2.1.1 Spanish

Spanish has differential object marking expressed by the marker a preceding a proper subset of its direct objects. This marker is often said to be triggered by animacy and definiteness, but the factors determining DOM are complex and there is a lot of variation across varieties of Spanish (see Jaeggli 1982; Suñer 1988; Brugè & Brugger 1996; Torrego 1998; 2010; Gutiérrez-Rexach 1999; Aissen 2003; Leonetti 2004; 2008; Rodríguez-Mondoñedo 2007; von Heusinger & Kaiser 2011; López 2012).

Example (1), repeated below, shows a paradigm of DOM in Spanish, with animacy distinguishing the morphologically unmarked definite direct object in (1a) from the DOM object in (1b). (1c) shows a ditransitive construction in which the indirect object (the recipient argument) a la mujer appears with the dative marker, homophonous with DOM in (1b).

The passive counterparts of the sentences in (2), repeated below, show that while the direct objects in (1a,b) can undergo passivisation, the indirect object in (1c) cannot. (2d) shows that a verb which takes only a dative object does not support passivisation of this argument either, indicating that a ban on passivising datives does not just hold in ditransitives.

These examples show that the properties that determine DOM for direct objects do not affect the ability of undergoing passivisation. Rather, direct objects differ from indirect objects in being able to undergo passivisation, independently of whether they would trigger DOM or not. The absence of DOM on the logical object after passivisation and expected if DOM is an allomorph of accusative. This is one indication that DOM, as accusative, is a case-marker of the direct object which is distinct from the marker of the indirect object in Spanish (see Montalbetti 1999 for a few apparent cases of datives passivising in varieties of Spanish, which do not affect this argument).²

2.1.2 Hindi

Hindi, too, has differential object marking: animate and specific direct objects get the case suffix or postposition -ko which is homophonous with dative case (Mohanan 1990; 1994; Butt & King 1991; Butt 1993; Bhatt 2007). (3a) shows that animate objects are case-marked and can be interpreted as definite and indefinite. (3b,c) show that inanimate objects can be case-marked or not, but case-marked nouns are interpreted as definite.³

Mohanan (1990) suggests that -ko is the exponent of two distinct cases: accusative and dative. One of her arguments comes from passivisation: accusative is not retained under passivisation, while dative is. (4) shows this for a monotransitive predicate, and (5) for a ditransitive predicate.

Mohanan’s (1990) translation of (5b,c) indicates which argument is the subject in each clause. While case-marking is identical in both sentences, Mohanan suggests that the sentence-initial phrase is the subject in both cases.⁴ Accusative and dative arguments thus behave alike in that both can be promoted to subject. However, accusative is not retained under passivisation, while dative is. Mohanan (1990: 121) suggests that this is natural, since “DAT is not associated with any particular grammatical function, but with a specific meaning”. In other words, it is not a structural Case, while accusative is.

The variety of Hindi described by Mohanan (1990; 1994) thus patterns with Spanish in that DOM is not retained under passivisation, and direct objects with DOM pattern with morphologically unmarked direct objects and not indirect objects. Mohanan (1990: 122–125) does point out, however, that there exist what she calls “accusative preserving dialects” of Hindi, too (cf. also Bhatt 2007).

For these varieties, Mohanan (1990) suggests that accusative, like dative, is an indirect case, i.e. a case that is not tied to a specific grammatical function and is therefore preserved when the grammatical function changes from grammatical object to subject. In contrast, in non-accusative-preserving varieties, accusative only appears on grammatical objects (cf. the distinction between structural and inherent Case in Chomsky 1981 et seq.).

The two cases accusative and dative still cannot be fully equated, however. The reason is that accusative, even in the accusative preserving varieties, is not determined by meaning, but can rather express a range of thematic roles. In (7), Mohanan (1990: 124) describes these as simply “affected”, “affected and undergoer of a change of state or location”, and “experiencer, and undergoer of change of state”.

In the accusative preserving varieties, accusative is, obviously, retained under passivisation.

However, arguments bearing the same thematic roles can appear as nominative subjects in active constructions with identical meaning.

The crucial point here is that accusative arguments, even in the varieties that retain accusative, express a variety of thematic roles, and that they are all logical objects. Where accusative is not retained under passivisation, accusative only appears on grammatical objects. Where it is retained, accusative only appears on logical objects but it is not restricted to grammatical objects. However, the grammatical subject of a passive is a derived position. Whenever accusative appears in a non-derived position, it is a grammatical object, in both varieties.

This is in contrast to dative, which can also be a grammatical and logical subject, but is associated with specific thematic roles (Mohanan 1990: 185–186) and can be a derived, as well as a non-derived subject and object.

Rajesh Bhatt (p.c.) also informs me that there are subtle semantic differences in the two types of passive in Hindi. In (10b), where the logical object retains -ko the predicate is interpreted as more agentive:⁵

Passivisation in Hindi mostly patterns with passivisation in Spanish: in the main variety Mohanan (1990) discusses, DOM is not retained under passivisation, as is natural if DOM is accusative and thus a structural case. Dative, on the other hand, is retained. The situation is less clear than in Spanish, as there are speakers who retain DOM on direct objects under passivisation, too. Importantly, however, even in these varieties, DOM and dative show some distinct behaviour. While dative can appear on non-derived subjects, as well as (indirect) objects, the exceptional accusative only appears on derived subjects.

2.1.3 Kashmiri

Differential object marking in Kashmiri appears in two domains. First, resembling Hindi and Spanish, Kashmiri marks specific, animate direct objects with a morphological case homophonous with dative, while non-specific, inanimate direct objects appear in their nominative form (Wali & Koul 1997; Bhatt 1999). With respect to passivisation, Kashmiri behaves like Hindi: while some speakers allow retaining accusative on passivised direct objects, such logical objects can also become nominative subjects. This is never possible with dative indirect objects. This is illustrated in the following examples.⁶

Wali & Koul explicitly compare (11) to indirect objects which also retain their dative case under passivisation. Like direct objects with DOM, such indirect objects do not control agreement either. Unlike direct objects, however, indirect objects never lose their case-marking.

Crucially, direct objects only retain their case-marking optionally: this means that direct objects behave as a natural class with respect to passivisation, independently of whether they would trigger DOM or not. Indirect objects are different: their dative case-marking is always retained, as shown in (12).

Second, differential object marking with personal pronoun direct objects is sensitive to aspect and the person of both the subject and the direct object. This is a so-called global case split: it is global because the properties of two arguments determine case-marking on one of them, rather than just the properties of a single argument (as in a local split; cf. Silverstein 1976; Malchukov 2008; Keine 2010; Georgi 2012; Bárány 2017 in general, and Wali & Koul 1997; Béjar & Rezac 2009 for Kashmiri). Descriptively, in the non-perfective aspect, personal pronoun objects appear in their dative form if the person of the direct object is higher than the person of the subject on the hierarchy in (13):

Examples are shown in (14). The case highlighted for each example indicates the case of the direct object. In (14b) and (15b), the object’s person is higher than the subject’s on (13) and therefore the object appears in a form homophonous with dative.

In this second domain of differential object marking, passivisation works in the same way. In (16a), the active sentence, the direct object (and the indirect object) are both in their dative forms. In (16b), the logical direct object is promoted to subject and is morphologically unmarked (cf. also Béjar & Rezac 2009: 65 who take this is an argument that DOM is a structural Case, but dative is an inherent Case in Kashmiri).

2.1.4 Interim summary on passivisation

Examining patterns of passivisation in Spanish, Hindi, and Kashmiri, we find that DOM objects behave like other direct objects, rather than indirect objects in each language. Direct objects can be promoted to subject and lose their case-marking, controlling agreement with the finite verb. None of these languages allow indirect objects to undergo the same process.

Hindi and Kashmiri differ from Spanish in that both have varieties in which DOM can be retained on the logical object. While retaining case-marking on the passivised object is possible for direct objects, this is not the case with indirect objects. These must retain their dative case.

2.2.1 Spanish

Spanish allows reduced relatives modifying a DP, as shown in (19). Both el libro ‘the book’ and la mujer ‘the woman’, understood as the logical object of visto/a ‘seen’ can be modified by a reduced relative clause.

In ditransitives, it is not possible to relativise the indirect object (the recipient), as shown in (20a), but it is possible to relativise the direct object (the theme), as shown in (20b).

The reasoning here is the same as with finite passivisation: if a direct object that can get the DOM marker is introduced as an indirect object, it should not be affected by the verb’s inability to assign accusative. Yet direct objects with and without DOM pattern together, independently of their ability to trigger DOM, and they pattern to the exclusion of indirect objects. Note also that it is not an inherent inability of ditransitive constructions to appear as reduced relatives, as (20b) shows: the direct object can be relativised in the context of an indirect object as well.

2.2.2 Hindi

According to Rajesh Bhatt (p.c.), reduced relatives in Hindi show the same pattern as reduced relatives in Spanish. DOs can be relativised, independently of their ability to trigger DOM, while IOs cannot. This is shown in (21).

While these data resemble the passives discussed in Section 2.1 to some degree, they show a clear-cut difference between direct and indirect objects: direct objects, whether or not they trigger DOM, can head reduced relatives, while indirect objects cannot.

2.3.1 Spanish

Spanish behaves like English (and Basque, see below) in allowing the subject and the direct object to control a depictive secondary predicate, but not the (goal) indirect object.

The examples in (23) illustrate the relevant patterns. First, (23a) has a reading in which the modifier rota ‘broken.F’ is a secondary predicate rather than a postnominal adjective. The object, being inanimate, is a direct object without DOM.

(23b,c) illustrate a pair of sentences with monotransitive predicates taking dative and accusative objects, respectively. As gender agreement on the secondary predicate borracho/a ‘drunk’ shows, only the DOM object in (23c), but not the dative in (23b) can control the secondary predicate (see also Bresnan 1982: 401 for this argument).

(23d,e) make the same point with ditransitives: again, it is only the direct object that can control a secondary predicate, but not the homophonous indirect object.⁸

These examples show that both the subject and the direct object can control the depictive secondary predicate borracho/a ‘drunk.M/F’, but the indirect object cannot. Like in passivisation, direct objects pattern together irrespective of whether they trigger DOM or not, to the exclusion of the indirect object.⁹

2.3.2 Basque

Odria (2014) shows that as in English and Spanish, Basque direct objects (and subjects) can control depictive secondary predicates, but not indirect objects. For those varieties of Basque which have DOM (marked with a suffix homophonous to dative), this is true of both unmarked and marked direct objects. (24a) shows this for the absolutive object umea ‘child.ABS’ while (24b) illustrates this with the DOM object zu-ri ‘you-DAT’.

2.5.1 Agreement and case-marking in Kashmiri

As briefly mentioned above, DOM in Kashmiri is not determined based on properties of the direct object alone, but it relies on properties of both the subject and the object when the objects are personal pronouns and the sentence is in the imperfective (Wali & Koul 1997).

The relative person of the subject and the object determines whether pronominal direct objects appear in their morphologically unmarked (nominative) or a morphologically marked form that is homophonous with dative (Kashmiri is split-ergative: subjects are ergative in the perfective). This is determined based on the hierarchy shown in (28).¹⁰

To give a concrete example, the case-marking on a second person direct object depends on the person of the subject. If the subject is first person, the object surfaces in its nominative form, if the subject is third person, the object surfaces in its dative form. The relevant examples are repeated below:

Informally speaking, the direct object is dative when its person is on the same level or higher on (28) than the subject’s. Otherwise, it is NOM.

Consider now the behaviour of indirect objects. Indirect objects never show a NOM/dative alternation, independently of the person of the subject, the direct object and the indirect object. This is shown in (29a,b). First, in (29a), the indirect object təm-is ‘s/he-DAT’ is dative, even though third person direct object pronouns appear in their nominative form when the subject is first person. Second, as (29b) shows, the indirect object is dative in the perfective aspect as well, even though the NOM/dative alternation on the direct object only appears in the imperfective.

In addition, direct objects also alternate between nominative and dative in ditransitive constructions. In (30), both internal arguments are pronominal. With the third person subject in (30a), the direct object tse ‘you.SG.DAT appears in the dative, as expected from the hierarchical Case-assignment rule. The indirect object me ‘I.DAT’ is also dative.

In (30b), however, with a first person subject, only the indirect object is dative — the direct object is nominative, as expected. This shows that effects of the person hierarchy only affect the direct object, but never the indirect object.

Note again that this NOM/dative split is language-specific and therefore the resulting argument about the distinct behaviour of direct and indirect objects only holds for Kashmiri. Nevertheless, the data in (30) again clearly shows that direct and indirect objects do not behave alike syntactically, even if they share their morphological exponent.

2.5.2 Nominalisations in Spanish

Case-marking in nominalisations differs from case-marking in the verbal domain. In English event nominalisations, which retain the verb’s argument structure, nominative and accusative are generally not assigned to the subject and the object, respectively (Chomsky 1970; Grimshaw 1990). Instead, these arguments are often expressed as pre-nominal possessors or post-nominal genitives introduced by the preposition of (see also Comrie 1976; Koptjevskaja-Tamm 2003; 2015 for cross-linguistic overviews of coding of arguments in nominalisations).

This property of nominalisations provides a further test of the similarity of DOM and dative arguments: in Spanish, DOM is generally absent in nominalisations (but see footnote 11), but dative is present on exactly the arguments that are assigned dative from a verb.

If both DOM and dative are exponents of identical syntactic structures and their cases have the same source, the expectation is that both (or neither) can be retained in nominalisations. This is not the case: DOM is, like accusative, unvailable in nominalisations, while dative is available.

Consider the sentences in (31). (31a) shows a transitive sentence with a definite, animate object, Juan, preceded by the DOM marker a. In (31b), the object of captura ‘capture’ is still Juan, but it appears with the preposition de ‘of’.¹¹

In (31), an animate, specific object triggers DOM when it is the internal argument of a verb. When nominalised, DOM is no longer an option and the internal argument has to be introduced by de.

We can compare this pattern to the nominalisation of a ditransitive predicate like entregar ‘to deliver, to hand over’, which takes a direct object that can be coded with DOM, and an indirect object that appears in dative. When nominalised, the only way to express the direct object is with de, but neither as a bare noun nor with a. This is shown for an inanimate direct object in (32) and for an animate direct object in (33). In both cases, the (animate) indirect object a Susana retains its dative a.

In addition, the pattern remains the same if the indirect object is inanimate as in the following examples. This shows that the distribution of a on direct and indirect objects is not governed by identical semantic properties.

The behaviour of DOM in passives and nominalisations thus indicates that its source is the finite, active verb: when it is passive, DOM is unavailable. When it is nominalised, DOM is not available either. Both these patterns are straightforwardly captured by treating DOM as accusative, a structural case that is assigned by the finite, active verb in a certain structural configuration.

In some cases, a fails to be retained in nominalisations too, however (as also pointed out by an anonymous reviewer). López (2018: 93) argues that a in nominalisations is a preposition and not dative case. He bases this on the fact that dative a does not actually appear in nominalisations. His reasoning goes as follows. (36a) is an applicative construction which features both the clitic le and the marker a. In the absence of the clitic, in (36b), the preposition para introduces the beneficiary. In the nominalisation of this structure, only para but not a is licit. Thus, López argues, dative a is not retained and a in (32)–(35) is actually a preposition.

For López (2018), this means that there are three homophonous markers spelled out as a: accusative, dative and a directional preposition. While I will gloss over the distinction between dative and the preposition in Section 3 and just contrast accusative a with dative a, it should be noted that neither the dative nor the preposition have the same distribution as DOM. DOM does not alternate with para and neither a dative nor a prepositional phrase can passivise like a DOM object.

That the dative or preposition a is retained in passives and (some) nominalisations thus suggests that it is not assigned by the finite, active verb in the same way that DOM is.

2.5.3 Object clitics in Spanish

Varieties of Spanish show a lot of variation in their object clitic systems. Iberian Castilian Spanish, among others, distinguishes accusative clitics, lo, la, masculine and feminine, respectively, from dative le, which is underspecified for gender. These clitics can double the respective arguments, though again there are differences across varieties of Spanish. The following examples from Rioplatense Spanish show clitic doubling of direct object with DOM:

For varieties which do distinguish dative and accusative clitics from each other, the argument is straightforward. Even though DOM and dative objects are homophonous, they are doubled by distinct clitics, suggesting a categorical difference between the two types (see also Jaeggli 1982 for this argument in different terms).

There is, however, a lot of variation in Spanish clitic systems in this respect. Most importantly, there are so-called leísta varieties (see e.g. Bleam 1999; Ordóñez 2012), which neutralise the accusative-dative distinction in clitics in favour of a single form le.

In addition to leísta varieties, there are varieties which generalise the accusative clitics to indirect objects, so-called laísta and loísta varieties. Without going into details, the existence of syncretisms going both ways, i.e. from dative to accusative, and from accusative to dative, can also provide support for a morphological view of DOM and dative homophony, unless one wants to argue that laísta and loísta varieties treat “dative” arguments doubled by accusative clitics as accusative arguments.

While (to the best of my knowledge) all other arguments mentioned above still hold for these varieties of Spanish, some varieties do treat direct objects with DOM and indirect objects in the same way with respect to clitic doubling. As noted by an anonymous reviewer, more detailed analyses could potentially support a syntactic view of the homophony of DOM and dative in such varieties (e.g. Ormazabal & Romero 2013), but the existence of such varieties is compatible with a morphological view of DOM and dative homophony as well.

2.5.4 Allomorphy in Palizzese

As a final empirical point in this section, I briefly turn to allomorphy in Palizzese, a Southern Italian variety. In Palizzese, a precedes animate, definite DOs, as well as IOs. When an argument preceded by a also has a definite determiner (u), the items fuse and are realised as o. This is illustrated in (38). In (38a), the direct object is inanimate, but definite, so it does not trigger DOM but only appears preceded by the determiner u. In (38b), the direct object is animate and definite and the fused marker o appears:

In ditransitives, o can mark the IO:

Dative o has a genitive allomorph. In (40), the regular dative marker is replaced by the genitive marker du (cf. (39)).

It is striking, however, that the genitive allomorph is not available for direct objects which trigger DOM, as shown in (41). Replacing the case-marker of a direct object with DOM by the genitive leads to ungrammaticality.

These data show that while direct objects and indirect objects can get homophonous marking, a certain type of allomorphy is only available on indirect objects. Again, if DOM objects and indirect objects have the same syntactic representation, it is unexpected that only indirect objects can appear with the genitive allomorph, but not direct objects.

3.1.1 Spanish

López (2012) argues that direct objects without DOM are incorporated into V and are licensed in that way. He further argues that direct objects with DOM are in a higher position than zero-coded objects, in a projection above vP he calls αP (see López 2012 for evidence for this proposal). When a direct object DP is in αP, it is assigned accusative, when it is incorporated it is not assigned Case at all.¹⁴

Assuming that López’s analysis is correct, we can model the relevant parts of a derivation as follows.

In (45), the DP in SpecαP and the DP in SpecVP are assigned Case by different heads, v and Appl, respectively. v, by assumption assigns accusative (46a), while Appl assigns DAT (46b). They are spelled out identically because of a single spell-out rule, shown in (47a).

This simple system derives the homophony straightforwardly.

3.1.2 Hindi

Next, let us consider Hindi, in particular the analysis suggested by Keine & Müller (2008); Keine (2010). These authors propose that differential object marking can be derived purely in morphology. In Hindi, the suffix -ko can appear on animate and specific direct objects as well as any indirect object. Keine & Müller argue that both direct and indirect objects are assigned case in syntax, but that the direct object’s case features can be modified by impoverishment rules (see also Keine 2010; Bárány 2017 for general discussion).¹⁵

As for Spanish, I assume that direct objects and indirect objects are assigned Case by different heads in Hindi (see Bhatt 2005; 2007; López 2012 for discussion).

I assume case suffixes to have a more complex representation in Hindi than in Spanish. This is because Hindi, as a split-ergative language, also has an ergative case (not shown here, but see Keine 2010; Bárány 2017). The relevant case-markers and their spell-out rules are shown in (49) and (50). As above, the idea is that a single spell-out rule, (50a), spells out both accusative and dative, even though they have different syntactic representations.

These examples from Spanish and Hindi illustrate how a morphological approach can capture syncretism in DOM. Independently of its source, DOM is an allomorph of accusative case that is syncretic with dative case. This syncretism is purely morphological, and modelled by assuming that there is a single underspecified vocabulary item that is inserted for both accusative (DOM) and dative.

Because the two cases have different feature specifications in syntax and are assigned by different heads (as is generally assumed), their distinct behaviour in syntax need not be explained on this approach: the differences follow from their different syntactic representation. On this view, the only property direct and indirect objects must have in common is their morphological form.

4.1.1 Icelandic case-marking

In Section 3.2, I briefly discussed Harðarson’s (2016) revised version of Caha’s case contiguity hypothesis. As shown there, and repeated in (56), Harðarson shows that accusative and dative case are syncretic in some types of nominal stems in Icelandic. In (56), this is true of drottning ‘queen’ and tunga ‘tongue’.

Icelandic does not have differential object marking. While accusative case has several allomorphs (see again (56)), the choice of these does not depend on referential or semantic properties of nominals but on the stem class they belong to.

In this respect, Icelandic differs from the languages discussed in Section 2. However, it is similar to them in another respect: certain stems show syncretism of accusative and dative. Icelandic is well-known for allowing a wide range of cases on subjects and objects (see e.g. Andrews 1982; Zaenen, Maling & Thráinsson 1985; Yip, Maling & Jackendoff 1987; Thráinsson 2007). While both direct (accusative) objects and indirect (dative) objects passivise, for example, the two cases nevertheless show some differences: accusative case is not retained under passivisation, but dative is (cf. Hindi discussed in Section 2). This is expected if accusative is a structural case, while dative is a lexical case (Yip, Maling & Jackendoff 1987; Thráinsson 2007: 181–185). These distinct patterns are analogous to the behaviour of direct and indirect objects as discussed in Section 2 for languages with DOM.

This shows is that syncretism of accusative and dative is independent of differential object marking. What is more, if one were to say that the forms drottning-u ‘queen-ACC.DAT’ and tung-u ‘tongue-ACC/GEN/DAT’ are syntactically identical or special in some other sense, it becomes difficult to explain why other stem types, exemplified by e.g. arm-ur and land in (56) do not show the same syncretism.

In sum, Icelandic shows a clear example of syncretism of accusative and dative case in some nominal stems, i.e. a purely morphological phenomenon, in a language without differential object marking. The syntactic differences between accusative and dative in the language are parallel to those discussed in Section 2, which I take to support the claim that Spanish, Hindi, etc. also exhibit accusative (DOM) and dative syncretism.

4.1.2 Accusative recipients

A number of languages show argument structure alternations in (certain) transitive verbs. In many cases, two internal arguments, e.g. a theme or patient argument and a recipient or goal argument, can appear in two different in case frames. (57) exemplifies this using the well-known English alternation between the so-called prepositional dative and double object constructions (see a.o. Oehrle 1976; Barss & Lasnik 1986; Larson 1988; Johnson 1991; Pesetsky 1995; Harley 2002; Beck & Johnson 2004; Rappaport Hovav & Levin 2008; Bresnan & Ford 2010; Bruening 2010; Hallman 2015; Harley & Jung 2015; Harley & Miyagawa 2018).

For present purposes, the relevant property of this alternation is that, while the two constructions are fairly similar in their meaning, they differ in their morphosyntactic properties. In (57a), it is the theme argument that is adjacent to the verb and appears in object case, while the recipient is introduced by a preposition. In (57b), it is the recipient argument that is adjacent to the verb and appears in object case. Here, the theme shows the same case-marking (although this is more difficult to show). The two types of constructions show “indirective” and “secundative” alignment, respectively, using the terminology of Haspelmath (2005); Malchukov, Haspelmath & Comrie (2010) (see also Dryer 1986).¹⁶

In addition to identical morphology (“ACC”), the patient in (57a) and the recipient in (57b) also show similar syntactic behaviour. For example, both can passivise, while the respective other argument in each clause cannot (or not as easily). This is shown in (58) and (59).

Crucially, both the theme and the recipient can passivise, but in different constructions. They are straightforwardly passivised when they are in accusative case and adjacent to the verb.

English is obviously not the only language that shows an alternation of this type (see e.g. Bresnan & Moshi 1990; Harford 1991; Alsina 1996; Rackowski & Richards 2005; Baker 2010; Ackerman, Malouf & Moore 2015; Sipőcz 2016; van der Wal 2018). The following examples show data from the varieties of the Uralic language Khanty, in which ditransitive alternations of this type are found, too (Nikolaeva 1999b; 2001; Dalrymple & Nikolaeva 2011).

The data in (60) illustrate this for Northern Khanty (other varieties of Khanty behave alike, cf. Csepregi 2015; F. Gulyás 2015).

Like English, Khanty allows both themes and recipients to be coded as accusative. In both types of construction, the other internal argument is coded in a different way. Crucially, the accusative-marked argument shows syntactic properties that the other argument lacks. These include triggering agreement with the verb, as shown in (60), being in a higher (VP-external) position, see (61), passivising, see (62) and heading reduced relatives, as in (63).

These examples resemble the Spanish and Hindi data discussed in Section 2 in the sense that arguments with different semantic roles can be expressed in morphologically identical ways. But they differ from those data in the crucial respect that in Khanty, recipients and themes with identical case-marking show identical syntactic behaviour (the exact opposite of what the Spanish and Hindi data showed).

The reason for this is that Khanty (and English) show a ditransitive alignment alternation such that whichever internal argument is coded with accusative behaves like a direct (or primary) object (cf. Dryer 1986) while the other internal argument is oblique and cannot passivise, etc. In the Spanish and Hindi data discussed earlier, DOM does not affect which internal argument is the direct object: both morphologically zero-coded objects and those with DOM are direct objects. Indirect objects (marked with DAT) show distinct behaviour.

Before concluding this section, I briefly turn to a similar alternation in Spanish. English and Khanty show two types of ditransitive alignment for certain verbs, including give. It is also well-known, however, that English does not allow all verbs to alternate between the double object and the prepositional dative constructions. The verb donate, for example, favours the prepositional dative construction, i.e. alignment preferences can be encoded lexically as well.

Such lexical preferences are also found in other languages, including Spanish. The verb armar ‘to arm’ or ‘to provide with weapons’ differs from the predicates discussed previously in that its accusative argument is a recipient rather than a theme or patient. The theme argument, if expressed, is coded by the preposition con ‘with’. This is shown in (64).

As in Khanty above, in Spanish it is also the accusative argument that shows the typical behaviour of direct objects (cf. Section 2). The recipient argument can now trigger differential object marking:

The ACC.DOM coded recipient can be passivised:

The predicate armar also allows its R argument to head a reduced relative:

The R argument’s case changes to GEN in a nominalisation, a is impossible:

Finally, the R argument of armar can control a depictive secondary predicate:

In sum, with the predicate armar which assigns its recipient argument accusative case, the recipient shows the same behaviour as zero- or DOM-coded theme arguments discussed earlier. The recipient of armar behaves strikingly differently from the recipient of the predicates discussed in Section 2. I take this to mean that the syntactic Case assigned to an internal argument, be it a theme or a recipient, strongly correlates with the argument’s syntactic behaviour. Morphological form, however, is not a good predictor of syntactic behaviour.

5.1.1 Syntactic analyses

Manzini & Franco (2016) propose that DOM and datives are not just morphologically, but syntactically identical. The core of their proposal is that DOs with DOM and IOs are both sisters of the same head, a preposition they term P(⊆) (or Q(⊆) if the element is an affix).

Manzini & Franco (2016: 211–215) characterise these heads in informal semantic terms, suggesting that they express an “inclusion” or a “part-whole” relation, and that they are related to possession.¹⁷

Manzini & Franco (2016) also suggest that syncretism of dative and accusative in Romance pronominal systems instantiates the same overlap. (73) illustrates this for Italian. The verb parlare ‘to talk’ takes a dative object, while colpire ‘to hit’ takes an accusative object. This distinction is marked with third person arguments in (73b,c), but the cases are syncretic in first and second person.

Manzini & Franco (2016: 210) suggest that the syncretism with first or second person is not morphological, but that the objects of both parlare and colpire are structurally represented as datives, i.e. as in (73c).

They adopt Svenonius’s (2002) proposal that accusative and dative case are related to the way a predicate structures (sub-)events. Svenonius (2002: 197) proposes that accusative is assigned when two subevents overlap and dative when they are distinct, and discusses a range of different classes of verbs. Manzini & Franco (2016) adopt this distinction and argue that in Italian a third person accusative, e.g. the object of colpire ‘to hit’, is syntactically a D, while a third person dative, e.g. the object of parlare ‘to talk to’, involves a D head embedded under Q(⊆). They suggest that such a dative structure can be paraphrased as “I caused him to be on the receiving end of some talk.” (Manzini & Franco 2016: 216).

Manzini & Franco (2016) further propose that first and second person pronouns must be embedded in the dative structure (involving Q(⊆)), independently of the predicate. This suggests, of course, that to hit me has a different event structure than to hit him, because the former, but not the latter, is based on the dative structure. Manzini & Franco do not spell out what this difference is, however:

There is no a priori reason why an argumental frame including a Participant internal argument should reflect a complex organization of the event with verbs like ‘hit’ in [(73)]—while the embedding of a 3^rd person argument does not in [(72a)]. However the lexicalization patterns of Italian suggest that this is exactly what happens. DOM datives are no morphological accident—nor do they reflect morphological regularities. They arise in the syntax and they reflect a slightly different structuring of the event structure with Participant internal arguments. (Manzini & Franco 2016: 218, emphasis mine).

Without specifying the semantic effects of the “slightly different structuring” of subevents, Manzini & Franco’s (2016) argument w.r.t. to the structures shown here is essentially morphological, as they suggest that it is the “lexicalization patterns” of first and second person arguments which motivate representing them as datives in (73).

Discussing why DOs and IOs do not behave alike under passivisation, Manzini & Franco (2016: 219–220) then suggest that indirect objects are embedded under P/Q(⊆) because of selectional requirements of the predicates they are an argument of, while DOM DOs require P/Q(⊆) because of their referential properties (e.g. animacy and specificity). Merging IOs under P/Q(⊆) is thus obligatory and happens under passivisation, too. The authors attribute the impossibility of passivising indirect objects to their “inherent case properties”, which make them unavailable for movement and promotion to nominative.

This acknowledges the fact that passivisation is possible for DOs with and without DOM and this means that even DOM arguments are not obligatorily merged with P/Q(⊆). Manzini & Franco (2016: 220) propose that an LF constraint rules out first or second person objects inside VP without Q(⊆): a way to escape this constraint is to move such arguments outside of the VP. This is exactly what happens under passivisation, although Manzini & Franco (2016) do not make it clear what exactly happens to the P/Q(⊆) head. They conclude their discussion of the distinct syntax of IOs and DOs with respect to passivisation by stating that the “parallelism between accusative and DOM depends on the fact that they are both structural cases (not selected by the verb) assigned VP-internally” (Manzini & Franco 2016: 222).

While Manzini & Franco frame their argumentation in semantic terms, they do not provide independent evidence of how the relationship between the predicate and its theme arguments with and without DOM differ semantically and in what way the former resemble indirect objects semantically or syntactically. Their proposal of how to capture differences in passivisation introduces an additional constraint to account for asymmetries between direct and indirect objects, but again, it lacks independent motivation.