Cross-linguistic regularities in the organization of case systems point to the existence of an abstract case sequence, with a demarcation between core and oblique cases (McCreight & Chvany 1991; Johnston 1996; Blake 2001; Caha 2009; 2017; Harðarson 2016; Bárány 2016; 2018; 2021; Starke 2017; McFadden 2018; Smith & al. 2018; Zompì 2019). However, various problems arise with the more precise ordering inside each of these classes or at the border between core and oblique zones. A well-known hierarchy is the one formulated by Caha (2009), as represented in (4). Here, the accusative, the locative and the genitive are adjacent, to the exclusion of the dative. Given that dom is a type of acc (thus explaining its similar syntactic behaviour with unmarked accusatives), it is not possible to explain its syncretism with dat, as the latter is not contiguous with acc. In a certain paradigm two forms cannot share a morphological property (A) across an intervening form which does not share the same property, but has instead property B. Bobaljik (2012) intuitively labeled this constraint as the *aba restriction, demonstrating it is also operative in other domains, such as suppletion.

Noting problems with acc-dat contiguity in other areas of the grammar, Harðarson (2016) proposes instead that more than one case hierarchy is active in human language. Some languages make use of the hierarchy in (5), where acc and dat are adjacent. However, Harðarson’s (2016) sequence in (5) in itself isn’t enough for Spanish. As we will see later in the paper, the a marker signaling dom and datives is also a locative. Although Harðarson (2016) does not explicitly discuss locatives, they need to be arranged outside gen, so as to also capture the dat-gen syncretism that is active in some of the languages he examines.

A different way to reconcile problematic data of this type has been formulated by Starke (2017). Instead of using different case hierarchies, Starke (2017) argued for the need to construct more complex case hierarchies, which contain more than one accusative and more than one dative, occurring on each side of the genitive. Starke (2017) labeled the two accusatives/datives Sacc/Sdat and Bacc/Bdat, where S stands for ‘structural/smaller’ and B for ‘bigger’. This is represented in (6). Thus, the apparently problematic syncretism involving accusatives and datives in Spanish is not a matter of distinct case hierarchies but of what type of accusative/dative a language selects. Spanish has both Sacc which is syncretic with nom and Bacc which is syncretic with Bdat. As these classes are adjacent in the enriched hierarchy, no violation of the *ABA restriction obtains.

The main proposal we will make is that enriched case hierarchies capture syncretism patterns with oblique dom more generally, beyond the dative-accusative syncretism discussed by Starke (2017). In this paper, we will examine other oblique strategies for DOM, beyond the dative, namely the locative (Romanian) and the genitive (Russian, Ossetic). We will also see that oblique dom is found not only in nominative-accusative, but also in ergative-absolutive languages, such as southwestern Basque, Hindi and Gujarati. We propose that the extended hierarchy in (7) derives all these patterns. The enriched sequence in (7) contains more than one case for direct objects, abbreviated here as A (standing for accusative/absolutive) and also more than one locative (following Caha 2009, et subseq.).

We will also show that an enriched case hierarchy straightforwardly accounts for another important trait of oblique dom, namely syntactic distinctions oblique dom establishes with respect to the unmarked accusatives, despite a general accusative profile (Section 4). Thus, two A categories are motivated in the syntax, with Ba being used for objects that need to be differentially marked due to their more complex internal make-up.

The structure of this paper is as follows. Section 2 introduces three main types of oblique dom (dat=dom, loc=dom and gen=dom), using a database of 11 languages from various families, namely Romance (Spanish, Catalan, Romanian, Neapolitan, Sardinian), southwestern Basque, Indo-Aryan (Hindi, Gujarati, Kashmiri), Slavic (Russian), Ossetic, both nominative-accusative and ergative-absolutive. Section 3 provides various diagnostics supporting the conclusion that despite oblique morphology, such objects do not have an oblique syntactic profile, but rather behave as structural accusatives (or absolutives). Section 4, in turn, examines syntactic differences oblique dom shows as compared to unmarked objects. Section 5 demonstrates that neither the hierarchy in (4) nor that in (5) can capture the accusative-oblique syncretism in a uniform manner cross-linguistically. In Section 6 the three oblique dom patterns are uniformly derived under the enriched case hierarchy in (7). Section 7 examines the differences between the current proposal and two recent accounts that make use of partially ordered hierarchies (Zompì 2019; Bárány 2021). Section 8 concludes.

As we have seen in the introduction, a well known example of the dative re-purposed for dom (dat=dom) comes from (varieties of) Spanish (Torrego 1998; Rodríguez-Mondoñedo 2007; López 2012; Ormazabal & Romero 2013a; 2013b; 2013c; Fábregas 2013; Bárány 2018; 2021; Muñoz Pérez 2020, among many others). In (8a), repeated from (1a), the animate definite direct object must be introduced by a preposition which is homophonous with the dative. The same oblique marker is ungrammatical with the inanimate object in (1b), repeated in (8b).

A second pattern employs a locative marker. For example, in Romanian, classes of nominals including highly referential human DPs can/must carry the locative preposition pe, as seen in the two examples below (Farkas 1978; Dobrovie-Sorin 1994; Cornilescu 2000; Tigău 2010; Mardale 2015; Irimia 2020; Hill & Mardale 2021; a.o.).1

A third sub-type of oblique dom uses the genitive, as in Slavic (Mel’čuk 1980; Brecht & Levine (eds.) 1986; Bossong 1991; 1998; Halle & Marantz 1993; Rappaport 2003; Wiese 2004; 2011; Glushan 2010; Lieb & Friederich 2011; Franco & Manzini 2017; a.o.). The masculine animate direct object must use the genitive in (10b), while the inanimate in (10a) shows up with accusative morphology (which is homophonous with the nominative). An example with the genitive of possession is in (10c).

An extensive use of gen=dom is found in the Ossetic family (also called Ossetian), see especially Thordarson (1989; 2009); Cheung (2008); Erschler (2009; 2012; 2021); Vydrin (2013); a.o.2 As seen in (11a) and (11b), personal pronouns and specific animates require obligatory dom using the genitive, while inanimates show an unmarked (nominative) form, as in (11c). The possessor genitive is illustrated in (11d).

Oblique dom is also widespread in ergative-absolutive languages (as seen in more detail in Sections 6.2 and 6.3). Here we will be illustrating case studies using the dat=dom stategy.3 For example, in southwestern Basque (those varieties that have been under contact with Spanish), higher animates and pronouns as in (12b) are introduced by dative morphology (‘dative overmarking’, see especially Austin 2006; Mounole 2012; Odria 2014; 2017; Rodríguez-Ordóñez 2016; Fernández & Rezac 2016; a.o.). Additionally, they trigger dative or absolutive agreement (depending on the variety). Inanimates, on the other hand, use absolutive morphology, which triggers absolutive agreement, as seen in (12a).

We present an example from a split ergative language too, namely Gujarati. Similarly to other Indo-Aryan languages, Gujarati shows aspect-based split ergativity. The perfective triggers the ergative marker on agents in transitive clauses, while subjects of intransitives and objects of transitives are marked with the absolutive and also trigger agreement on the participle, under a type of past participle agreement (ppa). Also similarly to other Indo-Aryan languages, Gujarati has oblique dom. Objects at the higher end of animacy and specificity scales must/can be encoded using dative morphology. Oblique dom is seen in both the perfective split-ergative alignment, as in (13b) and the imperfective nominative-accusative alignment, as in (14a); see especially Cardona (1965); Mistry (1976; 1997); Masica (1982); Magier (1983); Comrie (1984; 1985); Wunderlich (2012); Woolford (2006); Grosz & Patel-Grosz (2014); a.o. As we will also show later in the paper (Sections 3 and 6.2) a special property of Gujarati is that in the perfective (ergative alignment) oblique dom triggers ppa just like unmarked objects; thus, both (13a) and (13b) must show object agreement.

Passivization is a common distinguishing point between oblique dom and other obliques. Generally, oblique dom is subject to passivization, a syntactic operation which obliques tend to resist. We repeat here the Spanish examples presented in the introduction ((2) vs. (3)). As can be seen in (15b) both the animate dom-ed object and the unamrked inanimate one undergo a process of passivization, constructed periphrastically.

(Goal) datives, however, do not allow periphrastic passivization, as seen in (16b). Crucially, the absence of passivization with datives is not due to other internal argument interveners: the verb in (16c) has a sole internal argument specified as a a dative.4 Passivization of the latter is ungrammatical.5

The same split is seen in Romanian, a language where oblique dom uses locative morphology. The obligatory oblique dom on the object pronoun in (17a) and the unmarked accusative in (17b) are unified as classes which are subject to periphrastic passivization, as demonstrated by (17c).

Spatial (and, in fact, other types of) locatives as in (18) resist this process. Similarly to Spanish, the problem is not a potential direct object intervener; the verb in (18c) and (18d) selects the locative as its unique argument, whose passivization is equally ungrammatical.

In Romanian clitic doubling is another syntactic diagnostic which sets aside oblique dom from locatives. In general, accusative clitic doubling exhibits very subtle interactions with dom in the language (see especially Tigău 2010 or Hill & Mardale 2021); in (17a) we see that an object pronoun needs obligatory doubling, using the accusative6 morphology of the clitic. Locatives, on the other hand, do not permit clitic doubling. The example in (19) with an accusative clitic intended to double the feminine nominal embedded under the spatial locative pe is strictly ungrammatical.

Turning to the gen=dom strategy, we find a similar behaviour under periphrastic passivization. In Russian the animate genitive is subject to passivization, just like the regular accusative (as seen in (20)).7

Similar behaviour of unmarked objects and oblique dom under periphrastic passivization is also a trait of some ergative-absolutive languages, as we will see in (25) and (26) from Hindi.

Another diagnostic under which oblique dom patterns like unmarked accusatives, and not like true obliques is possibility of object agreement. In the Gujarati examples in (13), which we repeat here, illustrating the ergative alignment in the perfective, we see ppa in number and gender tracking not only an unmarked absolutive, as in (21a), but also oblique dom, as in (21b). Goal datives, or datives that are lexically selected by various verbs, as in (21c) are simply not grammatical with ppa; default, neuter inflection must instead be used.

The agreement facts in Gujarati present a complex picture, and not all details are relevant here; see especially Cardona (1965); Mistry (1976; 1997); Masica (1982; 1991); Magier (1983); Comrie (1984); Woolford (2006); Wunderlich (2012); Grosz & Patel-Grosz (2014); Joshi (2020); a.o., for extensive discussion and exemplification. What matters for our purposes is that in examples such as (21a) and (21b) there is clear indication of the presence of object agreement.

Moreover, Grosz & Patel-Grosz (2014) show (for Kutchi Gujarati) that object agreement is encoded on the participle, while subject agreement is tracked on the auxiliary (unless the subject is ergative). The example below contains both subject and object agreement (see also Bárány 2021).

We take the contrast between (21a)/(21b), on the one hand, and (21c), on the other hand, to indicate that ne is the syncretic spell-out of two categories with different syntactic natures. Additionally, we agree with the analysis of Gujarati agreement put forward by Grosz & Patel-Grosz (2014). The two authors motivate the presence of two probes, a higher one on T, which is responsible for subject agreement, and a lower one on Asp or υ, where agreement with the object is implemented. As both unmarked and marked objects show participle agreement, related to the lower probe, for Grosz & Patel-Grosz (2014) both these categories are connected to abstract Case licensing by Asp/υ. In next section, we provide a refinement of this hypothesis and propose that some aspects of oblique dom licensing, namely its [uC] feature, are related to υ, the same licenser involved in [uC] licensing of unmarked objects. The oblique morphology on the differentially marked objects is a result of the presence (and licensing) of additional features, beyond [uC].

Uniformity in agreement between unmarked objects and oblique dom is not restricted to Gujarati, in the set of languages examined here. We present another example from Neapolian (Italo-Romance, see especially Loporcaro 1998; 2010; Ledgeway 2000, or Ledgeway et al. 2019; a.o.), a language with systematic nominative-accusative alignment. We see in (23a) that unmarked objects trigger ppa, and so does oblique dom in (23b), which uses dative morphology. Goal datives (or any other types of inherent and lexical datives for that matter) do not permit ppa. As such, the speakers who can construct the goal using a full lexical dative (as opposed to just a clitic) expressly mention that participle agreement with the dative is not possible; this is seen in (23c).

Just like in Gujarati, the agreement facts of Neapolitan strongly support the existence of two probes: a higher one (probably in T) which tracks features on subjects, and a lower one which is responsible for object agreement. In (23a), we see first person subject agreement on the auxiliary and object agreement on the participle. Oblique dom, just like unmarked accusatives triggers agreement but only on the lower participle head.8 This supports the conclusion that unmarked and marked objects are syntactically similar, as types of accusatives, while the a marker on datives as in (23c) spells out a different syntactic structure. Additionally, unmarked objects and oblique dom are similar in another respect – as illustrated by (23a) and (23b), besides ppa, they can both trigger clitic doubling,9 but only under the accusative form of the clitic.

There are also languages in which oblique dom does not show object agreement, while unmarked accusatives/absolutives do. One such example in our database is Hindi (Saksena 1981a; 1981b; Comrie 1984; 1985; Mahajan 1989; Mohanan 2001; Bhatt 2003; 2005; Butt 2006; Bobaljik 2008; a.o). The unmarked, inanimate object in (24a) from the ergative-absolutive alignment in the perfective is tracked by ppa, and there cannot be agreement with the ergative, similarly to what we have seen for Gujarati. On the other hand, the human proper name object in (24b) which must be differentially marked with dative morphology, cannot trigger ppa. Default masculine agreement must be used instead. In this respect, oblique dom might appear to be similar to inherent datives, which do not trigger ppa. In (24c) the participle can only agree with the object badhaaii, and not with the indirect object dative.

However, despite the agreement facts, there are robust syntactic diagnostics under which unmarked absolutives and oblique dom pattern alike, to the exclusion of indirect object datives. One such test is passivization. As observed several times (Mahajan 1989; Bhatt 2003; Kidwai 2021; a.o.), both unmarked objects and oblique dom are possible under passivization; moreover, the latter can also preserve its differential marker.10 Compare (24a) with (25a) and (24b) with (25b). As seen in (26), indirect object datives cannot be passivized no matter whether they preserve dative morphology or not.

We have seen that some southwestern Basque varieties that have oblique dom show dative11 agreement on the verbal complex (see (12b), but also the examples below); absolutives, instead, are only possible with absolutive agreement (see Odria 2014; 2017; 2019; a.o.).

However, agreement seems to be a superficial diagnostic. As Odria (2014 et subseq.) and Fernández & Rezac (2016), a.o., have conclusively shown there are more solid diagnostics under which oblique dom patterns syntactically with absolutives, despite its oblique morphology. For example, oblique dom can host a depictive secondary predicate, as in (27a). The example in (27b) demonstrates that this is similar to absolutives. and the opposite of what indirect object datives do.

Another congruence point between Basque absolutives and oblique dom are the similar co-occurrence restrictions they give rise to.12 Odria (2014; 2017; 2019), or Fernández & Rezac (2016), a.o. have shown that there are speakers for whom oblique dom blocks the presence of an agreeing dative, as in (28).

Additionally, in all languages examined here, oblique dom is crucially different from obliques given its restriction to animacy, specificity, definiteness, etc. Regular obliques do not have this limitation. In fact, as the literature has discussed (Bárány 2021; Odria 2017; 2019; Irimia 2021a, among many others), oblique dom and obliques are distinguished in the syntax on the basis of many other diagnostics, such as the possibility of hosting nominalizations and reduced relative clauses, whose illustration we have left out here for lack of space. Table 1 summarizes the results obtained in this section (the locative and the genitive are collapsed under OBL).

Given that the accusativity diagnostics are real, the most straightforward account would take oblique dom to involve a type of syncretism (Johnston 1996; Blake 2001; Baerman & al. 2005; Bachrach & Nevins 2008; Glushan 2010; a.o.) caused by two syntactically distinct categories being spelled out with the same overt morphology, namely the differential accusative and the oblique. However, an analysis along these lines must be formulated in such a way as to capture another crucial property of oblique dom: despite its general accusative syntax, it does exhibit syntactic differences from unmarked accusatives.

In many languages, one important syntactic trait of oblique dom is that it can be found in a higher position than unmarked accusatives. This is common in many Indo-Aryan languages and has been analyzed formally for Hindi-Urdu (see especially Bhatt & Anagnostopoulou 1996; Baker 2021; a.o.). In (35a), the unmarked theme has to follow the dative, while in (35b) and (35c) the marked nominal has to precede the dative.18 López (2012) has argued that oblique dom is found in a higher position in other languages too, such as Spanish, etc.19

Another syntactic correlate of oblique dom are the co-occurrence restrictions it gives rise to, as we have already seen for Basque in (28). The data in (36) from (standard) Spanish indicate that oblique dom cannot co-occur with a dative which is clitic doubled (using the dative form of the clitic).20

There are also languages where the position of oblique dom or whether it involves raising is not straightforward. Nevertheless, there are other tests indicating that oblique dom does exhibit syntactic differences from the unmarked accusative, despite the general accusative profile of the two. We provide here examples from Romanian, where oblique dom is not possible in a configuration that contains a dative clitic interpreted as a possessor, as in (37a).

That this is not a PF restriction is indicated by the grammaticality of (37b), where we see the same form of the clitic; crucially, there is a difference in interpretation. The dative clitic in (37b) is interpreted either as a goal dative (a reading which is not possible in (37a)) or as a possessor on some other DP in the structure, which is not dom. Preliminarily, (37a) indicates a possessor raising analysis for the dative clitic, which is merged in the same position as the component accounting for oblique dom; thus, the two cannot co-occur as they compete for the same position. All these examples indicate that an analysis according to which the special morphology of oblique dom is just a matter of PF, for example via impoverishment of accusative features, is very difficult to support for the languages discussed here.21 To fully account for the data, two categories with accusative syntax are needed.

We briefly point out here that the accusative general profile, on the one hand, and syntactic differences with respect to unmarked accusatives, on the other hand, can be accommodated under accounts which connect dom to the licensing of specifications beyond Case per se, more specifically to grammaticalization of information structure categories such as secondary topics, perspectivization, sentience, discourse-relevant humanness, etc. (Dalrymple & Nikolaeva 2011; Belletti 2018; Hill & Mardale 2021; a.o.), as opposed to licensing just in terms of (interpretable) φ-features or uninterpretable Case ([uC]).

For example, a recent analysis by Irimia (2021a) takes grammaticalized humanness which is relevant in the discourse22 to involve the merge of a δ(iscourse) related feature in the extended (KP) projection of the nominal, beyond Case per se, as schematically shown in (38).23 This ties in with a line of research, going back to at least Jaeggli (1982), which employs the mechanics of additional licensing for dom, also reworking it as a problem related to the complexity of differentially marked categories. As δ-linking needs licensing, an additional (local) licenser must be made available beyond the regular licenser in the relevant domain (for example υ, etc.), which licenses just Case. The effect of this additional licenser at PF is the spell-out of special (oblique) morphology for dom.

To summarize, in this section and in the previous one we have established three important facts related to oblique dom: i) it does not have an oblique syntax; ii) despite its being an accusative at core, it does exhibit syntactic differences from unmarked accusatives and absolutives; iii) given its accusative profile, a plausible starting point is to attribute its surface identity with obliques to syncretism at PF. But the syncretism hypothesis has to be worked out in such a way as to also capture the syntactic differences with respect to unmarked accusatives.

That oblique dom also presents syntactic characteristics of its own, despite its general accusative/absolutive behavior, is relevant and a comprehensive account should derive both these facts. In our view, the syntactic and semantic evidence motivates the presence of more than one accusative/absolutive in the hierarchy. In turn, the more complex accusative (Ba, with B standing for ‘bigger’) which contains features beyond Case per se can be spelled out syncretically with a true oblique. This provides support for the enriched case hierarchy (modeled after Starke 2017) we briefly presented in the introduction in (7) and which we repeat here:

In Section 5 we show in more detail that none of the two non-enriched hierarchies in (4) or (5) can capture the syncretism patterns established by oblique dom for both nominative-accusative and ergative-absolutive languages.

Harðarson’s (2016) hierarchy in (45) avoids violating the *ABA restriction for Spanish dom and datives, as the accusative and the dative are contiguous in the case sequence. However, the Spanish syncretism patterns are more complex than the dat-dom homophony. On the one hand, the a marker is also a locative, as illustrated in (48); on the other hand, the language also contains a genitive, spelled out via the de marker. A fragment of Spanish case markers is given in Table 2.

Syncretism of dom with locatives is not uncommon cross-linguistically, as we have seen in the introduction.24 The same dat-loc-dom syncretism is found elsewhere in Romance, for example, in Catalan, which we illustrate in Table 3.25 But, as we can see in the two tables, these more complex patterns cannot be captured by the hierarchy in (45). As already mentioned, although Harðarson (2016) does not explicitly discuss locatives, in order to capture the data he is addressing, the locative must be placed outside the genitive, which in turn, includes the nominative and the accusative. Under this ordering, we obtain an *ABA violation. Rather, what we would need for Spanish is a case sequence of the type in (49), where the accusative (dom) is adjacent to both the locative and the dative. Following the remarks in Caha (2009) we can preliminarily assume that the locative is closer to the accusative and the dative follows the two. Given their adjacency, a syncretic spell-out rule can target the three markers.

Let’s examine now whether the same preliminary case sequence can be extended to other oblique dom languages, for example those that use the genitive. We repeat here the relevant data from Russian, where masculine animates use the genitive inflection. Inanimates, on the other hand, preserve the accusative which is homophonous with the nominative (in this declension).26 We have seen that Ossetic is another family where this type of syncretism is active, as in (11).

Russian is a language with rich case inflection and its intricate syncretisms have been well studied (see for example the contributions in Brecht & Levine 1986, a.o.). In Tables 4 and 5 we present two relevant paradigms which include the masculine animate. As it is clear, the preliminary hierarchy we introduced in (49) for Spanish incurs an *ABA violation, as the dative intervenes between the accusative and the genitive. A similar problem will arise for Ossetic, as demonstrated by the case paradigms in Table 6.27 What we need for Russian/Ossetic, instead, is a hierarchy more similar to the one proposed by Caha (2009). If we leave aside loc₁ (assuming that Russian/Ossetic does not lexicalize this category), we can capture the accusative (dom)-genitive syncretism. Subsequently, we can explain the differences between Spanish and Russian/Ossetic under the assumption that various languages can select various case hierarchies. Spanish uses the hierarchy in (49), while Russian/Ossetic Caha’s (2009) sequence in (51):

We have introduced a third type of oblique dom languages, those that use the locative, with a well-known case from Romanian. We have seen one example in (9a); we present two other sentences here. The proper name in (52a) needs obligatory differential marking, which builds on the locative pe. For some speakers, the proper name also needs obligatory clitic doubling, using the accusative inflection. The inanimate in (52b) cannot take the differential marker, in this context. Instead, it shows up with morphology which is homophonous with the nominative.28

The obvious means that comes to mind in order to capture the loc=dom syncretism is to use Caha’s (2009) full hierarchy, which we repeat in (55). The adjacency between acc and loc₁ does derive the facts without incurring an *ABA violation. Moreover, the placement of the dative and the genitive outside the accusative can also explain the dative-genitive homophony which extends to all nominals in the language, both in the inflected realization and in the prepositional form. In (53) we illustrate the inflected genitive/dative, while in (54) we see the prepositional variant (some types of prepositional dative are colloquial):

However, we immediately run into a problem, as Romanian also exhibits acc-dat syncretism, as shown in Table 7 which contains the clitic paradigms.30 The sequence in (55) cannot capture the acc-dat syncretism without *ABA, as the genitive and locatives are interveners. Another complication is given by the presence in the language of bare locatives, as in (56). These are syncretic with the bare accusatives (and thus with nominatives), as in (52b).31 This locative-accusative-nominative syncretism is morphologically distinct from the locative-accusative syncretism we have seen with dom, which uses the obligatory preposition pe.

To summarize, in Romanian we notice the following case syncretisms:

None of two hierarchies we have introduced can capture all the patterns. Caha’s (2009) sequence in (55), which we repeat in (57), gets the various dative-genitive syncretisms as well as the locative-accusative syncretism. But it cannot capture the accusative-dative syncretism, which will need a hierarchy more similar to Harðarson’s (2016) instead. But the latter does not capture the acc-loc syncretism.

Of course, one could hypothesize that Romanian needs to use two distinct case hierarchies. However, this solution misses some points, leaving aside its redundancy and lack of elegance. First, differential object marking interacts with clitic doubling. In contexts such as (52a) and many others, clitic doubling is obligatory for the greatest majority of native speakers. Secondly, a hierarchy such as (58) does not contain the locatives, which are crucial for differential object marking patterns. The question is whether a uniform case sequence could be devised in such a way as to capture all the syncretism patterns of Romanian and the other languages presented here (Spanish and Russian/Ossetic), and extend to ergative-absolutive languages, while at the same time including the locatives too.

Let’s start with the complex case of Romanian, a Romance language where oblique dom is built on a locative preposition and which exhibits the following syncretisms:

We have also seen that none of two case hierarchies containing only one accusative, repeated below, can capture the data. They both lead to *ABA violations. Caha’s (2009) sequence in (62) can capture the accusative-nominative as well as the genitive-dative syncretism, but cannot derive the accusative-dative syncretisms. Harðarson’s (2016) sequence in (63), on the other hand, cannot get the accusative-locative syncretism which is a hallmark of oblique dom in this language.34

Let’s see now what the enriched hierarchy predicts. We repeat it here, in (64), and we also insert Table 9 containing the relevant fragments from Romanian:

Table 9 and (65) illustrate very clearly how the syncretisms are captured: the nominative and the accusative are homophonous for both animates and inanimates. There is also syncretism between the accusative and loc₁ (for inanimates). As these three categories are adjacent, no *ABA violation obtains. Moving down the hierarchy, the dative-genitive syncretism is also captured. Sdat and gen are adjacent. Then we also obtain the locative-dom syncretism produced by loc₂ and Bacc, which are contiguous too. And lastly, the accusative – dative syncretism involves Bacc and Bdat.35 Thus, an enriched case hierarchy, containing several accusatives, datives and locatives does not run into any *ABA violations.

With this enriched case hierarchy we can also derive Spanish,36 where oblique dom is syncretic with both dat and the a- locative, and nom and Sacc use the same morphology, as in the updated fragment in Table 10 and (66).

And yet the same enriched hierarchy also captures Russian/Ossetic, where dom uses the genitive. The relevant Russian case fragment in Table 5 is repeated with adjustments in Table 11 and (67).38 Russian does not lexicalize Sdat. nom and Sacc are adjacent and syncretism can apply. With masculine animates the genitive is used in oblique dom. The presence of loc₂ cannot be motivated either, thus gen and Bacc are adjacent. A very similar picture is seen in Ossetic, as exhibited by Table 6. One observation about Russian is that it contains a category called the ‘prepositional’, which shares case morphology with the genitive. We follow the ordering in Caha (2009) under which the prepositional is found after the genitive (and thus after Sacc in our enriched hierarchy).39

Oblique dom is found in ergative-absolutive systems too, raising the question of whether an enriched case hierarchy is also applicable to these languages. Caha’s (2009), Harðarson’s (2016) or Starke’s (2017) hierarchies have been formulated for nominative-accusative systems; in general, work on case hierarchies in ergative-absolutive systems is less advanced.

However, some important remarks have been made by Blake (2001), who has formulated the hierarchy in (68), or Smith & al. (2018), as well as Zompì (2019), a.o. The latter two works do not assume strict case hierarchies, proposing instead that an important difference is made in Universal Grammar between core cases and oblique cases. For example, Zompì (2019) has motivated the organization of cases in (69), which is assumed to regulate both nominative-accusative systems as well as ergative-absolutive systems. In the latter the unmarked case is the absolutive (71), while in the former the nominative counts as the unmarked (70). The unmarked case as well as the accusative/ergative are set aside from obliques in that they are core, structural cases which are calculated using the dependent case algorithm (see also Section 7.1).

Let’s see now how oblique dom can be captured in an ergative-absolutive system, against this background. We will illustrate with two languages: Gujarati (Indo-Aryan) and southwestern Basque. As we have seen in Section 3 (Table 1) the puzzle with oblique dom is that it passes syntactic diagnostics which make it more similar to absolutives than to oblique datives.

In Gujarati certain types of higher animates must be introduced by a postposition which is homophonous with the dative; we repeat a relevant example in (72b). Extremely important is the fact that, in the perfective, oblique dom exhibits agreement patterns which are characteristic to absolutives, as seen in (72a) and (72b). In both examples, the past participle must show agreement with the direct object, irrespective of whether the latter is differentially marked or not (see especially Cardona 1965; Mistry 1976; 1997; Masica 1982; Magier 1983; Comrie 1984; 1991; Wunderlich 2012; Woolford 2006; Grosz & Patel-Grosz 2014; Joshi 2020; a.o.)

Indirect objects, or lexical datives more generally do not allow object agreement. Thus, in (72c), only default agreement is possible, as overt ergatives cannot agree either.

Under the assumption that oblique dom is a type of absolutive, not a type of ergative,40 the problem is that oblique dom (abs)-dat gives rise to an *ABA violation under the sequence in (68)/(71). Here, the ergative case zone intervenes between the absolutive and the oblique. As Blake (2001), Smith & al. (2018) or Zompì (2019) show, erg cannot be below dat/oblique cases. Thus, it will intervene between abs and dat. As a result, we cannot derive the syncretism, even under the assumption that cases are not ordered inside case zones. Ordering, however, must be assumed across case zones in Zompì’s (2019) system.41

In Zompì’s (2019) hierarchy in (69)/(70), the dom-obl syncretism might be derivable in nominative-accusative systems, as the accusative and the oblique are contiguous case zones (although we have seen numerous problems with a non-enriched case hierarchy in nominative-accusative languages). The challenge in ergative-absolutive languages is that oblique dom is seen in configurations that also show the ergative, thus in a straightforward ergative alignment. The way out could be to say that oblique dom is a type of accusative (and not a type of absolutive). This will entail that in languages like Gujarati the ergative, the absolutive and the accusative are all needed in the same alignment. A potential wrinkle is that oblique dom is possible on some types of subjects, as we briefly discuss in the next section with respect to dom passives. Such contexts are not trivial and more investigation is needed in order to safely establish whether oblique dom is a type of accusative or a type of absolutive.

Our proposal is that an enriched case hierarchy is also active in ergative-absolutive languages. Just like nominative-accusative languages, these systems too contain more than one case associated with direct objects (absolutive or accusative). Thus, we do not run into any of the problems raised by the hierarchies in (68)-(71). Similarly, these languages also give indication that more than one locative is active in the hierarchy. Additionally, there is more than one case for indirect objects too.

To evaluate the advantages of an enriched hierarchy, let’s take a look at a fragment of Gujarati case markers, as in Table 12.42 What we see here is that a hierarchy with two cases for the direct objects, as well as two datives captures the facts. The data support the following sequence, which contains both a smaller case (Sa) and a bigger case (Ba) for the direct object:43

We notice that Sa is homophonous with the unmarked case, while the ergative, loc₁ and Sdat are syncretic and contain more reduced morphology than other cases. This indicates that they are adjacent in the sequence, following Sabs. The next case is the genitive which introduces the morpheme -n. A second dative, as well as the differential marker contain the morpheme -n, to which further material is added. As these latter two categories are adjacent, their syncretism can be easily derived. A similar organization of cases is seen in other Indo-Aryan languages such as Hindi and Kashmiri.

A potential problem is raised by examples such as (72c), where we see the ergative containing the morpheme -n. Is this a possible violation of the *ABA constraint? A careful examination of these patterns indicates a purely phonetic status of -n, which is needed to facilitate the syllabification of a vowel added to the sequence kišor. In fact, the same -n is not grammatical in the construction of the ergative forms of the other two proper names in (72): *sudha-ne, *raj-ne (these two forms are well-formed, but not as ergatives).

A second example of oblique dom in an ergative-absolutive system comes from southwestern Basque. In examples (12a) and (12b) (repeated here in (74a) and (74b)), we illustrated two types of direct objects (Mounole 2012; Odria 2014; 2017; 2019; Fernández & Rezac 2016; a.o). On the one hand, there are the unmarked absolutives with absolutive agreement on the verb, as in (74a). On the other hand, there are the highly referential direct objects which can/must be signaled by dative morphology, as in (74b). The latter can also trigger dative agreement on the verb, although their syntactic behavior is absolutive, as demonstrated with an extensive set of diagnostics by Odria (2014; 2014; 2019) or Fernández & Rezac (2016), a.o.44

Let’s examine nominal inflection in the language, included in Table 13.45

The language treats Sa and erg separately from the other cases. Sa is unmarked, while erg uses the desinence -k, which is not seen with any of the other cases. The latter appear to use the morpheme -re with its allomorphs, suggesting that they have a common core, which is not shared with the ergative. Although more needs to be said about the more precise orderings of cases such as the genitive,46 the benefactive or the comitative,47 all that matters to us is that there must be a second absolutive/accusative which shares the same morphology with the dative. In a system that contains only one case for direct objects these facts won’t be derivable.

As already mentioned in Section 6, Zompì (2019) puts forward the restricted case hierarchy in (69), which we repeat in (75).

Here, the leading hypothesis is that cases are partitioned into classes, which are rigidly ordered relative to each other, although internal ordering within a certain class might not be strict. The hierarchy in (75) collapses both nominative-accusative and ergative-absolutive systems. The unmarked class contains the nominative (in nom-acc languages) and the absolutive (in erg-abs languages), while the next case zone contains the ergative (in erg-abs languages) and the accusative (in nom-acc languages). And lastly, the third zone contains the inherent/oblique cases.48 The logic is that the representation of an inherent case must contain that of a marked core case. The latter, in turn, must include the representation of an unmarked core case, according to the generalization in (76). Thus, the *ABA Generalization reduces to successive containment relations between the case categories.

One question that is relevant here centers around the status of oblique dom. Is it a type of absolutive or a type of accusative in an ergative-absolutive alignment? If oblique dom is a type of absolutive, the hierarchy in (75) does not capture the facts. It might appear desirable to say that oblique dom is always a type of accusative, and not a type of absolutive, even in erg-abs languages (see also Baker 2015; Sec.3.3).49 This would entail that erg-abs languages with oblique dom grammaticalize both the absolutive and the accusative, as direct object cases. One problem is that oblique dom appears to be preserved on certain types of subjects in many of the languages discussed here. For example, we see that in Hindi, oblique dom is preserved (in fact, in certain instances even required) on the sole argument under passive voice. An example is provided under (77).50

Additionally, oblique dom appears to pass subjecthood diagnostics in that it can control the possessive anaphor apnaa (only subjects can do so, see Mahajan 2010, a.o.). This behavior makes it more similar to absolutives. Baker (2015; 2021) proposes to account for the presence of dom in these types of passives under the assumption that such configurations contain an implicit agent. However, demonstrating that this implicit agent is syntactically active is no simple task. A quick examination of the most common diagnostics used to probe the presence of the implicit agent in the syntax (depictives, reciprocals, etc.) does not yield positive results for these configurations.51 Similar challenges are seen for other languages in which oblique dom is possible in medio-passives, for example the Spanish se_mp illustrated in (29b). Irimia (2021b) contains a detailed discussion on the interaction between dom and medio-passives, with a thorough illustration of all the diagnostics for both Hindi and Spanish. Of course, further research is needed into this aspect.52 In any case, the enriched case hierarchy we have proposed avoids these problems.

Another point refers to the precise organization of the case zones. The ergative and the accusative are assigned to the same class, to the exclusion of inherent cases. Their grouping is motivated by these two case categories being derived as dependent cases (Marantz 1991; Baker 2015; Levin & Preminger 2015; a.o.). In a nutshell, the mechanics of dependent case presupposes: a) cyclical assignment of case in the relevant domains (TP and NP/DP); b) scanning of the domain by a case-assignment algorithm, following the three strictly ordered steps (using Zompì’s 2019 model) in (78):

In the ergative-absolutive languages examined here the challenge is that erg and oblique dom (the accusative) are not dependent on each other. We have seen that in the ergative alignment, the object can appear unmarked; also, oblique dom is possible in the nominative-accusative alignment with non-perfectives, in the absence of the ergative. Oblique dom is a matter of properties of the nominal, and does not depend on a higher nominal (in fact, as we have just pointed out, it is not even clear that in examples with passivization such as (77) there is a higher implicit argument in the syntax). Baker (2021) proposes a solution for this problem using Hindi and southwestern Basque under the assumption that erg and oblique dom enter into the dependent case calculus in different domains (υP vs CP).53 However, it is not clear the same solution can be applied to the other ergative-absolutive languages, such as Gujarati, where both unmarked and marked objects appear to raise quite high.

Nominative-accusative languages pose their problems too. For example, in languages like Neapolitan where there are agreeing objects and agreeing objects that are also differentially marked, and where both classes can be clitic doubled using the accusative form of the clitic. It is not clear which of these accusatives enters the dependent case calculus and how to apply the mechanics of (78) unstipulatively. The analysis proposed here avoids these wrinkles too.

On the other hand, it is also true that Zompì (2019) has brought some counterarguments to enriched case hierarchies in the sense that they appear to both under- and over-generate. However, most of the problematic issues that have been pointed out refer to the place and nature of genitives which i) might not have a fixed position (see also Smith & al. 2018, a.o.); ii) in some languages might be syncretic with nominatives. We show that these aspects are not insolvable.

First, as also pointed out in fn. 46 the genitive is complicated in that it is both a verbal and a nominal case. It is clear that the nominal domain needs a separate case hierarchy from the verbal domain, as emphasized by Zompì (2019) himself. Secondly, the observation that the genitive might be spelled out with a bare form, similarly to the nominative, is not surprising. An extensive body of research has shown that (some types of) genitives have a structural nature, signaling the external argument in the nominal domain. Thus, in order to set this problem, a careful examination of the genitive in the nominal domain is necessary. In this work we have only focused on verbal cases, as dom does not appear in nominalizations in the languages in our database, and thus cannot provide the full solution to the genitive puzzle. Also note that one example of gen-nom syncretism Zompì (2019) points out comes from a honorific genitive in Icelandic (and only for 1pl under suppletion). But here, once again, there might be an independent explanation. Honorifics are categories with special syntax and a nominalization structure has been proposed for them (Hill 2014; a.o.).

Similarly, independent facts are relevant for the observation that enriched case hierarchies predict nom-Sdat syncretism, when Sacc is not lexicalized. But here, the problem is that there don’t seem to be that many languages which do not contain an accusative, due to the importance of this category. As Bacc tends to be linked to differential marking, due to its complex structure, a language would not easily lexicalize just Bacc.

Noticing that restricted hierarchies are not enough to derive the various syncretism patterns seen with accusatives, work by Bárány (2016; 2021) or Graf (2019) makes extensive use of case hierarchies which are partially ordered. A partial ordering of cases can avoid the need for more than one instance of the same case in a hierarchy, while still obtaining case contiguity. The main intuition is that a case sequence does not exhibit a total order, but is rather a partially ordered set (poset). Thus, cases in a particular hierarchy should not be obligatorily ordered with respect to each other.

Let’s look at a concrete illustration from Gujarati. In the case paradigm fragment in Table 12, we have seen the following: an absolutive which is spelled out unmarked; oblique dom which is homophonous with dat and a type of locative, all of them being spelled out as -ne; the ergative which is homophonous with another locative, both of them being spelled out as -e; the genitive which uses the morpheme -n and agreement morphology. The partially ordered hierarchy proposed by Bárány (2021) is in (80), while the vocabulary items which correspond to the Gujarati case markers are in (81). In (80), the order is partial in the sense that not all cases are ordered with respect to each other. erg and loc are ordered separately from acc and dat. Both these groups include the feature {A}, just like the unmarked absolutive and the genitive, and thus are comparable with these latter too.

As Bárány (2021 also shows, the partial ordering in (80) captures some characteristics of other simple (i.e., ‘non enriched’) case sequences. For example, case contiguity is weak as in Harðarson (2016); classes of cases, as opposed to individual cases, are the ones that get ordered, and dependent and oblique cases are not interleaved, similarly to Zompì (2019).

Generally, although case hierarchies in terms of partial ordering require more attention due to their potential, there are several reasons for which we have not assumed them here. First, some additional constraint needs to specified so as to block over-generation, as noticed by Bárány (2021) himself; to focus just on oblique dom, the system predicts syncretism with various types of obliques. The cross-linguistic picture is much more restricted. Secondly, as we have shown in the previous subsection, the divide ‘dependent/oblique’ is not as strict as it might seem. Thirdly, the presence of more than one case for direct objects predicts and captures better the syntactic distinctions between them (as discussed in subsection 4.1), despite a general accusative profile.

Fourthly, even under partial ordering, it is still necessary to have two datives54 in the hierarchy. For example, as we have seen in subsection 5.3 (ex. (53a) and (54a)), in Romanian there are both an inflected dative as well as a directional, prepositional (la) dative. They can both get doubled by the dative clitic, indicating that they are datives, and not just locatives. In order to capture both, we need two datives.55 The issue is important, as only one of these datives interacts with oblique DOM, which is not spelled out with the la marker. Two datives are also seen in Gujarati (Table 12), or other Indo-Aryan languages.

It could be argued that datives should be set aside from dependent cases such as the accusative/absolutive, because they are inherent. However, what we see cross-linguistically is that the separation between accusatives and datives is not as straightforward as expected. Of course, more needs to be said about this and the limits of this paper do not allow us to do justice to this matter. However, what we notice in Romanian, as well as western Romance is the accusative-dative syncretism for clitics, irrespectively of how dom is realized; clitics, however, interact with dom in non-trivial ways.