1.1 CD-ed objects: Adjuncts or arguments

This section presents the current state of affairs about analysis of Greek CD in the literature and shows that the empirical data discussed in this literature are compatible with different assumptions about the kind of position i.e. argument or adjunct, that CD-ed objects can occupy.

To start with, previous literature on CD has concluded that the doubled objects of Greek are arguments considering syntactic positions where adjuncts are excluded and testing whether CD-ed objects can occur in them. The syntactic positions that were tested are the subject positions of e.g. control or ECM, and it was observed that CD-ed objects can occur in them, as shown in (1). Given this, it was concluded that Greek CD-ed objects can be arguments (cf. Collins 1990; Schneider-Zioga 1994; Sportiche 1996; Anagnostopoulou 1999a).

However, most recently, Philippaki-Warburton et al. (2004) argue that constructions like the ones in (1) are compatible with an analysis of CD in which doubled objects are adjuncts with the clitic first merging in the argument position. Under this view, (1a) “must be analysed as an object control construction, where the DP [ti Maria] is the double to the clitic ‘tin’ and an adjunct in VP followed by the complement clause in subjunctive.” Philippaki-Warburton et al. also claim that examples like (1b) can be analyzed along similar lines as object control. While this is not the view that will be adopted in this paper, the data in (1) are compatible with the alternative presented in Philippaki-Warburton et al. (2004) challenging the conclusions of Collins (1990); Schneider-Zioga (1994); Sportiche (1996) and Anagnostopoulou (1999a).

Anagnostopoulou (2003) argues that Greek CD-ed objects can occur in argument positions on the basis of a distributional difference in doubling of Greek and Romance languages. Concretely, Anagnostopoulou shows that CD-ed objects can surface before postverbal focused subjects, as in (2). Anagnostopoulou also notes that the subject in this case can bear main sentence stress like in plain VOS orders.

On the other hand, she claims that the surface order in (2b) is never attested in Romance languages e.g. Peninsular Spanish and Catalan, which independently allow an object to be preceded by a doubling clitic. Anagnostopoulou (2003) argues that the CD-ed object in (2b) occupies a syntactic position, which is only available for arguments, and that this is also the position that undoubled objects occupy when they surface in the VOS order. In particular, the assumption that Anagnostopoulou (2003) adopts from previous literature for this order is that it is derived via leftward A-movement of the object from the argument position into a position higher than the post-verbal subject. She concludes that Peninsular Spanish and Catalan do not allow the surface order in (2) because these languages have only clitic right dislocation. Clitic right dislocated objects have distinct distributional properties because they do not occupy argument positions like CD-ed objects. Importantly, in a recent intonation study, Revithiadou & Spyropoulos (2009) show that in the VOS order, the object always forms a distinct prosodic unit from the verb if it is doubled by a preceding clitic. On the other hand, if the object is not doubled by a clitic it can form a prosodic unit with the verb. This finding suggests that CL_i-VO_iS order cannot be treated on a par with the plain VOS one (pace Anagnostopoulou 2003). Furthermore, Krapova & Cinque (2008: 29b) present data showing that the order in (2b) with a focused subject, exactly like in Greek, is allowed in Italian, as illustrated in (3).

As Krapova & Cinque (2008) point out, this finding poses a problem for the claim, e.g. in Anagnostopoulou (2003), that doubling of languages like Italian or other Romance languages e.g. Catalan or Peninsular Spanish, should be treated as a distinct syntactic phenomenon from CD.¹² Now, since the CD vs. right dislocation distinction of Anagnostopoulou (2003) is weakened in light of the Italian facts and the intonation study of Revithiadou & Spyropoulos (2009), (2b) cannot show conclusively whether the CD-ed object is an argument, therefore, the adjunct status of CD-ed objects (Philippaki-Warburton et al. 2004) is a possible alternative, which cannot be immediately ruled out. This view holds that the doubled object in (2b) is a left adjunct above Spec vP where the post-verbal subject is hosted. Again, the view of Philippaki-Warburton et al. (2004) will not be adopted but it cannot be rejected on the basis of (2).

Finally, let us consider the data in (4a) and (4b). These data illustrate that possessor extraction is possible out of undoubled definites (cf. Horrocks & Stavrou 1987). On the other hand, Philippaki-Warburton et al. (2004) and Revithiadou & Spyropoulos (2009) point out that CD-ed objects block possessor extraction for focus or wh-movement, (5):

Based on this observation, Philippaki-Warburton et al. (2004) and Revithiadou & Spyropoulos (2009) claim that the behavior of CD-ed objects with respect to extraction is not expected in an analysis in which the CD-ed objects are arguments, like the corresponding undoubled ones in (4). Note, however, that (4b) does not form real minimal pairs with (5b) and (5c). CD-ed objects are necessarily marked as –Focus whereas undoubled objects are not (cf. Warburton 1975; Kallulli 2000). More specifically, Warburton (1975: 568–569) argues that CD-ed objects cannot be stressed or focal. The fact that Greek CD-ed objects are marked as –Focus is illustrated in (6) where it is shown that the CD-ed DP cannot be combined with a focus marker like akoma-‘even’ (cf. Kallulli 2000):

Now, since the CD-ed DP has to be –Focus, I argue that (5b) is ruled out because the possessor argument hosted within the CD-ed DP cannot be marked as +Focus and undergo focus movement. Similarly, (5c) is ruled out because wh-items are obligatorily +Focus (cf. Horvath 1986; Brody 1990 i.a.). In other words, the possessor cannot undergo focus or wh-movement in which case it would have to be +Focus possibly because the whole doubled DP, that is, to aftokinito tu Jorghu in (5a), constitutes a –Focus domain. Given this, the syntactic position of CD-ed objects cannot be revealed by the data in (5b) and (5c) because the extraction restrictions in these examples are expected to arise regardless of the syntactic position of the CD-ed object i.e. adjunct or argument.

To sum up, the discussion so far must have shown that the question regarding the position in which doubled objects enter the derivation in Greek has not yet been sorted out. The following sections aim to bring further clarity in this discussion. This is achieved by using new evidence which is derived from a systematic investigation of the reconstruction properties of Greek doubled objects.

3.1 CD-ed objects are arguments

In this section I show that CD-ed objects enter the derivation as arguments. To start with, consider (13a) which shows that a plain undoubled direct object hosting a pronoun can be interpreted in the argument position where the pronoun is bound by an indirect object QP scoping below negation, as shown in the translation. In (13b), I show that binding into the CD-ed direct object is possible under identical conditions i.e. with the quantifier scoping below negation.

This new finding shows that CD-ed objects can be interpreted in the argument position exactly like undoubled arguments and be bound by a low scoping quantifier. In Section 3.4, I show that CD-ed objects undergo obligatorily an A-movement step into the middle field. Given this, I argue that the bound interpretation in (13b) becomes possible under total reconstruction of the CD-ed direct object from the middle field into a position below the indirect object QP, which is the argument position occupied by the undoubled object in (13a).⁹¹⁰

Importantly, previous literature has concluded that Greek CD-ed objects cannot undergo reconstruction. In particular, Alexiadou & Anagnostopoulou (1997) point out that the direct object can bind into a PP indirect object as shown in (14a). However, since PPs cannot be CD-ed in Greek, Alexiadou & Anagnostopoulou construct examples with CD of dative indirect objects as in (14b), where it is shown that pronominal binding is not possible in this case.

Based on the contrast between (14a) and (14b), Alexiadou & Anagnostopoulou conclude that the CD-ed indirect object in (14b) is interpreted in a position higher than the direct object due to (some kind of) movement that the CD-ed object undergoes to this higher position. In addition, they assume that pronominal binding is not possible into the CD-ed indirect object because it cannot undergo total reconstruction below the direct object. Note that the detail that Alexiadou & Anagnostopoulou (1997) do not discuss is that the two sentences in (14) do not form a minimal pair. In (14a), the indirect object is a PP whereas in (14b), it is a dative DP. Moreover, pronominal binding is not expected to be possible in (14b) because bare direct object QPs cannot bind into undoubled dative indirect objects in the first place (cf. Anagnostopoulou 2003). This latter fact is illustrated below:

Given the above, I argue that the data presented in Alexiadou & Anagnostopoulou (1997) do not pose any challenge to the conclusion that CD-ed direct objects can undergo reconstruction.

3.2 CD-ed XPs do not move higher than the subject

This section examines whether CD-ed objects can be interpreted higher than the subject. Thus, I compare only for the purposes of this section the interpretive properties of CD-ed objects to these of C(litic) L(eft) D(islocat)ed ones.¹¹ CLLD-ed objects are used as a benchmark case, as they surface in the left periphery above the subject, and as shown in (16), can be interpreted in the left periphery.

In (16), the CLLD-ed object is combined with a relative clause that contains a proper name i.e. pu o Janis evghale sto Parisi, and the proper name is shown to be able to corefer with the silent subject of ksechase. This shows that the CLLD-ed object can be interpreted in the left periphery, allowing the relative clause to undergo Late Merge with it, which in turn is responsible for bleeding Condition C, as the proper name ends up being interpreted outside the c-command domain of the silent subject pronoun in (16). If CD-ed objects could occur in the left periphery either in the specifier of a TopicP or as adjuncts to a left peripheral projection, like CLLD-ed objects, they should be able to bleed Condition C when they combine with a relative clause. If they do not bleed Condition C, this will in turn suggest that CD-ed objects do not move higher than the subject or that they cannot adjoin to any projection higher than the subject. Note that the latter would be totally unexpected if CD-ed were DP adjuncts as they should in principle be allowed to freely adjoin XPs at different levels. Interestingly, I show that when a CD-ed object combines with a relative clause containing a proper name, coreference between the proper name and the silent subject is blocked, (17).¹²

Similar facts have first been reported for Italian in Cecchetto (1999); Cardinaletti (2002) and for Dutch in Koopman (2007: 171). I take these facts to provide evidence that CD-ed objects are not adjuncts in Greek, and that if they undergo any movement step, this movement step must be targeting a position below the position where silent subject pronouns are hosted. In Rizzi (1986), this position has to be Spec TP where pro undergoes movement in order to be identified. More recently, the idea that pro is in Spec TP had been defended in Rizzi (1997); Cardinaletti (2004); Holmberg (2005) and Roberts (2010a). Following this previous literature, I assume that subject pro is indeed in Spec TP. Moreover, I assume that pro can only be interpreted in Spec TP. In other words, pro cannot undergo reconstruction to Spec vP where it originates. I suggest that this is so as a result of the fact that A-moved pronouns do not undergo reconstruction. In Sportiche (2017), the fact that A-moved pronouns do not reconstruct was shown on the basis of the following pair from English (see also Angelopoulos & Sportiche 2018):

(18a) is ruled out under a coreference reading between he and John. However, if the raised pronoun could undergo reconstruction, it should be binding theoretically equivalent to (18b). The fact that it is not suggests that A-moved pronouns do not reconstruct.¹³

3.3 CD-ed objects do not move higher than dative clitics

In this section I examine the interaction with respect to Condition C between CD-ed direct objects and dative clitics in order to determine the relative height of movement of the first with respect to the latter. In previous literature, Condition C has been argued to rule out coreference between a dative clitic and a definite description hosted within an undoubled direct object, (19a). This literature also reports that if the direct object is CD-ed, Condition C is obviated, as in (19b).

Based on this contrast, Alexiadou & Anagnostopoulou conclude that Condition C is obviated in (19b) because the doubled object undergoes (some kind) of A-movement across the dative clitic.¹⁴ With that said, I would like to argue that there are confounds in the minimal pair in (19), and as a result of this, that it is not clear whether CD-ed objects move higher than the dative clitic. Let me clarify, first, that genitive and dative are morphologically syncretic in Greek. Thus, (19b) in the absence of the two clitics is ambiguous between (20a) and (20b).¹⁵

In (20a), tis Marias functions as possessor forming one constituent with the direct object to vivlio. Simiomeno is a secondary predicate taking to vivlio tis Marias as subject. In (20b), tis Marias is a single constituent functioning as a dative indirect object. In this case, simiomeno is extraposed into a position after its subject, that is, to vivlio.¹⁶ Now, if two clitics are present, they can enter dependencies with different arguments. For instance, the accusative clitic to can associate with to vivlio tis Marias, which forms a single constituent, as in (21a). Or, to can associate with the direct object to vivlio in which case tis Marias functions as a dative argument associating with the dative clitic tis, as in (21b).

Importantly, Condition C between tis Marias and tis is not expected to arise in (21b) because the two associate via CD. Anagnostopoulou (1994: 127–128) who also observes this structural ambiguity notes that (21b) needs to be suppressed in order to diagnose clearly if Condition C is bled in (21a) or not. Yet, Alexiadou & Anagnostopoulou (1997) treat (19b) as one unambiguous string and conclude without further discussion that “the condition C effect disappears once the direct object to vivlio tis Marias is doubled.”¹⁷ However, given the confound in these examples, this conclusion is not supported by these data. In fact, I show that Condition C is not expected to be bled in CD of direct objects, and, therefore, that the conclusion that Alexiadou & Anagnostopoulou (1997) report is a result of the confound discussed in Anagnostopoulou (1994). Concretely, let us assume that dative clitics are interpreted in Spec CL_DATP and that this position is lower than the subject position, as in Sportiche (1996) (see discussion in Section 3.4.1 for comparison with different analyses of clitic constructions and for more details of Sportiche’s analysis).¹⁸

Let us also take (23a) as a benchmark case showing that the dative clitic c-commands the undoubled direct object.

In (23a), coreference between the clitic and the proper name in pu o Janis evghale sto chionodhromiko is blocked due to Condition C. If there is a copy of the CD-ed direct object higher than Spec CL_DATP in (22),¹⁹ Late Merge of the relative clause with this higher copy should bleed the Condition C effect seen in (23a), as the proper name would not be in the c-command domain of the dative clitic. (23b) shows that coreference between the dative clitic and the proper name in the relative clause is blocked after CD suggesting that there is no higher copy of tis fotoghrafies higher than Spec CL_DATP, which would allow Late Merge of the relative clause. I conclude based on this that CD-ed direct objects do not move higher than Spec CL_DATP in (22). Having clarified this, the next section shows that although CD-ed objects do not move higher than the position in which dative clitics are interpreted, they still undergo movement into a lower vP peripheral specifier.

3.4 CD-ed objects obviate WCO

It has been known for a long time in several languages such as Lebanese Arabic (cf. Aoun 1981) or Greek (cf. Agouraki 1992; Alexiadou & Anagnostopoulou 1997; 2000; Anagnostopoulou 2003) that CD can obviate WCO effects. This effect is illustrated in the following examples, where a pronoun embedded within a subject cannot be bound by an undoubled quantifier in the direct object position, (24a), while if the quantifier is CD-ed, the bound interpretation becomes possible, (24b).

In what follows, I present an account for these facts assuming that the CD-ed object undergoes XP/X_max movement to a vP peripheral specifier. I argue that this movement step is an A-movement step from the argument position. Subsequently, I discuss that competing approaches that do not take CD-ed object to undergo XP/X_max movement from the argument position fall short in accounting for the absence of WCO in CD and the correlation with Hindi or Japanese A-scrambling.

4.1 CD as Feature movement

In previous literature, it was argued that CD can obviate WCO assuming that CD-ed DPs form an A-chain with the clitic. The clitic which is the head of the chain is assumed to move to T via feature movement, and it is argued to bind from this position into Spec vP where the subject can undergo reconstruction, as shown in the LF representation below (cf. Anagnostopoulou 1999b; 2003; Marchis & Alexiadou 2013 i.a.).

Notably, the feature movement analyses never explicitly formulated Binding Theory in terms of features, therefore, it is unclear on what grounds the assumption that features can bind can be evaluated. At any rate, I argue that the real challenge the feature movement account faces is that it relies on the rather doubtful assumption that features can act as binders in the first place. Feature movement was first motivated in Chomsky (1995) to account for the agreement facts in the expletive constructions of English and Italian shown in (27b) and (27a).²²

Here, Chomsky (1995) argues that features of the associate undergo covert movement into T where they check the ø-features of the verb. As Chomsky (1995) supports, these features can bind and license PRO in cases like (28a). Chomsky further observes that, by contrast to PRO, reflexive binding cannot be licensed in expletive constructions, as shown in (28b).

Cardinaletti (1997: 525) points out other instances from previous literature where binding and control are subject to distinct licensing requirements (cf. Rizzi 1986) and concludes that “both agreement and control are triggered by simple feature movement, whereas binding requires that the antecedent be the whole category (not just some features).” Similarly, Lasnik (1999: 183) shows that pronominal binding is not allowed in expletive constructions, (29), and concludes like Cardinaletti (2008) that feature movement cannot license binding.

Based on the conclusions of Cardinaletti (1997) and Lasnik (1999), I reject the feature movement analysis of CD, as there is no independent motivation for feature movement playing a role in binding.

In the next section, I turn to the grammatical distribution of clitics. Building on Angelopoulos & Sportiche (2018), I show that there is no independent motivation for the doubling clitic to have referential import that matters for binding. This suggests even more strongly that the clitic should not be able to bind into anything, like it was shown to be the case with the raised features in English expletive constructions. Based on this finding, the next section also argues that the long head movement analyses of CD must be ruled out because as with the feature movement analysis, they cannot account for the fact that CD-ed objects exhibit interpretive properties as these of objects e.g. of Hindi, that have undergone movement into the middle field.

4.2 CD as Long Head movement

Having pointed out preliminary observations about the issues with respect to binding that the feature movement accounts face, I discuss next the most closely related analyses presented more recently in Rezac (2008); Roberts (2010b) and Preminger (2019). These works assume that CD-ed objects are arguments and that they stay in situ, as in the feature movement analysis. Furthermore, they propose that instead of feature movement, the D-head of the doubled object undergoes long head movement into T or other lower heads. These approaches do not explore WCO in CD, however, since the doubled XP stays in situ, the fact that WCO is obviated in CD can make sense only if the doubling clitic, which is the D head of the CD-ed object has referential properties and thus, can be interpreted after long head movement in a higher functional domain. The issue that arises with this view is that Greek clitics do not have interpretive import. This fact has been shown in previous work on the basis of cases in which a reflexive is CD-ed, as in (30).

In (30), the accusative clitic doubles an anaphor bound by the subject (cf. Iatridou 1986). If clitics had interpretive import and could enter referential/binding dependencies, they should have the index of the anaphor which is the element that the clitic associates with in (30). Nevertheless, if this was true, the clitic should give rise to a Principle B violation because it would be locally bound by the subject, as illustrated in (31).

Collins (1990) and Angelopoulos & Sportiche (2018) argue that accusative clitics do not give rise to Principle B violations in (30) because they are expletive elements, like expletive determiners, lacking referential import. In Angelopoulos & Sportiche (2018), this claim also finds further support in the fact that clitics in Greek and French can double elements that lack reference such as embedded clauses or predicates e.g. adjectives, in French.²³ Under their analysis, which I adopt, only the element that the clitic associates with has referential properties. In (30), this element is the anaphor, and in plain cliticization, this element is a silent pronoun:

Furthermore, in Angelopoulos & Sportiche (2018), the silent pronoun does not remain in situ. Instead, it has to move to the specifier of the phrase headed by the clitic for indentification purposes, exactly like silent pro has been argued to be identified by T in null subject languages (cf. Rizzi 1986).

Now, since the clitic is an expletive determiner and lacks reference, it should not be able to bind or change anything with respect to WCO, therefore, the analysis of Rezac (2008); Roberts (2010b), and Preminger (2019) predicts that the CD-ed and plain objects should be interpreted in situ and that they should not differ in terms of interpretive properties. However, this prediction is wrong since, as we saw, CD-ed objects can obviate WCO effects with the subject.

4.3 CD-ed objects as adjuncts

Under the account in Philippaki-Warburton et al. (2004), the doubled DP is base generated as adjunct e.g. to vP in (33), and the clitic first merges as an argument of the verb and undergoes X_min/X_max movement to some higher projection. Here, the doubled DP should not be able bind into anything under the assumption possibly that “adjunct” positions do not have the properties of A-positions (see Angelopoulos et al. 2018 for data showing that binding from adjuncts is not possible in Greek). Moreover, the clitic cannot bind or be interpreted, because accusative clitics, as we discussed, are expletive determiners lacking referential import. The issues that arise here is that, like in previous analyses, the fact that WCO is obviated in CD cannot be accounted for.

5.1 CD and Information Structure

CD-ed definites have been argued in extensive previous literature of Greek to “[…] resemble pronominals in that they cannot be understood as novel, a fact which can be taken to suggest that they obey the Prominence Condition” (cf. Anagnostopoulou 1999a: 771). The Prominence Condition was proposed in Heim (1982) and was argued to apply to pronouns like him in (34).²⁴

In (34), him being subject to the Prominence Condition can only refer to Schubert, who has been previously asserted. On the other hand, the definite in (35) is different and can also have the author of the book about Schubert as referent despite the fact that there is no previous mention of an author.

Anagnostopoulou (1994; 1999a; 2007) argues using evidence from examples like (36) that Heim’s Prominence Condition also applies to CD-ed objects.

It is argued that in the presence of the doubling clitic, the definite ton singrafea–‘the author’ can only be interpreted as familiar and like him in (34) it can only have Arthur Miller as its referent, which has been previously asserted. On the other hand, if the clitic is absent, Anagnostopoulou claims that the definite is free to pick the author about the book for Arthur Miller or Arthur Miller himself as a possible referent. Moreover, Anagnostopoulou also argues that CD-ed objects are obligatorily interpreted as familiar because the doubling clitics are contentful. This claim is explored in more depth in Marchis & Alexiadou (2013) who building on Anagnostopoulou’s claims propose that Greek accusative and dative clitics are contentful determiners encoding familiarity in the sense that they are subject to Heim’s Prominence Condition. I argue that the judgments this previous literature reports do not hold uniformly with objects of different predicates. For instance, I show that the direct object of skotoso–‘kill’ in (37) exhibits different behavior.²⁵²⁶

In (37), the CD-ed definite is free to pick the author of the book as referent despite the fact that there is no previous mention of the author.²⁷ This suggests that depending on the lexical semantics of the verb, there can be cases in which CD-ed objects do not have to be subject to the Prominence Condition. Furthermore, note that there is a very sharp contrast if instead of a CD-ed DP in (37), there is a bare clitic:

Here, it is quite clear that the sentence is strongly ruled out if the pronoun in (38) has the non-familiar reading under which it corefers with the author of the book. This contrast between (37) and (38) is very sharp and this fact is quite unexpected if Heim’s Prominence Condition applies invariantly both to CD-ed objects and clitics. With this conclusion in mind, I turn next to more data challenging the idea that CD-ed XPs are always subject to the Prominence Condition. For instance, let us consider the well-formed examples in (39) and (40) where it is shown that generic definites can undergo CD.²⁸

To hrima–‘money’ or tin alghevra–‘algebra’ are generic definites, still, they can be CD-ed although they do not need to have been asserted previously.²⁹ Further support to the assumption that CD-ed objects do not always obey the Prominence Condition comes from indefinites as in (41), that can undergo CD although they do not require any previous mention:

In (41a) and (41b), ena sokolataki–‘small chocolate’ and enos sovaru fititi–‘a serious student’ can be CD-ed and they do not need to have been asserted previously. Sokolotaki in (41a) can refer to any any kind of small chocolate e.g. white or black. Similarly, sovaru fititi in (41b) refers to any serious student and not to a particular one that has been previously mentioned. Notably, the fact that indefinites can be doubled by a clitic has been observed since very early in the literature (see Anagnostopoulou 1994 and references therein). Kallulli (2000) also shows that CD-ed indefinites have the same distributional properties like CD-ed definites e.g. in ECM constructions, small clauses and object control. Moreover, Kallulli (2000), who observes that CD-ed objects are not uniformly subject to the Prominence Condition, like I do here, proposes an alternative analysis according to which CD-ed objects in Greek are uniformly marked as –Focus. In addition, she proposes that –Focus is a syntactic feature that CD-ed objects carry and license via movement into the middle field. Here, I adopt from her analysis the main idea that movement in CD has to take place in order to license a syntactic feature that relates to their –Focus interpretation. Interestingly, scrambled objects e.g. in the Germanic languages or Hindi, have been argued in previous literature to associate with certain interpretative properties, which, as with CD-ed objects, are licensed via movement to a middle field syntactic position (cf. Broekhuis & Corver 2016 i.a.). Moreover, scrambled objects have also been shown to obviate WCO (cf. Mahajan 1990; Miyagawa 2009 i.a.), which like I argued for CD, is due to A-movement of the scambled object into the middle field.³⁰ I argue that these similarities between A-scrambling and CD of objects in Greek are not accidental. Instead, they suggest that movement of CD-ed XPs is the correlate of the XP/X_max movement step of A-scrambling, as has been proposed in a number of previous works (Sportiche 1996 i.a.).

5.2 Clitics are not +familiar or –focus

As discussed in the previous section, previous analyses have proposed that clitics have semantic import. For instance, clitics have been assumed to be lexically specified as +familiar (cf. Marchis & Alexiadou 2013) or as –Focus in which case they can license the –Focus property of CD-ed objects via Spec head (cf. Kallulli 2000). In this section, I argue that if we take into consideration more predicates, it becomes quite clear that clitics are not intrinsically specified with any of these two properties. The predicates I focus on here are experiencer predicates of Class II or III in the typology of Belletti & Rizzi (1988). These predicates allow CD of the experiencer, as shown in (42):

(42) shows that a dative experiencer, that is, tu Jani, can be doubled by a clitic, which can surface before or after the experiencer. In both cases, it is crucial that despite the presence of the doubling clitic, the experiencer can be +Focus, as shown by the fact that it can combine with akoma ke–‘even’, which is a focus particle.³¹³² Moreover, I show in (43) and (44) that doubled experiencers, dative or accusative, can be wh-items.³³

The clitics used to double experiencers e.g. tu and ton in (42), (43) and (44), are identical in form and distribution to the clitics used in doubling of arguments with distinct theta roles e.g. themes, therefore, it cannot be argued that experiencers associate with special clitics. Moreover, doubled experiencers combining with similar focus particles or being realized as wh-items are not only available in Greek. For instance, Krapova & Cinque (2008: 268–270) show that doubled experiencers in Bulgarian can be +Focus or wh-items and hence, propose that doubling of experiencers constitutes a distinct syntactic phenomenon. Given the above, I conclude that Greek clitics are not intrinsically specified as +familiar or –Focus (pace Kallulli 2000; Marchis & Alexiadou 2013) and as a result of this, that the interpretive properties of CD-ed objects or the position in which they license their interpretive properties should be dissociated from the clitics and the clitic position. Building on this conclusion, the next section provides an account for the optional presence of clitics with –Focus DPs.

6.1 Optional CD

In previous literature it has been argued that direct object CD is an optional phenomenon (cf. Anagnostopoulou 2007 i.a.). Building on this claim, this section aims to show that this optionality is not unrestricted. In particular, building on findings of recent literature, I tentatively argue that the presence or the wholesale absence of clitics in CD is constrained and that the constraints that apply are syntactic. Let me start the discussion by taking into consideration sentences like (36) from Anagnostopoulou (1994). Recall that in her analysis clitics mark doubled DPs as familiar. Now, since familiar definite DPs do not have to be CD-ed, as we saw in (36), Anagnostopoulou (2007) argues that CD of direct objects is an optional phenomenon.³⁴ Similarly, citing data from Agouraki (1993); Kallulli (2000: 20) explicitly argues that “it is not the case that for the direct object DP to be interpreted as [–Focus], it has to be clitic doubled (e.g. Greek).” Let us consider some of the data Kallulli considers:

In (45), the subject is a wh-item, that is +Focus and the rest of the sentence is marked as –Focus. Kallulli (2000) argues that despite the fact that the direct object is –Focus in this example, it does not have to be CD-ed and concludes that CD is less strict—and hence optional I add—in Greek. Interestingly, I show next that a crucial difference between the CD-ed –Focus DPs and the undoubled ones is that only the first can be shown to move to the middle field. For instance, let us consider the following minimal pair.

In both examples, the direct object is a definite DP, ton zografo–‘the painter’ referring to a previously introduced individual, that is, ton Yanni Tsarouchi. Moreover, the direct object is –Focus, as it is part of a yes/no question, where the verb only is focused (see Kallulli 2000 for similar examples). (46a) shows that WCO with the subject can be obviated if the object, ton zografo, is CD-ed. On the other hand, the direct object cannot bind the pronoun in the subject, thus, (46b) is ruled out under the indicated bound reading. In Section 4.2, I argued in light of the WCO facts that doubled DPs undergo movement above the vP from where they can bind into a reconstructed subject. Nonetheless, since undoubled –Focus DPs do not behave similarly with respect to WCO, it makes sense to conclude that, if they have to move to a certain syntactic position where they can license their interpretive properties, they do not move as high as the doubled ones. Concretely, I propose that there are two distinct syntactic positions differing in syntactic height where CD-ed DPs can license the –Focus property. The lower one is indicated as YP in (47b) and is situated between VP and vP. The higher one is shown as XP in (47a) and it is higher than the vP, like I argued before.

Let us also assume that clitics are base generated in the middle field between XP and TP in (47a) like in Sportiche (1996), and that they can be present if and only if they can Agree with an object DP.³⁵ In (47a), the direct object undergoes movement to Spec XP higher than the vP phase. Moreover, after movement, the direct object is in the same phase with the clitic. Being in the same phase, the clitic and the direct object can stand in an Agree relation. On the other hand, the direct object does not escape the first phase i.e. the vP in (47b), therefore, it is not in the same phase with the clitic. If this is correct, the prediction is that the clitic should be present in the sentence corresponding to the syntactic structure in (47a), though not in (47b). Indeed, (46a) where the clitic is present corresponds to (47a) whereas (46b) where the clitic is absent corresponds to (47b). Based on these facts, I (tentatively) conclude that indeed clitics are present if they are local enough —possibly in the same phase— with their associate. Increasing the distance between two as in (47b) results in the wholesale absence of the clitic.

Interestingly, Preminger (2009) focuses on CD in Basque and observes that the presence of doubling clitics depends on locality. In fact, he argues that CD always conforms to a locality condition. In Basque, he proposes that this condition (roughly) is the clausemate condition. Preminger argues that dative morphemes in Basque enter a CD dependency with their associate. Under this view, dative clitics are present on the Basque auxiliary only if they are in the same clause with their associate. On the other hand, if this local relation is broken, Preminger claims that this results in the wholesale absence of the dative morpheme. In Greek, the locality condition between the clitic and its associate needs to be defined in more local terms as it is not sufficient for clitics to be in the same clause with their associate. With this in mind, let us now turn to issues and open questions that future research needs to address. For instance, it is important to understand the locality conditions imposed by Greek CD in more precise terms. Or, it is crucial whether clitics are optional or whether they are syntactically present but silent.³⁶ Lastly, it is an open question at this point why only doubled experiencers but e.g. not doubled themes, can be wh-items or combine with focus particles.

6.2 Residual Issues

In this section I discuss a residual issue that has to do with extraction out of CD-ed objects. As discussed in Section 1.1, possessor extraction has been argued in two previous works to be ungrammatical (cf. Philippaki-Warburton et al. 2004; Revithiadou & Spyropoulos 2009). On the other hand, an anonymous reviewer points out that Anagnostopoulou (2007: 19c–d, 23c) presents grammatical sentences showing that CD-ed objects permit possessor extraction:³⁷

Indeed, (48c) is grammatical, however, one needs to control for the fact that tis Marias in (48a) or tinos in (48c) can also function as dative arguments as I discussed previously. In the latter function, (48a) is interpreted as I put the dress on Mary and (48b)–(48c) as on whom did you put the dress?. In order to control for the ambiguity, I presented (48b) and (48c) to fifteen native speakers in different contexts that were compatible with tinos as possessor or indirect object. All speakers had strong judgments that tinos in (48c) cannot be a possessor while they all found it acceptable if tinos is interpreted as on whom.³⁸ On the other hand, they reported that both readings are available in (48b). These results confirm the intuition presented in Philippaki-Warburton et al. (2004) and Revithiadou & Spyropoulos (2009) that CD-ed object block possessor extraction.