The Role of the Absolutive Object in Morphological Accessibility

Certain types of noun phrases are more accessible than others with respect to core syntactic operations (e.g., Keenan and Comrie 1977, 1979, Moravcsik 1978). Most recently, accessibility has been defined as a hierarchy of morphological case (Bobaljik 2008, Deal 2016, 2017), in which unmarked arguments are most accessible for (a) φ-agreement (Bobaljik 2008) and (b) Ā-movement (Deal 2016, 2017), followed by dependent-marked arguments and oblique arguments (1).

The goal of this squib is to demonstrate a previously unnoted difference between Ā-movement and φ-agreement, regarding accessibility of ergative DPs. On the basis of data from ergative languages that, in certain environments, retain ergative marking on subjects in the absence of an absolutive object, I show that absence of an unmarked coargument affects the (in)accessibility of a marked (i.e., ergative) subject with respect to Ā-movement, but not with respect to φ-agreement. This has key consequences for theories about the source and nature of syntactic ergativity, namely, by suggesting that restrictions on the movement of an ergative DP—in contrast to restrictions on agreement with an ergative DP—do not arise from the properties of the ergative directly but are triggered by an absolutive coargument.

The squib is organized as follows. Section 2 lays out the background of morphological accessibility in ergative languages, and section 3 presents data from languages in which ergative case is retained in the absence of an absolutive object, demonstrating a difference between Ā-movement and φ-agreement. Section 4 discusses the implications of this difference.

In ergative languages, the argument bearing absolutive case is, typologically speaking, most accessible as a target for φ-agreement. As discussed by Bobaljik (2008), the ergative subject cannot control verbal agreement in certain languages, such as Hindi-Urdu (2a); the verb agrees in φ-features with the absolutive object instead. Only if the subject is absolutive—as in the intransitive sentence in (2b)—does it trigger verbal agreement. Bobaljik (2008) proposes that Hindi-Urdu agreement is determined by the hierarchy in (1), wherein only unmarked arguments are accessible as targets of φ-agreement probes (e.g., T⁰).

An analogous scenario is observed in Ā-movement: in a subset of morphologically ergative languages that exhibit “syntactic ergativity,” the ergative argument cannot undergo extraction in environments such as relative clauses and wh-questions, but the absolutive argument can.

Q’anjob’al (Mayan) (3) is one such language.¹

Extending Bobaljik’s (2008) account, Deal (2017) proposes that accessibility with respect to Ā-movement is also governed by the hierarchy in (1): in languages that ban extraction of the ergative subject, such as Q’anjob’al, the syntactic probe triggering Ā-movement (e.g., C⁰) can only agree with a goal bearing unmarked (i.e., absolutive) case features, and cannot target a DP bearing marked case (e.g., ergative, accusative, or oblique) features. Analogously to agreement in Hindi-Urdu, in which only unmarked arguments can be targeted by agreement probes, only unmarked arguments can be targeted by Ā-probes in languages like Q’anjob’al. Deal’s proposal is advantageous in that it successfully unites the syntactic mechanisms that give rise to syntactic ergativity with those that trigger the agreement pattern in (2), thereby providing an explanation of syntactic ergativity in terms of (1): namely, that syntactic probes are sensitive to the case-featural properties of their goal.²

I show, however, that despite surface similarities, Ā-movement and φ-agreement are in fact not truly analogous with respect to the hierarchy in (1). The absolutive object can be shown to play a crucial role in the restriction on movement of ergative subjects: in syntactically ergative Mayan languages such as Q’anjob’al, the ergative DP may be extracted when no absolutive object is present.³ In contrast, the absence of an absolutive object does not render the ergative subject a viable target for φ-agreement in languages such as Hindi-Urdu. In the next section, I examine the behavior of ergative subjects in the absence of an absolutive object; I first discuss Ā-movement and then turn to φ-agreement.

This section focuses on a subset of ergative languages that allow ergative marking when absolutive case is absent. This patterning may arise under a number of circumstances: (a) where ergative case is retained when the transitive object is caseless; (b) where ergative case is retained when the transitive object is case-licensed by an alternative strategy; and (c) in so-called split-S ergative languages, wherein subjects of unergative verbs are marked as ergative (in contrast to subjects of unaccusatives, which are absolutive). In these environments, differences in the behavior of the ergative subject offer insight into how its accessibility is affected by the presence of an absolutive coargument. I focus first on movement, drawing together several previous observations from existing Mayanist literature, and then assess the corresponding facts for agreement, introducing novel data from Hindi-Urdu and Basque.

Across many syntactically ergative Mayan languages, ergative case is retained in the absence of an absolutive-cased object; one such language is K’iche’. As observed by Aissen (2011), while ergative subjects cannot extract in the presence of a full DP object (4a), extraction is permitted when the object is bare (4b).

Henderson and Coon (2017) note an analogous pattern with Kaqchikel light verb constructions of which the complement is a bare nominal borrowed from Spanish (e.g., manifestar ‘protest’; see (5b)). Whereas ergative subjects cannot usually extract in transitive constructions in Kaqchikel (5a),⁴ they can extract in light verb constructions (5b).

Furthermore, the ban on movement of ergative subjects in Q’anjob’al, as in (3), no longer holds when the object is reflexive (6) (see discussion in Pascual 2007 and Coon, Mateo Pedro, and Preminger 2014). The ability of the ergative subject to undergo extraction when the object is reflexive is also reported for Q’eqchi (Berinstein 1984), Popti’ (Craig 1977; see also Ordóñez 1995), K’iche’ (Mondloch 1981; see also Stiebels 2006), and Kaqchikel (Henderson and Coon 2017).

Coon, Mateo Pedro, and Preminger (2014) propose that reflexive objects in Mayan are caseless, noting crucially that, whereas word order in Q’anjob’al is usually strictly VSO, VOS order is obligatory when the object is reflexive (see also Clemens and Coon 2018); this requirement for verb adjacency suggests that the object is a bare NP. Such an analysis is not straightforward, however. Aissen (2017) observes that reflexive objects do not pattern distributionally with caseless NP objects; notably, nonfinite clauses, which permit only caseless NPs as objects (7a), do not permit reflexive objects (7b).

Thus, reflexive objects are unlike both absolutive-cased DP objects (in that they must appear adjacent to the verb) and caseless NP objects (in that they cannot occur in nonfinite clauses, as in (7)). There are at least two ways of accounting for these facts, however. One possibility is that reflexives are indeed bare caseless NPs (as in Coon, Mateo Pedro, and Preminger 2014), but that (7b) is ruled out on independent semantic grounds, wherein Mayan nonfinite clauses require that objects be predicate-denoting (which reflexives are not).⁵ A second possibility is that reflexives are full DPs (as in Aissen 2017)—such that they cannot appear in NP-only environments like (7)—and are thus subject to case-licensing requirements. This then would leave the question of why VOS order obtains. A possible scenario is that Mayan reflexives are not licensed via assignment of absolutive case; instead, licensing takes place via Morphological Merger with the verb (Baker 1988, Massam 2001, Coon, Mateo Pedro, and Preminger 2014, Levin 2015, Van Urk 2019). Morphological Merger occurs through a process of Local Dislocation (Embick and Noyer 2001; see discussion in Van Urk 2019) in which the object incorporates into V⁰ at PF and thereby escapes the Case Filter.⁶ As a result, unlike absolutive-cased objects, reflexive objects must occur immediately postverbally (in order for Morphological Merger to apply).⁷ Crucially for present purposes, when absolutive case is not assigned—either to reflexive objects (6) or when the object is bare ((4b), (5b))—the ergative subject can undergo Ā-movement.

Still more evidence that Ā-movement is not conditioned by case marking comes from Ixil. Along with several other Mayan languages in the Yucatecan, Cholan, and Poqom subbranches (see Zavala Maldonado 2017), Ixil exhibits a split-ergative pattern: whereas an ergative-absolutive alignment is observed in the perfective aspect, nonperfective aspects yield a so-called extended ergative pattern (Dixon 1979), wherein all subjects—transitive (8a) and intransitive (8b)—are indexed on the verb by the same prefix.

There is debate in Mayanist literature about the status of the “ergative” prefixes in (8), however. Among other analyses, Coon (2013) argues that they are genitive prefixes, and Zavala Maldonado (2017) analyzes them as nominative prefixes. For present purposes, the crucial point is that a contrast obtains between transitive and intransitive predicates with respect to subject extraction: the transitive subject in (8a) cannot be extracted, as in (9a), but the intransitive subject in (8b) can, as in (9b) (see also Assmann et al. 2015).

Taken at face value, the Mayan facts seemingly cannot be tied to the notion of extraction as regulated by the case-featural properties of the subject, per (1), and thereby call instead for a theory of syntactic ergativity that ties Ā-movement of the ergative subject to properties of the object (e.g., Coon, Mateo Pedro, and Preminger 2014, Assmann et al. 2015). However, it is also possible that morphological markedness per (1) is determined on a relational as opposed to absolute basis. This would mean that an ergative DP is treated as “marked” for the purposes of featural accessibility only if an unmarked competitor argument is also present (within some domain). Conversely, in the absence of an absolutive DP, the ergative—now the least-marked DP—is treated as featurally “unmarked” (and is therefore accessible).

The scenario outlined above establishes a key prediction regarding φ-agreement: if morphological markedness is relational with regard to (1), then the availability of ergative DP extraction in the presence versus absence of an absolutive object should be paralleled in the agreement patterns in languages such as Hindi-Urdu (see (2)). In other words, the absence of an absolutive object should allow for the ergative DP to trigger φ-agreement. As I will now discuss, however, this is not the case.

This section assesses agreement in languages in which ergative subjects cannot trigger agreement and that have “split-S” alignments, wherein subjects of both transitive and unergative verbs are marked ergative. Like many Indo-Aryan languages,⁸ Hindi-Urdu meets these criteria: ergative subjects do not trigger agreement (see again (2)), and—in perfective aspects—ergative case can mark unergative subjects (10). However, agreement in such constructions is always 3SG masculine, regardless of the φ-features of the subject; the 3SG feminine subjects in (10) do not trigger feminine agreement.

As is the case in many languages, 3SG masculine agreement in Hindi-Urdu is the agreement default (Bhatt 2005), arising through the failure of an agreement probe to target any nominal whatsoever (see Preminger 2011, 2014). The masculine agreement marking in (10) can therefore be taken to indicate lack of agreement. The most straightforward conclusion to be drawn is that ergative unergative subjects in Hindi-Urdu are not targets for φ-agreement. However, there is an alternative possibility: namely, that the 3SG masculine verbal morphology in (10) does not reflect default agreement; rather, it reflects agreement with a covert absolutive cognate object (see, e.g., Hale and Keyser 1993). This alternative seems unlikely, however, since the cognate objects of the verbs in (10) are themselves feminine: when these are overt, as in (11), feminine agreement surfaces.

Thus, if the verbal agreement in (10) were agreement with a covert cognate object, we would expect it to be feminine as opposed to masculine. The masculine agreement that surfaces on unergative verb forms can therefore be taken to be true default agreement. As a result, the absence of an absolutive object in Hindi-Urdu does not render the single ergative argument a viable target for agreement.

The pattern for Hindi-Urdu, wherein ergative unergative subjects do not trigger agreement, is widespread among Indo-Aryan languages, as exemplified by Kashmiri (12) and Marathi (13): notice that, while the verbs in (12a) and (13a) φ-agree with nominative subjects, they do not φ-agree with ergatives (see (12b), (13b)). Indeed, Bhatt (2007:19) notes that “agreement with ergative subjects . . . as a last resort” is unattested in Indo-Aryan.

Finally, consider Basque. Unlike ergative arguments in Indo-Aryan, those in Basque generally do trigger agreement. In transitive constructions such as (14), the clause-final auxiliary φ-agrees with both the ergative subject and the absolutive object. Both unmarked and dependent-marked arguments are thus (simultaneously) accessible for absolutive and ergative agreement, respectively.

However, Basque exhibits a pattern known as “ergative displacement” (see, e.g., Laka 1993), in which a morphosyntactically ergative DP exceptionally triggers absolutive agreement (15). Ergative displacement arises when (a) tense is nonpresent, (b) the ergative DP is 1st or 2nd person, and (c) the absolutive DP is 3rd person. Thus, under these conditions the ergative argument can be said to be inaccessible for ergative agreement and is indexed by absolutive agreement instead.¹⁰

Basque also exhibits a split-S alignment: subjects of unergative verbs bear ergative case marking and typically trigger normal ergative agreement, as in (16).¹¹

On the basis of long-distance agreement phenomena, iterative constructions, and absence of certain cognate DPs, Preminger (2012) argues that unergative verbs in Basque lack implicit objects: by this analysis, the sentences in (16) are not covertly transitive. Now consider displaced agreement again: if the lack of ergative agreement in constructions such as (15) is dependent on the presence of an absolutive coargument, then ergative displacement should not apply to subjects of unergative verbs. However, absence of an absolutive object makes no difference in this regard: a 1st/2nd person unergative subject in a nonpresent tense still cannot trigger regular ergative agreement (17). Instead, the same 3rd person absolutive agreement prefix surfaces as for the transitive subject in (15).

Thus, ergative displacement is triggered even in the absence of an absolutive coargument. The agreement facts in Basque hereby yield the same conclusion drawn from Indo-Aryan: absence of an absolutive coargument does not affect accessibility of an ergative argument for ergative agreement. Following from these observations, the strong typological generalization to be made is that no language should exhibit verbal φ-agreement with ergative arguments, only in the absence of an absolutive-cased object. This generalization contrasts with the Mayan movement data in section 3.1, in which the absence of an absolutive-cased object allows for Ā-movement of the ergative. I turn now to the implications of this contrast.

The contrast between Ā-movement and φ-agreement in ergative languages—whereby absence of an absolutive DP renders an otherwise inaccessible ergative DP accessible for Ā-movement (as in Mayan) but not for φ-agreement (as in Indo-Aryan and Basque)—minimally suggests that Ā-movement and φ-agreement are not analogous with respect to morphological case.¹² Regarding how best to formally characterize these facts, there are three options. One is that Ā-movement and φ-agreement are parameterized differently with respect to the morphological case hierarchy in (1): extraction—at least, for the Mayan languages discussed in section 3.1¹³ —is regulated on a morphologically relational basis, and agreement is regulated on an absolute basis. This might seem reasonable in view of a key contrast between the languages discussed in section 3.1 and those discussed in section 3.2: the syntactically ergative languages in section 3.1 are head-marking, while the absolutive-only agreement languages in section 3.2 are dependent-marking.¹⁴ According to Coon (2017), however, the syntactic mechanisms that give rise to ergative head-marking in Mayan are the same as those that give rise to ergative dependent-marking in languages like Hindi-Urdu (i.e., are a reflex of an inherent Agree relation between v⁰ and an external argument¹⁵). In view of this, it seems unlikely that accessibility would differ according to whether an ergative alignment is manifested via head-or dependent-marking.

A more promising approach is one in which the hierarchy in (1) governs only one of Ā-movement or φ-agreement. I spell out the two options in (18).

I leave the choice between these two options open for future research.

I am very grateful to Kinza Mahoon for judgments of Hindi-Urdu sentences, and to Saioa Lazarra for judgments of Basque. For helpful feedback and comments, I thank Susana Béjar, Lauren Clemens, Diane Massam, the reviewers and audience at NELS 48, and three anonymous LI reviewers.