2.1.1 The subject of have is an external argument

Starting from possessor have-sentences, it has already been presented that they can express various facets of possession (6,7). A closer look at them reveals that a human possessor may exhibit properties of an intentional agent. This discloses that the subject possessor is an external argument. These properties become evident as the subject passes two criteria for agenthood found in Alexiadou et al. (2015); it licenses modification by the adverb skopima/epitiðes ‘deliberately’ and controls a purpose clause (12).

In addition, have-sentences host non-human or non-animate possessors only if they are in an expected or standardized (if not part-whole only) relationship with the entity denoted by the object (13a). If a non-animate entity relates to a second non-animate entity that does not constitute an integral part or a standardized associate, a PP that includes a pronoun co-indexed to the subject must be added (13b). This type of sentence is often called a locative have-sentence.⁵

In other words, non-human entities fit into a plain have-sentence only if (a) they are in an inalienable possession relationship, i.e., they are inseparable by definition, or (b) they are expected to be associated with each other. Interestingly, this signals that the subject of plain have-sentences (human or non-human) is licensed only if it is closely related to the object and controls it, i.e., only if it is somewhat “agentive” towards it. Along these lines, this could serve as a first indication that the subject of have is an external argument.⁶

This conclusion is fortified by the fact that the subject of have can be read as an experiencer or a causer, two interpretations typical of external arguments, particularly those introduced by Voice.

In (14), the nominative subject bears the role of the experiencer of the situation described in the dependent clause. Crucially, the licensing of modification by skopima ‘deliberately’ and the acceptability of a purpose clause suggest that the subjects are external arguments introduced by have, not arguments of the CP-complement that have been raised to a sentence-initial position.⁷

The same conclusions apply to causer have-sentences as they also pass the tests for agenthood.

This behavior of possessors, experiencers, and causers supports the hypothesis that they are external arguments. The compatibility of these sentences with ability modals substantiates this hypothesis further.

Hackl (1998) notes that verbal passives, which arguably have an external argument that stays implicit, are acceptable under an ability modal, while stative or adjectival passives, which lack such an argument, are not. Alexiadou (2018) extends this criterion to object-experiencer verbs. This criterion predicts that predicates with external arguments should be acceptable under an ability modal. This prediction is indeed borne out in have-sentences as presented below.

To illustrate this, consider first that, according to Giannakidou and Staraki (2013), Greek distinguishes lexically between two types of modals. The language uses the impersonal bori as an epistemic possibility modal, i.e., like English might, and personal boro, as an expression of ability or deontic modality. Therefore, the criterion tailored for Greek predicts that a verb has an external argument if it is acceptable under personal boro in its reading as an ability modal. The behavior of Greek exo-sentences in (16) implies that they do have an external argument:

Moreover, the argument structure of the nominalization of possessive sentences provides additional evidence for the external status of have’s subject. To give some background, Markantonatou (1995) and Alexiadou (2001) show that the internal argument of a nominalization in Greek bears genitive case while the external argument of a nominalization deriving from a transitive verb merges in an apo ‘by’-phrase. The fact that a by-phase introduces the possessor in the nominalization of have suggests that it is an external argument.

Keep in mind that the prefixed form katoxi nominalizes the copula as plain *oxi ‘lit.*hav-ion’ is unattested in the language. This nominalization is licit only in the context of possession or ownership. Crucially, katoxi is a nominalization of exo even though there is a prefixed verb kat-exo in Greek which also means ‘possess’. This is evidenced by the fact that (17b) is a paraphrasis of the sentence in (17a) that can contain only the verb exo. Katexo is unacceptable in this case (see more in Kampanarou, 2023.):⁸

As presented below, the possessor cannot appear in genitive, i.e., as the internal argument of the nominalization (18a). Only the possessee can. The example in (18b) shows that the possessor surfaces in an apo-phrase, the Greek equivalent of a by-phrase. This implies that the possessor is an external argument (see more in 3.1).

Finally, there are modal have-sentences that constitute cases of obligatory control; the subject of have must be co-referring with a dropped subject in the embedded clause (19). These sentences differ from the above in that their subject is also thematically related to the verb of the na-clause complement.

As a detailed discussion of modal exo-sentences would require a comparison with other modals in Greek, an in-depth study is left for future research. Nonetheless, I refer to some key characteristics implying that these sentences involve Control instead of Raising, i.e., that their subject is merged as an external argument.

First, Bhatt (1998) argues against a control-analysis and claims that modal have-sentences involve Raising based on English and Hindi data. However, the distribution in these languages is the complete opposite of Greek, as described so far. An important piece of evidence for the author is that in English sentences, the subject of have may not be the bearer of the obligation (20a). As presented in (19) and (20b), this does not hold for Greek.

The behavior of the idiom to milo pefti kato apo ti milja, lit. ‘the apple falls under the apple tree = like father like son’ provides further support for a control-analysis. If the subject of modal exo was raised to this position from within the na-clause, the non-compositional meaning of the idiom should have been preserved. However, this is contrary to fact (21a). Hypothesizing Control instead of Raising is also justified, according to a reviewer, through examples like (21b), which show that the subject of a second modal (boruse ‘could’), i.e., o Janis ‘John’, controls at the same time the subject of exi ‘has’ and ɣrapsi ‘writes’.⁹

The observations above suggest that the subject of exo is introduced as an external argument, even in its use as a modal. However, it remains to be explored whether it is an argument of the exo that merges in v or a different one that merges as a modal. If the first scenario is true, the fact that the verb still surfaces with Active Voice morphology implies that its subject is external and merged in [Spec, VoiceP]. The following section capitalizes on this idea.

To sum up, this section has shown that the subject in possessor, experiencer, and causer have-sentences is introduced as an external argument based on (a) the fact that the subject of exo realizes thematic roles typically assigned to external arguments, (b) the compatibility of have-sentences with ability modals, and (c) the fact that the possessor appears in a by-phrase in the nominalization of the possessive exo-sentence. A question that remains open is whether modal have-sentences should be exempted from this generalization, even though preliminary evidence suggests they should not.

At this point, it must be specified that claiming that the subject of exo is external is not an issue debated in the literature, as a reviewer confirms. However, further substantiating this claim was the first essential step towards arguing that have is the transitive counterpart of BE. The second step is to show that have’s external argument is introduced by a transitive Voice head and, hence, that have is the allomorph that merges in the presence of such Voice head. The following section is dedicated to this.

2.1.2 The external argument of have is introduced by Voice

Even though have’s subject can bear the whole range of thematic roles a Kratzerian Voice head can assign, the fact that this subject can be an intentional causer, as evidenced in (15) above, is crucial. Drawing on previous work, Alexiadou et al. (2015) argue extensively that intentional DP-causers, as o Janis ‘John’ in (22), are constituents introduced by Voice.

Furthermore, the morphological makeup of experiencer have-sentences, an example of which is added below, is also indicative of a Voice projection:

Even though the status of the experiencers is generally controversial, the subject-experiencer in the above cases is in direct parallel with the subject-experiencers of Belletti and Rizzi’s (1988) Class I psych verbs for which there is a consensus in the literature. Class I verbs are unambiguously stative and behave like transitive predicates (Hale and Keyser 2002). Implementing the work of Belletti and Rizzi (1988), Pesetsky (1995), and Anagnostopoulou (1999) within a de-compositional model stemming from Kratzer (1996), this type of experiencer is a version of a State Holder, introduced by a (stative) Voice head and assigned to the DP that occupies its specifier position in the configuration presented in (24).¹⁰ This implies that this Voice head is syntactically and semantically transitive because it introduces a specifier position and assigns a thematic role to it accordingly (Schäfer 2008; Wood 2015).

Anagnostopoulou and Sevdali (2020) suggest that the case pattern in sentences like (23) acts as a supplementary argument for the same hypothesis. The authors explain that, generally, Greek experiencers are not introduced only by Voice. They are most often introduced by a High Applicative (v_APPL). When this happens, they get assigned dative (morphological genitive) (25a) or nominative case. However, introduced-by-v_APPL experiencers receive nominative case only if they do not have a theme DP-object (25 b,c) (Anagnostopoulou and Sevdali 2020: 25, ex. 51):

Consequently, based on their conclusions, we can argue that the nominative on the experiencer subject of exo is not the case of an experiencer introduced by v_APPL because the verb has a DP as a theme object, e.g., ton idravliko ‘the plumber’ in (23). We must assume that Voice is present and introduces this type of experiencer.

Moreover, the fact that the verb bears Active Voice morphology indicates not only that the subject surfaces in [Spec, VoiceP] but that it externally merges in this position.

To give some background, it is widely accepted in the literature that adopts Kratzer’s (1996) de-compositional view of the verbal structure that morphological Active Voice (henceforth Act) in Greek is assigned either in the absence of the syntactic projection of Voice (i.e., in unaccusative verbs) or in the presence of a φ-featured item in [Spec, VoiceP]. Building on Embick (1998) and much subsequent work, Alexiadou et al. (2015) hold specifically that Act in Greek appears either as default in the absence of a syntactic projection of Voice (26a) or when a regular (i.e., thematic and non-expletive) Active Voice head appears (26b). In contrast, non-Active Voice morphology (henceforth NAct) appears in (26c), i.e., when a Voice head with no specifier (marked as ‘{-D}’) is projected.

The rule in (27) captures the condition that determines NAct. This rule outlines that the syntactic head of Voice is instantiated by NAct whenever it lacks a specifier. As extensively argued in the literature, Greek is a language in which NAct is highly syncretic as it is used for reflexives, reciprocals, anticausatives, middles, deponents, and passives (Manzini et al. 2016; Oikonomou & Alexiadou 2022). In this system, the non-active version of the Voice head is most appropriately described as Middle Voice.¹¹

Turning our focus back to Active Voice, Oikonomou (2011) is more specific on (26b) and claims that only external merge to its specifier position can lead to Act if Voice is projected.

As non-existential have-sentences have an external argument, they are not unaccusative. Since they have an external subject and an internal object, they behave as typical transitive verbs. Therefore, the fact that they surface with Act can only mean they have a Voice head with a projected specifier position filled via external merge. Consequently, it must be assumed that the external argument of have is introduced by Voice. Notably, the Voice system of Greek implies the obligation for external merge. Other languages may fill the same position via internal merge (see, for instance, Szabolcsi’s 1981; 1983 seminal hypothesis on the ‘possessor that ran away from home’).

In summary, this section supported the view that the external argument of have is introduced by a Kratzerian transitive Voice. Apart from the thematic roles the subject of exo realizes and the agentive properties it exhibits regardless of the theta-role (see the discussion in 2.1.1), the nominative-accusative case pattern in experiencer exo-sentences and the Active Voice morphology on the copula in all its non-existential uses constitute evidence for the presence of a syntactically and semantically transitive Voice head. The following section shows that the morphological characteristics of existential have-sentences in Greek fortify this conclusion.

2.2.1 Case assignment and agreement

Delveroudi (1992) and Kampanarou (2021a) show that have-based sentences constitute the most common way to introduce new discourse referents, i.e., to achieve the existential function. In this type of sentence, Greek exploits an invariable verb form of have, namely exi, which bears third-person singular marking. The post-verbal nominal is marked with accusative case.

Greek is not unique in exhibiting an existential construction with a non-agreeing have-copula, as this pattern is recurrent within the Romance family. Bentley et al. (2015) and Cruschina (2015) describe that in most cases whereby a Romance language uses have in existentials, the nominal surfaces in the accusative, and the verb appears invariant, non-agreeing with this nominal. French and Spanish offer examples of this distribution in Romance; in (29), the verbal item retains a third-person singular marking even though the sentence’s unique nominal argument is plural (see 2.2.3. for the exception of Brazilian Portuguese and Serbian).

The idiosyncrasy of the accusative case marking on the unique nominal argument of the sentence is also attested in Icelandic, outside the context of existentials. The so-called stray accusative constructions, as named by Haider (2019), are part of a transitivity alternation. In this language, there are verbs that, in their transitive use, mark their objects with accusative (30a). When the same verbs appear in what looks like an intransitive (anticausative) use, the unique argument of the construction still surfaces with accusative creating a ‘stray accusative’ construction (30b).

Schäfer (2008) analyzes the above as the covert version of German es-constructions like the ones in (31).

For Schäfer (2008), es in German is a “defective” expletive, i.e., an expletive with reduced φ-features that cannot be questioned or focused but can be merged in the position of an external argument. The covert version of this expletive, i.e., what Haider (2019) labels a quasi-argument, occupies the same position in Icelandic. For both researchers, the presence of es in German and the quasi-argument in Icelandic is obligatory for case reasons (among other reasons).

Even more interestingly, Wood (2017), who investigates the whole set of Icelandic sentences with ‘stray accusative’ subjects, concludes that accusative subjects in this language are never thematic arguments of morphologically intransitive verbs. This is what he calls the Accusative-Subject Generalization (ASG).¹² To see whether such a hypothesis can be maintained for Greek, we should first consider how Case is assigned in this language.

Initially, Anagnostopoulou and Sevdali (2020) argue that accusative in Greek qualifies as a dependent case following Marantz (1991) and Baker (2015). Specifically, it is assigned under the rule in (32) (Anagnostopoulou and Sevdali 2020: 1025):

In addition to the above, Alexiadou and Anagnostopoulou (2021) contend that nominative qualifies as the unmarked case in the TP domain and is not assigned under agreement with T. However, agreement always goes with nominative when both nominative and accusative are present, as reported at least as early as Philippaki-Warburton (1970) (see also Catsimali 1990; Spyropoulos 1999; Kotzoglou 2013).

Therefore, the fact that the post-verbal nominal in exi-sentences is marked with accusative (28) presupposes that there is a higher constituent that acts as a case competitor. More specifically, as the rule in (32) suggests, the latter must appear within the TP domain in a c-commanding position.

Out of all the overt items, the only potential case competitor would be the locative prepositional phrase since it is the sole constituent that includes a case-marked item, i.e., a determiner phrase (DP) like to trapezi ‘the table’ in (28a) or to patoma ‘the floor’ in (28b). However, an adverb could easily replace the prepositional phrase:

In the presence of the adverb, the post-verbal nominal is still marked with accusative case. Therefore, the locative constituent does not provide the competitor for the dependent accusative. In fact, Baker (2015: 185) extensively argues that PPs do not qualify as competitors for the dependent case algorithm (see also Anagnostopoulou and Sevdali 2020). The author attributes this to the preposition delineating a phase that makes the DP that follows it inaccessible to case assignment processes.

This means there is no overt case competitor for the accusative nominal in exi-sentences. Thus, a covert (phonetically null) item of a (pro)nominal status must be postulated instead. The (pro)nominal status is mandatory so the item can be case-marked.

The data from Icelandic and German above hints that this item is also an expletive. Additional cross-linguistic evidence regarding existential constructions corroborates this assumption. In particular, Francez (2007) claims that expletives are not universally available, i.e., they are not available cross-linguistically. However, when a language has them, they are mandatory in the context of existentials. This means they are obligatorily used to mark the different functions among locational sentences, e.g., between existentials and locatives. In contrast with existentials that introduce new discourse referent (34a), locatives establish the location of a presupposed entity (34b).

Cross linguistically, expletives are not only used in BE-sentences, as in English. The French and Spanish examples in (29) and the Catalan sentence in (35) illustrate that expletives are also exploited in have-sentences.

Greek is a language with no overt expletives but has an expletive pro in its inventory (Iatridou & Embick 1997). In the seminal work of Rizzi (1986), expletive pro is the emptiest null pronoun as it bears no theta-role and no features that need to be identified (see also Haider 2019).

Although Rizzi assumes that such a null pronoun exists in all VS(O) orders, Alexiadou and Anagnostopoulou (1998) convincingly show that this does not hold for VS(O) orders in Null-Subject languages, including Greek (see also Philippaki-Warburton 1987). In this paper, the authors put forward a parametric difference in the EPP-feature checking between English-type languages and Greek-type languages: English must have a DP moved to [Spec, TP] to check this feature, while Greek need not; EPP in the latter case is checked through the V-to-T-movement, and this suffices because the verb bears morphological marking for person and number. Thus, a thematically expletive pro cannot be postulated for unaccusative verbs in Greek.

In the same paper, the authors introduce a second parameter that creates a sub-division within English-type and Greek-type languages. This parameter refers to licensing [Spec, TP] as an A- or an A’-position. That is, languages are further differentiated depending on whether their [Spec, TP] position is a place for derived subjects (see also Biberauer 2009).

An argument used by the authors to support the first parameter is a difference in the Definiteness effects: English unaccusatives exhibit such an effect, as they preclude definite DPs in the post-verbal position, while Greek unaccusatives do not (36).

The case of existentials, though, is different as both languages exhibit the same Definiteness Effect (henceforth DE): definite noun phrases that denote presupposed entities are disqualified from the post-verbal position (37).

This suggests that the DE in existentials is different from the DE in unaccusatives. Bassaganyas-Bars and McNally (2019) show that the DE in existentials is triggered for pragmatic and semantic reasons. Existential sentences are used to introduce new discourse referents. The referents must be pragmatically novel and of a particular semantic type. The denotation of presupposed entities is not allowed. Definite determiners or strong quantifiers, in the sense of Milsark (1974), that lead to this type of denotation are excluded from this context.

Alexiadou and Anagnostopoulou (1998: 512–513) agree on this point. They claim that “We take the non-universality of DR (=DE) effects in unaccusative, unergative, and transitive constructions as evidence that DR effects are syntactically triggered in these contexts. For existential contexts, we maintain that DR effects are semantically triggered as these contexts assert existence. […] In other words, we do not subscribe to proposals (cf. Hoekstra & Mulder 1990, Moro 1997, among others) according to which DR effects in unaccusative, unergative, and transitive constructions should be treated on a par with DR effects in existential there contexts.” Furthermore, for Greek have-constructions where the verb shows third-person singular Agreement, and the nominal exhibits DR effects and accusative case marking, the authors “[…] have reason to assume that they are null expletive constructions.”. Therefore, the accusative case marking in exi-sentences and the fact that the DE in them is not the same as the DE in unaccusatives leads us to postulate an expletive pro in exi-sentences.

Consequently, the structure of the exi-sentence in (38a) should be as in (38b).

Since pro is the highest argument in the TP domain it is assigned the unmarked nominative case. Then, according to Alexiadou and Anagnostopoulou (2021), it controls agreement. As expletive pro yields default agreement, which is always manifested as third-person singular marking (Iatridou & Embick 1997), the verb exo always appears as exi. Thus, the accusative case marking on the post-verbal nominal constitutes evidence of an expletive pro.

At this point, an anonymous reviewer wonders whether the fact that the unique overt argument of an exi-sentence bears accusative case questions the validity of the hypothesis that accusative is a dependent case and, thus, its use as an argument for postulating pro. Instead, they propose that Chomsky’s approach to Case as Agree fares better in this context since the accusative can be captured as the manifestation of Agree with v (or Voice if necessary).

Although exi-sentences could be considered an environment supporting the existence of agreement with v/Voice in Greek, I hold that the analysis advocated in this paper is more cohesive. First, it is crucial that regardless of the case-assignment mechanism, pro would have to be independently postulated to explain why exi surfaces invariably. This means that even if the accusative were assigned via Agree with v/Voice, pro would still have to be posited to regulate the third-person-singular agreement on the verb (cf. Preminger 2014). As a parallel, this is argued to happen with weather verbs (39), which also appear invariant, and when licensing overt arguments, they mark them with accusative case.

As further evidence, witness that any order whereby the nominal surfaces pre-verbally (a restriction that applies to both existentials (40a) and weather verbs (40b)) is ungrammatical, as a second reviewer suggests. This sharply contrasts with Greek unaccusatives (40c), as already shown in Alexiadou & Anagnostopoulou (1998). Such a contrast follows if pro is present only in the former two cases but not in the latter.

Moreover, cross-linguistic evidence points toward the presence of an expletive. Apart from the discussion regarding the cross-linguistic examples in (29–31) and (35), arguing for a phonetically null case competitor is not unknown to the literature, though not uncontroversial. Wood (2017) postulates a covert expletive clitic-like item for Icelandic ‘stray-accusatives’, while Baker & Bobaljik (2017) posit a covert cognate object for ergative unergatives in several languages.

Finally, apart from the fact that the Agreement-with-v/Voice hypothesis would undermine the postulation of a uniform case-assignment mechanism in the language, by assuming Agree, we would have to accept that Agreement precedes Case, i.e., Agree is an operation that is triggered to assign case. However, recent work by Bobaljik (2008), Preminger (2014), and Baker (2015), among others, suggest that Case precedes Agreement, i.e., the latter is dependent on the former and not the other way around. Alexiadou & Anagnostopoulou (2021) provide evidence that this also holds for Greek.

Therefore, I maintain that assuming expletive pro and a dependent-case status for the accusative in exi-sentences not only accounts for the morphological make-up of these sentences but also preserves the generalization of a uniform case-assignment algorithm in the language.¹³

Then, given that pro merges in the structure, it remains to be seen where it is inserted. In virtue of the hypothesis that [Spec, TP] in Greek is not a position for introducing constituents, we should postulate a lower locus of insertion for pro. The following section argues that pro is introduced in [Spec, VoiceP].

2.2.2 Active Voice morphology

Recall that based on the rules proposed by Alexiadou et al. (2015) and their specification by Oikonomou (2011), Act appears either in the absence of Voice or if Voice is projected and an item externally merges to its specifier position. Therefore, the active Voice marking on exi means that either of the above conditions is met. This section suggests that Voice is present and [Spec, VoiceP] is filled by pro.

To argue for it, consider that the absence of Voice would make the structure unaccusative.¹⁴ The accusative case marking on the post-verbal nominal suggests that this copula cannot be unaccusative. Myler (2016) concurs with the view that have is never unaccusative judging by (a) its passivizability, (b) -able affixation availability, (c) contrasts with raising verbs, and (d) assignment of Genitive of Negation in Polish. Piotrowska (2021) adds to his argumentation, which is not presented in detail due to space reasons, evidence from Swedish, Danish, and Norwegian.

It is the case then that Act on exi can only derive if a Voice projection is present and has its specifier filled by an item that externally merges in this position. Given that pro has been shown to exist independently, I propose that this is the element occupying this position.

Unlike non-existential exo-sentences, the expletive pro does not receive a thematic role, i.e., it does not need thematic integration and is merged as a true expletive. This is confirmed by the fact that in exi-sentences, the adverbial material that identifies the semantic import of the external arguments is unacceptable (41). In this case, Voice is semantically intransitive or athematic:

Complementary evidence against the assumptions that pro could be introduced higher in [Spec, TP] or lower in [Spec, vP] strengthens the hypothesis that pro cannot be inserted in any position other than the specifier of Voice.

On the one hand, the TP cannot be the domain in which pro is introduced because, on the basis of phase theories derived from Chomsky (2000), pro must be merged before the spell-out as it determines the Voice morphology of the verb. Since Voice constitutes a phase sent to spell-out (Alexiadou et al. 2015: 111 –112, ft 19, 21), pro must be merged in this position and not late-inserted (Embick 2010; Bobaljik 2012). Moreover, similar subject expletives like German es and the expletive quasi-argument in Icelandic, as well as French il, that are not of a locative nature (like English there, French y, etc.) are also considered to be merged in the same position according to Longa et al. (1998) and Czinglar (2002). Last, recall that Alexiadou and Anagnostopoulou (1998) argue that [Spec, TP] in Greek is an A’-position, i.e., a place for derived, not first-merged subjects (see also Kotzoglou 2013).

On the other hand, we can neither assume that pro is merged in the vP-domain because (a) Alexiadou and Anagnostopoulou (1998) show, based on the distribution of adverbials, that pre-verbal subjects (including expletive pro) are introduced in a position higher than the lexical projection of the verb, i.e., VP or vP, and (b) assuming a de-compositional analysis of the verbal phrase means that v is separate from Voice and unrelated to argument structure. Alexiadou et al. (2015) and Panagiotidis et al. (2017) argue extensively for this, particularly for Greek.

Finally, the merge of an expletive pro could be seen as a strategy for making a construction impersonal, as an alternative to passivization. The language opts for using the expletive null pronoun instead of marking the structure as passive.

Recall that passive in Modern Greek requires NAct. However, exo has a gap in its inflectional paradigm; it does not have a non-active form as *exome. It, therefore, mandates the merge of a thematically inert empty pronoun to take part in an impersonal construction while using its unique active forms. Since the lack of a non-active form naturally follows from the Vocabulary Insertion rules proposed in this article, it further corroborates the present analysis.

It is the case, then, that have in Greek projects Voice even in its use as an existential, and it has pro being externally merged in the specifier position. The nominative subject occupies the same position in non-existential have-constructions, i.e., in possessor, experiencer, and causer exo-sentences.

In the latter case, Voice is syntactically and semantically transitive. In exi-sentences, the Voice projection is syntactically transitive while semantically intransitive. Thus, to answer a question posed by a reviewer, this completes the typology of Voice proposed by Alexiadou et al. (2015), as it constitutes an example of a type considered unattested in Greek. Specifically, the authors discuss causative verbs. They conclude that three out of four Voice types are exploited in causatives: a thematic active Voice in the structure of transitive causatives (42a), a thematic non-active Voice in all Greek passives (42b) and an athematic non-active in Greek marked anticausatives (42c). Athematic active Voice was argued to appear only in Romance se-anticausatives and German sich-anticausatives (43). The current proposal suggests that the latter type of Voice is also attested in Greek exi-existentials.

To sum up, Section 2.2 demonstrated that exi-existentials involve a syntactically transitive Voice head by drawing evidence from the case and agreement pattern as well as the Voice morphology they exhibit. In other words, have was argued to be transitive even in its existential/impersonal use. To further substantiate this, I briefly consider some additional cross-linguistic evidence.

2.2.3 Discussing some possible cross-linguistic counterexamples

At this point, a reviewer wonders whether the distribution in languages like Brazilian Portuguese (hereafter BP) or (occasionally) Serbian, whereby the post-verbal nominal in have-existentials surfaces with nominative case marking, creates a problem for the current analysis that uses case as indirect evidence for have’s transitivity. To answer this briefly, although the analysis I adopt does not immediately predict such cases, I propose that it can account for them without creating doubts about its fundamental claims. In other words, nominative post-verbal nominals in some languages do not imply that accusative case-marked nominals in other languages cannot constitute evidence for transitivity. Even though space reasons prevent us from investigating these data thoroughly, I review some key points to show how an analysis in this framework would work. I use BP as an example.

First, note that BP existentials use the invariant form tem, which is part of the paradigm of the have-copula ter ‘hold, have’ (Creissels 2014) (44). The situation is very similar to Greek.

As argued in Avelar (2009a), the interpretation of tem as an existential appeared when the syncretism in the verbal forms arguably led the language to lose its pro-drop status. In this period, BP started diverging from a typical pro-drop language status towards a status where null referential subjects became rarer and rarer, i.e., a status where referential subjects needed to be overtly realized. That is, although, in a previous stage, the sentence in (45) would be interpreted as a transitive possessive sentence with a dropped subject meaning ‘(S)he has several pants inside the closet’, in this second stage, the only interpretation of the sentence that survived is the existential one. European Portuguese (EP) never entered this new stage and retained the original interpretation.

Avelar (2009a) points out that, in this new stage, BP replaced the original referential argument with the locative phrase itself. In a Freeze (1992)-style analysis, the locative prepositional phrase is initially introduced within a P(reposition)-based predication layer. Then, due to the EPP, it is raised to the subject position, i.e., to [Spec, TP]. Although he does not specify it, his analysis predicts that a phonetically null locative argument is raised to the subject position in the absence of an overt locative phrase. If this is true, there is evidence that the nominative case on the nominal/pronoun is assigned following the principles of the Dependent Case algorithm as proposed by Marantz (1991) and Baker (2015). Since there is no competitor for the nominal within the sentence, it is assigned the unmarked case, i.e., nominative. Crucially, nominative case marking does not mean that there is nothing above or below the nominal but nothing that qualifies as a case competitor.

Moreover, there is supporting evidence that tem is transitive in BP, even in its use as an existential in this second stage. As reported again by Avelar (2009b), the personal pronoun você can appear as the subject of tem without taking away the existential interpretation of the sentence:

For reasons explained in the same paper, você must first be introduced in a thematic position and then raised to the TP. For Avelar, você is introduced in [Spec, PP], leaving open the question of how the overt locative, e.g., em Nova York or em times europeus in (46), is introduced in the structure. Under current assumptions, você could be merged in [Spec, VoiceP]. This would be compatible with its requirement to be first-merged in a thematic position, and would allow reserving the same predicative structure postulated in the absence of você.¹⁵

Therefore, as with Greek, independent evidence leads to the conclusion that tem-sentences in BP are transitive, i.e., they include a Voice head with a projected specifier. Whether the nominative case in the post-verbal nominal can be used as an argument for this needs to consider how case assignment is determined in this language. What is crucial is that the distribution does not contradict the analysis pursued in this paper and the claim that the accusative case is an argument for the transitivity of exi-sentences in Greek.

Consequently, as Section 2 has argued that have includes a transitive Voice head in all its guises, it supports the idea that it is the transitive copula. Specifically, the fact that all its stative uses are unified due to the presence of Voice allows us to abandon a multiple-have hypothesis and assume that there is only one have in the language. This partially confirms that BE and have are related via transitivity. To fully argue for it, it must be shown that BE is the non-transitive variant of have. The following section tackles this issue.