2.1.1 The phenomenon

The focus of our interest here is clitic doubling as exemplified in (5a), where the doubled DP (here ton fititi) occupies an postverbal position. CLD is not to be conflated with clitic left dislocation (CLLD) (5b), where the doubled DP appears in a higher, left-peripheral position (see Angelopoulos & Sportiche 2021 for recent discussion). Greek is a clitic doubling language par excellence, and has figured prominently in debates on the phenomenon ever since the influential work of Anagnostopoulou (2003).

We will assume going forward that the doubled DP in CLD occupies an argument position and is thus not base-generated in a right-dislocated non-argument position. This conclusion is well-supported in Greek and beyond; among other crucial pieces of evidence, one finds the observation that Greek ECM subjects can be freely doubled, as in our (12) further below. See Angelopoulos (2019: 3) for further recent discussion.⁴

Space considerations prevent us from going into more detail on the conditions governing clitic doubling in Greek. We thus limit ourselves to noting the following points. Firstly, Greek CLD usually targets DPs high on the referentiality/topicality scale (Anagnostopoulou 2017a, see also section 2.2.4 below on information structure and WCO), but the precise conditions remain imperfectly understood, as quantified DPs and even indefinites can be doubled under certain conditions (see Angelopoulos 2019). Secondly, holding givenness/topicality constant, CLD is optional outside of intervention configurations (see sections 2.2.4., 2.2.5. below). Finally, CLD in Greek can target both direct and indirect objects, but not subjects (like other pro-drop languages, Greek lacks subject clitics); IO and DO doubling generally behave alike, with the exception that doubling of some IO, but not DO, third-plural DPs is difficult under specific circumstances (see Kouneli & Kushnir to appear, whose account is discussed briefly in Paparounas & Salzmann to appear). For more details on the distribution of CLD, we refer the reader to authoritative publications on CLD more generally (Anagnostopoulou 2017a) and on CLD in Greek specifically (Anagnostopoulou 1994; 2003; Angelopoulos 2019; Angelopoulos & Sportiche 2021).

2.1.2 Theories of clitic doubling

A major goal of theories of clitic doubling is to account for the emergence of the clitic: what mechanism ensures that the features of a DP occupying an argument position are indexed by a higher element thematically linked to that DP, and bearing the same case? Different answers to this question define different theories of CLD (see Anagnostopoulou 2017a for a recent overview).

In this section, we briefly summarize the major camps of CLD theories, before arguing that FC CLD points to just one of them, namely, the family of pure Agree-based approaches, as the correct one for Greek. Throughout, we illustrate the different analyses with trees for the structure of the simple clitic doubling example in (5a) above; to keep the exposition focussed on doubling, we do not show V-to-T movement and externalization of the subject DP in our trees.

In Big DP approaches, the clitic and the doubled DP underlyingly form a constituent: under the assumption that anaphoric dependencies are encoded derivationally, the two are coindexed precisely because they originate in the same big DP constituent. While approaches vary in terms of the precise internal structure they assign to the big DP (e.g., Uriagereka 1995; Nevins 2011; Arregi & Nevins 2012), they uniformly postulate that the clitic strands the DP in the course of the derivation by moving to a verbal projection, as schematized in (6), which is based on the structure of Uriagereka (1995: 81):

In a different class of movement-based approaches, clitics are treated as secondary realizations of the D head introducing the doubled DP. In a first subtype of such an approach, the doubled DP undergoes A-movement to a peripheral position within the vP; the D head subsequently amalgamates downward with the verbal head whose specifier hosts the doubled DP, via rebracketing/m-merger (Harizanov 2014; Kramer 2014). Under this analysis, illustrated in (7) below, it is crucial that only the lower copy of the moved DP and the rebracketed D-head are realized. A-movement of the doubled DP is understood here as an instance of object shift/scrambling, given the doubled DPs’ semantic/pragmatic profile as specific/definite/topical. These approaches thus seem well-positioned to explain the interaction of clitic doubling with intervention and weak crossover, as detailed in section 2.2.4; however, as we argue in that section, the relevant facts are merely compatible with movement, but do not speak in favor of it.

A second subtype of the movement-based approach to CLD treats the clitic as arising by means of long head movement (e.g. Řezáč 2008; Roberts 2010; Preminger 2009; 2019; cf. Anagnostopoulou 2003 on feature movement). On this approach, an Agree relation between v and the object DP triggers movement of the head of that DP to v, with the clitic being the realization of the moved D head (Preminger 2019: 31–33). Both the moved D and the DP are realized at PF, yielding doubling (Preminger 2019: 20), see (8).

Despite important differences between them, the theories outlined thus far share movement as a crucial property.

The last type of approach treats clitic doubling as a type of object agreement. The idea goes back to at least Suñer (1988), where clitics are treated as base-generated agreement markers on the verb that enter a chain with the doubled argument DP. Under a more modern understanding of agreement, this type of approach can be recast by taking the clitic to be the realization of φ features which have been copied via Agree from the argument DP onto a functional head. Such an approach is sketched in (9) below, where we label the probing head as F for convenience (while v would be a possible alternative, we choose a higher head here for compatibility with our analysis developed below). Crucially, this approach involves only feature copying (or sharing), but does not involve movement.

Before concluding this overview of approaches to clitic doubling, it is important to note that many analyses are hybrid in practice, incorporating components of more than one theory. Quite a few in fact include (A-)movement in addition to their arguably main ingredient.

For instance, the Big DP approaches in Uriagereka (1995) and Nevins (2011) involve an object-shift-like phrasal movement step preceding the attachment of the clitic to the verb via head movement or m-merger.

A-movement components are also found in agreement approaches. In Sportiche (1996), clitics are treated as independent functional heads in the extended projection of the verb that agree with the doubled DP, which undergoes covert movement to the specifier of the clitic head. More recently, in Angelopoulos (2019: 21), the doubled DP is said to undergo scrambling-like A-movement to the specifier of a functional head above vP. This movement is followed by Agree with a clitic head, which is situated below T (the A-movement step enabling the DP to become accessible to Agree).

2.2.1 FC CLD: Data

Against this background, consider firstly the fact that Modern Greek shows first conjunct agreement (FCA) with postverbal subjects: in (10a), a coordination of a singular and a plural noun can result not only in resolved 1pl agreement on the finite verb, but also in 2pl first conjunct agreement. In (10b), where the order of conjuncts is flipped, 2pl agreement has become ungrammatical, while 1sg agreement is now grammatical.

Crucially, Greek allows first conjuncts to not only trigger agreement on the finite verb, but also to be clitic-doubled. In (11a), alongside the resolved features of the entire coordination (3pl and default m), clitic doubling can index just the features of the first conjunct (3sg.m); it cannot index the features of the last conjunct (3sg.f). Once we reverse the order of conjuncts (11b), the masculine singular clitic becomes ungrammatical, and the feminine clitic grammatical.⁵⁶

In Modern Greek, person and number participate in (first conjunct) agreement, while person, number and gender participate in (FC) CLD. In both cases, person resolution proceeds according to the hierarchy 1st>2nd>3rd, and number resolution always leads to plural features. If the conjuncts mismatch in gender and agreement/doubling targets the entire &P, gender resolution takes place; gender resolution patterns are complex, see Adamson & Anagnostopoulou (to appear) for resolution in agreement with coordination, but in our examples will lead to masculine features.

Both FCA and FC CLD are only possible if the &P is in postverbal position. If it occurs in preverbal position, viz., Spec,TP for preverbal subjects or a left-peripheral A′-position as with preverbal objects, only resolved agreement/doubling are possible. We address this fact in section 4.3 below.

2.2.2 Implications for theories of CLD

Recall now that, as outlined above, most approaches to clitic doubling involve movement of some constituent in the generation/placement of the clitic. Crucially, under these approaches, FC CLD will lead to a violation of the Coordinate Structure Constraint (CSC, Ross 1967), which bans extraction of individual conjuncts and asymmetric extraction from individual conjuncts; FC CLD is thus incorrectly predicted to be ungrammatical under movement-based approaches (cf. Legate 2014; Kalin & Weisser 2019 for a similar argument for differential subject and object marking, respectively).

Under the Big DP analysis, movement of the clitic, associated only with the first conjunct, to the verb would amount to subextraction from a single conjunct, and thus a CSC violation. Similarly, an A-movement + rebracketing analysis would need to postulate asymmetric A-movement of the entire first conjunct to, e.g., Spec,vP, again in violation of the CSC. Finally, under the head-movement approach, there would be asymmetric head-movement of the D head of the first conjunct to the verb, an instance of subextraction also in violation of the CSC. Importantly, the CSC is not at stake in the first place under an Agree-based approach, which only involves feature-copying. Thus, FC CLD favors an Agree-based approach to Greek CLD, since only this type of analysis is compatible with the CSC.⁷⁸

Further refinements to the argument against movement from the CSC are presented in Paparounas & Salzmann (to appear). There, we show that (i) the CSC holds in the relevant environments in Greek, viz., for A-movement of subjects to Spec,TP.; and (ii) our argument holds also under alternative conceptions of the Big DP hypothesis where the clitic head is generated outside of &P, and under more refined formulations of the CSC such as Bošković (2019; 2020), where minimal violations of the constraint are possible if the extractee originates at the edge of the first conjunct. We provide data based on coordinated ECM clauses with FC CLD of the ECM-subject in the first conjunct, which is crucially not at the edge as it is preceded by an adverb modifying the ECM clause, as shown in (12):

Examples like (12) would still be ruled out by the theory in Bošković (2019; 2020) and thus confirm that our argument based on the CSC stands.

2.2.3 Further arguments against movement: Binding

As argued in more detail in Paparounas & Salzmann (to appear), binding-theoretic considerations provide additional crucial evidence against the involvement of movement in Greek clitic doubling.

Data from Condition C and Condition A shows that clitic doubling neither feeds nor bleeds binding, providing strong support for the conclusion that the doubled argument (or, under the head-movement approach, parts of it) does not undergo movement. The configurations of interest are abstractly represented in (13), where it is always the lower argument (DP2) that is clitic-doubled: under A-movement based approaches to clitic doubling, then, DP2 is taken to undergo A-movement, and this is what is schematized in (13).

If, for instance, DP1 contains an unbound anaphor, movement of DP2 across it should create the correct binding configuration. Similarly, if DP1 is an R-expression and DP2 contains the same R-expression, a Condition C effect should be avoided if DP2 moves across DP1. Conversely, if DP2 is an anaphor bound by DP1, doubling of DP2 is predicted to destroy the binding relationship. Similar considerations hold for Condition C: if DP2 is an R-expresssion co-indexed with an R-expression embedded within DP1, movement of DP2 should trigger a Condition C effect.

Crucially, however, none of these predictions are borne out when we clitic-double the lower argument of the relevant examples. Beginning with Condition A, (14) shows that clitic-doubling the direct object does not enable this argument to bind a subject anaphor, contrary to the predictions of an account where doubling the direct object amounts to A-moving it to the edge of the verb phrase (see also Angelopoulos & Sportiche 2022: ex. 34b). Note that nominative anaphors are not independently ungrammatical in Greek (see Angelopoulos & Sportiche 2022: section 6.2).

(15) shows the other side of the same coin: clitic doubling the anaphor does not cause it to raise above its antecedent and violate Condition A.⁹

The Condition A data thus shows that clitic doubling has no effect on binding; Condition C data reinforce this conclusion.

In (16a), where an IO R-expression c-commands another instance of itself in the DO, doubling the DO does not alleviate the Condition C violation; similarly in (16b) for the subject and the IO.

Finally, clitic doubling also does not create a Condition C effect. A-movement-based accounts predict that the doubled version of (17a) should be a Condition C violation, since the doubled DO should have raised to c-command the IO; but this prediction is not borne out. (17b) patterns similarly.

The binding data thus strongly suggest that the doubled DPs stay in situ: there is no evidence that the doubled DPs occupy anything but their base-generated positions for the purposes of binding.

Note that this observation calls into question the assimilation of clitic doubling in Greek(-type languages) to instances of overt A-movement such as A-scrambling in German: as we argue in more detail in Paparounas & Salzmann (to appear), a hallmark of A-scrambling and other bona fide instances of A-movement such as raising is that they can interact with Condition A and C, unlike Greek clitic doubling, which never does.

The fact that CLD does not affect binding follows straightforwardly under a pure Agree approach. Agree only copies phi-features and is well-known to have no effect on semantic interpretation (see, e.g., den Dikken 1995 on binding in expletive constructions).

2.2.4 Reanalyzing previous arguments for movement

In Paparounas & Salzmann (to appear), we discuss empirical arguments previously adduced in favor of movement: it has been argued that weak crossover (WCO) and defective IO intervention can be alleviated under clitic doubling. These effects of amelioration prima facie do not follow from an Agree-based account; but we show that the evidence can be insightfully reanalyzed.

It has been observed that clitic doubling repairs WCO violations in Greek: examples like (18), where the IO contains a bound pronoun and the DO is a quantified DP, are only grammatical if the DO is doubled. This effect has been interpreted as crucial evidence in favor of A-movement of the doubled DP in clitic doubling (Anagnostopoulou 2003).

In Paparounas & Salzmann (to appear), we argue that the alleviation observed under doubling is in fact properly independent of any movement, and is instead plausibly reducible to the information-structural correlates of doubling. It has independently been shown that information structural manipulations, especially givenness/D-linking/topicality of the binder, improves WCO violations (see Safir 2017: 23–24 for a recent overview and Eilam 2011: 150–160 for detailed discussion). Crucially, this also holds for Greek, where WCO can be alleviated under D-linking and focus, without any doubling.

(19) illustrates one such case. (19a) is a standard instance of WCO that is repaired by doubling (19b). Importantly, however, this instance of ungrammaticality can also be alleviated without doubling (or indeed any obvious instance of A-movement): the example improves considerably if the wh-phrase is D-linked (19c), and becomes perfect if D-linking is combined with focus on parts of the phrase containing the bound pronoun (19d).

Importantly, since CLD is associated with givenness/topicality in Greek, WCO alleviation under doubling is a subtype of the more general phenomenon of information-structurally-based WCO alleviation, with no crucial role necessary for movement. From this perspective, information structure is the hidden third variable at play: it governs alleviation of WCO both under clitic doubling (19b) and without it (19c-d). This view, we argue, provides a more unified view of WCO alleviation in Greek compared to that offered by A-movement-based accounts of clitic doubling. Though accommodating (19b), these accounts have no way of linking this type of example to movement-free alleviation (19c-d), thereby missing the generalization that the information-structural correlates of doubling are precisely those that can be independently shown to alleviate WCO.

The second prominent argument in favor of movement comes from intervention effects in IO-nominative configurations, which were shown in Anagnostopoulou (2003: 45) to be alleviated by clitic doubling, see (20):

The effect of CLD in (20) follows if movement takes the intervener out of the way before T probes for the subject. We show in Paparounas & Salzmann (to appear) that the interaction in (20) can also be captured under an Agree-based approach and without movement if the notion of activity predicated of a DP refers not to case assignment but rather to whether the DP has been involved in φ-Agree. While not the majority view, this perspective on activity has been fruitfully applied in previous literature (see, e.g., Georgi 2013: 167, Kalin & van Urk 2015: 673, Oxford 2017). As long as the probe that generates the clitic is discharged before T probes, it can target the IO and deactivate it. Subsequent probing by T can skip the IO and access the low nominative. We argue that this activity-based view of intervention fares at least as well as the competing movement-based account.

Since these issues are addressed in great detail in the above-mentioned paper, we do not take them up any further here; we will briefly return to intervention in section 3.3 below where we evaluate the implications of a high clitic probe for theories of the PCC.

2.2.5 Distribution of clitic doubling

A final issue addressed in more detail in Paparounas & Salzmann (to appear) concerns the distribution of clitic doubling. Recall from the brief remarks in section 2.1.1 that doubling is most likely with highly referential/topical DPs, typically definites; although the situation in Greek is in fact more complicated (with even non-specific indefinites being capable of undergoing doubling in some cases). Any attempt to correctly circumscribe the range of DPs that can undergo doubling will also have to reckon with the issue of optionality: all things being equal, clitic doubling is hardly ever obligatory, even with the most doubling-eligible of DPs.

Approaches postulating A-movement of the doubled DP seem prima facie well-suited to addressing the distribution of doubling. Under such approaches, whether doubling occurs or not depends on whether the step of object shift that brings the clitic into the locus of m-merger or the domain of an Agree probe applies; if object shift does not apply (because the relevant DP does not bear the necessary features, related to topicality, for instance), then m-merger/Agree will not be able to ‘see’ the DP in its base position, for independent locality reasons. An approach based purely on Agree, on the other hand, owes an account of why certain DPs seem to not be eligible goals for the clitic-generating probe.

Before offering just such an account, we note that, contrary to the simple narrative just sketched, the distribution of clitic doubling is in fact just as much of a puzzle for movement-based theories as it is for Agree-based ones. A first reason for this is that any parallels between the A-movement step purportedly involved in doubling and bona fide cases of scrambling/object shift are at best imperfect and stand alongside striking binding-theoretic countervidence to the alleged parallelism, as argued earlier in this section. Secondly, for languages like Greek, movement-based accounts would need to invoke object shift even for elements otherwise expected to never undergo this process, since even (non-specific) indefinites can sometimes be doubled.

We thus consider the distribution of clitic doubling to be a puzzle for analyses of the phenomenon more generally, including movement-based ones; we will not attempt to shed new empirical light on this issue here, and will limit ourselves to showing how this kind of distribution can be implemented syntactically on our Agree-based account.

A proposal compatible with a pure Agree approach is the licensing approach to Differential Object Marking (DOM) by Kalin (2018; 2019). The underlying idea of this approach is that, in languages with DOM, only DPs with certain features need to be licensed, where licensing takes place via Agree. In Kalin (2019) this is technically implemented as follows: the features that require licensing, e.g, [specific], are associated with a so-called derivational time bomb; unless such a time-bomb is defused, viz., the feature it is associated with enters Agree, the derivation crashes. While the Agree probe on T is taken to be obligatory (because in the relevant languages, subjects agree irrespective of their semantic/pragmatic properties), an economy principle restricts the presence of a secondary licensor, viz., an Agree probe on v: a secondary licensor is only merged if necessary for convergence.

This logic can be directly extended to clitic doubling, which is thus treated as a variant of DOM. Concretely, objects with certain semantic/pragmatic properties, e.g. [def, top] etc., will carry a derivational time-bomb. A derivation will consequently only converge if there is a probe that agrees with this DP. If an object has no such feature, no licensing via Agree is required; a secondary licensor (in our case on T, see section 3.3 below) is thus not possible, and no CLD arises. Note that such an approach does not provide a deeper understanding of the distribution of CLD and has nothing to say about the optionality other than that the time-bomb is optional in some cases, which is little more than a restatement of the facts. This is, however, an explicit implementation of the distribution of CLD in syntactic terms that is compatible with an approach to CLD that solely relies on Agree. We will briefly come back to the optionality issue in our discussion of PCC repairs in section 3.4 below.

3.2.1 The problem of insufficient features and/or movement

A popular family of approaches treats the PCC as an intervention effect (Béjar & Řezáč 2003; Preminger 2009; 2019; Coon & Keine 2021, i.a.). The general idea underlying these approaches is that one probe has to license two internal arguments; given the limited features of the probe, not all person-number combinations are possible in double object constructions.

Concretely, in these analyses, the probe on v is split into a person and a number probe with the person probe invariably applying first. In a scenario with both an IO and a DO present, the structurally higher IO blocks person Agree with the DO. After person Agree between v and the IO has applied, the IO cliticizes, viz., head-moves to v. This removes the IO as an intervener (it is assumed that its trace is invisible). Thereafter, the number probe is discharged. Given that the IO no longer intervenes, the number probe on v can target the direct object, which subsequently also undergoes cliticization/head-movement to v. The basic configuration underlying intervention accounts is illustrated in (32), DO cliticization not being indicated for ease of representation (in what follows, we use the star notation for Agree probes that need to be discharged and bullet features for structure building; π = person, # = number):

Next to movement of the IO and the split-probe, intervention accounts have a another component to implement the PCC-restriction. In Béjar & Řezáč (2003); Preminger (2009; 2019), the so-called Person Licensing Condition (PLC) requires first and second person arguments to enter person Agree. As a consequence 1st/2nd person direct objects will not be licensed in PCC configurations because the person probe on v has already been discharged through Agree with the IO. A third person direct object, however, is possible because its person feature does not require licensing. Since the split probe is located above both arguments and the person probe is discharged first, the person restriction only affects the DO.

In Coon & Keine (2021), this other component is different. The implementation of the PCC does not rely on the Person Licensing Condition, which they reject. Rather, in a 3>1/2 configuration, the PCC violation obtains because the person probe (in fact, different segments of it) agrees with both objects. The authors assume that the required ensuing clitic-movement operation fails as it would have to involve both clitics moving simultaneously, which by assumption is ruled out in this work (cf. also the implementation of the PCC in Arregi & Nevins 2012: 64–69).

Thus, in intervention accounts, the implementation of the PCC-restriction crucially involves movement of the arguments, which clashes with the evidence against movement discussed in sections 2.2.2 and 2.2.3.

Intervention accounts are also incompatible with our FC CLD data because they crucially rely on movement for the generation of the clitics. Note that in these intervention approaches, the result of the Agree operation initiated by v, namely the features copied unto v, remains unrealized. In fact, the features copied from the objects would not be sufficient to obtain fully specified clitics given that person and number probe separately and interact with different arguments. With respect to the IO, it is usually assumed that either no features are copied from it (Béjar & Řezáč 2003: 54; Preminger 2009: 656–657; Preminger 2014: 50–54) or only the person feature is (Coon & Keine 2021). With respect to the DO, the failure to interact with the person probe may at first seem unproblematic as long as 3rd person is interpreted as the absence of person. However, note that clitics in Greek are also specified for gender, which none of the PCC probes probe for. Thus, no gender feature is copied onto v in these approaches. Therefore, even if a person feature is copied from the IO and a number feature from the DO, v will lack the features necessary to realize fully specified clitics. Both the IO and the DO clitic must consequently arise in a different way; in a clitic doubling scenario, this will require one of the three movement-based approaches discussed in section 2.1.2 above; the approaches just mentioned in fact assume head-movement. However, this is clearly not compatible with the result obtained above, viz., that clitic doubling cannot involve movement: recall that FC CLD is possible with both coordinated IOs and DOs. Note that even if Agree were to generate fully specified clitics in these approaches, e.g., by employing different/more complex probes, they would still not be compatible with our FC CLD data since movement of the IO is a crucial component of the intervention account: without movement of the IO, the DO cannot be accessed.

Thus, intervention accounts of the PCC are incompatible with our data for two reasons: they require movement both for the implementation of the PCC and the generation of the clitics.¹⁴

The problem with Agree not copying sufficient features to generate fully specified clitics also arises in approaches where movement as such is not crucial in the implementation of the PCC. In one such approach, Anagnostopoulou (2003: 286–291), v also bears probes for person and number with [person] probing first. The PCC arises because once person features have been copied from the IO to v, only the number feature remains on v to Agree with the DO; under the assumption that the absence of person-Agree results in third person, the DO is consequently restricted to third person, while local person DOs cannot be licensed. Intervention is not an issue in this approach because the IO is assumed not to have a (accessible) number feature, which is why the number feature on v can target the DO across the IO. The basic configuration underlying this PCC-implementation is shown in (33):

As in the intervention approaches, since Agree only involves partial feature copying from the objects, this will not be sufficient to generate IO and DO clitics specified for person, number and gender. Rather, Anagnostopoulou (2003: 211–215) proposes that in a clitic doubling scenario, the clitics arise via feature movement to T, which in modern terms would be recast as head-movement. Consequently, to obtain fully specified clitics, the approach crucially requires movement after all and is therefore also incompatible with our evidence against movement in clitic doubling.¹⁵

The problem with insufficient features copied via Agree also arises in the approach by Pancheva & Zubizarreta (2018: 1299–1306). In this approach, the relevant probes are on Appl, which introduces the IO and undergoes person-Agree with both the IO in its specifier and the DO within its VP complement. Appl is assumed to have a valued interpretable person feature [+proximate], which agrees with the IO. This feature is used to encode the fact that the IO must be a perspectival holder. Additionally, Appl has an unvalued uninterpretable person feature that agrees with the DO. The basic configuration underlying this approach is illustrated in (34):

The implementation of the PCC restriction is handled by means of a constraint that stipulates that the person features on Appl must not be identical. Since the interpretable feature licensing the IO must be [+proximate] (to encode the IO’s role as a perspectival holder), the uninterpretable features copied onto Appl, viz., those copied from the DO, cannot contain [+proximate]. This restricts the DO to (obviative) 3rd person arguments since they bear [–proximate]. Like in the approach by Anagnostopoulou (2003), the PCC is not implemented by means of movement but by specific assumptions about the Agree probe on Appl. In this respect, the approach encounters no difficulty with our evidence against movement from FC CLD. However, even though the authors explicitly state (p. 1299) that their approach is compatible with the clitics either being the result of Agree or being present in the syntax already, in the latter case matching features on Appl, the fact that the [+proximate] feature on Appl is valued and interpretable implies that it cannot copy more (person) features from the IO. Consequently, Agree will not be sufficient to generate fully specified clitics if the IO is 1st or 2nd person, given that, in addition to [+proximate], they bear the features [+participant] and [+/– author]. Thus, the clitics in a CLD configuration have to arise in a different way in addition to Agree. Again, all other options discussed above are conceivable (Big DP, head-movement or A-movement + rebracketing). Since all these options require movement, this approach is also incompatible with our FC CLD data.¹⁶

The problem that Agree copies insufficient features to obtain fully specified clitics also arises in the approach by Foley & Toosarvandani (2022: 35). Because of its complexity, we will abstract away from many details and focus on the aspect that is of relevance to our discussion: this approach to the PCC differs from previous accounts in that Agree only targets the first/IO argument (followed by cliticization of the IO). The person restrictions affecting the DO are instead reanalyzed as restrictions on cliticization: it is proposed that the DO can cliticize onto the probe only if it contains a subset of the probe’s features (cf. Roberts 2010), which, simplifying somewhat, is relativized to probe for [participant]. Since this is not the case in a X>1/2 configuration (the notion of subset adopted in this approach is somewhat involved), such combinations are ruled out. In a X>3 configuration, however, the DO does contain a subset of the probe’s features because third person matches the probe’s values in a specific sense: it does not match any of them, but the empty set also qualifies as a subset of every set. Regardless of the merits of these restrictions on cliticization, it should be clear that the Agree operation alone will not be sufficient to generate fully specified DO clitics given that there is no feature copying from the DO. Consequently, this approach (which admittedly does not discuss clitic doubling) will require an additional mechanism to obtain the clitic in a CLD configuration (Big DP, head movement, A-movement + rebracketing). Again, these implementations all require movement and are thus incompatible with our FC CLD data.

All the approaches surveyed so far were shown to be incompatible with our evidence against movement because the Agree operations involved did not copy enough features from the objects to generate fully specified clitics. However, one can also find the reverse problem: there are approaches where Agree as such is sufficient to generate fully specified clitics but where movement is necessarily involved in the implementation of the PCC. In Baker (2011: 885), the person restriction is implemented as follows. By assumption, agreement in person requires a spec-head relationship. In a double-object configuration, v can probe twice but has only one EPP-feature and therefore attracts the closest goal, the IO, with which it can agree fully, viz., also in person. Subsequent Agree with the DO cannot be in person given the lack of a spec-head-relationship. Thus, v only agrees in number (and possibly gender) with the DO; hence, under the assumption that third person corresponds to the absence of person, DOs are limited to third person. The basic logic of the approach is illustrated in (35):

The Agree operations initiated by v here would be sufficient to generate fully specified clitics (given that v can probe twice), and as such this analysis would require neither a Big DP nor a derivational account. However, since the approach requires movement of the IO to Spec,vP for person agreement to be possible, it is not compatible with our FC CLD data after all (recall that the IO can also involve a coordination, see ex. (31)).

Table 1 provides an overview of the approaches discussed in this subsection focusing on the components that require movement (viz., the generation of clitics or the PCC):¹⁷

3.2.2 Compatible Agree approaches

In the approaches discussed in the previous section, a problem arises in the context of our FC CLD data because these approaches require (i) a movement approach for the generation of the clitic and/or (ii) movement of the objects in the implementation of the PCC restriction. Unlike in these approaches, there are a few Agree-based approaches to the PCC that can generate fully specified clitics while not requiring movement at all. They are thus compatible with the FC CLD data and more generally with our evidence against movement.

Nevins (2007: 296) proposes a multiple Agree approach where the probe on v targets both objects simultaneously. Here, the PCC restriction is implemented by means of restrictions on Agree. In languages displaying the strong version of the PCC, probing for person is relativized to contrastive values of [author]. In (3>1/2) configurations, a violation obtains because a third person argument has a non-contrastive value for [author] (author is only contrastive in the context of [+participant]). It thus acts as an intervener between v and the DO and multiple Agree fails given a stipulated condition on contiguous Agree. In X>3 configurations, no issues arise because here it is the lower argument, the DO, that bears a non-contrastive value for [author]. Consequently, this does not affect the interaction between v and the IO and no violation of contiguous Agree obtains. The basic configuration underlying this approach is illustrated in (36):

Combinations of local persons (1>2/2>1) are ruled out by an additional condition called Matched Values, which is violated in this case because there are conflicting contrastive values for [author]. Thus, since the clitics arise purely via Agree and the PCC is implemented by means of restrictions on probing, the approach is compatible with our evidence against movement of the objects.¹⁸

Another approach compatible with our FC CLD data is Adger & Harbour (2007: 25–26). In this approach, v agrees with the IO, while Appl agrees with the DO. Appl is specified to select an IO bearing [participant], which includes 1st and 2nd person arguments as well as animate third person arguments; this rules out inanimate third person IOs. The PCC restriction is implemented as follows. A stipulation prevents Appl from probing for features that it requires for its specifier (cf. Pancheva & Zubizarreta 2018); consequently, given that Appl selects a specifier bearing [participant], it cannot probe for [participant], but only for number. This limits DOs to 3rd person arguments (third person resulting from the absence of [participant]). If it were to probe for [participant] and Agree with a local person DO, the IO would have to be inanimate, resulting in an uninterpretable structure. The basic configuration presupposed in the approach is shown in (37):

While the approach is discussed in the context of clitics only, it would be compatible with an Agree-only approach to clitic doubling since, having two phi probes, it can generate fully specified (IO) clitics (the absence of [participant] on the DO clitic will result in 3rd person) and does not involve movement of the objects in the implementation of the PCC.¹⁹

The last approach that can generate fully specified clitics by means of Agree is Deal (to appear). Here, the probe has a so-called interaction condition for φ, which means it agrees with all objects bearing phi-features. This, in principle, allows the probe to interact with and copy features from several arguments. Probing stops once the probe’s satisfaction condition is obtained: in the PCC context, the probe is satisfied once it encounters the feature [participant]. The probe is located on Appl and first targets the DO within VP. Thereafter, the search space of Appl is extended to include the IO in its specifier (cf. the cyclic Agree approach by Béjar & Řezáč 2009). The basic configuration assumed in the approach is shown in (38):

In a PCC-obeying context, e.g., 2>3, the probe will first copy the φ-features from the 3rd person DO. Since the probe is not yet satisfied, it continues to probe and targets the local person IO in its specifier and copies the features from there, too. This results in fully specified phi-feature sets. A PCC violation obtains if the DO already satisfies the probe, viz., if it contains [participant]. The probe is deactivated and no clitic IO can be licensed (the assumption being that clitics can only be generated via Agree). Since this approach does not involve movement in either the generation of clitics or the implementation of the PCC, it is compatible with our FC CLD data.²⁰

Thus, of the eight types of approaches surveyed so far, only three are compatible with our evidence against movement of the objects. This is because they use Agree to generate fully specified clitics and implement the PCC by means of constraints on this operation itself.