Variation in differential object marking: On some differences between Spanish and Romanian

Although differential object marking (DOM) has been studied from a multitude of perspectives, research into the types of variation it allows in closely related languages is still needed. This article examines DOM from the point of view of (micro)variation, focusing on two genetically related languages, namely, standard Spanish and Romanian. Both identities and points of divergence are discussed in detail and argued to result from a morphosyntactic parameter in the realm of differential marking. Following López [López, Luis. 2012. Indefinite Objects Scrambling, Choice Functions and Differential Marking. Cambridge, MA: MIT Press], the Spanish special marker signals certain types of nominals which undergo short scrambling in order to have Case licensed. In Romanian, however, DOM is equated with an additional operation beyond the valuation of Case.


Introduction
Splits in the morphosyntactic realization of objects are very common crosslinguistically as instantiations of the phenomenon labeled differential object marking (DOM; Givón 1984, Comrie 1989, Bossong 1998, Lazard 2001, Aissen 2003, de Swart 2007, López 2012, Ormazabal and Romero 2013a, Bárány 2017, Kalin 2018, Levin 2019. A typical illustration is the animacy-based object contrast in Romance (Niculescu 1965, Rohlfs 1971, Roegiest 1979, Bossong 1991, Torrego 1998, encoded via a typologically common adpositional strategy. DOM has been addressed both descriptively and theoretically, especially in individual languages. What is less understood, however, are the limits of this phenomenon, especially at the microvariation level. This article takes a step toward filling in this gap. Two prominent DOM varieties in Romance, namely, standard Spanish and Romanian, are compared in a comprehensive manner. A battery of formal diagnostics and contexts are examined, some of which are either novel or not usually discussed in the literature (such as strong quantifiers [SQs], elliptical contexts, bare quantifiers, weak quantifiers, DOM in comparative structures, DOM in partitives, etc.). The main observation is that the complexity and richness of the existing patterns as well as the wealth of data present a real touchstone for a unified analysis.
On the one hand, the investigation reveals non-trivial similarities between the two languages, motivating a common structural source. For example, DOM is sensitive to animacy/humanness, affecting nominals that are high on the animacy/definiteness scale (personal pronouns, proper names, etc.). Similarly, strong/universal quantifiers with an inbuilt animacy restriction need DOM in both Spanish and Romanian. Among other important DOM properties shared by the two languages, one can also list: There are several well-defined contexts where direct objects cannot be used without the a-marker. These mainly signal humanness/animacy, with some configurations where the accusative preposition is also required on inanimates. As inanimates are also relevant for DOM in Romanian, we address them separately, in Section 2.3.
We start with personal pronouns, as in (1). Here the accusative a is obligatory. Note that in this context, clitic doubling is also necessary, but the clitic can only take accusative morphology. The dative form of the clitic is ungrammatical: (1) 'Mónica *(lo) vio *(a) él // 'Mónica le vio a él.² Monica CL.3SG.M.ACC saw DOM he // Monica CL.3SG.M.DAT saw DOM he Intended: "Monica saw him." standard Spanish In (2) we see a proper name, and in (3-a) a run-of-the-mill referential definite animate. These classes must be accompanied by the differential marker. Inanimates, as in (3-b), are ungrammatical with DOM in regular transitive clauses of this type: (2) Juan visitó *(a) Pedro.
He encontrado *(a) la niña.³ have.1SG found DOM DEF.F.SG girl "I have found the girl." b.
He encontrado (*a) el libro. have.1SG found DOM DEF.M.SG book "I have found the book." Standard Spanish (Ormazabal and Romero 2013a, ex.1a,b) Turning now to another obligatory context, we see in (4) SQs which either modify an overt animate determiner phrase (DP), as in (4-a) and (4-b), or have an understood animate antecedent, as in (4-c).
Inanimates are not well-formed with differential marking in these contexts, as shown in (4-d). The behavior of SQs is a locus of variation from Romanian, as we show in Section 2.2: However, if the differential marker is dropped, the definite can only be interpreted non-referentially/non-specifically. Thus, the sentence in (i) can only refer to a type of girl, not to a specific girl. As we will see in the next section, this is an important difference from Romanian, where animate definites can obtain a specific interpretation even without the differential marker: (i) He encontrado la niña (que buscas). have.1SG found DEF.F.SG girl that search.2SG "I have found the type of girl (you are looking for)." standard Spanish Animate negative quantifiers tend to require special marking crosslinguistically, when functioning as objects (Bossong 1998, a.o.). Spanish is no exception, as we see below in (5-a) and (5-b). Existential quantifiers which have an animacy restriction on their domain of quantification also need the special marker, as in (5-c): Standard Spanish a. Juan no amó *(a) ninguna mujer. Juan NEG loved DOM no.F.SG woman "John loved no woman." (López 2012, ex. 25d, p.13) b. Juan no vio *(a) nadie. Juan NEG saw DOM nobody "John didn't see anybody." c.
Maria insult.PST.3SG DOM a.M.SG colleague "Maria insulted a colleague." (García García 2005, ex. 16) Another context where differential marking is obligatory with animates comes from clause union (see López 2012 for pioneering discussion). In such configurations, the nominal normally acts as a shared argument of two predicates. We provide below two examples from nominals with adjectival ECM-small clauses (SCs). In these contexts, even if the shared argument is not (necessarily) interpreted specific,⁴ it still needs DOM. As we discuss later in the article (Sections 4 and 5), this restriction is due to a specific syntactic configuration in these contexts.
Considero *(a) un estudiante intelligente/necessario por consider.1SG DOM a.M.SG student intelligent.M.SG/necessary.M.SG for un proyecto. a.M.SG project "I consider a student intelligent/necessary for a project." b.
Considero (*a) un libro interessante. consider.1SG DOM a.M.SG book interesting.SG "I consider a book interesting." A preliminary summary of the Spanish obligatory contexts is given in Table 1. As a methodological background, we follow the procedures in recent parametric comparison research, especially Roberts (1998Roberts ( , 2012Roberts ( , 2017, the Parametric Comparison Method Longobardi 2005, Longobardi andGuardiano 2009) or Ledgeway et al. (2019). Descriptively, we take the contexts/properties/characteristics discussed below to act as (surface) parametric points and thus as comparanda for the examination of the two languages discussed here. In Section 4, we show that most of the surface patterns ("parameters") can be reduced to more basic featural or configurational primitives in the grammar. A semantic/ morphosyntactic specification set positively for DOM is indicated with +, while the absence of the a marker for certain morphosyntactic specification is encoded as −. Contexts set as + also indicate obligatory differential marking.

Obligatory differential marking in Romanian
The use of a preposition⁵ to signal certain classes of objects is a very robust differential marking strategy in Romanian too, as discussed in numerous contributions (Niculescu 1965, Farkaş 1978, Dobrovie-Sorin 1990, 1994, Mardale 2009, Cornilescu 2000, Tigău 2010, 2011, Hill 2013, Pană-Dindelegan 2013, Avram and Zafiu 2017, Irimia 2018, 2020. The data, however, show some intricate loci of variation with respect to standard Spanish, which have yet to be exhaustively classified and analyzed. The obligatory contexts we begin with are the pronouns and proper names. Pronouns with direct object function must be clitic doubled, and the case morphology on the clitic can only be accusative, as we see in (8). This is similar to Spanish. Note, however, that Romanian diverges from Spanish in that clitic doubling is possible with differentially marked objects, more generally, not just with the pronouns.⁶ In (8-b), we see an object proper name which is clitic doubled.⁷  5 This preposition can have a locative meaning (on), as shown in (i-a), but is not homophonous with the dative. The latter preserves inflectional morphology in Romanian, as seen in (i-b). Historically, the differential marker was a directional.  I-a dat cartea fetei. CL.3SG.DAT-has given book.DEF.F.SG girl.DAT.F.SG "S/he has given the book to the girl." Romanian 6 There are also contexts where clitic doubling is ungrammatical with DOM, for example, negative animate quantifiers. For lack of space we do not address clitic doubling in this article. 7 For some speakers, in fact, proper names seem to be ungrammatical without accusative clitic doubling.
Maria CL.3SG.ACC-has seen DOM Ion "Maria saw Ion." Romanian Turning now to referential and specific definite DPs, we get another difference. In Spanish, these classes need the obligatory prepositional marking, as illustrated in (3-a). In Romanian, on the other hand, animate definites can have a referential/specific interpretation even without the differential marker. Thus, both (9-a) and (9-b) are possible: It is only animates that can take the differential marker in these types of contexts. Inanimate objects, no matter whether their interpretation is referential/specific, are ungrammatical with the differential marker in regular transitive clauses, as demonstrated by the example in (10). This is the exact same state of affairs as in the standard Spanish sentence in (3-b).
(10) (*L-) au lăudat (*pe) romanul clasic. CL.3SG.M.ACC-have praised DOM novel.DEF.N.SG classical.N.SG "They have praised the classical novel." Romanian Another DOM-inducing example we mentioned for Spanish contains SQs accompanying an animate object, as in (4-a) or (4-b). In Romanian, DOM is possible with animates in this context: Ion a văzut (*pe) toate filmele. Ion has seen DOM all.N.PL movie.N.PL.DEF.N.PL "Ion has seen all the movies." Romanian The differential marker also extends to elliptical contexts; here, an important difference is seen with respect to standard Spanish. If in Spanish examples of the type in (4-c) the presence of accusative a entails an animate understood referent, in Romanian the differential marker must be used irrespectively of animacy. These contexts are part of a larger class where the DOM preposition is obligatory irrespectively of animacy and will be further addressed in Section 2.3.
Am văzut *(pe) cineva. have.1 seen DOM somebody "I saw somebody." Romanian In both languages, the animate wh element who is ungrammatical without the accusative preposition, when used as an object:⁹ (14) a. *(Pe) cine ai văzut? b. *(A) quien has visto? DOM who have.2SG seen DOM who have.2SG seen "Who have you seen?" Romanian "Who have you seen?" Spanish The major difference between the two languages affects the negative quantifier in its adjectival form. The first sentence in (5) shows that the latter requires differential marking in Spanish, when its DP is animate. In Romanian, this restriction does not hold, as we show in more detail in Section 2.5 (where optional DOM is discussed).
The presence of demonstratives also appears to have an effect on the use of DOM. Both Romanian and Spanish patterns are alike in requiring DOM if demonstratives accompany animate nouns: (15) a. Am văzut-o *(pe) această fată.
have.1 seen-CL.3F.SG.ACC DOM this.F.SG book "I have seen this book." Romanian  9 As these quantifiers cannot be analyzed in terms of specificity, we have unambiguous evidence that DOM is not a specificityinducing mechanism. In Section 3, we present further motivation for this conclusion.
have.1 seen DOM that.F.SG street "I have seen that street." standard Spanish However, a difference arises with respect to demonstratives in elliptical constructions. The differential marker is obligatory with an animate antecedent in standard Spanish, as we see in (16). In Romanian, on the other hand, the demonstrative requires differential marking in elliptical structures irrespective of animacy. Relevant examples are provided in Section 2.3.

(16)
He visto a aquella. have.1 seen DOM that.F.SG "I have seen that one." Spanish, animate antecedent only The other two conditions for DOM in Spanish are telicity (6-c) and syntactic configurations where DPs act as shared arguments, such as (7-a). Telicity does not play a salient role in Romanian (with respect to the prepositional accusative), and animate shared arguments can be used without the differential marker in SCs, as seen in (17) Before proceeding with the inanimates, we summarize the obligatory DOM contexts for both Romanian and Spanish, as restricted to animacy. The differences are signaled in boldface. We have also tagged the relevant examples, so that they can be easily tracked ( Table 2).

Obligatory differential marking and inanimates
Although highly sensitive to animacy, DOM interacts with inanimates in very subtle ways across Romance varieties. Restricting our attention to standard Spanish, both Torrego Salcedo (1999) and García García (2007) discuss examples of the type below, which are generally subsumed under the telicity-inducing contexts.¹⁰ We follow the same grouping here.
(18) a. Un adjetivo acompaña/califica a un sustantivo.¹¹ a.M.SG adjective accompanies/qualifies DOM a.M.SG noun "An adjective accompanies/modifies a noun."  10 See also Fish (1967) and Weissenrieder (1991) for a comprehensive list of predicates involved in these patterns, as well as Martín (1999Martín ( , 2005, or Rodríguez-Mondoñedo (2007), a.o. 11 A clarification is in order here. These predicates are part of classes that appear to select lexical case morphology (dative, genitive, etc., as opposed to accusative) in many languages (see especially de Hoop 1996). A legitimate question would be whether the a marker indicates the dative in these examples. We have unambiguous evidence that a signals accusative morphology here, as clitic substitution only tolerates the accusative form in the relevant interpretation: While telicity plays no role in Romanian, the obligatory use of differential marking irrespective of animacy is extremely robust in equatives. Two relevant examples are provided below; for more discussion, see Pană-Dindelegan (2013) or Irimia (2018Irimia ( , 2020.  Additionally, Romanian exhibits many other contexts where animacy plays no role in regulating differential marking, with the DOM preposition extending across the board to all direct objects (Cornilescu 2000, Pană-Dindelegan 2013, Avram and Zafiu 2017, Irimia 2020. Prominent are nominal ellipsis contexts, headed by demonstratives, as in    13 TV commercial for a type of cheese. 14 CEL developed from the endophoric distal demonstrative acel/acela, the latter descending from the Latin complex demonstrative ecce/eccum (an adverbial reinforcer) + illum (distal demonstrative). 15 Romanian has preserved inflectional morphology for the genitive, which is (partly) homophonous with the dative. Importantly, the genitive is subject to a strict adjacency constraint which requires it to immediately follow a noun with suffixed definiteness (i-a). The genitive is ungrammatical with forms that do not contain overt definiteness, for example, the indefinite in (i-b). Structures like (i-b) can only be made grammatical by the addition of a linker (abbreviated LK and whose spell out form is a) to which definiteness is suffixed, as in (i-c): See also Cornilescu (1992), Giurgea (2008), a.o., for further discussion.
Variation in DOM: on some differences between Spanish and Romanian  433 We have already presented another animacy-insensitive context, namely, the one led by the SQ tot, under nominal ellipsis.¹⁶ We repeat here a relevant example: Le-am văzut *(pe) toate. CL.F/N.PL-have.1 seen DOM all.F/N.PL "I saw them all." Romanian (animate or inanimate) In Romanian, animacy can be overridden in yet other contexts, such as some types of partitives (25-a), D-linked pronouns (25-b) as well as relative clauses (25-c). These configurations do not trigger differential marking in standard Spanish with inanimates.  To summarize, both Romanian and standard Spanish exhibit a differential marking strategy which affects animates under certain conditions. Descriptively, when animacy is at stake, this special morphology follows the referential hierarchies known as scales (Aissen 2003, a.o.); DPs ranked at the higher end of scales are more prone to carry the prepositional marker (pronouns, proper names, human DPs, etc.). However, one important problem with scales is that animacy must be overridden in welldefined structural configurations.

(26)
Person/Animacy: 1/2 < 3 < proper name < human <animate < inanimate (27) Specificity/Definiteness: pronoun < name < definite < specific indefinite < non-specific There are also non-trivial differences in the DOM profile of the two languages. On the one hand, Romanian animate referential definites are possible without the prepositional marker, and indefinite animates can obtain specific readings in the absence of DOM. Telicity effects, as well as "prepositional accusative" selection by  16 The elliptical SQ is ungrammatical with differential marking under some types of propositional ellipsis or when it tracks non-individuating entities in the singular; in these contexts the universal quantifier shows up in the neuter singular form tot). We present two contexts below: Nu cred (*pe) tot ce îmi spui tu. not believe DOM all.N.SG what I.DAT tell.2SG you "I don't believe everything you are telling me." b.
Ion mănâncă (*pe) tot (din farfurie). Ion eat.3SG DOM all.N.SG (from plate) "Ion eats everything (off the plate)." 17 Some speakers do omit the differential marker in relative clause contexts. Descriptive grammars attribute such omissions a sub-standard status, and the lack of differential marking is strongly frowned upon. certain predicates, irrespective of animacy are also not operative in Romanian. On the other hand, Romanian exhibits a large group of contexts, mainly related to nominal ellipsis and D-linking, where DOM is obligatory irrespective of animacy. These latter contexts are not matched by the Spanish data. In Table 3, we summarize the results obtained from the comparison of obligatory DOM in both Romanian and Spanish.

Where the differential object marker is ungrammatical
According to the literature, in Spanish there are five important contexts that do not tolerate the differential marker: (i) existential configurations, (ii) complements of predicates like tener "have" under an individual-level interpretation, (iii) bare plurals, irrespective of animacy, (iv) quantifiers with a restriction on inanimacy, and (v) objects of unaffected predicates with inanimate subjects. Romanian shows a very similar picture, and we illustrate the two languages in parallel. In (28), we provide an existential sentence,¹⁸ which in Spanish is constructed with the verb haber "have": En el patio hay *(a) un niño. in DEF.M.SG yard have.3SG DOM a.M.SG boy "There is a boy in the yard." (López 2012, ex.44, p. 20) An individual-level interpretation with a possessive transitive predicate like tener "have" is seen in (29-a). Note that, if the predicate is constructed as stage level, expressing a temporarily restricted eventuality,¹⁹ the differential marker is possible (29-b): (29) a. María tiene *(a) tres hijos.
Maria has DOM three children "María has three children." b. María tiene (a) un hijo en el ejército.
María has DOM a.M.SG son in DEF.M.SG army "María has a son in the army." (López 2012, ex.45, p. 20/ex. 48, p. 21) We now turn to Romanian existential clauses, as in (30-a), where the prepositional accusative is indeed impossible. Transitive possessive clauses in (30-b) also have the same behavior as in modern Spanish. Under an individual-level reading, differential marking is out. The stage-level interpretation allows the differential marker with animates, which moreover can only have a specific reading. This is the same picture as in standard Spanish: (30) a. Sunt (*pe) copii la scoală.
be.3PL DOM children at school "There are children at school." b. Ion are (*pe) copii.
Ion have.3SG DOM children "Ion has children."  18 A question could be why we are discussing existential predicates, given that in (28) the DP un niño ("a boy") is normally assumed to be a subject. As the differential marker only affects objects, its absence in contexts like (28) would be as expected.
However, various discussions have demonstrated that DPs in existential sentences (ES) have an object behavior, at least in some varieties of Spanish (and across Romance, more generally). For example, Rodríguez-Mondoñedo (2007) shows that ES DPs exhibit object agreement and can be resumed by object clitics in non-standard Spanish. Similar observations can be found in Fernández-Soriano and Táboas-Baylin (1999: 1754-9), Suñer (1982), and Torrego (1984), a.o.
19 See Bleam (2005) or Klein and de Swart (2011) for further discussion about the differences between the two readings and the ways in which such distinctions affect the presence/absence of the differential marker.
c. Ion îi are pe copii. John CL.3M.PL.ACC have.3SG DOM children "John has the children (known to the speaker/hearer)."

Romanian
Two examples with bare plurals where the Spanish a-marker is ungrammatical are given below. As already mentioned, animacy is irrelevant in these configurations. For many speakers, (31), which contains an animate DP, is as ill-formed as the inanimate in (32): search.1SG DOM book.PL "I am looking for translators." " I am looking for books." Romanian bare plurals can be constructed with the prepositional accusative for some speakers. Sentence (33) contains an example:²⁰ Table 3: DOMstandard Spanish and Romanian (preliminary)

Contexts and characteristics DOM Spanish DOM Romanian
Personal pronoun ( 11) (Non-adjectival) Neg/Exist quantifiers with animacy restriction +(5) + ( 13-a) Adjectival Neg/Exist quantifiers with animates  20 In Romanian, definiteness morphology (which is suffixal) is blocked under a preposition, if the DP does not contain overt modification. The differential marker, being prepositional, is subject to this constraint, as we show in (i). Crucially, the definiteless noun in (i-a) is interpreted as definite, as indicated in the translation. This raises the question of whether what looks like a bare noun in (33) is a true bare noun or a concealed definite: In (34), we present two examples with quantifiers whose domain of quantification is restricted to inanimates. DOM is ungrammatical in these contexts in Spanish: (35) He visto (*a) algo. not have.1.SG seen DOM nothing have.1SG seen DOM something "I haven't seen anything." " I have seen something." Spanish (García García 2007, ex. 2a) As the two examples below show, the same pattern holds in Romanian: (36) Nu am văzut (*pe) nimic.
(37) Am văzut (*pe) ceva. not have.1 seen DOM nothing have.1 seen DOM something "I didn't see anything." " I saw something." Romanian Differentially marked objects are also impossible in various transitive sentences with an inanimate subject. The two Spanish contexts below illustrate an important contrast, which can be summarized as followsunaffected predicates with inanimate subjects do not permit special marking on their animate objects. The same restriction is active in Romanian.  Table 4 summarizes the ungrammaticality contexts presented in this section.


We know that example (33) contains a bare noun because both definites and bare nouns are possible with DOM in the plural, if the noun is modified. Thus, example (ii), which contains overt suffixed definiteness, is also possible:

Optional DOM
The most relevant cases of optionality are seen with indefinites. A typical remark made about standard Spanish is that animate indefinites do not obligatorily need the differential marker. Traditionally, the presence vs absence of the marker has been connected to specificity. As the received wisdom goes, if the nominal is interpreted as non-specific, DOM cannot be used. However, if the animate indefinite is interpreted as specific, DOM is assumed to be obligatory. Contrasts like the one between (40-a) and (40-b) are clear in the language: (40) a. María buscó a una gestora che habla alemán.
María searched a.F.SG manager that speaks.IND/speaks.SUBJ German "Maria looked for a manager that could speak German." Spanish (López 2012, ex. 38a, b, p. 18) In (40-a), the prepositional accusative tolerates the indicative mood, which is connected with a presuppositional (specific) interpretation of the DP. In other words, (40-a) can mean that María has looked for a specific manager who actually speaks German. The non-DOM-ed DP, illustrated in (40-b), on the other hand, is only possible with the subjunctive mood. As we have mentioned above, the subjunctive mood is a marker of non-specificity (and non-actualization) in Spanish (see especially Rivero 1979). Thus, the animate object una gestora "a manager" cannot be interpreted as specific in the Spanish example (40-b).
The PE prepositional marker is also optional with animate indefinites in Romanian. But the presence of PE is not always connected to specificity. The specific reading might be prominent in examples like (41-b), but it can be missing in many other contexts, which are summarized in the next section. Moreover, differently from Spanish, non-DOM animate indefinites can receive a specific interpretation without PE. Thus, the contrast between (41) and (40): (41) Romanian a. Ion admiră o femeie. Ion admires a.F.SG woman = "Ion admires a woman." (unknown to the speaker/hearer) = "Ion admires a specific woman." b. Ion (o) admiră pe o femeie.
Ion CL.3F.SG.ACC admires DOM a.F.SG woman "Ion admires a (specific) woman." Other contexts of optionality come from existential quantifiers that alternate between the so-called weak and strong readings (see especially Milsark 1979, Diesing 1992or de Hoop 1996. The weak reading is related to non-specificity, while the strong one normally entails specificity. The specific interpretation on the animate needs differential marking (42-b): (42) a. Juan vio muchas chicas (en la calle). Juan saw many.F.PL girl.F.PL in DEF.F.SG street "Juan saw many girls (on the street)." b. Juan vio *(a) muchas chicas (en la calle).
Juan saw DOM many.F.PL girl.F.PL in DEF.F.SG street "Juan saw the many specific girls (on the street)." Spanish (43) a. Am văzut multe fete.
CL.3F.PL.ACC-have.1 seen DOM many.F.PL girl.F.PL "I saw many (specific) girls." Romanian This type of alternation can be resolved as follows: specific readings can be subsumed under the already existing SQs; non-specific interpretations need to be encoded as DOM restriction in the table containing ungrammatical contexts (Table 4).
Lastly, the differential marker is optional with Romanian definite animates, as we have seen in examples like (9-a) and (9-b).
A conclusion frequently drawn on the basis of contrasts seen with inanimates and strong/weak quantifiers is that DOM is a specificity-inducing mechanism. This statement, however, requires more attention. A careful examination of the data indicates, in fact, that DOM is insensitive to specificity in the two languages discussed here (see also López 2012).

DOM and specificity
That differentially marked objects are not restricted to specificity can be demonstrated by a variety of diagnostics. For example, as Rodríguez-Mondoñedo (2007) or López (2012) have shown, objects introduced by a can act as antecedents for a discourse anaphor, similar to the non-a objects. The same observation is valid for Romanian.
Additionally, both types of objects can appear in configurations that contain the subjunctive mood, a marker of non-specificity (Rivero 1979). Moreover, the differential marker can also co-occur with the quantifier cualquier-"no matter who." Note that in Spanish the latter quantifier tolerates only the subjunctive mood: (44) a. María buscó a una gestora cualquiera que María looked-for DOM a.F.SG manager no-matter-who that *habla/✓hable alemán. speaks.IND/speaks.SUBJ German. "Maria looked for a manager (no matter who) that could speak German." (López 2012, ex. 38a, p. 18) Spanish b. L-am întrebat pe un om oarecare.
CL.3M.SG.ACC-have.1 asked DOM a.M.SG man no-matter-who "I have asked a random man." Romanian Insensitivity to specificity also predicts that differentially marked objects do not need to take the widest scope. This is actually borne out by the data; the same conclusion is also supported by López (2012), some of whose examples we address below. To better understand the point, we can start the discussion by first exemplifying contexts where differentially marked objects do take wide scope with respect to other quantifiers.
The sentence in (45) contains an a-marked indefinite which can be interpreted as having wide scope over the conditional. One of the possible readings of (45) is that Bert will get offended, if Lud invites a specific philosopher (that might be known just to Bert and Lud).²¹ (45) Si Lud invita a un filósofo, Bert se ofenderaá.
if Lud invites a.DOM a.M.SG philosopher, Bert SE offend.FUT.3SG = "If Lud invites a philosopher (no matter which one), Bert will be offended (Bert wants to be the only philosopher at the party." (∃ >→) = There is a specific philosopher such that if Lud invites him, Bert will be offended.
(→ >∃) ( López 2012, ex. 25e, p. 13) Given that the narrow scope is also possible, we cannot conclude that the differential marker is an absolute scope widening mechanism.²² Contributions by Leonetti (2003), Rodríguez-Mondoñedo (2007) as well as López (2012) have brought to the fore numerous examples with differentially marked quantifiers in object position which cannot take wide scope over higher quantifiers (e.g., those in subject position). An example with a DOM universal quantifier restricted to narrow scope is provided below: (46) Un hombre ama *(a) toda mujer.
A.M.SG man loves DOM every.F.SG woman "A man loves every woman." (López 2012, ex.7, p We have concluded that if tod modifies an animate noun, the differential marker is always obligatory, as in (4). However, despite the obligatory presence of DOM, the quantifier cannot take wide scope with respect to the existential in the subject position. It seems warranted to follow López (2012) on the observation that accusative a can widen scope only if merges with an indefinite. Thus, a cannot be unambiguously associated with specificity or widest scope.
 21 That differentially marked indefinites can take scope over conditionals is problematic under classical theories of Q(uantifier) R(aising). As discussed in detail in Reinhart (1997), QR equates the scope of an operator to its c-command domain and is sensitive to islands. Conditionals are islands for movement, as shown in the examples below, where extraction out of a conditional results in ungrammaticality.
(i) *Which dress 1 will Maria donate if she buys e 1 ? (ii) Which dress 1 will Maria donate after buying e 1 ?
This suggests that the mechanism regulating wide-scope DOM readings out of conditionals is of a different nature. Reinhart (1997) proposes instead a semantic procedure, choice function (CH). Its basic role is to turn an indefinite from its "regular" <e, t> type (as a property) to type <e>. The choice function is in fact a variable that can be bound by an existential quantifier. As the position of the latter is not fixed in the tree, a DP containing a choice function can receive various scopes depending on the precise location where it is bound. We use below the entry from López (2012, ex.15a, p.7) to exemplify the representation of an indefinite taking scope above the conditional.
(iii) If Lud invites a philosopher, Bert will be offended.
∃f [CH(f) ∧ [Lud invites f (philosopher) → Bert will be offended]] "There exists a (choice function that picks out a) philosopher such that if Lud invites the philosopher picked out by the choice function, Bert will be offended." For further discussion about choice functions, see Reinhart (1997), Kratzer (1998) as well as Winter (2007). We, however, do not provide a more detailed discussion about indefinites and whether other mechanisms might, in fact, be more adequate than choice functions. Romanian indefinites (can) take wide scope and show insensitivity to islands even in the absence of the differential marker. The exact nature of these mechanisms requires a comprehensive discussion, going beyond the data at hand and the space in this article. It also does not necessarily solve independent problems raised by definites, which also trigger a variation locus between Romanian and Spanish, and we do not discuss it further.
22 Similarly to what we saw about specificity, the differential marker induces specificity in some contexts but is also wellformed in non-specific configurations.
In fact, as Rodríguez-Mondoñedo (2007), López (2012), a.o., and others have noticed, there are also contexts where differentially marked objects can only be interpreted with narrow scope and must take scope under negation. A relevant example is (47), which conveys that there is no woman that John loved.

(47)
Juan no amó a ninguna mujer. Juan NEG loved DOM no.F.SG woman "John loved no woman." Spanish (López 2012, ex. 25d, p. 13) * ∃>¬ ¬>∃ Absence of wide scope and specificity with DOM is clear in Romanian, too. We provide the example below from Cornilescu (ex. 32), where a non-specific, narrow scope interpretation of the DOM indefinite is natural and easily available: Fiecare parlamentar l-a numit secretar pe un every member of parliament CL.3M.SG.ACC-has appointed secretary DOM a.M.SG prieten. friend "Every member of parliament has appointed a friend (of his) as a secretary." These types of contexts confirm the following two assumptions with respect to Romanian and Spanish: (i) differential marking is not a specificity-inducing mechanism;²³ (ii) the differential marker is not a scopewidening device either. We can add these specifications to our list of similarities and differences, as illustrated in Table 5, which encodes two other DOM properties: syntactic behavior of the marked objects and their interpretation below the external argument (the latter is discussed in Section 4.3). We summarize below (in Tables 7 and 8) the accusative diagnostics for both Spanish and Romanian.²⁴

Summary
The parameters summarized in Table 5 encode both important similarities and several differences with respect to differential marking in Romanian vs standard Spanish. We begin with the similarities signaled in Table 6.
While differential marking is crucially sensitive to animacy in both languages, there are also syntactic contexts where this restriction must be liftedtelicity (6-c), as well as aspectual restrictions imposed by certain classes of predicates in Spanish (18), or nominal ellipsis contexts with SQs,, and possessive determiners (23-b) in Romanian. When it comes to animates, both languages obligatorily mark personal pronouns and proper nouns, the animate negative quantifier, wh elements restricted to animacy, the universal quantifier with animate DPs, demonstratives with animate DPs as well as strong readings of weak quantifiers with animate DPs. In both languages, the prepositional accusative is not a specificity or a wide-scope-inducing mechanism. On the morphosyntactic side, both languages use what looks like "noncanonical" prepositional morphology to signal objects in prototypical transitive clauses. Crucially, such objects syntactically behave like true accusatives, as opposed to datives or adjunct PPs (Tables 7 and 8).
The differences, in turn, are grouped together in Table 9. The most salient contrast is that Romanian referential definite animates are possible without the prepositional accusative. This is precisely the contrast illustrated by (3-a) and (9-b). Moreover, animates do not have to be obligatorily marked in syntactic contexts such as SCs, as seen in (7-a) from Spanish when compared to the Romanian (17). Second, the inventory of animacy-insensitive differential marking is not homogenous in the two  23 We are not claiming that additional morphology on the object is always insensitive to specificity crosslinguistically. In some languages, differential morphology or object agreement can be restricted just to specificity (e.g., Turkic, Bantu, Neo-Aramaic languages, etc.). 24 Some of which have been modeled after Bárány (2018).
languages. We have mentioned that Romanian inanimates are not robustly affected by telicity, differently from standard Spanish examples of the type in (6-c) or (18). On the other hand, there are several configurations in Romanian which require differential marking for both animates and inanimates, the

Contexts and characteristics DOM Spanish DOM Romanian
Personal pronoun +(1) + ( 8-a) Proper name +(2) + ( 8-b) Referential definite animate +(3-a) ( 9-b) Animate SQs +(4-a) ( 11) Non-adjectival Neg/Exist quantifiers with animacy restriction +(5) + ( 13-a) Adjectival Neg/Exist quantifiers with animates +(5-a) − Animate (indefinites) in telic contexts +(6-b) − Animate shared objects in clause union +(7)     most salient being nominal ellipsis with SQs (12), the augmented demonstratives (21-b), the adjectival (22b) and genitival (23-b) determiners, relative pronouns (25-c), D-linking (25-b) as well as certain types of strong partitives (25-a). These latter contexts are strictly sensitive to animacy in standard Spanish. The comparison of two closely related varieties provides important hints into the nature and limits of complex categories such as animacy, referentiality, and morphological case. One of the challenges is finding a formal mechanics that is flexible enough as to derive both the similarities and the differences. In Section 4, we propose an analysis which takes into account a typology of argument licensing strategies in the two languages. The account is couched in the generative formal tradition, more specifically minimalism.

Case and incorporation
In the previous sections, we have illustrated several contexts where the oblique differential marker is obligatory, irrespectively of its "canonical" features. For example, we have seen that in standard Spanish SCs (or more generally clause union configurations, López 2012), animate nominals require the presence of the differential preposition, irrespectively of whether their interpretation is specific or not. In the example below we repeat a DOM indefinite which can be interpreted as non-specific, especially if the embedded predicated is intensional ("necessary"): (49) Considero *(a) un estudiante intelligente/necessario por un Consider.1SG DOM a.M.SG student intelligent.M.SG/necessary.M.SG for a.M.SG proyecto. project "I consider a student intelligent/necessary for a project." Spanish Crosslinguistically, SCs are contexts that regulate the distribution of nominals; the latter might need certain pieces of structure in order to be grammatical in such configurations. One plausible way to explain this (morphological) restriction is to equate it to a more abstract requirement on nominals. Generally, this more abstract need is related to Case.
In recent formal analyses, the larger structure in differential objects has been associated with the presence of an uninterpretable Case ([uC]) feature on the nominal, as schematically represented in (51) vs (50). Following strict minimalist conditions on narrow syntax derivations as well as legibility conditions imposed by interfaces where uninterpretable (Case) features cannot be read, the latter must be eliminated under checking by appropriate heads in the sentential domain (Chomsky 2000 et subseq.).
Under several accounts (Ormazabal and Romero, 2007, 2013a, López 2012, Kalin 2018, Levin 2019, what unifies special marking on certain classes of objects is precisely their requirement to enter into a (ϕ-) relationship with functions heads (v, T, C, etc.) in the clausal spine. This permits the valuation of their [uC] feature. Nominals that do not take the differential marker are generally assumed to lack an [uC] feature, even being predicates of type <e,t>. They do not need licensing and, in fact, undergo (pseudo-)incorporation with V²⁵ (Ormazabal and Romero 2013a, Bárány 2018, a.o.). The split is illustrated below: (50) (pseudo)-incorporation (51) Differentially marked objects

López (2012)
However, assuming that all nominals that do not show up with differential marking lack [uC] (and must undergo incorporation) is problematic, at least for languages like standard Spanish. As we have seen throughout the article, Spanish nominals come in various sizes. They can be bare, can show an indefinite or definite morpheme, and, additionally, can present the differential marker. The latter is independent of definiteness/indefiniteness. If we examine contexts that signal the presence of an [uC] feature on the  25 For Romance it is the D head that (pseudo-)incorporates into V (see López 2012 for extensive discussion).
nominal, such as SCs we just mentioned, we notice a more complex picture. As already seen in (7), inanimates are used in SCs without the differential marker. Crucially, what cannot be missing is the definite/indefinite marker. We repeat the relevant example here: (52) Considero (*a) un libro interessante. consider.1SG DOM a.M.SG book interesting.SG "I consider a book interesting." Spanish Contexts like (52) indicate that at least some types of non-differentially marked nominals must contain an [uC] feature. This is exactly the same conclusion López (2012) reaches. The more specific claim is that such nominals check their Case after their D incorporates into V and raises to υ. Differentially marked nominals, on the other hand, require short scrambling to the specifier of α, an intermediate functional projection between V and υ, as shown in (54-a). In that position, the objects carrying an [uC] feature can be probed by v and have their accusative feature valued.
[uC] valuation is based on feature sharing, one of the outputs of the Agree operation. This operation can value or covalue all the features specified on the two items that are part of the relevant operation. As a result of its application, uninterpretable features ([uf]) are valued and removed from the derivation. The formalization of Agree used by López (2012) is as in (53); a restriction López (2012) introduces is that Agree (a, b) must be strictly local, in the sense that it can only affect the specifier of the complement of the probe.

DOM -Accusative case checking
For López (2012), the case feature in K is associated with a choice function switching the semantic type of the nominal from <e, t> to <e>,²⁶ as shown in (55). Moreover, the choice function can only be interpreted in a position above (existential closure at) VP.

(55)
Choice function in K (López 2012: ex. 13, p. 78) The oblique preposition is inserted at PF when certain conditions are met, for example in the context of an animate feature, as illustrated in (56), or in the context of a telic feature (which, as discussed in the previous section also affects inanimates of the type in (6-c) or (18)), as seen in (57):  26 For quantifiers, the type is more complex.

Romanian DOM: some problems
Turning now to Romanian, with respect to contexts that signal the licensing of an [uC] feature on nominals, such as SCs,²⁷ we noticed that differential marking is not obligatory with definite animates. We repeat in (58-a) a relevant example. Differential marking is obligatory with those nominals which must carry it even in regular transitive clauses, irrespective of SCs. In (58-b), we see a sentence with the animate negative quantifier in an SC; as expected, DOM is obligatory. This demonstrates that DOM is not a reflex of SC structure in Romanian. Crucially, what cannot be missing with object nominals of the type in (58-a) are the definite or the indefinite markers.²⁸ Knowing that these latter elements are normally in D 0 , the reasonable conclusion to make is that the [uC] feature is in D 0 in Romanian. This provides initial support to an analysis along the lines we just presented above from López (2012). Thus, differential marking is not simply the distinction between case-marked and caseless (incorporated) nominals. However, the Romanian data demonstrate that not all the predictions made by López' analysis, as sketched above, hold. There are differences with respect to the position of differentially marked objects and with respect to syntactic effects involved in DOM.
Another observation López (2012) made is that differentially marked objects cannot c-command the external argument (EA). In typical examples like (59), the differential object quantifier cannot bind a variable in the postverbal subject.
 27 Assuming that object (pseudo-)incorporation can only take place in a complement position to V, the nominal in SCs is prevented from undergoing this operation. No matter whether consider-A contexts are seen as projecting small clauses or as constructing V-A complex predicates (Williams 1983, a.o.), the nominal will not be in a complement position to V at any point in the derivation. 28 As expected, mass nouns which can be used bare as complements of transitive predicates must carry the definite marker in SCs, irrespectively of whether they receive a definite interpretation. The contrast below is telling: Ion consideră miere-*(a) sănătoasă.
(ii) Ion mănâncă miere. Ion considers honey-DEF.F.SG healthy.F.SG Ion eats honey "Ion considers honey healthy." " Ion eats honey." Romanian (59) Ayer vio su * i padre a cada i niño. yesterday saw his father DOM every boy "Yesterday his father saw every boy." Spanish (López 2012, ex. 13 adapted) The same observation holds for Romanian differential objects which are not clitic doubled:²⁹ (60) A văzut tatăl său * i pe fiecare i copil ieri. has seen his father DOM every boy yesterday "Yesterday his father saw every boy." Romanian These facts provide strong support to the conclusion that differentially marked objects do not c-command the external argument. This implies that, in non-dislocated configurations, they are not found in a position above the external argument. However, the Spanish ones are not found in situ either. López (2012) assumes that Spanish differentially marked objects undergo short scrambling to the specifier of α (the locus of both accusative and dative case assignment in a language like Spanish). This explains why the Spanish differentially marked objects do c-command the indirect object. In the Spanish example below, we notice that a differentially marked object allows a quantifier-variable interpretation when paired with the indirect object. That is, a reading under which prisoners are matched to their own sons is possible. This precise reading seems to be absent if the direct object shows up without the differential marker, according to López (2012).
(61) Context: What did the enemies do? The enemies delivered X to Y and Z to W, but… Los enemigos no entregaron a su i hijo a/∅ ningún i prisionero. the enemies NEG delivered.PL DAT his son DOM/∅ no prisoner "The enemies did not deliver any prisoner to his son." (López 2012, ex. 18, 41) For Spanish, short scrambling to the specifier of α, as in (54), explains quantifier binding from DOM into the IO and DOM presence in a position that c-commands the IO.
Analyzing Romanian now, if we test the binding relations between direct objects and indirect objects, the facts are not as clear as in Spanish. Using the same context as in (61), several conclusions emerge. First, binding from the differentially marked object into the IO does not go through for most speakers (62-a). The quantifier-variable interpretation improves (at least for some speakers) if the differentially marked object is clitic doubled, using the accusative clitic. Accusative clitic doubling and differential marking must be distinguished as two separate mechanisms (see also Cornilescu (2000), a.o.);³⁰ as such, these examples indicate that although DOM is lower than the EA it is not necessarily higher than the IO. Crucially, binding from the IO into the DOM-ed nominal does go through, irrespectively of the linear position of the IO and the differentially marked object: (62) a. Inamicii nu (i-)au înmânat fiului său * i pe enemies.the.M.PL not CL.3SG.DAT-have delivered son.DAT.M.SG his DOM niciun i prizonier. no prisoner "The enemies did not deliver any prisoner to his son."  29 Note that binding of a variable inside the subject by the DOM quantifier becomes possible if the latter is also clitic doubled, using the accusative form of the clitic: (i) L-a văzut tatăl său i pe fiecare i copil ieri. CL.3M.SG.ACC-has seen father his LOC=DOM every boy yesterday "Yesterday his father saw every boy." Romanian As expected, differentially marked objects that are subject to clitic doubling under right dislocation also allow binding into the EA. Clitic doubling is independent of DOM and we do not discuss it further in this article. 30 Remember that there are several contexts in Romanian where differential object marking is required, while accusative clitic doubling is impossible. Among these are the animate negative quantifier, illustrated in (13-a) or the animate wh element, presented in (14). b. ?Inamicii nu i l-au înmânat enemies.the.M.PL not CL.3SG.DAT CL.3M.SG.ACC-have delivered fiului său i pe niciun i prizonier. son.DAT.M.SG his DOM no prisoner "The enemies did not deliver any prisoner to his son." c. (L)-a dat pe copilul său i fiecărui i părinte.
CL.3M.SG.ACC-has given DOM child.the.N.SG his each.DAT.SG parent "S/he has given each child to his/her parent."

Romanian
To summarize what we see in Romanian: (i) there are diagnostics supporting the connection of DOM with case licensing of nominals, as in (58-b); (ii) however, examples with definite animates in SCs allow us to safely conclude the [uC] feature is introduced in D 0 and is independent of differential marking; (iii) differentially marked objects do not take scope over EAs and IOs. The last property indicates that the position of Romanian DOM is not easy to identify. But this creates a tension with the first propertyit is clear that DOM is related to nominal licensing, but it does not seem to be the only possible licensing mechanism for nominals. The question is how exactly to tell DOM apart.
Let's see if we can apply a morphological explanation to DOM. Under this hypothesis DOM and the non-DOM definites carrying an [uC] feature have the same syntax. What sets DOM aside is just a special spell out form triggered by some features like animacy, etc., whose set turns out to be slightly different in Romanian as opposed to Spanish. Knowing that DOM indeed shows variation in the class of elements that it can signal, this idea could be motivated. The problem with this hypothesis is twofold; on the one hand, it makes wrong predictions when it comes to syntactic effects triggered by DOM; on the other hand, it leaves many DOM contexts unexplained, such as the apparent optionality with definite animates.
One important observation with respect to (Romanian) DOM is that it gives rise to co-occurrence restrictions. For example, in Romanian, DOM-ed nominals are not possible with possessor dative forms of (reflexive pronominal) SE. We illustrate a relevant example in (63): Ion îşi ajută (*pe) un prieten. Ion SE.DAT.SG helps DOM a.M.SG friend Intended: "Ion helps a friend of his." Romanian These types of effects, which remind the Person Case Constraint (PCC),³¹ cannot be derived in the morphology. This implies that DOM gives rise to syntactic effects which are distinct from the result of [uC] valuation. In Section 4.4, we present an analysis that, on the one hand, can reconcile all these properties and, on the other hand, is able to formally derive the differences from Spanish.

Licensing beyond Case
The gist of the proposal, building on Jaeggli (1982),³² is that Romanian DOM involves an additional licensing operation beyond the valuation of [uC] itself. At Phonetic Form (PF), the result of this additional licensing operation is the spell out of the oblique marker on differentially marked objects. The conclusion we draw from Romanian is that, in general, D 0 acts as a phase edge, and requires licensing, as it hosts the [uC] feature. Any additional feature will require additional licensing. Additional features are contributed  31 See Ormazabal and Romero (2007) or Ormazabal and Romero (2013a) for a detailed discussion about animacy-related oblique DOM and PCC effects in Spanish varieties. 32 We take from Jaeggli (1982) the intuition that an important aspect of DOM resides in the need of more than one (intepretable) feature inside the nominal to be obligatorily case licensed. Kalin (2018) contains a proposal that equates DOM to the need of obligatory licensing of nominals with interpretable features, but the system developed there does not capture the Romanian data. either by: (i) a [+PERSON] specification on a sentience-defining projection above D 0 , which encodes the difference between animates and inanimates as well as the relevance of animates in the discourse (see also Cornilescu et al. 2017) or (ii) by features on projections that are merged above D 0 , and which are needed in the construction and interpretation of various classes of quantifiers, nominal ellipsis, etc. Therefore, a nominal can contain more than one feature that requires licensing (see also d'Alessandro 2017, a.o. for other languages). The result is that a DP can enter into multiple licensing operations such that all the features can get licensed (see also Béjar and Rezac 2009). We also take at face value the recent hypotheses that Universal Grammar (UG) contains more than one licensing strategy, namely, licensing linked to ϕ features and δ licensing. δ is a grammaticalized discourse specification linked to an A-feature (Miyagawa 2010, Belletti 2018. The two features will be abbreviated as ϕ and a δ. Another aspect that needs clarification is the role of PERSON features. Following Bernstein (2008), Longobardi (2008), a.o., definiteness ("strong" D, more generally) is connected to a [PERSON] feature, signaling that a nominal is not a predicate but rather a participant in the event. On the other hand, researchers such as Richards (2008), Cornilescu (2000), or Adger and Harbour (2007), a.o. have connected a [PERSON] feature to animacy, as shown in Table 10. As both views seem to be correct, we propose that they can be reconciled in the following way: the multifaceted behavior of [PERSON] is the result of the position in which this feature merges. For example, if merged in the Sentience projection above D 0 (as in 64, following and adapting Zubizarreta and Pancheva (2017)), it will be interpreted as related to animacy.
Connected to these observations, we also make use of a second line of research into DOM, namely, approaches that link it to a syntax-pragmatics interface phenomenon; DOM is generally assumed to signal topical DPs or the so-called secondary topics (Leonetti 2008, Iemmolo 2010, Dalrymple and Nikolaeva 2011, d'Alessandro 2017, Belletti 2018, Khouja 2019. In Romanian, the connection to secondary topics is not straightforward in all instances differentially marked objects can be insensitive to information structure. For this reason, it seems more plausible to assume that animacy can be a separate category, but relevant to the syntax-pragmatics interface. From this perspective, the above mentioned connection of animacy to [PERSON] feature, possibly further connected to individuation (see also Ledgeway et al. 2019, a.o.) makes better predictions. Putting together decompositions in the higher DP left periphery (Ihsane and Puskás 2001, Harley and Ritter 2002, Ihsane 2008, Kučerová 2018, we propose an expanded DP structure as in (64): (64) Crosslinguistically, a locus of variation is given by whether sentience bundles topic/specificity features or not (Mursell 2018, a.o.). In Romanian, sentience can be realized without topicality. If D 0 licenses a referential interpretation of its associated noun (following Longobardi 2008), it is clear that sentience requires a referential category; but it is not a marker of specificity or presuppositionality, per se. However, given that D 0 contains an [uC] feature that requires valuation any A-feature above D 0 will require a separate Agree/licensing operation. This implies that the additional feature will require the presence of an additional licenser, beyond the regular probe which values the [uC] on D 0 (following Jaeggli 1982, see also Kalin 2018).
We use a more refined decomposition of the architecture of transitive predicates. Following Legate (2014), we assume that their structure involves not only an abstract light verb (υ) but also Voice, the functional projection introducing the external argument. Similarly to López (2012), we also make use of the abstract functional category α³³ between υ and voice, as in (65): (65) The model assumed here presupposes some other ingredients. First, Agree can be established at a distance (see also Chomsky 2000, a.o) and can apply cyclically, so as to ensure the valuation of all relevant features (Béjar and Rezac 2009, a.o.).³⁴ Second, there can be more than one structural accusative (Starke 2017) and, thus, more than one probe involved in the checking of accusative case. And third, DOM is linked to a last resort licensing mechanism (Jaeggli 1982, Kalin 2018. In a nutshell, under our account, DOM differences between standard Spanish and Romanian are a result of the following macro-parameter: Before moving on to the detailed analysis of (micro)variation, a few remarks are in order about Cornilescu's (2000) alternative account, which also attributes an important role to [+PERSON] in the Romanian prepositional DOM. Although different in both the mechanics and the analysis, the discussion in this article is indebted to some crucial intuitions in Cornilescu (2000). The gist of Cornilescu's (2000) proposal is that DOM is connected to semantic gender in Romanian. Besides categories that obligatorily contain a semantic gender specification (the wh-cine "who", the negative Q nimeni "nobody"), there are also instances where grammatical gender features are interpreted as semantic gender. For Cornilescu  33 We assume that this is a functional head in the low VP discourse-related periphery (see especially Belletti 2004, a.o.) that can license (features on) internal arguments, which are not otherwise licensed by υ or Voice. In some languages, for example Spanish, α bundles applicative and telic functions (see López 2012). In Romanian, it seems to be a discourse-related head that acts as a last-resort licenser for features that are left after the initial agree relation by υ, which is the main licenser in this domain. 34 Thus, we agree with accounts under which the Activity Condition does not hold.
(2000), animacy as a reflex of semantic gender equals to featural specification as [-Neuter]. More specifically, the preposition pe is viewed as a functional head specified with the feature [-Neuter], according to the criterion in (66): (66) The preposition PE is not selected in the accusative if a DP is specified for semantic gender as [-Person], that is, [+NEUTER] or [α PERSON]. PE is the obligatory otherwise. (Cornilescu 2000, 4') It follows that, for Cornilescu (2000), DOM is expected to be ungrammatical with propositional ellipsis, as the latter cannot be reduced to semantic gender. This seems to go through in some contexts, such as (73). However, once propositional ellipsis is examined in more detail, it appears that DOM is not necessarily sensitive to the absence of a [+NEUTER] feature. The example below, obtained from ordinary speech, has the uninflected form asta tracking propositional ellipsis but also the prepositional ACC.³⁵ Ei, n-o mai cred pe-asta. wow, not-CL.3F.SG more believe DOM-this.DFT "Wow, I don't believe this (i.e., that the criminals have escaped again)." Romanian Another problem is how to extend this analysis to DOM partitives of the type in (25-a), which contain an overt lexical head, irrespective of animacy. The present analysis does not equate [+PERSON] to semantic gender. In fact, a [+PERSON] specification can have a variety of sources. What is relevant for Romanian [+DOM] is not the animate semantic gender but rather an additional feature beyond [uC] in D 0 .

Addressing (micro)variation
We can now explain how the macro-parameter discussed above derives both similarities and differences with respect to DOM in standard Spanish and Romanian. We will start with contexts where special marking is obligatory on the object. As we have seen in the previous sections, these classes are not completely uniform in the two languages. While it is true that there are many points of convergence, we also notice important differences. Let's address first commonalities in obligatory marking. We repeat the relevant table below, namely Table 6 (repeated as Table 11).
Among these properties, an important part support the accusative syntax: doubling by accusative clitics, the accusative form of tonic pronouns as well as other accusative syntactic diagnostics (summarized in Tables 7 and 8). The analysis proposed above explains the accusative behavior under the assumption that differential marking is the spell out of accusative [uC] and [PERSON] features on the relevant nominals. The position of the α licenser, below the EA introducing head, derives lack of binding into the EA.
Moving on with personal pronouns and human proper names, these are classes where animacy³⁶ is obligatory and which cannot be of type <e, t> when argumental. Under López' (2012) assumptions, these  35 The additive operator mai seems to have a contribution to the presence of PE. This is even clearer in contexts which also have an additional additive operator, spelled out by şi "and": Am mai auzit-o şi pe asta. have more heard-CL.3F.SG.ACC and DOM this.DFT.AUG "Now I heard even this." In fact, a preliminary investigation shows that native speakers tend to prefer PE marking on the propositional demonstrative in this context. 36 In both Romanian and standard Spanish, tonic forms of third person pronouns cannot be used with inanimate reference (as direct objects).
categories are correctly predicted to be ungrammatical without DOM in Spanish. For Romanian, the obligatory animacy of these classes and its relevance in the discourse motivate the assumption that they contain a [+PERSON] feature merged above D 0 . Given that the initial licenser (υ) checks the [uC] feature on D 0 , an additional licenser is needed for the [+PERSON] feature. Thus, obligatory marking is derived. Animate SQs receive a similar explanation. With respect to their containing a feature above D 0 , note that such quantifiers must be constructed with definite nominals in both Romanian (11) and Spanish (4-a). In their elliptical animate reading, what gets deleted at PF is only the definite nominal.
The non-adjectival Neg/Exist quantifiers with animacy restriction include elements such as nobody or the strong variant of somebody. The negative nobody has an animacy restriction in its domain of quantification. Moreover, it is specified as a universal quantifier. Thus, it can be assumed that the animacy component, read as [+PERSON], is merged above D 0 . This predicts obligatory DOM for both Romanian and Spanish. In the same line, following canonical assumptions in the literature (Milsark 1979, de Hoop 1996., for a few classical titles), the strong variant of positive quantifiers is not of <e, t> type. It can also be safely assumed that they contain a [+PERSON] feature (equated with animacy) merged above D 0 . Thus, differential marking is needed. The same reasoning extends to strong readings of weak Qs with animates (DPs or under ellipsis), to direct object animate wh elements and to the demonstratives with an animacy restriction. Turning now to possessive have, López (2012) has convincingly argued that this predicate can project two types of structures: a locative one where the direct object is of type <e, t> and a transitive one where the direct object is of type <e>. This predicts that the latter sub-type, if animate, can take DOM.
Let's examine now the equatives, where DOM can be obligatory, irrespectively of animacy. One important observation is that in these configurations the preposition is not a lexical marker introducing the standard. These comparatives are not phrasal in Romanian or Spanish. Unambiguous evidence comes from obligatory grammatical function tracking. For example, in both languages the differential marker disappears if the antecedent and the standard function as subjects.³⁷ Grammatical function tracking is sufficient to demonstrate both the structural nature of DOM and the clausal (non-phrasal) status of the equative in Romanian. Moreover, this type of case identity restriction is seen in other VP ellipsis contexts. Some authors have proposed that case identity is in fact one of the conditions of the licensing of ellipsis (see Chung 2003, and more recently Takita 2015). The condition Takita (2015) introduces is the following: (68) Case-oriented syntactic identity (Takita 2015: ex. 24) If a DP is extracted from an ellipsis site, and if the head that case licenses the DP is contained in the ellipsis site, the case licensing head in the ellipsis site must have an identical head in the antecedent that case licenses the correlating DP.
Another observation, made by Irimia (2018), is that these comparatives have reduced clausal structure. The interpretations provided by native speakers signal the use of the so-called evasion strategies.³⁸ The two evasion strategies seen with these comparatives are: (i) reduced syntactic structure (absence of C, T projections) in the comparative which forces obligatory mismatched temporal/aspectual interpretations (i.e. readings of the type "as one would V DP obj ") and (ii) copular clause ("as if it were a DP"). Thus, an equative comparative projects one of the structures below: As the case-licensing head is contained in the ellipsis site, the object will have to undergo extralicensing so its case licensing head is similar to that of the antecedent. An additional δ head is recruited which will license the object as a last-resort strategy (confirming the last-resort nature of the differential preposition). The structure of reduced comparatives makes plausible the assumption that the additional licenser is a low Topic head (see Belletti 2004 for the low information-structure periphery). Thus, a secondary licensing account also predicts these puzzling data at least for Romanian. In Spanish, such objects must undergo obligatory scrambling to a position where their [uC] must be checked. This, once again, predicts the presence of the differential marker (Table 11).

Differences in obligatory DOM contexts
As illustrated in Table 9 repeated as Table 12 the most important differences in obligatory DOM contexts result from: (i) obligatoriness of the differential marker with animate definites which are interpreted referentially in Spanish, as opposed to optionality in Romanian, (ii) DOM obligatoriness with inanimates in telic contexts in Spanish, and (iii) DOM obligatoriness with inanimates in a variety of nominal ellipsis contexts in Romanian.
The hypothesis that DOM results from an additional licensing operation in Romanian explains its optionality with animate definites (which receive a referential interpretation). The differential marker tracks an animacy specification (encoded as [+PERSON]) only if the latter is relevant in the discourse and merged in the derivation. Structurally, the [+PERSON] feature is introduced above D 0 , beyond the [uC] feature. In Spanish, animate definites with a referential interpretation will be obligatorily marked, as they are never of <e, t> type and cannot undergo incorporation. Thus, there is no other option for them. The same explanation holds for animate DPs in clause union contexts.
Turning to Spanish telic predicates, the short scrambling operation opens the possibility for their objects to be interpreted in the specifier of α. The scrambling operation is made available by the aspectual structure of the relevant predicates; telicity requires direct objects to be of a certain semantic type (<e, t> is excluded) and appears in a position above VP (see also López 2012 for detailed references). In Romanian, on the other hand, short scrambling does not signal DOM; from this it follows that telicity is not relevant to the special marker.
An independent parameter underpins the distinct behavior of animate indefinites with specific readings. In Spanish, these contexts require obligatory differential marking, as these objects are animate  and at the same time subject to a choice function which shifts their type from <e, t> to <e>. In Romanian, however, indefinites can obtain a specific reading even without the presence of a differential marker and irrespective of animacy. This is due to the fact that Romanian indefinites can be subject to other interpretive mechanisms beyond choice functions. For example, Romanian indefinites can undergo quantifier raising more easily. Clitic doubling of non-pronominal DPs also results from yet a different parameter; as discussed by Ormazabal and Romero (2013b), a.o., clitics have a distinct structural makeup and a different nature in the two languages (pronouns vs determiners). The remaining points of variation group together contexts where DOM is obligatory only in Romanian. We will show below how the additional licensing hypothesis proposed for Romanian derives them. We begin with nominal ellipsis contexts with demonstratives, as in (70): (70) a. Ai cumpărat (*pe) paltonul acela?
CLT.3SG.M/N.ACC-have bought DOM this-N.SG.AUG "I have bought this (one)." Romanian (referring to the coat) In (70-a), the demonstrative is used as an adjective, modifying the DP paltonul. As the head noun is inanimate, prepositional ACC is not possible. In (70-b), the demonstrative is used as a pronoun in a nominal ellipsis construction, which contains a null pro head and tracks an inanimate antecedent. Nominal ellipsis of this type requires obligatory prepositional ACC, irrespective of animacy. Two other categories in the language show the same constraint as the demonstratives. These are the so-called adjectival demonstrative CEL, and the genitive linker a which contains a suffixed definite determiner under nominal ellipsis (see especially Cornilescu 2005, Cornilescu and Nicolae 2012 for a detailed discussion of these categories). The examples in (71) and (72) illustrate the similarity with the elliptical demonstratives:³⁹ Referential definite animate +(3-a) ( 9-b) Adjectival Neg/Exist quantifiers with animates +(5-a) − Animate (indefinites) in telic contexts +(6-b) − Animate shared objects in clause union +(7) In the case of indefinites, the index j must obey the novelty condition (the context cannot already have a referent (x j )). The new referent that is introduced into the domain will have "(x i )), the referent of the NP, as its subset" (Enç 1991). Given that the index responsible for the specificity of indefinites is interpreted above D (where definiteness is), it results that in this configuration there are two features that require licensing (definiteness and specificity). Thus, an additional licenser must be made available. The same structure is seen with the so-called D-linked wh elements as in (25-b), which are interpreted as specific partitives. Unsurprisingly, they require obligatory DOM, irrespectively of animacy, as they too contain more than one feature that requires licensing.

Optional and ungrammatical DOM
In Section 2.5, we have listed the contexts where the prepositional marker appears to be optional. Typical representatives are definite animates in Romanian as well as the weak determiners in both languages, when used adjectivally with an animate lexical noun. In the latter environments, if the head noun is animate, the prepositional marker can be absent without giving rise to ungrammaticality. With prepositional DOM, the quantified DP is normally interpreted as "specific". Under nominal ellipsis, DOM is only possible if the understood antecedent is animate. When the marker is absent, a weak, nonspecific interpretation arises. The alternation between a "specific" (also called "strong") and a non-specific ("weak" reading) is typical of weak quantifiers (see Milsark 1979, de Hoop 1996. For Romanian, we propose that the alternation is connected to the type of structure these elements project. Weak quantifiers do not have a full DP structure, while the strong ones do (see also Cornilescu 2000). On SQs, a [+PERSON] feature merged above D 0 will be interpreted as a restriction on the domain of quantification of Q, resulting in an animacy restriction. As a result, the nominal will end up containing two features and an additional licenser will be needed. A similar type of explanation can also be attempted for the relative pronoun care "who/which". In normative grammars, its use is banned without PE irrespectively of animacy. However, in colloquial speech, the prepositional accusative marker is easily omitted: For Spanish, the strong reading entails that such quantifiers are not of type <e, t>. Thus, they will have to raise to the specifier of α and, if animate, they will carry the differential marker.
Turning now to configurations where DOM is ungrammatical, Section 2.4 ( Table 4) has listed the following: objects in existential sentences, objects of possessive have with individual-level readings, inanimate bare plurals, negative/existential quantifiers with an inanimacy restriction, and objects of unaffected predicates with inanimate objects. Here the object nominal can only be of <e, t> type, excluding DOM.

Conclusion
In this article, we have taken some steps towards probing the limits of variation in DOM. We have provided comprehensive descriptions of DOM contexts in two robust prepositional accusative languages, Romanian and standard Spanish. We have noticed that this phenomenon shows both non-trivial similarities and steady points of variation. On the similarities side, in none of the two languages is DOM connected to specificity or a wide-scope-inducing mechanism. Moreover, sensitivity to animacy is a constant. The differences boil down to extensions to inanimates as well as points of optionality.
On the theoretical side, we have proposed that the full microvariation picture can be best explained as deriving from a morphosyntactic parameter. Following López (2012), Spanish DOM signals the presence of features such as animacy/telicity on nominals whose [uC] specification needs valuation via short scrambling. In Romanian, on the other hand, DOM exhibits syntactic differences with respect to other nominals with [uC] features and which must equally escape (pseudo)incorporation. Romanian DOM signals instead the licensing of an additional feature beyond [uC] in D 0 . This simple parametric choice has non-trivial consequences on the categories that show differential marking in each of the two languages, explaining both similarities and differences in a systematic manner.