Oblique differential object marking and types of nominals

Abstract

The question of whether differentially marked objects should be linked with Case licensing or some other mechanism in the grammar has given rise to numerous debates. Addressing contexts of differential object marking (DOM) with oblique morphology, this article shows that, while the Case licensing approach might be adequate for varieties of Spanish, oblique differential marking rather signals an independent licensing operation, beyond Case, in languages like Romanian, Gujarati or Mandarin Chinese. This additional mechanism, relevant at the syntax-semantics-pragmatics interface, tracks the role of grammaticalized animates or how the speaker relates to other entities in the discourse. Additionally, the data examined here indicate that objects can come in a variety of sizes and structures, with distinct licensing constraints, such that the divide Case licensed/unlicensed or Case licensing/(pseudo-)incorporation is not enough.

Résumé

La question de savoir si les objets marqués de manière différentielle doivent être liés à au licenciement casuel ou à un autre mécanisme de la grammaire a donné lieu à de nombreux débats. Abordant les contextes de marquage différentiel d'objets obliques, cet article montre que, bien que l'approche qui lie le marquage différentiel avec le licenciement casuel puisse être adéquate pour les variétés d'espagnol, le marquage différentiel oblique signale plutôt une opération de licenciement indépendante, au-delà du licenciement casuel, dans des langues comme le roumain, le gujarati ou le chinois mandarin. Ce mécanisme supplémentaire, pertinent à l'interface syntaxe-sémantique-pragmatique, suit le rôle des arguments animés grammaticalisés, ou comment le locuteur se rapporte à d'autres entités dans le discours. De plus, les données examinées ici indiquent que les objets peuvent être de tailles et de structures variées, avec des contraintes de licenciement distinctes, de sorte que la distinction ‘licencié par le cas/non-licencié’ ou ‘licencié par le cas/(pseudo-)incorporation’ n'est pas suffisante.

1. Introduction

An important reflex of the internal structure of nominals is their behaviour in sentential syntax. Cross-linguistically, a relevant split has been observed between nominals that do not project a full structure (being NPs or NumberPs) and bigger nominals with additional functional heads, such as D⁰ or K⁰ (Bittner and Hale 1996, Massam 2001, Danon 2006, Ghomeshi 2008, López 2012, Levin 2015 for distinctions between KPs and DPs).¹ In many languages, more complex structure might induce interpretive and/or scrambling correlates, generally connected to the presence of a (uninterpretable) Case² feature that needs valuation in syntax. Reduced structure, instead, can be limited to non-specific interpretations, prototypical or number neutral readings, with the nominal and the verb sometimes even forming a unit. Under one possibility, the nominal head is incorporated into a verbal head (via head movement, or base generated X⁰ adjunction, Baker 1988), and they form together an X⁰ constituent. The literature has also identified a less radical process under which a nominal without functional extended projections is base generated as a complement to V and forms a minimal VP with the latter. This process, labeled pseudo-incorporation by Massam (2001), has been discussed for many languages, under many diverse realizations (Farkas and de Swart 2003, Dayal 2011, the contributions in Borik and Gehrke 2015).

In this article I examine issues related to nominal structure and size which are harder to accommodate under this broad split (namely, licensing in terms of Case vs (pseudo-)incorporation). I restrict my attention to phenomena in the class of differential object marking with oblique morphology (henceforth oblique DOM). A long-standing intuition associates the oblique marker with Case licensing on higher functional heads, such as D⁰, K⁰, forcing such nominals to escape (pseudo-)incorporation (Kornfilt 2003, Ghomeshi 2008, López 2012) and undergo obligatory licensing in the syntax (Ormazabal and Romero 2013a, 2013b; Kalin 2018).

The data I am concerned with come from Romance (Spanish, Romanian), Indo-Aryan (Gujarati), and Sino-Tibetan (Mandarin Chinese). A problem in these languages is that the morpho-syntactic encoding of their objects cannot be easily reduced to the split (pseudo-)incorporation vs. oblique DOM. One issue is that, besides oblique DOM, there appear to be other structural objects that come with positional and more general licensing constraints, similar to uninterpretable Case, suggesting that they too escape (pseudo-)incorporation and need licensing, even if they do not carry oblique DOM. These observations raise two questions, which will be addressed in this article: i) how to analyze oblique DOM so as to distinguish it from other non-incorporating objects; ii) what precise strategies correlating nominal size and licensing are possible in human language.

I propose that these, more complex, nominal patterns can be reconciled under the hypothesis that oblique DOM signals a separate licensing operation, beyond uninterpretable Case, on nominals with an enriched structure. This additional mechanism, relevant at the syntax-semantics-pragmatics interface (but distinct from topicality or specificity in some of these languages), is connected with the valuation of an interpretable feature, tracking the role of animates or how the speaker relates to other entities in the discourse. As the initial licenser (v ⁰) in the relevant domain (below TP) is needed for Case, any additional features will use an additional licenser (following Jaeggli 1982). The languages under discussion recruit a functional projection from the low discourse-related layer in the vP (following Belletti 2004 or Pancheva and Zubizaretta 2018), which will value any features left behind by the initial licenser, resulting in the spell-out of oblique DOM. The work also builds on recent discussions about a type of A licensing, related to δ (discourse) features (Miyagawa 2017, Belletti 2018, Mursell 2018), extending it beyond topics. I also show that reducing oblique DOM to a morphological operation cannot account for the various syntactic effects it gives rise to. Another line of research the article contributes to is the exploration of a syntax-pragmatics interface layer in the high periphery of nominals (such as the discourse information-bearing Speech Act domain, in Ritter and Wiltschko 2019, or Hill and Mardale's 2021 multi-layered DP).

The structure of the article is as follows. In section 2, I introduce data illustrating oblique DOM and other direct objects (DO) in standard Spanish. I review two prominent accounts: i) oblique DOM equated with Case, anti-incorporation and obligatory licensing (Ormazabal and Romero 2013a, 2013b), and ii) oblique DOM as obligatory raising (López 2012). In section 3, I present a related oblique-DOM prominent language, Romanian, where these two hypotheses appear to be problematic. Although similar to other objects that cannot be analyzed under pseudo-incorporation and which appear to require licensing, Romanian DOM is not easily identified via raising; moreover, it cannot be equated just with a morphological operation, as it gives rise to important syntactic effects. The hypothesis that DOM is a syntactic, A-related licensing mechanism beyond Case can capture its behaviour in a non-stipulative way. In section 4, I turn to Gujarati; in this language oblique DOM, similarly to other objects which appear to equally escape pseudo-incorporation, co-occurs with object agreement, which results from an independent nominal licensing mechanism. The idea of an additional licensing operation for oblique DOM explains these two co-occurring overt realizations (agreement and oblique DOM) for Gujarati direct objects. In section 5, I extend the same analysis to the ba marker in Mandarin Chinese, deriving its scrambling to a yet higher position than other objects which, similarly, cannot undergo (pseudo)-incorporation due to their complex structure. Section 6 summarizes the results and concludes with further brief remarks about the role of oblique DOM in the realm of licensing operations, and remaining questions.

2. Differential object marking in standard Spanish

Spanish direct objects come in a variety of sizes (Torrego 1998; Leonetti 2003, 2008; Bleam 2005; Laca 2006; Rodríguez-Mondoñedo 2007; López 2012). As illustrated below in (1a) they can be bare (especially if plural), can take an overt indefinite morpheme or the definite one. Additionally and mainly regulated by animacy, object nominals with overt (in)definiteness morphology can surface with a preposition which is homophonous with the dative, instantiating a type of oblique DOM. The preposition is obligatory with animate definites, as in (1b); with animate indefinites (1c), as the traditional wisdom goes, the preposition is restricted to a specific interpretation (Rivero 1979). In this article, I am precisely interested in investigating the nature of objects that take this prepositional marker and their syntactic relation to other direct objects.

Splits in the morpho-syntactic encoding of direct objects, based on features such as animacy, specificity, topicality, etc. are not rare cross-linguistically (Givón 1984; Comrie 1989; Bossong 1991, 1998; Aissen 2003; Rodríguez-Mondoñedo 2007; Haspelmath 2008; López 2012; Bárány 2017). In order to capture the complexities of the (standard) Spanish data, it is generally assumed that both an animacy and a specificity scale are necessary (forming a multidimensional DOM system):

Recent formal work has, however, demonstrated that an account in terms of scales reveals several shortcomings. A serious problem is that the prepositional marker must override its ‘canonical’ animacy and specificity features in a variety of contexts, which can instead be unified structurally (López 2012; Ormazabal and Romero 2013a, 2013b.). For example, Ormazabal and Romero (2013a, 2013b) note that nominals in Exceptional Case Marking (ECM)-type contexts take obligatory oblique DOM in Spanish varieties, even irrespectively of animacy. López (2012) has, likewise, presented examples with obligatory DOM in the absence of specificity or animacy.³

In order to provide a comprehensive account for these types of patterns, a more abstract condition on nominals has to be assumed, going beyond the encoding of animacy and specificity. Therefore, recent formal research has equated oblique DOM with the presence of an uninterpretable Case (uCase/uC) feature (Chomsky 1995, et seq.) which requires licensing in sentential syntax, thus forcing the nominal to escape (pseudo-)incorporation (see especially López 2012; Ormazabal and Romero 2013a, 2013b).

An important observation in this direction was made by Ormazabal and Romero (2007; 2013a, 2013b) and takes into account interpretational shifts that lack of DOM triggers in some contexts. Staying with (continental) varieties of Spanish for now, DOM is not possible in a configuration that also contains an indirect object which is clitic doubled.⁴ An example is provided in (4), from Ormazabal and Romero (2013b, ex. 2b).⁵ If both the indirect object (IO) and its dative (dat) clitic double are to be kept, then DOM must be removed.

Ormazabal and Romero (2013b:157) mention the following with respect to the interpretation of the animate DOM-less direct object in examples similar to (4):

the availability of (4) [O&R's 2b, my note] is extremely restricted. Sentences like (4) – [O&R's 2b, my note] are only grammatical with nouns such as sick people, soldiers, slaves, kids, etc.; nouns whose referents are regularly treated as entities lacking free will. The range of animate nouns that can appear without dom in this context is, more or less, the same one that allows incorporation in polysynthetic languages….⁷

In fact, DOM ungrammaticality is confirmed in contexts that flag (pseudo-) incorporation⁸ cross-linguistically (such as existential clauses, individual level have predicates, transitives with bare plural objects, see Bleam 2005; López 2012; Ormazabal and Romero 2013a, 2013b), as also summarized in Table 1 in subsection 3.2.

Table 1: DOM in Spanish and Romanian (fragment)

Based on remarks along these lines, Ormazabal and Romero assume that unmarked nominals either undergo complex predicate formation with the verb (plurals of type <e,t>, etc.) or simply stay unlicensed (definites), grouping together both NPs and DPs, as in (5a). DOM-ed nominals, on the other hand, have a more complex structure, projecting a KP layer which hosts the accusative Case feature, as in (5b). The latter needs to be valued in the syntax via raising to a position above VP, blocking pseudo-incorporation.

1. (5)

As further discussed by López (2012), DOM as an anti-incorporation mechanism can also explain the obligatoriness of the oblique marker in Small Clauses (SC) and other clause union contexts (see also fn. 3). Generally, only objects in a complement position to V can undergo (pseudo-)incorporation (Baker 1988, Massam 2001, López 2012). In SCs, on the other hand, the shared nominal is never found in a complement position to V, no matter whether these constructions are seen as projecting a reduced/small clausal structure, as in (6a) or as constructing complex predicates, as in (6b):

1. (6)

2.1 López (2012) – DOM and other nominals with Case

Although the unlicensed/licensed divide can explain the many DOM exceptions to the scales, it also leaves some questions unanswered. For example, assuming that SCs block nominal (pseudo-)incorporation and require licensing on the nominal, the prediction would be that inanimate objects should always show DOM in this context. This appears to be borne out in several varieties of Spanish; but standard Spanish speakers judge DOM ungrammatical on inanimate nominals in SCs, as shown in (7). There is, however, an important restriction – bare forms of nominals are completely ungrammatical and thus, (in)definite morphology is obligatory on the shared argument.⁹

To summarize, SCs show the following: i) DOM appears to be obligatory regardless of specificity (see López 2012) on animates (which moreover cannot be bare); ii) inanimates do not allow DOM (in standard Spanish), but cannot be bare. The obligatory presence of overt (in)definiteness can be taken as indication of a licensing condition in these anti-incorporation contexts too; only nominals of a certain size are allowed. This, in turn, suggests that the basic split (pseudo-)incorporation vs. Case marking is not enough. López (2012) motivates a similar conclusion.

Thus, López (2012) has a more nuanced take on the (pseudo-)incorporation vs. Case licensing issue and its relevance to DOM. The author explicitly mentions that the grammar of Spanish contains other nominals which have a structural [uCase] feature, besides the DOM-ed ones. Objects that carry the oblique DOM preposition are a sub-type of the Case licensed structural accusatives; what sets them apart is their obligatory raising to a position above VP in order to have their [uCase] valued. For López (2012), other (unmarked) objects with a structural [uCase] feature are licensed only by v⁰. In the case of definites, it is the definite functional head that incorporates into V and is licensed after V raising to v⁰.¹⁰

López (2012) provides arguments from binding, demonstrating that marked nominals are above the IO, but below the external argument (EA). Their accusative [uCase] is valued in a position above VP but below the EA. This is shown in (8a), where the DO raises to the specifier of an intermediate head α (which bundles aspectual and applicative features); in that position it can be probed by v ⁰. This short scrambling operation explains DOM presence in a position c-commanding the IO, and thus binding from DO into IO.¹¹

DOM-ed nominals contain a KP layer, where the Case feature, associated with a choice function (f), is housed (8b). According to López (2012), f switches the semantic type of the nominal from <e, t> to <e> (or a more complex type for quantifiers), and can only be interpreted in a position above VP.

Case valuation is understood as feature sharing under the Agree operation, which values or co-values all the uninterpretable features (uf) features on v ⁰ and DO, as in (9). López (2012) further assumes a locality restriction on Agree; the probe (v ⁰) can at most reach the specifier of its complement.

1. (8)

2. (9) Agree   (a[f], b[uf]) → (a[f], b[f])       (López 2012: 35)

The differential preposition is inserted at PF if certain conditions are met, for example if an animate feature is present, as in (10a), or whether α is specified as telic, as in (10b).

1. (10)

As already mentioned, López (2012) shows that there can be other nominals specified with a structural [uCase] feature, but which do not show the differential marker; nevertheless, all objects with [uCase], no matter whether differentially marked or not, have an accusative syntactic behaviour (hence the traditional label prepositional accusative for DOM). This observation is confirmed by a variety of diagnostics, such as the possibility of passivization, etc. (see also Bárány 2018). However, the more precise licensing mechanism for non-DOM-ed (and non-incorporated) direct objects with [uCase] needs further attention. An observation is that not all oblique DOM systems can be subsumed under the two theories presented above. In the next section, I turn to Romanian, a language which exhibits one of the most complex DOM systems not only within Romance, but also cross-linguistically.

3. Differential object marking in Romanian

Similarly to Spanish, Romanian oblique DOM is sensitive to animacy (Niculescu 1965; Dobrovie-Sorin 1990, 1994; Cornilescu 2000; Tigău 2011; Irimia 2020, a.o.). As illustrated in (11a), the animate definite can be introduced by a preposition which is homophonous with a locative.¹² In this type of context, an inanimate cannot take the differential preposition, as shown in (11b). The marked objects can (and in some configurations, must) also be clitic-doubled, using the accusative form of the clitic.¹³

Example (11a) illustrates a first difference from Spanish:Romanian DOM is optional with referential definite animates, as opposed to obligatory in Spanish (1b). One cannot conclude, however, that DOM is optional in Romanian. The challenge is that there are many configurations where absence of DOM results in ungrammaticality, as in (12). They illustrate: i) the negative quantifier nimeni (‘nobody’); ii) the animate wh-element cine (‘who’);¹⁴ iii) the elliptical demonstrative in its augmented form;¹⁵ iv) the D-linked element care (‘which’). The latter two are also part of a large class of contexts where DOM is obligatory irrespective of animacy (and specificity). In turn, in (12e) we see the object of a psych-verb (interesa) with oblique DOM and clitic doubling that appear to be needed for most speakers.¹⁶ In general, although Romanian and Spanish oblique DOM share a general profile, there are also important differences. In fact, given its many complications, Romanian DOM is still in need of an adequate formal explanation.

3.1 Romanian DOM and anti-incorporation

Given the obvious problems with accounts in terms of scales, we need to test the predictions of the two formal analyses introduced in the previous section for Spanish: i) oblique DOM as a licensing mechanism on nominals with a structural [uCase] feature, which need obligatory licensing in syntax (Ormazabal and Romero 2013a, 2013b); ii) DOM as a subclass of nominals with [uCase], signaled via overt raising (López 2012).

A conclusion is clear. Contexts signalling nominals (of type <e,t>) which might undergo complex predicate formation with V, or which can stay unlicensed in Spanish or cross-linguistically (such as existential clauses, individual level have predicates, transitives with bare plural objects,¹⁷ mentioned in section 2), do block oblique DOM in Romanian too (as seen in Table 1). For example, ungrammaticality of oblique DOM with possessive have under individual-level readings is illustrated below.¹⁸

Turning now to contexts that signal anti-incorporation, namely SCs (or clause union more generally), we notice an important difference from Spanish. Remember from the discussion in section 2 (and fn. 3) that Spanish animates generally require DOM in these contexts. This was one of the arguments for associating oblique DOM with a more abstract licensing condition, namely [uCase]. In Romanian, however, animate definites are possible without DOM in SCs; but, crucially, nominal distribution is not completely free – just like in Spanish, bare nominals¹⁹ are strictly excluded, as in (14). Also note that categories which need obligatory DOM outside SCs also need to have it here (15). This latter observation supports the conclusion that SCs are not only anti-incorporation contexts, but also signal (in these languages) an abstract licensing need on the nominal.²⁰

In Romanian, some bare nominals can be found in contexts that are similar to complex predicate formation, more generally (be it pseudo-incorporation or some other mechanism); thus, one can assume that these latter classes lack an [uCase] feature, as they are probably NPs or NumPs. Definites and specific indefinites, on the other hand, (can) project a larger structure as DPs. The obligatoriness of overt (in)definiteness morphology in SC contexts such as (14) in turn indicates that that this piece of morphology is associated with a structural Case feature, which allows it to escape complex predication formation with V (under pseudo-incorporation, etc.).²¹ The [uCase] feature is located in D⁰, under most accounts (see especially Giusti 1993). But this entails that: i) oblique DOM (which must be linked to a KP layer) signals some licensing constraint on the nominal independently of [uCase] per se, and ii) the split pseudo-incorporation vs. [uCase] licensing is not enough, thus confirming López’ (2012) observations. Moreover, as actually seen in the various examples presented in the article, the Romanian differential preposition is independent of definiteness²² or indefiniteness morphology and is linearized in a position that precedes these morphemes.

3.2 Romanian oblique DOM and raising

Let's evaluate now the other observation made by López (2012), namely that DOM signals a subclass of nominals with [uCase], which must undergo overt raising to an intermediate position between VP and v ⁰. Romanian confirms that non-clitic-doubled oblique DOM is generated below the EA, as no binding into the EA is possible (see the various examples in Hill and Mardale 2021). However, if we examine binding and c-command relations between DOs and IOs, it cannot be concluded that DOM-ed objects are higher than IOs. In the example in (16a), binding from DOM into IO does not go through.²³ Binding from IO into DOM is fine, as demonstrated in (16b). This difference from Spanish indicates that Romanian differential objects can be lower than both the IO and the EA. Of course, this, in itself, does not prove that Romanian oblique DOM does not need raising. It could be that it raises, but to a position below the IO, which is still above VP. The problem is that unmarked nominals show the exact same behaviour, and it is moreover not easy to map this position.²⁴

Table 1 contains both similarities and differences regarding Spanish and Romanian DOM.²⁵ As Romanian DOM is not obligatory on definite animates in SCs (anti-incorporation contexts), and raising to Spec, α is not motivated, some other explanation needs to be found for Romanian oblique DOM beyond obligatory raising (for [uCase]).

3.3 Romanian DOM is not derived in the morphology

To summarize, Romanian objects indicate the following: i) some classes, such as the bare nominals, might pass diagnostics indicating pseudo-incorporation; ii) other classes, such as (certain types of) nominals with overt (in)definiteness morphology, as well as the prepositional accusatives, escape this process; instead, they give evidence for licensing in syntax – they contain a [uCase] feature, which can explain their (positional) restrictions; iii) as (in)definites with or without differential marking have a different shape on the surface, the question is what accounts for the presence of the differential marker with certain types of non-incorporating DOs; iv) an explanation according to which DOM signals just those nominals with [uCase] which undergo short scrambling to a position above the IO does not easily go through in the language. Although raising cannot be completely ruled out, in this regard, marked and unmarked objects behave similarly.²⁶

Some classical and some more recent discussions (Halle and Marantz 1993, Keine and Müller 2008, Keine 2010, a.o.) have explored a morphological explanation for oblique DOM. Both oblique DOM and other [uC] objects are seen as having the same syntax as licensed objects; the only difference is given by the (obligatory) application of a morphological operation (e.g., Impoverishment) which removes the accusative case features and inserts an oblique marker, in the environment of certain features, such as animacy (for example, the schematic, simplified rule [acc] → [+obl]/_[+animate]).

Connecting oblique DOM to morphological Impoverishment explains the syntactic properties it shares with other objects with [uC] and might give the right results for some oblique DOM languages. But it proves problematic for Romanian; although DOM and other accusative objects with [uC] undergo Case licensing and can occupy the same position, there are syntactic properties they do not share. Illustrating with one example, Romanian DOM gives rise to co-occurrence restrictions with certain types of dative clitics.²⁷ The ungrammatical sentence in (17a) contains DOM and a dative clitic with a possessive reading. An unmarked DO is well-formed, as in (17b).²⁸

In (17a) ungrammaticality obtains under the possessor reading of the dative reflexive clitic. This same clitic can have many other interpretations in Romanian, such as the (high) applicative one. If we test a configuration in which the dative clitic cannot be interpreted as a possessor on DOM we obtain grammaticality. In (18), the negative quantifier, which takes obligatory DOM, does not allow a possessor reading of the dative possessor clitic.²⁹ These two examples demonstrate that the DOM restriction is not a morphological one and that DOM has a different syntax than the unmarked objects with a [uCase] feature. As this syntactic difference cannot be unambiguously tied to a different position for DOM in Romanian, we need to explore other explanations into its nature.

3.4 DOM as an additional licensing operation beyond Case

The solution I propose here links oblique DOM to an additional licensing operation beyond [uCase]. In a nutshell, I start from the assumption that the extended projection of nominals can contain other features that are merged above the functional projection which houses [uCase]. Building on and extending observations going back to Jaeggli (1982), if the main licenser in a domain can value only [uCase], an additional licenser will be needed for an additional feature beyond [uCase]. The differential marker results from the activation of this additional licenser.

I build on decompositions in the higher left periphery of DPs (Ihsane and Puskás 2001, Harley and Ritter 2002, Ihsane 2008, Hill and Mardale 2021), hypotheses related to stacking nominal speech act structures above the DP (Ritter and Wiltschko 2019), as well as on observations regarding the importance of notions related to sentience and perspectivization in human language (Kuno and Kaburaki 1977, Speas and Tenny 2003, Pancheva and Zubizarreta 2018, Sundaresan 2018). It is not uncontroversial to assume that D⁰ is a phase edge in Romance, and the locus where interactions with the discourse are made available. In Romanian, referential nominal structures that project a D⁰ layer must escape incorporation, as they contain a [uCase] feature. What is relevant here is that, beyond Case, an expanded nominal structure can contain various other features, generally related to discourse (δ) and speech-act specifications. The structural organization of the discourse-related nominal periphery is not a trivial issue, and I cannot give an exhaustive cartography here. In (19), I only provide a very basic schema, including only specifications (such as Sentience) which are relevant for DOM in the languages discussed here.

If animates are salient in the discourse and other participants/the speaker relate to them, the Sentience layer is projected on the nominal. In other words, these types of animates are entities to which the speaker relates and which the speaker acknowledges. Following several hypotheses about the encoding of animacy in syntax (Cornilescu 2000, Adger and Harbour 2007, Rodríguez-Mondoñedo 2007, Richards 2008, Bárány 2017) we can assume that discourse-salient animates³⁰ of this type are signaled in syntax by an interpretable [person] feature merged in Sentience. Thus, the nominal in (19) contains both [uCase] as well as δ-related [person] which need licensing. The differential marker results from the impossibility of the main licenser to license the two features; as a result, an additional licenser must be used.³¹ To explain the observation that in Romanian, DOM-ed objects can be found in a relatively low position, I follow Pancheva and Zubizarreta (2018) in assuming that the additional licenser is recruited from the domain with discourse related specifications in the vP (see also Belletti 2004, Jayaseelan 2001, Tsai 2015, a.o).³² For Pancheva and Zubizarreta (2018), notions such as empathy, which they collapse with perspectivization and viewpoint, are encoded as an interpretable feature in the Appl⁰ head, situated above V⁰ but below the EA. This might explain the use of dative as DOM in languages like Spanish. As Romanian uses a locative preposition instead, I make recourse to a more general α projection with δ-specifications.

Decomposing the low verbal domain into v and Voice, the latter introducing the EA (see Legate 2014), we obtain the configuration in (20a). Here v⁰ values [uCase], but cannot value δ-[person]. It is the discourse-related projection α that values the ð feature. The result of the latter operation is the spell-out of oblique DOM, as in (20b).^{33, 34}

The hypothesis put forward here shares some intuitions with a long standing analysis for Romanian DOM in terms of the Kayne/Jaeggli Generalization.³⁵ The latter was proposed to account for contexts such as (12c) and (12d) where oblique DOM also needs accusative clitic doubling. The generalization linked DOM to a last resort Case checking mechanism to avoid a violation of the Case Filter – the accusative clitic absorbs the case from V, leaving the nominal caseless. We have seen, however, that accusative clitic doubling is not obligatory with DOM across the board (and in fact, in some contexts it is ungrammatical). In my analysis, what triggers the activation of the additional licenser is not the clitic, but the initial [uCase] feature, which also needs licensing.

1. (20)

This account presents clear similarities with formalisms that link DOM to an information-structure strategy beyond Case.³⁶ Generally, oblique DOM is correlated with topics (the so-called secondary topics, see especially Dalrymple and Nikolaeva 2011; Leonetti 2003, 2008 for Spanish; Iemmolo 2010 for western Romance; or very recently, Onea and Mardale's 2020 E-topics).³⁷ Topic-hood finds a correlate in the observation that, in some languages, differentially marked objects can only be found in overt dislocation configurations, which exclude focus. However, an account in these terms is hard to extend to Romanian (and the other languages discussed here), where dislocated topics are insensitive to the oblique marker. Moreover, for many speakers, DOM does not have the phonetic correlates of topics; and, for all speakers, it is well formed under focus. In (21b) the differential marker is necessary on the argument which is not given, while in (22) we see DOM-ed animates under contrastive focus.^{38, 39}

Isolating sentience as a separate category, and the presence of generalized δ-features beyond [uCase] derive DOM insensitivity to topicality; moreover, it is a more plausible starting point to an explanation regarding co-occurrence restrictions of the type seen in (17), which are equally not easy to derive under DOM as topic.⁴⁰

4. Differential object marking in Gujarati

Similar types of problems are seen in Gujarati, an Indo-Aryan language which exhibits aspect-based split ergativity (Cardona 1965, Mistry 1976, Magier 1983, Woolford 2006). We are mostly interested in examining perfective paradigms, where object agreement is indicated overtly as an inflectional suffix on the verbal stem. In (23a–c) we see three examples with the unaccusative predicate come. The only argument is a subject that can only take the absolutive form and agrees with the perfective stem in gender and number, as seen from an examination of subjects with different genders:

In the perfective, direct objects exhibit a more complex morpho-syntactic behaviour (Cardona 1965; Mistry 1976, 1997, 2004; Comrie 1984; Woolford 2006; Wunderlich 2012; Grosz and Patel-Grosz 2014 for Kutchi Gujarati; Joshi 2020 for Surati Gujarati). Transitive predicates require the external arguments (the agents) to be marked with ergative case (the ergative postposition -e). Agreement can either be with the object or show up in a default form, depending on the interpretation (and the structure) of the latter. First, in some contexts and for some speakers, a number-neutral reading and a non-referential interpretation appear to be possible, while agreement with the object is blocked. In example (24) the perfective stem instead shows default (neuter gender) inflection, irrespective of the gender of the direct object (DO).⁴¹

A referential interpretation of the nominal is possible if the perfective stem shows object agreement, which is realized in gender and number. In the examples I present in (25), there is object agreement and the object has to be interpreted either as definite⁴² or indefinite. When the plural marker is present on the direct object, as in (25b), a referential plural reading is obtained and object agreement is obligatory (when the subject has ergative case).⁴³

As I also mention later, this split is captured in formal accounts under the assumption that the agreeing objects contain a [uCase] feature which requires valuation. The non-agreeing objects (for those speakers who accept them) can be assumed to be NPs which do not contain a Case feature undergoing pseudo-incorporation with the verb.

This basic split is challenged by the existence of a third class of objects. Similarly to Romanian and Spanish, certain types of animates (especially those at the higher end of the animacy and referentiality scales) can/must take an oblique postposition, as in (26), under yet another instantiation of oblique DOM. Just like in Spanish, the DOM postposition is homophonous with the dative case maker -ne. However, despite their oblique appearance on the surface, DOM-ed objects must show agreement just like the bare, non-incorporating absolutives in (25). Lack of agreement with a differentially marked object would result in ill-formedness/ungrammaticality in the examples in (26).

As such, oblique DOM does not appear to have the syntax of indirect/oblique objects, which never trigger agreement on the perfective stem.⁴⁴ The absolutive syntax of DOM is confirmed by other diagnostics (passivization, lack of Case preservation under nominalization, etc.) under which DOM-ed objects and agreeing bare absolutive objects pattern alike, to the exclusion of indirect objects. Moreover, their structural absolutive nature matches the structural accusative status of such objects in Romanian and Spanish. We are left with the same question: where is the distinction between the agreeing absolutive and DOM to be located?

4.1 A licensing operation beyond [uCase]

Object agreement in Indo-Aryan has received a great deal of attention. I will be presenting below two main hypotheses initially formulated for Hindi-Urdu, but which can be applied to Gujarati too. For Mahajan (1989), the perfective participles in examples like (25) or (26) cannot assign accusative Case, as they are deficient. The Case of the objects in transitives embedded under perfective participles is assigned instead by an Agr⁰ head together with the finite tense projection. Case assignment is followed by object raising to Spec, AgrP, as illustrated in (27):

Bhatt (2005), on the other hand, has provided various diagnostics dissociating object agreement from Case assigning deficiency. The objects of transitives receive accusative Case from v ⁰, irrespective of whether the participle is perfective or imperfective. Object agreement results instead from the need to value the uninterpretable φ-features of T. Bhatt (2005) proposes an AGREE⁴⁵ operation under which the goal XP does not have to be active, that is, have unvalued Case features; instead, AGREE can/must target Case licensed objects. In examples like (25)/(26), given that ergative morphology renders the EA non-available, the non-incorporated direct object is the nearest argument with visible interpretable φ-features T⁰ can use to value its uninterpretable φ-features. Pseudo-incorporated objects, as in (24), do not permit this type of agreement in Gujarati.

No matter which of the two accounts above is to be used, it is clear that oblique DOM has to be linked to a distinct operation beyond [uCase]/AGREE. The hypothesis I have entertained in this article derives these empirical facts without problems. Oblique DOM results from the matching of a δ-related (person) feature, beyond [uCase], by a discourse-related sentience functional head in the low verbal projection; this is a distinct licensing operation, dissociated from the valuation of [uCase]. The need for an additional licenser for δ is also motivated by the observation that the perfective stem can only show overt gender and number agreement, but not person. In fact, as the literature mentions, object agreement in person is not available with auxiliaries either (see also Bhatt 2005). We see in (28a) that the present auxiliary displays person agreement with the subject, but in (28b) person agreement with the object is not possible.

One question that comes to mind is whether this is an instantiation of Baker's (2011) SCOPA,⁴⁶ which postulates special structural configurations for person agreement, presumably not met in contexts like (28). However, this issue requires more attention, as demonstrated by languages such as southern Basque, where absolutives agree in person. Higher animate objects carry dative morphology, under a type of oblique DOM, as in Spanish/Gujarati. Dialectally, they can show either person dative agreement (Fernández and Rezac 2016, Odria 2019, a.o.) or co-occur with person absolutive agreement (dative displacement varieties, Odria 2017).⁴⁷ The special shape of DOM indicates an important difference between φ-related person features and δ-related person. They both need licensing (Odria 2017, 2019), but by distinct types of licensers.⁴⁸

5. Differential object marking in Mandarin Chinese

The last case study I will be addressing comes from Mandarin Chinese, whose objects appear to be subject to complex positional (and licensing) restrictions.

Mandarin Chinese is a typical analytical language; there is no overt case morphology or agreement, and overt definiteness and number morphology are also lacking. The default word order is SVO, but objects can also be found in a preverbal position. Distinct locations for objects do have interpretive consequences. For example, the postverbal nominal in (29) can be interpreted either as a singular or a plural, and it can get either an indefinite interpretation or a definite one:

Despite the absence of overt definiteness, nominal phrases can contain various pieces of morphology, such as demonstratives, numerals, classifiers, etc.⁴⁹ Such nominals can either precede or follow the verb, as seen in (30).

Yet, direct objects can surface with an even more complex structure. In a preverbal position, some classes of nominals must or can be preceded by the ba ⁵⁰ marker. The ba construction (‘baziju’, ‘the disposal form’) has been extensively studied in both descriptive and formal studies, encompassing a vast literature (Li 1990, 2006; Zou 1993; Liu 1997; Sybesma 1999; van Bergen 2006; Yang and van Bergen 2007; Huang et al. 2009; Kuo 2010; Paul 2015; Sun 2018). Similarly to Romanian or Spanish, its conditions of use are extremely complex. Providing an exhaustive account is beyond the scope of this article, which has a much more modest goal. I am interested in examining the syntactic relationship of ba nominals to other nominal phrases, supporting the hypothesis of a licensing operation beyond [uCase] for DOM.

Generally, animate nouns and pronouns must be preceded by ba, as seen in (31a–b), while with inanimates the marker is optional, as in (31c).⁵¹ However, DPs that contain the numeral/indefinite morpheme and are found in a preverbal position must be preceded by ba, irrespective of animacy. This is illustrated in (31d). Sensitivity to both animacy and definiteness/specificity indicates that the ba construction instantiates a bi-dimensional type of oblique DOM (just like in Spanish, Romanian and Gujarati).

These examples make it clear that connecting the ba construction with raising to a preverbal position is not enough to derive its nature, as there are speakers who accept objects in preverbal position without DOM, as in (31c). Similarly, linking DOM to a preverbal position and animacy is not sufficient either – in (31d) we see that ba is obligatory even with inanimates that have a certain type of structure.⁵² As in the other languages I have discussed, an account in terms of scales (Animacy/Specificity) does not explain the data in a non-stipulative way. I show below that connecting ba with the distinction Case licensed vs. (pseudo-) incorporation is also not sufficient.

5.1 Ba and (pseudo-)incorporation

One property of ba DPs is undisputed in all grammars – such objects are never possible in a postverbal position. Native speakers confirm this observation, unanimously indicating that examples such as (32) are clearly ungrammatical with ba:

The question is then why the postverbal position is incompatible with ba DPs. This is not a simple problem. As Huang (2018) notices, nominals that can appear in a postverbal position are not homogeneous, syntactically nor semantically. One important split is given by nominals that can precede or follow postverbal directional phrases (DirP), such as three years/three days in the examples in (33).

Following Huang et al. (2009), Huang (2018) assumes that there are two argument positions inside VP, namely Spec, VP and as a complement to V. DirPs are left adjoined to the first projection of V, as seen in (34). Another observation Huang (2018) makes is that bare NPs with a non-referential interpretation are only possible to the right of the DirP, as in (33a). Nominals with a larger structure, encompassing demonstrative, numeral, classifier projections, etc. are ungrammatical in that position and must, instead, precede the DirP, as seen in (33c).

1. (34) (based on Huang 2018: 204)

The examples in (35) demonstrate that the bare post-DirP NP can only take narrow scope (35a) and has a number neutral reading (35b).⁵³

These properties could, in principle, be explained under the assumption that the bare post-directional nominal undergoes noun incorporation with V. However, as Huang (2018) also shows, in these contexts we are rather dealing with the process of pseudo-incorporation (in Massam's 2001 terms). This conclusion is supported by two main diagnostics: i) bare post-directional nominals are phrase-level NPs and can be modified, as in (36); ii) Mandarin Chinese presents contexts of true head level incorporation, where N⁰ left-adjoins to V and constructs a complex predicate, as in (37):

Huang's (2018) proposal is that bare post-directional nominals as in (33a) and (36) undergo pseudo-incorporation with V. Then V undergoes raising to a position above the directional phrase. More complex nominals (possibly DPs) as in (33c), on the other hand, contain a [uCase] feature, cannot compose directly with V and thus must raise to a position where they can get their [uCase] valued (see (34)). But if DPs such as those in (33c), that cannot take the -ba marker, have a [uCase] feature and must scramble to a position where their [uCase] can be valued, how exactly are the ba DPs to be explained? One could entertain the hypothesis that non-ba-marked DPs such as in (33c), even though escaping incorporation with V, do not have a [uCase] feature and stay unlicensed. But then the questions are: i) why do they need to raise above directional adverbials? ii) why can non-ba nominals raise even further and show up even in a preverbal position, as we see in (31c)? In examples such as (33c), in order to capture the position of DPs as preceding DirPs, following the structure in (34), it must be the case that V raises to v. The DP can have its Case feature valued in situ, in the specifier of VP. But, then, in order to capture the preverbal placement of the non-ba marked DPs in (31c), it must be the case that such objects raise even higher, above V.⁵⁴ This implies that a separate operation is needed for the ba objects, such as those in (31a–b) and (31d). Note that the ba marker is not optional in most contexts, such as with animate DPs, pronouns, etc. One way to interpret examples like (31c) is that they do not signal ba-optionality. Instead, we are dealing with [uC] objects which must scramble to a v ⁰ preceding position either as a result of an EPP feature on v ⁰ (objects showing up bare) or a result of a different operation (the ba-ones).

I propose that this second operation is connected with a feature that is present in the composition of ba objects. Similarly to what we have seen in the other oblique DOM languages I have discussed, one instantiation of this additional specification is a δ-related ([person]) feature, which requires additional valuation by a discourse-related functional head, namely the α (Sentience) related projection, in vP domain.⁵⁵ This is illustrated in (38). This type of reasoning captures repeated remarks made in the literature with respect to ba objects being connected to affectedness, prominence, as well as their insensitivity to specificity in some contexts (Sybesma 1999, Sun 2018 for detailed discussion). As we have also seen for Romanian, Spanish (and Gujarati), oblique DOM is not a marker of specificity. It results, instead, from a licensing operation needed to value a [person] feature or other discourse specifications on complex nominals, beyond Case. It also captures the observation that ba objects are not to be understood in terms of a scrambling operation to the left periphery, for reasons of topicalization. As discussed in the literature and as confirmed by native speakers, dislocated topics are possible without ba in Mandarin Chinese.⁵⁶

Of course, much more needs to be said about the ba marker in order to derive its non-trivial interactions with telicity, transitivity, genericity, etc.; many of its properties are also seen in the other languages examined in this article, suggesting a common core. Here, I was mainly interested in showing that this type of DOM is an operation beyond [uCase] and that the split (pseudo-)incorporation vs. [uCase] licensing is not enough.

1. (38)

6. Concluding remarks

In this article I have examined some instances of (animacy-based) oblique DOM against two main theoretical backgrounds: i) oblique DOM as a nominal licensing strategy on objects specified with a structural Case feature which needs valuation in syntax (López 2012; Ormazabal and Romero 2013a, 2013b; a.o.); and ii) oblique DOM as connected with information structure (topic) specifications, beyond Case (Leonetti 2003, 2008; Iemmolo 2010; Dalrymple and Nikolaeva 2011; Belletti 2018). I have shown that in some languages, such as Romanian, Gujarati and Mandarin Chinese nominals show a more complex behaviour than what the split licensed/unlicensed would predict. In Romanian there are DPs (definites and certain types of indefinites) which need [uCase] licensing, independently DOM. In Gujarati, oblique DOM co-occurs with object agreement, which results from an independent licensing operation. And in Mandarin Chinese, DOM-ed objects are spelled out in a higher position than other scrambled DPs which give equal indication of [uCase] licensing.

I have also demonstrated that reducing oblique DOM to a morphological operation is not an adequate explanation, as these objects are syntactically distinct from other (accusative) Case licensed DPs, either positionally (Mandarin Chinese) or with respect to co-occurrence syntactic restrictions DOM gives rise to (Romanian, Spanish). These observations appear to provide support to the second line of analysis. However, in none of languages examined here is oblique DOM similar to (left dislocated) topics; for example, it also appears under focus or on elements which are not given. Building on hypotheses of DOM as an additional licenser on the same nominal (following Jaeggli 1982), I have proposed instead that the differential marker signals an additional licensing operation beyond [uCase] on nominals with complex structure. For example, animates that are relevant in the discourse are linked to a δ-related ([person]) feature which is valued by a δ-related functional projection in the mid verbal domain (Pancheva and Zubizaretta 2018), below the EA.

This account makes various predictions and raises several questions. As the relevant licensing operation is independent but can co-occur with [uCase] licensing mechanisms, oblique DOM contexts discussed here are similar to case stacking configurations. This seems to be borne out in Romanian (as well as across Romance), where pronouns have independent accusative case morphology, but they additionally require the DOM preposition, and accusative clitic doubling. Similarly, Sun (2018) discusses stacking configurations with DOM and pronominals across Chinese varieties.

The observation that the difference between differentially marked nominals and unmarked nominals is not just the split structural Case vs. (pseudo-) incorporation/unlicensed has been discussed for other families of languages, matching the picture presented here. For example, various works have shown that across Altaic, (certain types of) unmarked nominals give evidence of the presence of structural [uCase], similarly to differentially marked objects (Taylan 1986, Kamali 2015, Öztürk 2005).

Two questions that need further investigation are the precise status of the δ–related ([person]) feature and restrictions to objects. If oblique marking were simply a reflex of highly complex nominal structure, we would expect to see it on subjects/EAs too (signalling additional licensing in the TP domain for features beyond [uCase]). While it is indeed true that DOM can appear with certain subjects across Romance, Indo-Aryan or Mandarin Chinese, it is impossible with various types of agents. A preliminary answer to this problem could start from the observation that T is a different type of licenser than v ⁰, in that it can contain the relevant type of δ–related feature that is more easily transmitted from C (Miyagawa 2017), such that recourse to an additional licenser is not needed. A better understanding of the nature of subjects where the differential marker is permitted is thus needed. Another possibility is that subject licensing interacts with information-structural notions in more restricted ways, permitting only certain types of topics, which might clash with the specifications in the δ-layer, so that the latter must remain unlicensed. Another big picture question relates, of course, to the notion of nominal licensing itself, its typology and its precise nature.