactic and semantic level of representation. This can be modeled under a separationist approach to the argument structure, whereby thematic relations are separated from the main event encoded in the verbal predicate and are instead introduced as separate predicates (see A. Williams Reference Williams2015 for a detailed discussion and Pietroski Reference Pietroski2018 for an implementation in Fodorian semantics).Footnote ⁸ These predicates are then coordinated with the main predicate by means of conjunctive semantics. In sum, the object is not a content argument of the verb; it is added to the verb in the transitive and AF configuration but not in AP (25b).

For simplicity, we adopt the ‘full absence’ approach (25b), under which the internal argument in AP clauses is absent at both syntactic and semantic levels, as this modeling option is the most economical. At this point, nothing particularly hinges on that choice. In Section 6.1, we briefly discuss an alternative existential closure approach, as proposed by Coon (Reference Coon2019) for Chuj. While this approach encounters certain challenges when applied to the Kaqchikel data, it is not incompatible with our analysis.

4.3. Reconsidering the antipassive

The considerations presented in this section suggest that antipassives are inherently intransitive, with no internal argument projected. If this result is on the right track, it suggests that an approach that treats antipassivization as the ‘demotion’ of an internal argument is untenable for Kaqchikel. Beyond Kaqchikel, our results also necessitate reevaluating such a dominant view of the antipassive.

While many scholars define antipassivization primarily by its effect on the internal argument (Polinsky Reference Polinsky, Coon, Massam and Travis2017: 309; Basilico Reference Basilico2019: 192; Heaton Reference Heaton2020: 132), our analysis suggests that any ‘demotion’ of the internal argument is merely incidental rather than definitional. These findings align with both Coon (Reference Coon2019) and Aldridge (Reference Aldridge2012); the latter proposes ‘to connect antipassive to syntactic intransitivity, rather than forcing it to be analyzed as a derived construction in which the internal argument has been demoted to adjunct status’ (p. 195).

A crucial argument in support of this position comes from the fact that unaccusatives, which have a quintessential internal argument but lack an external one, never antipassivize. Back to Kaqchikel, the following data point illustrates this gap.Footnote ⁹

Beyond Kaqchikel, we have not found any examples of antipassivized unaccusatives either (assuming that a given language has structural diagnostics of unaccusativity). Based on our discussion so far, we conclude that this is a principled typological gap rather than an accidental lack of data attestation. We contend that it follows from the properties of antipassive, whose function is to manipulate the EA, not the internal one.

As an interim summary, we have argued that objects are not projected in the syntax of antipassive constructions. The antipassive is therefore a genuine intransitive construction, one lacking an internal argument. The interpretation involving an internal argument is just an entailment of the intransitive predicate. This is in line with other researchers who have also emphasized the intransitivizing function on the antipassive: Aldridge (Reference Aldridge2012), Coon (Reference Coon2019), Heaton (Reference Heaton2017, Reference Heaton2020), and Chung (Reference Chung, Clemens, Gribanova and Scontras2025).

We now turn to the differences between the syntactic structure of the antipassive (and comparatively, AF) and that of the transitive. The differences are twofold: the composition of the verbal spine and the status of the external argument in clausal structure.

5.1. Intransitive and transitive predicates

We propose that both antipassive and AF clauses lack the Voice projection altogether and that v_ITV and v_AF serve two functions: introducing a new relation corresponding to the external argument (EA) and projecting a syntactic argument. The resulting vP is fully saturated and is merged as a complement of a higher functional head outside of the thematic domain (recall that, following Mayanist literature, that head is Infl).

While AP_null and AF both lack a VoiceP, their respective v heads differ in case-assigning properties. The v_AF head takes as its complement a VP with an internal argument, expressed as either an overt DP or a pro, which it licenses. Note that only one absolutive morpheme appears on the verb, which can be explained in terms of a general mechanism governing morphological realization of person and number features (consider Watanabe Reference Watanabe2017). In contrast, the antipassive v head is incompatible with an internal argument; thus, its properties are identical to that of the unergative v head. It is worth noting that most unergative stems in Kaqchikel end with -Vn (28), the same exponent as in the antipassive.Footnote ¹⁰ Similar to AP_null, unergatives in Kaqchikel are incompatible with (cognate) objects, which can be explained by the absence of Voice and Infl being the only source of Case licensing.

Unlike antipassive and AF clauses, transitive clauses involve a larger thematic structure, as outlined in (29). Here, VoiceP dominates v_TV and projects an EA that saturates the generalized Agent role.

A question remains: How does v_ITV ‘know’ to combine with a VP without an internal argument, while v_AF and v_TV always combine with a VP with an internal argument? Our answer lies in syntax. It has been proposed that the transitive and the AF v is equipped with a special feature that triggers the raising of the internal argument (e.g., Coon et al. Reference Coon, Baier and Levin2021); if the latter is not projected, the feature cannot be checked, and the derivation crashes. In contrast, the antipassive v lacks such a feature, which in principle makes the presence of an internal argument optional. At the same time, if the internal argument of AP_null were projected as a DP or pro, it would remain unlicensed because of v_AP’s Case deficiency. Hence, the only option is that the internal argument is absent. (We assume a Case-licensing approach to DPs, but this proposal can also be easily modified to an account whereby DPs are licensed via agreement.)

The novelty of our proposal is that we extend the Voice-less analysis to all morphosyntactically intransitive predicates (except for passives; see below). Another key component of our approach is the two base positions available for EAs – spec,vP, as in (27), and spec,VoiceP, as in (29). This flexibility may seem to over-complicate the analysis if one takes into account only the active transitive, AP_null and AF clauses. However, as we expand the dataset and consider together all the patterns of verbal derivation attested in Kaqchikel, our approach captures them in a more straightforward and elegant way, compared to alternative analyses whereby the vP-VoiceP are bundled together and/or all the EAs are always base-generated in the same position.

5.2. The inventory of v and Voice heads

The following idea informs our analysis of the thematic domain: Only v is capable of introducing a new argument at the semantic level, even when that head does not project it in syntax.Footnote ¹¹ In contrast, the main function of Voice is to manipulate the preexisting argument structure, specifically targeting the EA role (see Pylkkänen Reference Pylkkänen2008 and Harley Reference Harley2013, Reference Harley, D’Alessandro, Franco and Gallego2017 for conceptually similar ideas). This entails that VoiceP is only added to the structure when there is an unsaturated relation; otherwise, its presence is redundant and results in an uninterpretable structure. Accordingly, v and Voice emerge as conceptually distinct functional categories.

The contrast between vP and VoiceP is reminiscent of the PossP-FP/AgrP split in the nominal domain (consider É. Kiss Reference É. Kiss2002 and Dékány Reference Dékány2018 on the structure of Hungarian possessive DPs). In these environments, the lower head is semantically enriched: v and Poss both introduce a new thematic relation. In contrast, the higher functional head is purely syntactic in nature, designed to license an argument. Additionally, both the verbal domain and the nominal domain appear to be parameterized: While many languages require a split structure, some prefer to bundle the relevant projections (consider Pylkkänen Reference Pylkkänen2008; Harley Reference Harley2013, Reference Harley, D’Alessandro, Franco and Gallego2017; and Coon et al. Reference Coon, Baier and Levin2021 for Mayan; see also footnote 16).

The proposed inventory of v and Voice heads is shown in Table 3. To account for the selectional properties of a particular head, we adopt the system put forward by Bruening (Reference Bruening2013), whereby a head is equipped with certain selectional features that it needs to check by combining with dependents of particular categories. For example, [S:V,N] means that the functional item combines with a complement of the verbal category and further requires a nominal dependent in the specifier position. Every v takes a root projection as its complement, which we mark as VP, for simplicity.

Table 3. The inventory of v and Voice

As outlined in Section 5.1, v_ITV and v_AF (the intransitive and the AF functional heads) introduce a new agentive argument in semantics and require that a DP be merged in their specifier position to saturate that thematic role. In addition, v_AF is equipped with the case feature [ABS], which it checks with a c-commanded nominal. Treating the antipassive v and v_AF as distinct functional elements is critical because some Mayan languages have antipassives but no AF (the opposite is possible but not attested in Mayan). For instance, within the K’ichean branch, v_AF appears to be absent in Q’eqchi’; yet Q’eqchi’ has antipassives (Berinstein Reference Berinstein, Hinton and Munro1998). To complete the picture, the inventory includes the unaccusative functional head, v_Unacc: an unspecified v, which can be construed as a verbalizer sensu stricto (recall that Kaqchikel unergatives/unaccusatives differ in their agreement patterns). While this inventory is constructed to be universal, individual languages may not realize all its members. For example, the set of intransitives may only have v_Unacc, as Coon (Reference Coon2013) proposes for Ch’ol.

Going back to the transitive structure in (29), we propose that v_TV introduces a new thematic relation (Agent), but that head is defective in that it does not project a syntactic argument. This leaves the transitive vP unsaturated and the EA relation must be taken care of for the derivation to be interpretable at LF. Under our proposal, it is Voice that does the job.

Voice manipulates the preexisting EA relation by projecting a DP to match it (or by dealing with an existing variable, as in passives). We distinguish Voice_TV and Voice_Pass. Voice_TV introduces a DP that binds the Agent variable; see Harley (Reference Harley2013). Voice that projects an EA in the specifier position is further equipped with a case feature and assigns ergative to the DP under a spec-head relation (29), in line with accounts that treat ergative as an inherent case, as in Woolford (Reference Woolford1997), Legate (Reference Legate2002, Reference Legate2008), Aldridge (Reference Aldridge2004), Laka (Reference Laka, Johns, Massam and Ndayiragije2006), Coon (Reference Coon2013), Polinsky (Reference Polinsky2016), etc.

The role of Voice_Pass is to existentially close the EA (see Bruening Reference Bruening2013; also, Alexiadou, Anagnostopoulou & Schäfer Reference Alexiadou, Anagnostopoulou, Schäfer and Frascarelli2006 and Alexiadou & Doron Reference Alexiadou and Doron2012 on passive as a type of Voice). This treatment of Voice_Pass is similar to the proposal in Coon (Reference Coon2019) for the Chuj passive.Footnote ¹² As with the inventory of v heads, individual languages may lack some types of Voice; for instance, Basque and Georgian do not have passives.

5.3. Combining v and Voice: Possible and impossible derivations

We argue that VoiceP is added to the structure only when needed, that is, when v introduces a new thematic relation but does not project the required argument. We therefore expect the saturated intransitive vPs (unergative/AP, AF, and unaccusative) to be incompatible with Voice_TV and Voice_Pass. This prediction is borne out.

First, in Kaqchikel (as well as in many other ergative languages), the subject of unaccusatives, unergatives, antipassives, and AF is invariably absolutive.Footnote ¹³ This is unsurprising under the assumption that ERG is assigned by Voice under a spec-head relation. Although in principle Voice_TV can select an intransitive vP as its complement, it will need to project a DP in spec,VoiceP position; this DP will lack a thematic role (since the vP is already saturated and Voice itself does not introduce a new thematic relation), and the derivation will yield an uninterpretable result.

Second, only transitive predicates undergo passivization. Voice_Pass dominates v_TV and manipulates the EA relation; consider (30a), derived as (30b), where Voice_Pass existentially binds the Agent variable.

As (31) shows, passivization is restricted to transitive verbs.

The lack of passives formed from AF predicates is particularly striking. Despite the absence of ergative marking, AF is semantically transitive (Aissen Reference Aissen, Shklovsky, Pedro and Coon2011), with two argument DPs. As we show below, a passive VoiceP can be added to a causativized structure with two DP-dependents. Why could it not be added to AF? Our analysis offers an explanation: Only a transitive or causative vP is ‘deficient’, which in turn means that it allows/requires a VoiceP to manipulate the EA relation. Since v_AF projects an external argument in its specifier position, a vP in AF is fully saturated. Combining it with Voice_Pass is impossible, because the latter must manipulate a preexisting variable.

Finally, the split vP-VoiceP analysis outlined in this paper provides an explanation for why morphological causativization is restricted to unergative and unaccusative predicates.Footnote ^{14 ,} Footnote ¹⁵ Consider some examples of the Kaqchikel morphological causative, expressed with the suffix -isa:

Crucially for our discussion, morphological causativization does not apply to transitive, causativized predicates or to passivized transitive verbs (instead, a biclausal structure has to be used):

Splitting vP and VoiceP allows us to accommodate this restriction by manipulating the selectional properties of v_Caus. Causativizable intransitives are all vPs, and non-causativizable transitives and passives are VoicePs; while vP recursion is allowed, a vP cannot be merged on top of a VoiceP.Footnote ¹⁶ In contrast, an analysis that blends vP and VoiceP together requires a more complex explanation, especially since the restriction cannot be reduced to predicates with one vs. two syntactic arguments (since passives cannot be causativized) and Agentive vs. non-Agentive (as both unaccusatives and unergatives can combine with the causative).

To recapitulate, adding a vP on top of a VoiceP is prohibited, but adding a VoiceP on top of a transitive vP is required. Under the assumption that v_Caus is similar in its properties to v_TV, we expect causative verbs to undergo passivization. This prediction is borne out; such passivization is highly productive and shows no exceptions.

The split vP-VoiceP analysis captures the acceptable transitive, passive, and causative patterns and rules out the ungrammatical examples, while the bundled approach would require serious adjustments. Under the bundled approach, v_ITV and v_AF would remain the same, but v_TV would be responsible for projecting an Agent and assigning ERG to it, combining the functions of our v_TV and Voice_TV (36). This is an essence the analysis that was proposed for Chuj antipassives by Coon (Reference Coon2019); see also Coon et al. (Reference Coon, Baier and Levin2021).

To account for the causativization of unergatives and passivization of causatives, a bundled approach would need to allow vP/VoiceP recursion, which in turn would make causativizing passives and passivizing antipassives available, contrary to the data.

5.4. Relativized external argument positions

The idea that EAs can occupy different structural positions has been articulated by a number of researchers, e.g., Tollan (Reference Tollan2018), Tollan & Massam (Reference Tollan and Massam2022), and Polinsky (Reference Polinsky2016) for Polynesian; Tollan & Oxford (Reference Tollan, Oxford, Bennett, Hracs and Storoshenko2018) for Algonquian; McGinnis (Reference McGinnis, Roy, Boneh, Harbour and Matushansky2022, Reference McGinnis2023) and Nash (Reference Nash, Siegal and Boneh2020) for Georgian; Krishnan & Sarma (Reference Krishnan and Sarma2023) for Malayalam; Anand & Nevins (Reference Anand, Nevins, Johns, Massam and Ndayiragije2006) for Hindi; and Basilico (Reference Basilico, Suet-Ying and Satoru2023) for Halkomelem. Under our account, the difference in the base-position is manifested in Case. The high EAs receive ERG from the Voice head that projects them, in a spec-head configuration (Massam Reference Massam, Chung, Finer, Paul and Potsdam2009, Tollan Reference Tollan2018, Tollan & Massam Reference Tollan and Massam2022). VoiceP is absent when the EA is merged lower, in spec,vP, requiring it to be Case-licensed by other means (such as by the higher Infl).

Structural evidence for the differential positions of EAs across languages sometimes comes from scope facts, e.g., the interaction of EAs with matrix negation; see Anand & Nevins (Reference Anand, Nevins, Johns, Massam and Ndayiragije2006) on Hindi and Schmidt (Reference Schmidt2003) on West Greenlandic.Footnote ¹⁷ In Kaqchikel, the EA positions are invariably within the thematic domain, below the matrix negation and tense, and we have not observed any scopal differences. The main empirical evidence for differentiated EA positions comes from causativization (see Tollan Reference Tollan2018 on this diagnostic). Causatives of intransitives are formed morphologically (see Section 5.3). Meanwhile, causatives of transitives require the use of periphrastic causative. Recall that under our approach, a causative v cannot take an unsaturated transitive vP as its complement, nor can it be added on top of VoiceP. Assuming that all EAs are base-generated in spec,vP, we would expect the causative v/Voice to be equally compatible with both intransitive and transitive vPs, contrary to fact. To account for the observed causativization pattern, we would then need to stipulate a [±transitive] feature, whose real meaning would remain obscure; for instance, it could not be connected to the presence of an EA or internal argument in the vP. Recognizing two different EA positions, a higher spec,VoiceP for transitives, and a lower spec,vP for unergatives, antipassives, and AF straightforwardly accounts for the differences in Kaqchikel causatives.

An alternative analysis, whereby the EA is always generated in spec,vP and then, in active transitives, is Case-licensed by a higher Voice (either covertly or by raising to spec,VoiceP), faces further challenges: non-uniformity of Voice heads, ‘incompleteness’ of vPs, and the look-ahead problem.

First, an account that allows for a single EA position struggles to bring Voice_TV and Voice_Pass together; Voice_TV must take a fully saturated vP as its complement, but Voice_Pass can only select an ‘incomplete’ unsaturated vP (Bruening Reference Bruening2013). (We propose that both Voice_TV and Voice_Pass combine with the same transitive vP, which introduces an Agent role but does not itself project an argument.) This gives rise to further questions about the general nature and the distribution of such ‘incomplete’ vPs: Can an unergative vP be incomplete too and, if so, why do unergatives in Kaqchikel (and elsewhere) generally resist passivization?

Next, if in active transitives the Agent/Causer is externally merged in spec,vP, we encounter a look-ahead problem. The transitive vP is already fully saturated and, in principle, does not require a VoiceP: the EA can receive ABS from Infl, and the internal argument can be expressed as an oblique phrase or suppressed, which would create an antipassive structure. However, antipassive configurations without the antipassive marker are not attested in Kaqchikel.

Finally, allowing VoiceP with transitive Voice equipped with [ERG] to be added on top of a saturated transitive vP leads to overgeneration. On that analysis, nothing prevents VoiceP from being added on top of an intransitive vP. Accordingly, we need to explain why there are no transitive constructions with antipassive or AF morphology and an ergative subject.

Our approach explains the obligatory presence of ERG in active transitive clauses and its obligatory absence in intransitive and AF constructions by linking it to the Voice head and assuming that a VoiceP is projected only when there is an unsaturated EA role to be dealt with. It captures all the relevant data without overgeneration.Footnote ¹⁸

To conclude, we have argued for a different structural composition of active transitives and passives on the one hand, and unergatives, antipassives, and AF constructions, on the other. The latter all lack the Voice projection, and, as a consequence, the two types of structures differ in the position of their EA.

6.1. Antipassives cross-linguistically

In terms of the distribution of internal arguments, Kaqchikel presents two extremes. In AP_null, the logical object is suppressed completely and interpreted existentially, via an entailment of the predicate; recall from Section 4.2.2. that we tentatively adopted a separationist approach to argument structure (A. Williams Reference Williams2015).

In contrast, in AF the logical object is projected as a definite/specific DP. Despite the absence of ergative marking, AF is akin to the active transitive clause in that both EAs and internal arguments are projected as DPs. In other words, the logical object is not ‘demoted’; thus, no parallels with the null antipassive arise (Aissen Reference Aissen, Coon, Massam and Travis2017a, Reference Aissen, Ostrove, Kramer and Sabbagh2017b, i.a., going back to Smith-Stark Reference Smith-Stark and England1978).

Our account of AP_null can potentially be extended to other languages where the internal argument appears to be suppressed in the AP configuration. Coon’s (Reference Coon2019) analysis of Chuj AP_null is quite similar to ours in that the internal argument can also be syntactically absent in that construction. For Chuj, Coon (Reference Coon2019) proposes that the internal argument is existentially bound with the help of a special functional head introduced in the structure on top of the vP/VoiceP. In Kaqchikel, an approach in terms of existential closure presents some challenges. First, in Chuj, the relevant functional head is spelled out overtly and appears not only in antipassives but also in the passive. In Kaqchikel, there is no such morphosyntactic evidence. Second, if a functional head similar to that present in Chuj antipassive is introduced to the structure in K’ichean languages, its distribution will need to be carefully regulated. For instance, this head would be redundant in unergatives, and it should be banned in AF. It would also need to be optional to account for K’iche’ oblique antipassives (see Section 6.2) and even in some dialects of Kaqchikel (see Heaton Reference Heaton2017 and Ranero Reference Ranero2021). While these potential problems are not insurmountable, the approach outlined here captures the relevant data in a more straightforward way. Since antipassives appear to vary across languages, it is possible that both ours and Coon’s (Reference Coon2019) analysis of Chuj are needed to account for the existing variation. This highlights how ‘antipassive’ encompasses multiple distinct phenomena rather than a single unified construction. The next subsection expands on this variation.

6.2. Antipassives as a heterogeneous class

In addition to comparing Kaqchikel AP_null with the AF construction, we may also compare absolutive and oblique antipassives. As discussed in Section 3.2, the latter type is marginal for our Kaqchikel consultants but has been reported in other varieties of the language. Consider Kaqchikel examples in (37), where the oblique antipassive suffix is identical to the AF one (hence the denotation af(ap) in the glosses).

Oblique antipassives are common in other Mayan languages, such as K’iche’ (38). In contrast to Kaqchikel, K’iche’ oblique antipassive is marked the same way as the absolutive (null) antipassive (and not as AF) and does not require Agent-fronting.

Oblique antipassives are still awaiting their analysis. At this point, we can anticipate several analytical options, assuming that antipassive-like constructions are similar in lacking the Voice projection but differ depending on the presence and nature of the internal argument. Options (i) and (ii) below are equally appropriate with respect to the K’iche’ data in (38).

1. (i) Just like AP_null, oblique antipassive does not project an internal argument at all. The oblique dependent is associated with the antipassive predicate via conjunctive semantics (cf. Pietroski Reference Pietroski2018), which, however, would need to be elaborated. If this is the case, the difference between absolutive and oblique antipassive is rather minimal. It is worth noting that no languages with productive antipassive require that the oblique argument be expressed in the antipassive (Polinsky Reference Polinsky, Dryer and Haspelmath2013, Reference Polinsky, Coon, Massam and Travis2017); it always appears optional, which may be an indication of structural similarities between the two types of antipassive.

2. (ii) The v head in the oblique antipassive is similar to that in the absolutive antipassive in that it has no Case/[D] feature and cannot select a DP. However, instead of being absent in both syntax and semantics, as in AP_null, the internal argument in the oblique antipassive is projected as either a PP with an nP complement (oblique) or an nP (which would subsume cases of incorporation or pseudo-incorporation).

3. (iii) Oblique antipassive is structurally identical to AF. Its v head selects/can license an internal argument, but the internal argument may alternate between a DP and an oblique phrase. Such an approach is close to the proposal advanced by Ranero (Reference Ranero2021) for Kaqchikel and Aldridge (Reference Aldridge2012) for Tagalog. It allows us to explain the Kaqchikel pattern illustrated in (37) but gives rise to questions concerning the distribution of DP/PP internal arguments in transitive constructions.

Option (iii) leads us to examine the relationship between DP and oblique objects. If the DP argument in AF can alternate with an oblique phrase, do these arguments occupy the same structural position? And if yes, why is that alternation not attested in active transitive clauses?

Summarizing options (i)–(iii), we anticipate the following distribution of structures that lack Voice (see Table 4).

Table 4. Possible variation in constructions lacking VoiceP

This typology requires further scrutiny, but it aligns with the notion that antipassives are cross-linguistically a heterogeneous group. As an aside, Voice-less vPs serve to circumvent the Ergative Extraction Effect (Aissen Reference Aissen, Coon, Massam and Travis2017a) in individual languages, but the use of such structures for extracting an EA does not motivate their existence. Even in the absence of Voice-less vPs, a language may find a way to extract an EA.

6.3. Approaches to antipassives: By way of comparison

While a thorough overview of the existing literature on antipassives would require a separate article, three main approaches can be compared with our proposal.

According to some analyses, antipassive directly manipulates the theta-grid of the predicate. Baker (Reference Baker1988) is a well-known example of such an account, according to which the AP morpheme absorbs both the ACC/ABS case and the thematic role normally licensing the internal argument; the antipassive becomes a mirror image of the passive (e.g., Baker, Johnson & Roberts Reference Baker, Johnson and Roberts1989). If applied to the Kaqchikel data, this approach is not without problems. Although it still allows us to bring together antipassives and intransitives (Section 5.1), that is only possible under an assumption that all unergative verbs are hidden transitives (otherwise the function of an AP morpheme in such predicates remains unclear), and we may expect them to co-occur with an internal argument if the AP suffix is absent. We have not observed such alternations (see atin ‘bathe’ – atinisaj ‘bathe (caus)’ = ‘wash’, but no *ati(j) ‘wash’).

The next class of approaches to antipassives relies on the manipulation of the verbal projection. Some researchers suggest that antipassive allows object licensing, but that its structural locus is not the same as in active transitive clauses. In the antipassive, a new licensing head is present: either a special v/VoiceAP, ApP, or Asp(ect) (Alexiadou Reference Alexiadou, Hall, Hirotani and Tamanji1999; also Johns Reference Johns and Faarlund2001, Schmidt Reference Schmidt2003, Yuan Reference Yuan2018, and references on Inuit therein). The analysis that we put forward for Kaqchikel AF falls into this category: the AF v head not only projects the EA but also Case-licenses an internal DP argument.

Other scholars choose to constrain the licensing properties of v; see Coon (Reference Coon2019) on Chuj and Aissen (Reference Aissen, Shklovsky, Pedro and Coon2011) on incorporative antipassive in K’iche’, and Aldridge (Reference Aldridge2012) on Tagalog. Our analysis of AP_null is conceptually close to the proposals by Coon and Aldridge. However, we view the restricted licensing behavior of antipassives as a byproduct of the antipassive/unergative v, not as its defining characteristic; a similar analysis is advanced by Basilico (Reference Basilico, Suet-Ying and Satoru2023), who views antipassives in Halkomelem Salish as Voice-less unergatives. As discussed in Section 5, the v_ITV itself projects an EA, which results in the vP being fully saturated; as a result, VoiceP is unnecessary. Because there is no VoiceP, no ergative case can be assigned, and the EA is licensed by Infl. In turn, the internal argument is left unlicensed and is not projected. In other words, although the presence of v_ITV causes the internal argument to be absent, there is nothing in the featural specification or selectional properties of v_ITV that would categorically preclude it from combining with a VP that contains an internal argument or require it to combine with a Theme-less/Patient-less VP. This approach allows us to treat antipassives as a subclass of intransitives and also adequately explains why there are no antipassives of unaccusatives; if antipassivization consisted of some kind of demotion of the internal argument, that gap would be puzzling.

Our analysis and the existing analyses of antipassives share the conception that the subject of absolutive antipassives is merged in an EA position. Unlike many existing approaches (see Aissen Reference Aissen, Shklovsky, Pedro and Coon2011, Imanishi Reference Imanishi2014, Coon Reference Coon2019, and Coon et al. Reference Coon, Baier and Levin2021 on Mayan and Yuan Reference Yuan2018 on InuktitutFootnote ¹⁹), we propose that the thematic domains in active transitive clauses on the one hand, and the AP and AF clauses on the other, are of different sizes and have different structures: the former include a VoiceP, and the latter are Voice-less vPs. A similar approach whereby (some) antipassive-like structures are Voice-less was put forward by Ranero (Reference Ranero2021) for Kaqchikel. Ranero suggests the VoiceP is projected but later gets removed; this analysis is motivated by a desire to avoid the look-ahead problem. Here we propose that the Voice head has a very specific function (i.e., to manipulate the EA variable) and that, similar to some other functional heads, Voice is not projected when not needed. In addition to this analytical difference, our data suggest that the account should not be limited to AF (as proposed by Ranero) but should also be extended to AP_null; our consultants did not replicate the difference between null antipassives and AF in elliptical contexts reported by Ranero.