2.1 Simple crossover effects

It is well established that local scrambling in Hindi is not subject to WCO (Déprez 1989, Mahajan 1990, Gurtu 1992, Mahajan 1994). This is illustrated in (5), where scrambling of the object har laṛke-ko ‘every boy-acc’ over the subject uskii behin-ne ‘her/his sister-erg’ makes binding of the subject-internal pronoun possible, a binding that is impossible in the absence of scrambling.

At first glance, it appears that scrambling is clearly subject to SCO. If the pronoun c-commands the launching site, binding is impossible, as (6) shows. This restriction holds regardless of whether the pronoun is a regular personal pronoun (us-ne) or a reflexive (apne aap-ne).

But caution is in order in interpreting (6). In particular, binding in (6) is already ruled out for reasons independent of SCO. First, the pronoun us-ne is subject to Condition B, which is plausibly violated if us-ne is bound by har laṛke-ko. Second, the reflexive apne aap is subject oriented. The sentence in (6) involves binding of apne aap by a scrambled object, violating the subject orientation. As a result, (6) is correctly ruled out even if scrambling is not subject to SCO. Therefore, the contrast between (5) and (6) by itself does not establish that Hindi scrambling differs with respect to SCO and WCO.

2.2 Secondary-crossover effects

It is possible to circumvent the problems that arise in the interpretation of simple SCO configurations such as (6) by investigating secondary-crossover effects (Van Riemsdijk & Williams 1981, Safir 1984, Postal 1993a). In such configurations, the quantifier that binds the pronoun is not the moving element itself but is embedded inside the moving element (in the examples that follow, it is a possessor). As we now show, in such configurations a systematic contrast arises between WCO and SCO.

Like English, Hindi allows inverse linking, where the possessor of a DP binds a pronoun c-commanded by the container DP (see May 1977, Higginbotham 1980, Safir 1984, Ruys 2000, and May & Bale 2006 for general discussion of inverse linking):

Again as in English, possessors do not c-command out of their container DP. In (8), no Condition B effect obtains, which indicates that the possessor Ram-kii does not c-command the pronoun us-ko.

Furthermore, the cases of binding by a possessor in (7) do not involve possessor raising of har laṛke-kii ‘every boy-gen.’ While Hindi does allow possessor raising in some cases, DPs that bear ergative case (-ne) or accusative case (-ko) do not permit possessor raising out of them, as shown in (9) and (10), respectively (see Alok 2016 for related discussion).

Against this background, we now turn to secondary crossover. The examples so far have involved binding by a possessor in a base-generated configuration. Binding by a possessor may also be fed by scrambling, as (11) shows.

Here, the possessor har laṛke-kii ‘every boy-gen’ inside the moved DP har laṛke-kii behin-ko ‘every boy’s sister-acc’ binds the pronoun us-ke ‘he-gen’ inside the subject us-ke dost-ne ‘his friend-erg.’ Because the moved DP bears accusative case (-ko) and such DPs do not permit possessor raising out of them (see (10)), we can rule out the possibility of possessor raising having applied in (11). Thus, (11) demonstrates that Hindi scrambling does not give rise to secondary WCO, just as it does not give rise to standard WCO.

A puzzle arises when we consider (12), which involves a minimally different configuration from that of (11). In (12), the bound pronoun us-ne ‘he-erg’ is not embedded inside the subject but is itself the subject. In this case, the binding is ungrammatical.

In other words, (12) shows that scrambling is subject to secondary SCO. Crucially, the various confounds that arose with examples of apparent simple SCO such as (6) do not arise with (12). In particular, Condition B is not violated in (12) because the pronoun is not bound by a c-commanding DP (this is seen most clearly in the nonmovement counterpart in (8)).

In what follows, we adopt the view that possessor binding is direct: the bound pronoun is bound by and hence coindexed with the possessor (har laṛke-kii ‘every boy-gen’ in the examples above). An alternative, pursued by Chierchia 2023, is to coindex the pronoun with the container DP itself (e.g., coindex us-ne ‘he-erg’ with har laṛke-kii behin-ko ‘every boy’s sister-acc’) and then treat the pronoun as an E-type pronoun (Evans 1977, 1980): a pronoun that for every sister returns that sister’s brother(s). We will not pursue this line of account here; see appendix A in the supplementary material for discussion. Given our argument that the possessor does not undergo raising out of the container DP, our view that the pronoun is bound by the possessor requires that binding does not demand c-command. See Barker 2012 for general arguments to this effect, and see Kobele 2010, Barker & Shan 2014, and Bumford & Charlow 2022 for specific analyses.

The relevant structure of (12) is schematized in the following.

The impossibility of binding in this configuration poses a clear puzzle. First, we saw on the basis of (5) and (11) that scrambling may feed pronominal binding (i.e., there are no WCO effects). Second, we know that possessors may bind outside their container DPs in the absence of movement (7) and after movement (11). Third, the trace in (13) is not coindexed with the subject pronoun, so there is no transparent Condition C effect with respect to the trace. It would seem, therefore, that all the requirements for binding are satisfied in (12)/(13), and yet binding is impossible. Particularly significant is the contrast between (11) and (12). Binding of a pronoun by a possessor inside a moved DP is possible if the pronoun does not c-command the launching site of the DP (11) but not if the pronoun c-commands the launching site (12). No analogous restriction holds in the absence of scrambling (7). This strongly suggests that the contrast between (11) and (12) involves the fact that these structures are the result of scrambling.

The key takeaway of the contrast between the secondary-WCO configuration (11) and the secondary-SCO configuration (12) is that the distribution of SCO does not match the distribution of WCO in Hindi scrambling: secondary SCO arises in configurations that do not display secondary WCO. This finding provides clear empirical evidence that SCO is at least partially the result of a mechanism distinct from those that underlie WCO. An account that treats SCO and WCO in the same way (e.g., Van Riemsdijk & Williams 1981) does not give rise to this split.

Before we proceed, we note that the divergence of secondary SCO and secondary WCO in Hindi scrambling differs from both English A- and Ā-movement, where the two correlate. As shown in (14) and (15), Ā-movement displays both secondary WCO and secondary SCO (see Higginbotham 1980, Van Riemsdijk & Williams 1981, Safir 1984, Postal 1993a, Safir 1999, Ruys 2000, Safir 2017), whereas A-movement displays neither (unsurprisingly given that A-movement does not show crossover effects in general).

In other words, then, Hindi scrambling behaves like English A-movement with respect to (secondary) WCO but like English Ā-movement with respect to (secondary) SCO, as summarized in (16).³

An analysis of the Hindi scrambling facts thus requires accounts of SCO and WCO that explain why they part ways in the way they do and what conditions their distribution. The Hindi facts also connect to important debates on the relationship between scrambling and the A–Ā distinction. It is well known that scrambling displays a “mixed” behavior with respect to traditional A- and Ā-properties, including issues of locality, parasitic-gap licensing, and WCO. How scrambling relates to A- versus Ā-movement has been the subject of considerable debate, ranging from analyses that treat scrambling as A-movement or Ā-movement (e.g., Saito 1985, Fanselow 1987, Saito 1989, Fanselow 1990, Mahajan 1990, Müller & Sternefeld 1993, Mahajan 1994, Müller & Sternefeld 1994, Müller 1995) to analyses that treat it as a genuinely mixed type of movement (e.g., Webelhuth 1989, 1992, Browning & Karimi 1994, Dayal 1994). The distribution of properties in (16) adds a novel empirical dimension to this debate. In addition, it deepens questions about the extent to which scrambling can be treated as pure A- or Ā-movement and about the extent to which certain properties of scrambling may be derived from other properties of scrambling.

4.1 Condition C connectivity

We take as our analytical starting point the observation that the SCO facts correlate with the distribution of Condition C in Hindi. As in English, R-expressions are subject to Condition C in Hindi and hence must be globally A-free, per the following definitions.

An R-expression in the possessor position of an object thus must not be coindexed with a pronoun in subject position, as in (20a). Crucially, scrambling does not amnesty such Condition C violations, as (20b) shows. That is, coindexation is still ruled out in (20b) despite the fact the R-expression is no longer c-commanded by the pronoun after scrambling.

In other words, Hindi scrambling displays Condition C connectivity with possessors.

Because Condition C applies only under c-command, Condition C connectivity under scrambling arises only if the pronoun c-commands the launching site. As shown in (21), if the pronoun (us-kii ‘she-gen’) is embedded inside another DP that is crossed by the scrambling, coindexation is possible.

Assuming reconstruction, the contrast between (20b) and (21) is unsurprising given the standard c-command-based formulation of Condition C. It is worth noting, however, that this contrast correlates with the contrast between secondary WCO and secondary SCO that we saw earlier. Secondary-SCO configurations are analogous to configurations that result in a Condition C effect under scrambling. This is schematized in (22): a secondary-SCO effect corresponds to DP-gen in (22) being a quantificational DP, while a Condition C effect corresponds to DP-gen being an R-expression. Conversely, configurations in which the pronoun is embedded inside another DP result in neither a WCO effect nor a Condition C effect: see (23).

Because the distribution of SCO thus corresponds to that of Condition C connectivity, the Hindi data strongly suggest not just that SCO can be the result of a constraint unrelated to WCO but more specifically that SCO is best analyzed as a Condition C effect.

Despite the clear empirical connection between SCO and Condition C, the traditional trace-based Chomsky 1981 account of SCO in terms of Condition C is insufficient. On this account, the trace of Ā-movement behaves like an R-expression and is subject to Condition C. As it stands, this account does not extend to cases of secondary SCO because the SCO effect arises not with respect to the moving element itself but instead with respect to the possessor of the moving element. Thus, in (22), the trace t₂ does not violate Condition C because it is not coindexed with the pronoun. Therefore, to obtain a Condition C violation in (22), the trace must have additional internal structure, including at least the possessor DP and the information that it is coindexed with the pronoun.

In the sections that follow, we first develop an account of Condition C connectivity under scrambling that extends to such cases, and then we show how such an account immediately derives the distribution of SCO in Hindi.

4.2 Case, late merge, and Condition C

To overcome the lack of internal structure in the launching site on a trace-based account, it is standard to appeal to the copy theory of movement (Chomsky 1995).⁶ The key advantage of conceiving of movement as creating copies is that the internal structure of the moving expression is represented in the launching site. This allows an account of Condition C connectivity that arises with respect to elements contained within the moved expression, as shown in (24), where ⟨[DP-gen₁ … ]⟩ represents the unpronounced lower copy.⁷

While a copy-theoretic account is therefore promising, the simplest copy-theoretic account—according to which all movement creates a complete copy of the moved expression in the launching site—is too strong. In particular, movement types seem to differ in their propensity to induce Condition C connectivity in complex ways. In English, Ā-movement shows a greater degree of Condition C connectivity than does A-movement (Chomsky 1993, Sauerland 1998, Fox 1999, Takahashi 2006, Lebeaux 2009, Takahashi & Hulsey 2009, Safir 2019, Thoms & Heycock 2022). For example, R-expressions inside argument clauses show Condition C connectivity with Ā-movement but not with A-movement, as illustrated in (25) and (26).

A common intuition that the literature on this contrast has pursued is that A-movement leaves an impoverished representation of the moved expression in the launching site whereas Ā-movement leaves behind a more complete representation of the moved expression (Sauerland 1998, Fox 1999, Bhatt & Pancheva 2004, Takahashi 2006, Takahashi & Hulsey 2009, Stanton 2016, Safir 2019, Thoms 2019, Thoms & Heycock 2022).

In what follows, we adopt Thoms 2019’s and Thoms & Heycock 2022’s external-remerge account of the contrast in (25) and (26). Thoms and Thoms & Heycock propose that English A-movement allows the launching site to contain only an NP (for them nP, a distinction that we abstract away from here), with the DP portion merged later, before the moved element is merged in its landing site (this idea, though implemented quite differently, goes back to Sportiche 2005). More specifically, following Citko 2005, De Vries 2009, Johnson 2011, 2012, Poole 2017, Citko & Gračanin-Yuksek 2021, and others, this model assumes that constituents can be externally remerged, yielding a multidominant structure with two root nodes. The two root nodes are then merged with each other, yielding a single root node. Unlike other conceptions of late merge (like Lebeaux 1988 and 2000’s Adjoin-α or Takahashi & Hulsey 2009’s wholesale late merge), external remerge is not countercyclic in the sense that at least one of the two elements being merged is always a root node. As such, it obeys De Vries 2009’s Root Condition on Merge, stated in (27).

Thoms & Heycock 2022 accounts for the absence of Condition C connectivity for arguments with English A-movement (25b, 26b) by means of the derivation in (28) below. First, Thoms & Heycock assume, with Borer 2005, Moulton 2009, Lohndal 2012, Adger 2013, and Alexiadou 2014, that arguments of nouns are specifiers of a ModP projection between NP and DP. Second, they assume that A-movement permits a derivation in which only an NP is merged in the premovement position. Third, they assume that nominal material above NP may be externally merged on top of the NP and the resulting constituent merged into a higher position.

The resulting derivation for Condition C obviation under A-movement is shown in (28). First, only the NP is merged in the premovement position, lacking all adjuncts and arguments (28a). Second, the NP node is remerged with Mod, creating a structure in which the NP node has two mothers and the structure as a whole has two root nodes (28b). Note that this step complies with (27) because the Mod head is a root node. In the third step, Mod introduces nominal arguments in its specifiers (linearized to the right in this derivation), and the DP layer is merged above ModP (28c). Lastly, the resulting DP is merged into the landing site of A-movement, creating a single-root structure again (28d).

The crucial feature of the structure in (28d) is that the R-expression inside the PP is part of the externally remerged material. As such, it is represented in the landing site of the A-movement step but not in the launching site, and it is correspondingly not c-commanded by the coindexed pronoun. Condition C is therefore respected in (28).⁸

As with all multidominance theories of movement, questions arise as to how to determine which position an element is linearized in. Because all the movements we consider in this article are overt, it suffices to say that it is the highest occurrence of the multidominated element that is pronounced. See Johnson 2012 and Poole 2017: 135–138 for a linearization algorithm for multidominance structures. Since the question is not different in nature from analogous issues that arise under the copy theory of movement (see, e.g., Nunes 1995, 2004), we will not consider these questions further here.

If left unconstrained, external remerge would permit Condition C obviation across the board. But as we saw, Ā-movement shows Condition C connectivity in these cases (25a, 26a). This means that an external-remerge derivation as in (28) must be unavailable for Ā-movement and that Ā-movement must require the full DP structure to be present in the premovement position:

Why is external remerge of this kind available for English A-movement but not for English Ā-movement? Building on Takahashi 2006 and Takahashi & Hulsey 2009, Thoms & Heycock’s propose that this follows from considerations of case (also see Gong 2022, 2025). In particular, they assume that DP is subject to the Case Filter. This entails that the DP layer must be added before case is assigned:⁹

In English, A-movement feeds case assignment. It is therefore possible to late merge a DP layer in an A-movement step, as long as case is assigned to the landing site of this A-movement step. By contrast, Ā-movement applies to DPs that have already been assigned case. It is therefore not possible to late merge a DP layer to the landing site of an Ā-movement step, as this DP layer would remain without case, violating the DP Case Filter (also see Takahashi & Hulsey 2009 and Thoms & Heycock 2022 for arguments that when case is not an issue, Ā-movement as well may utilize late merge). It is also not possible to merge the DP layer early and to late merge the ModP layer after Ā-movement has applied. This derivation would require sandwiching the ModP between the NP and the DP and as such would involve a merge step that does not apply to a root node, in violation of the Root Condition (27).

Because late merge of the DP layer is thus the only way of obviating Condition C with arguments and the DP layer is subject to the Case Filter (30), Thoms & Heycock 2022’s account derives the contrast between English A- and Ā-movement’s ability to obviate Condition C violations (as does Takahashi & Hulsey 2009’s account, albeit in a somewhat different way). The account of Hindi in the next section will extend this analysis to scrambling.

Finally, as it stands, Thoms & Heycock’s account seems to require a complete representation of the moved expression in the launching site of Ā-movement, which would be too strong. It is standardly recognized since Freidin 1986, Lebeaux 1988, and Lebeaux 2000 (building on Van Riemsdijk & Williams 1981) that English Ā-movement does not induce Condition C connectivity with respect to R-expressions inside adjuncts. Thus, we observe Condition C connectivity with argument clauses, as in (31a), but not with relative clauses, as in (31b).

The traditional account of such effects is due to Lebeaux 1988 and 2000 and involves late merge of adjuncts. Lebeaux’s core proposal is that adjuncts are not required to be present in a moving constituent before the movement applies (though they are permitted to). That is, adjunction may apply freely, either before or after movement takes place, and it does not have to apply to the root node. It is thus possible to add a relative clause to a moved constituent after the movement has taken place, but arguments must be present before movement takes place. Applied to (31b), the relative clause can be merged after Ā-movement, in which case John is not c-commanded by he and Condition C is obeyed. The ungrammaticality of (31a) follows because argument clauses must be present in the launching site, creating a Condition C violation. Lebeaux derives this difference between adjuncts and arguments from the θ-criterion; Fox 1999 suggests a type-theoretic account.

In Thoms 2019’s and Thoms & Heycock 2022’s system, the late addition of adjuncts cannot be dependent on external remerge of DP because Ā-movement does not have access to a derivation that late merges the DP. Thoms & Heycock do not integrate adjuncts into their system, but there are a few options. One is to maintain Lebeaux’s account: adjuncts can be countercyclically added to the moved constituent; that is, adjunction is not subject to the Root Condition (27).¹⁰ An alternative is to assume that adjuncts are added to the DP shell. This permits external remerge of a DP with a relative clause, thus obviating Condition C effects with respect to R-expressions within this relative clause. This is shown in (32), with linear order not represented.

Note that this analysis requires that relative clauses can attach after merging D—that is, that D and the NP may form a constituent that excludes the relative clause. Structures of this kind are also adopted by Hunter 2015 and Safir 2019; a semantics for them is proposed by Bach & Cooper 1978 and more recently Charlow 2020. For the sake of concreteness, we will assume in what follows that Condition C obviation with adjuncts in English Ā-movement is the result of the derivation in (32), though nothing crucial hinges on this.

5.1 Condition C connectivity

Recall from section 4.1 that scrambling induces Condition C connectivity with possessors:

While Thoms & Heycock do not discuss the status of possessors in their system, their analysis extends to (33) rather straightforwardly if (i) Hindi scrambling does not allow late merge of the DP layer and (ii) possessors are introduced below the DP layer.

As for (i), recall that on Thoms & Heycock’s account, DP late merge is possible only if the movement feeds case assignment. A general fact about scrambling in Hindi is that it does not affect a DP’s case (see Keine 2018). In other words, the case a scrambled DP bears is the same as the case it would bear had scrambling not taken place. This is illustrated in (34) and (35). The examples in (34) show that the object Ram must bear accusative case -ko, regardless of whether scrambling takes place. An analogous observation is made in (35) for the object of the verb milaa ‘meet,’ which bears instrumental case.

Such case connectivity provides clear evidence that a DP’s case feature is determined before scrambling takes place; equivalently, that scrambling takes place after case is assigned. In conjunction with the DP Case Filter in (30), this entails that scrambling requires the DP layer to be present before scrambling applies, as the DP layer would otherwise remain caseless.¹¹

As for (ii)—the position of the possessor DP—we assume that possessors are introduced in a DP-internal PossP projection (see, e.g., Szabolcsi 1983, Kayne 1993, and Safir 1999 for arguments that possessors originate below D). Because the DP layer dominates the PossP layer, the Root Condition (27) requires that the PossP layer must be merged before the DP layer. Given that the DP layer must be present before scrambling applies (as just established), it follows that the PossP layer must be as well. The morphology of the genitive case marker is consistent with possessors being introduced below the locus of case. The genitive marker agrees in number and gender with the container DP’s head noun, and importantly it appears in an oblique form if the container DP is overtly case marked. Thus, in (36a), where the container DP bears unmarked case, the genitive case marker (with a masculine singular head noun like ‘son’) takes the form -kaa; by contrast, in (36b) the container DP bears a case marker, and the genitive marker of the possessor takes the oblique form -ke.

While the precise mechanism that underlies this case concord deserves further study, we interpret the facts in (36) as indicating that the possessor DP appears below the D head that contains the case information of the container DP, as shown in (37). The oblique form of the genitive marker then appears if it occurs in the domain of a D with certain case features.

In conjunction with these assumptions about scrambling and the location of the possessor DP, Thoms & Heycock 2022’s account derives the fact that scrambling does not obviation Condition C violations with possessors (33). As we now show, both early merge and late merge of the possessor are ungrammatical in this case. The early-merge structure is shown in (38). Here, the DP layer and the possessor DP Sita-ke ‘Sita-gen’ are merged in the base position. Sita-ke is thus c-commanded by the pronoun, violating Condition C.

The corresponding late-merge derivation is given in (39). This derivation involves merging only the NP in the base position, with external remerge of PossP and DP. In this structure, Condition C is obeyed, but the DP Case Filter (30) is violated, as the late-merged D does not receive a case feature in the landing site of scrambling. The derivation in (39) is therefore ungrammatical as well.

Of course, merging only the DP layer late (with PossP present in the base position) will not converge either, as it will produce a violation of both Condition C and the Case Filter.

Thoms & Heycock 2022’s external-remerge account of Condition C connectivity with arguments can thus be extended to possessors and scrambling. It derives the fact that scrambling does not obviate Condition C effects with possessors from the independently motivated fact that scrambling does not feed case assignment. Thus, scrambling patterns like English Ā-movement with respect to Condition C connectivity precisely because it shares with English Ā-movement its relationship to case: the moving element receives case before the movement applies, prohibiting late merge of DP and PossP.

5.2 SCO

We now turn to the (secondary-) SCO facts in section 2.2 that posed the initial puzzle. The crucial contrast in need of explanation is repeated in (40). Scrambling gives rise to a secondary-SCO effect (40a), but not to a secondary-WCO effect (40b). That is, binding of the pronoun is possible only if the pronoun does not c-command the launching site of scrambling.

As noted in section 4.1, the grammaticality contrast in (40) is clearly analogous to Condition C connectivity, which likewise arises only if a coindexed pronoun c-commands the launching site. We therefore analyze (40a) as a Condition C effect. To establish the connection between SCO and Condition C, we assume that quantified DPs are R-expressions for the purposes of binding theory (e.g., Chomsky 1981: 115–116) and hence subject to Condition C. The structure of (40a) is then analogous to that of (33). The early-merge derivation of (40a) is given in (41). Because har laṛke-kii ‘every boy-gen’ is subject to Condition C, (41) violates Condition C.

As with (33), an alternative derivation in which the DP layer and the possessor DP are late merged (not shown here; see (39) for (33)) violates the DP Case Filter (30).

By contrast, if the pronoun is embedded inside the subject DP and hence does not c-command the lower occurrence of har laṛke-kii ‘every boy-gen,’ as in (40b), Condition C is not violated:

The crucial difference between SCO and WCO is therefore accounted for not by appealing to whether or not binding from the landing site is possible (it is in both cases) but by invoking the structural relationship between the pronoun and the representation of the moved element in the launching site. This account hence not only derives the difference between SCO and WCO in Hindi scrambling but also explains why SCO travels with Condition C connectivity across movement types.

5.3 Late merge of adjuncts

The analysis we propose here attributes the same late-merge options to Hindi scrambling as to English Ā-movement, derived from the fact that both do not feed case assignment. Given that late merge of adjuncts is possible for English Ā-movement (section 4.2), we expect that scrambling patterns the same way. This expectation is borne out. Like English Ā-movement, Hindi scrambling obviates Condition C violations with relative clauses:

The account of (43) is analogous to the treatment of English relative clauses in (32): the relative clause can late merge onto the DP, resulting in (44). This structure obeys both the Case Filter and Condition C. As before, for typographic reasons (44) does not represent the linear order.

There is some Hindi-internal evidence that relative clauses attach very high, in particular higher than the locus of case. If a DP bears a case marker and a relative clause, the relative clause must follow the case marker, as in (45a), rather than the other way around, as in (45b).

It is possible, of course, that the linear position of the case marker in (45) does not reflect its syntactic position. But to the extent that it does, the ordering in (45a) indicates that relative clauses are merged above the locus of case (D, on our account). It is this high attachment site that enables the external-remerge derivation in (44).

Finally, given that adjuncts may merge late, it is crucial for our account that possessors are not adjuncts. This view is entirely standard. While possessors are typically not arguments of the head noun, they are introduced as specifiers (and hence arguments) of a functional projection (Abney 1987). See Safir 1999 for an argument that possessors are not adjuncts in English. Thus, possessor DPs cannot be late merged because they are arguments of PossP; PossP in turn cannot be late merged because it is not an adjunct but a projection in the nominal spine. In combination with the fact that they are generated below D, this derives the fact that possessors do not have access to a late-merge derivation with scrambling.

5.4 Extension to reciprocal binding

The account proposed here also allows us to make sense of another asymmetry, which arises with reciprocal pronouns and scrambling.

As (46) shows, the reciprocal pronoun ek duusre can appear either directly as an argument of the verb or as a possessor. In both cases, it must be bound by a c-commanding antecedent (in (46), the subject).

Scrambling may feed binding of a reciprocal pronoun inside the subject (Jones 1993: 80, Bhatt & Dayal 2007: 289, Bhatt 2016: 515, Keine 2018: 6), but it may not do so if the reciprocal pronoun is itself the subject (A. Kidwai 2000: 5, Bhatt 2016: 515, fn. 4). That is, we observe contrasts like (47). In (47a), the reciprocal pronoun is the possessor of the subject DP ek duusre-kii maaõ-ne ‘each other’s mothers-erg.’ The object is scrambled over this subject, which enables binding of the reciprocal by the object (the sentence is ungrammatical without scrambling). By contrast, in (47b), the reciprocal is itself the subject. Here, scrambling of the object does not enable binding of the reciprocal, and the sentence is hence ungrammatical (as it is if no scrambling takes place).

Conflicting conclusions have been drawn from the data points in (47). On the one hand, Bhatt & Dayal 2007 and Bhatt 2016 conclude from (47a) that scrambling lands in an A-position, which enables binding. On the other hand, A. Kidwai 2000 concludes from (47b) that scrambling does not land in an A-position as otherwise binding should be possible. It seems clear, then, that focusing exclusively on the properties of the landing site of scrambling will not provide an account of (47), simply because the landing site is the same in (47a) and (47b).

We propose instead that the contrast in (47) follows from the properties of the launching site, as a Condition C effect. The structure for the grammatical baseline case (47a) is given in (48). Because, by assumption, scrambling targets an A-position, the scrambled object may bind the reciprocal pronoun inside the subject (Bhatt & Dayal 2007, Bhatt 2016, Keine 2018).

The structure for the ungrammatical (47b) is provided in (49). Here, the reciprocal pronoun c-commands the launching site, inducing a Condition C effect.

Our analysis of SCO in Hindi thus extends to the reciprocal contrast in (47). The resulting account offers, then, a unified explanation of the empirical patterns of (i) SCO, (ii) Condition C connectivity, and (iii) binding of reciprocal pronouns.

5.5 Scrambling of pronouns

The traditional Chomsky 1981 account of SCO postulates that movement that is subject to SCO leaves behind a kind of trace that is subject to Condition C (a so-called “variable”). While the account proposed here likewise attributes SCO to Condition C, it fundamentally differs from Chomsky’s in that we do not assume that such movement leaves behind a special silent element. Instead, Condition C connectivity arises with respect to the lower occurrence of the moved element. In addition to conceptual advantages (in particular adherence to the Inclusiveness Condition: see Chomsky 1995), we also saw in section 4.2 an empirical argument against a trace-based account: a copy-theoretic or multidominance-based account derives secondary SCO because the occurrence in the launching site contains information about the internal structure of the moved expression, while a trace would not.

In this section, we briefly investigate another distinctive prediction of the account proposed here. The prediction arises for cases in which what is scrambled is a pronoun.¹² On a trace-based approach, the movement should still leave behind a variable and hence display Condition C effects with respect to higher pronouns. By contrast, the account developed here predicts that scrambling of a pronoun does not give rise to a Condition C effect because the occurrence in the launching site remains a pronoun and is hence not subject to Condition C. In other words, the nature of the scrambled expression should affect whether Condition C obtains or not.

As (50) shows, this prediction is borne out. The baseline structure in (50a) shows that in this construction the matrix subject may corefer with a pronoun inside the nonfinite clause but not with an R-expression—a standard Condition C effect. Scrambling of the object does not alter the coindexation options, as (50b, c) show.

The fact that Condition C connectivity obtains in (50c) but not in (50b) is quite puzzling on a trace-based account of Condition C connectivity (as noted by Barss 1988 and Büring 2005; see footnote 12). Because both (50b) and (50c) involve scrambling, the trace left behind would be identical. If the trace is subject to Condition C, the coindexed pronoun us-ne ‘he-erg’ should in both cases result in a Condition C violation and hence ungrammaticality; if the trace is not subject to Condition C, then both cases should be grammatical. By contrast, the account proposed here immediately derives this contrast: the occurrence in the launching site corresponds to the moving element, and so, as schematized in (51), it is subject to Condition C only if the moving element is an R-expression.

While it is not possible to conduct this sort of test for SCO (given that testing for SCO requires binding from the landing site, which in turns requires that the scrambled element is not a pronoun), we take the contrast between (50b) and (50c) to be strong evidence for a copy-theoretic or multidominance-based approach to Condition C effects in scrambling.