2.1 Clause-internal movement

Clause-internal movement in Hindi exhibits A-properties (Déprez 1989; Mahajan 1990; 1994; Gurtu 1992; Kidwai 2000). This is illustrated with weak crossover obviation in (10), where movement of the object har bacce-ko ‘every child-acc’ over the subject us-kii mãã-ne ‘his/her mother-ERG’ enables binding of a subject-internal pronoun.⁴

The same conclusion can be reached on the basis of reciprocal binding (Bhatt & Dayal 2007). The reciprocal pronoun ek-duusre ‘each other’ requires a c-commanding antecedent in an A-position. Clause-internal movement may provide such an antecedent (11).

Clause-internal movement can thus be A-movement in Hindi.⁵⁶ Whether it can also be Ā-movement is difficult to determine with certainty, but I will assume that it can be. First, as a reviewer mentions, clause-internal scrambling can reconstruct for variable binding:

If A-movement resists such reconstruction, then (12) provides evidence that clause-internal movement may also be Ā-movement in Hindi (see Dayal 1994a: 241 and Bhatt 2003: 12). Interpreting (12) is made more difficult by mounting evidence that A-movement may in principle reconstruct (Barss 1986; Romero 1997; Fox 2000; Sportiche 2006; Lebeaux 2009). If so, then (12) is uninformative as to the movement type involved.

One other difference between A- and Ā-movement is that only Ā-movement is able to license parasitic gaps (Engdahl 1983). It is difficult to identify parasitic gaps in Hindi due to its free argument drop, but Manetta (2016) presents evidence in favor of parasitic gaps in Hindi, and she furthermore shows that such parasitic gaps can be licensed by clause-internal movement. If this is the case, then clause-internal movement not only exhibits A-properties, but Ā-properties as well. Following the line of analysis in Mahajan (1990; 1994), I will assume that clause-internal movement in Hindi can in principle be either A- or Ā-movement; that is, it can target an A- or an Ā-position. On this account, examples like (10b) and (11b) are grammatical because this movement allows an A-movement parse, and hence no crossover violation results. Parasitic gaps are allowed because the movement dependency also allows an Ā-movement parse.

2.2 Movement out of nonfinite clauses

Movement out of nonfinite clauses has received less designated attention in the literature on Hindi, but seems to exhibit the same properties as clause-internal movement. It is not subject to weak crossover (13) (Keine 2016; to appear), and it may lead to binding of reciprocal pronouns (14). This makes it clear that this movement can be A-movement, and that nonfinite clauses are transparent for A-movement out of them. For notational convenience, I will label nonfinite clauses in Hindi as TPs, a claim that will be justified in section 4.1.

2.3 Movement out of finite clauses

Movement out of a finite clause contrasts with the configurations presented so far in that it does not display any A-properties (Déprez 1989; Mahajan 1990; 1994; Gurtu 1992). The example in (15) demonstrates that extraction out of a finite clause is possible, but it does not enable binding of the subject-internal pronoun us-kii ‘his/her’ by the moved object har bacce-ko ‘every child-ACC’. Unlike movement out of nonfinite clauses, then, extraction out of a finite clause is subject to weak crossover and hence invariably Ā-movement. I will notate finite clauses as CPs, as will be justified in section 4.1.

Reciprocal binding supports this conclusion. In (16), movement of raam aur prataap-ko ‘Ram and Pratap-ACC’ cannot result in binding of the reciprocal. This also implies that the movement is necessarily Ā-movement.

In sum, the locality facts in Hindi are identical to English:⁷

A unified account of Hindi and English is therefore highly desirable. The question is whether a case-based or a position-based approach affords such a unified account. To address this question, the next two sections will investigate the case and positional properties of A- and Ā-movement in Hindi. On a case-based account, (17) should be correlated with the distribution of case, whereas a position-based account predicts that (17) should mirror the positions that are involved. The next section will investigate the distribution of case in the relevant constructions and argue that it does not correlate with the locality facts. A case-based account therefore fails to extend to Hindi and hence misses a generalization. Section 4 will then argue that positional properties do correlate with the locality facts, and that this provides the desired unification with English.

3.1 Case connectivity

Crossclausal movement in Hindi can affect elements with a variety of different structural or lexical cases, but such movement must preserve the case of the moving element. This is illustrated in (18). As (18a) shows, a proper name in direct object position must be marked with the accusative case marker -ko; other cases are impossible. Significantly, this case marking must be preserved whether the element is moved out of a nonfinite clause (18b) or out of a finite clause (18c).

The same holds for lexical cases like the instrumental -se. In (19), the verb mil ‘meet’ assigns instrumental case to its object and this case has to be preserved under crossclausal movement.⁸

This restriction generalizes to other cases. Furthermore, all case connectivity facts discussed here also hold if the movement is unambiguously A-movement. Two examples are provided in (20), where A-movement out of a nonfinite clause takes place (as evidenced by reciprocal binding) and again case connectivity is obligatory. This holds for structural as well as lexical cases.

Case connectivity provides evidence that the case of the moving element is assigned in the embedded clause, i.e., before crossclausal movement takes place. The fact that case connectivity is obligatory then implies that case must be assigned before such movement takes place. This conclusion holds for extraction out of both finite and nonfinite clauses and for both A- and Ā-movement.

3.2 Possessor movement

Strong independent support for the conclusion in the preceding section comes from left-branch extraction of possessors out of their host DPs. Such movement is possible in Hindi, and it may be A- or Ā-movement, but it is subject to the locality constraints demonstrated in section 2, as (21) and (22) show. (21) demonstrates that a possessor may be A-extracted out of a nonfinite clause, illustrated with weak crossover obviation in (21a) and with reciprocal binding in (21b).

Possessor extraction out of a finite clause is possible, but it must be Ā-movement, as shown in (22). It is subject to weak crossover (22a) and may not lead to reciprocal binding (22b).

Possessor extraction thus patterns exactly like extraction of verbal arguments with respect to the locality of A- and Ā-movement. Importantly, possessor movement exhibits obligatory case connectivity: moved possessors must retain their genitive case, whether they are moved out of nonfinite clauses (23a) or finite clauses (23b):

As Bhatt (2005) emphasizes, genitive case in Hindi is only assigned by nominal structure.⁹ It therefore must be assigned to the possessor in (23) inside the container DP in the lower clause, i.e., before the left-branch extraction—and hence crossclausal movement—takes place.

Further evidence for this conclusion comes from genitive agreement. In Hindi, the morphological form of the genitive marker agrees in φ-features with the head noun of the container DP. This is illustrated in (24), using the two nearly synonymous words khat ‘letter’ and ciṭṭhii ‘letter’. The two words differ in their gender: while khat is masculine, ciṭṭhii is feminine. In (24a), the head noun is khat and the genitive marker on the possessor consequently bears the masculine singular form -kaa. In (24b), by contrast, the head noun is ciṭṭhii and the form of the genitive marker is correspondingly the feminine singular form -kii.

Crucially, this agreement persists under crossclausal movement of the possessor, irrespective of whether the movement leaves a nonfinite clause (25) or a finite clause (26):

Obligatory genitive agreement also obtains for possessor movement that is unambiguously A-movement. For agreement between the genitive marker and the head of the container DP (i.e., khat or ciṭṭhii in (25) and (26)) to be established, genitive case must be assigned inside the container DP, hence before left-branch extraction and crossclausal movement take place. The fact that agreement is obligatory implies that genitive case assignment in fact must take place in the container DP, and hence before movement. This holds for movement out of finite and nonfinite clauses alike.

3.3 The challenge for case-based accounts

The previous sections have presented converging evidence for the conclusion in (27):¹⁰

Combining the locality facts discussed in section 2 with the generalization in (27) yields the picture in (28): Both A- and Ā-movement are possible out of nonfinite clauses, but only Ā-movement may leave a finite clause, and in both configurations, case is obligatorily assigned to the moving element before the crossclausal extraction step takes place.

This state of affairs is problematic for an account that attempts to attribute the availability of A-movement to the Activity Condition or to considerations of case more generally. The Activity Condition (6) prohibits A-movement of a DP after the DP has been assigned case. This would incorrectly preclude A-movement in both (28a) and (28b) and hence predict that all crossclausal movement in Hindi must be Ā-movement. The Activity Condition is thus overly restrictive.

One might then explore the view that the Activity Condition for some reason does not apply to Hindi, perhaps along the lines of recent work on Bantu where it has been argued that case never deactivates a DP or that the relevant languages simply lack case altogether (see Carstens 2010; 2011; Obata 2010; Obata & Epstein 2011; Diercks 2012; Carstens & Diercks 2013). But such an account would be too unrestrictive for Hindi, as it would leave a DP active for A-processes indefinitely. It is then the ban on A-movement out of finite clauses that would remain unaccounted for (precisely the configuration that is grammatical in Bantu).

In general, because the case properties of the moving element are identical in (28a) and (28b), considerations of case fail to distinguish between the two configurations. The observation that A-movement is possible in (28a) but not in (28b) therefore cannot be attributed to case. This indicates that it is some other factor that regulates the possibility of A-movement in the two configurations.

Before moving on to position-based accounts of hyperraising, it is instructive to consider a more abstract version of the Activity Condition, according to which Activity of a DP is determined on the basis of “abstract Case”. For instance, dative subjects in Icelandic undergo A-movement to Spec,TP despite their lexical-dative case marking (see, e.g., Sigurðsson 2004 and references there). While Nevins (2005) takes this as evidence against the Activity Condition, another line of response, explored by Bejar & Massam (1999), Chomsky (2000; 2001), Bošković (2002), and Richards (2008), among others, is that dative subjects additionally bear an abstract structural Case feature and are correspondingly assigned nominative Case in Spec,TP.¹¹ The apparent violation of the Activity Condition is then resolved. Let us consider how such an account fares with respect to the Hindi data discussed here.

One general concern about such an approach is that it severs the distribution of abstract Case from any morphological manifestation of case. Recall that possessors may be A-moved out of their host DPs, as in (21). On the abstract-Case account, this fact would entail that a possessor’s abstract Case does not have to be assigned within the host DP, but that it can be assigned in a higher clause, hence after possessor extraction has taken place. The evidence in sections 3.1 and 3.2—which argued that case in Hindi is invariably assigned before crossclausal movement takes place—would then entail that the distribution of abstract Case is unrelated to that of morphological case. This would remove much of the predictive content of a case-based account. Unless there is some independent way of diagnosing the distribution of abstract Case, abstract Case conceived of in this way runs the risk of becoming an analytical diacritic whose sole purpose is to constrain the effects of the Activity Condition.

Second, the Hindi data also present an interesting empirical challenge for the abstract-Case account. Let us suppose, following the logic of the abstract-Case account, that possessor A-movement out of a nonfinite clauses such as in (21) indicates that abstract-Case assignment to the possessor can indeed be deferred until it reaches the higher clause. A problem arises because possessors are, of course, optional in Hindi, i.e., DPs may or may not contain a possessor DP. This in turn means that the purported abstract-Case assigner in the higher clause in (21) must be optional as well, because accounts involving abstract Case typically invoke the Inverse Case Filter, the requirement that a case feature on a functional head be assigned if it is present (e.g., Bošković 2002)—a requirement that may itself follow from the general need of uninterpretable features to be deleted by the end of the derivation (Richards 2008). But if the presence of the abstract-Case assigner is optional, this opens up the structural possibility in (29), which involves embedding of a finite clause. Here the possessor DP (Poss-DP) does not receive abstract Case in its containing DP and the embedded finite clause does not contain an abstract-Case assigner for it either; rather, the Case assigner is located in the higher finite clause (indicated as ‘H⁰’ in (29)). In this case, the possessor’s abstract-Case feature remains unvalued until it enters the matrix clause. As far as Activity is concerned, then, A-movement into the matrix clause should be possible in (29). However, as we have seen on the basis of (22), this is not the case: only Ā-movement out of a finite clause is allowed.

The general problem here is that the very mechanism that allows for a delay in valuing a DP’s abstract Case if the embedded clause is nonfinite, hence allowing A-movement of possessors out of nonfinite clauses (as in (21)), would also allow such a delay if the embedded clause is finite, incorrectly allowing for ungrammatical A-movement out of finite clauses (as in (22)). In the absence of a principled explanation for why such a delay is possible in one configuration but not the other, it is hence difficult to see how the locality facts could be derived from abstract Case either.¹² This again points to the conclusion that it must be some factor other than case that underlies the ban on A-movement out of finite clauses (28).

The discussion so far has focused on the Activity Condition, which stipulates a direct link between case and the ability to undergo A-movement. The challenge posed by the Hindi data also applies to approaches that link a DP’s ability to undergo A-movement to its case properties only indirectly. One example which was mentioned in section 1 is Obata (2010) and Obata & Epstein (2011), who develop an account on which case assignment to a DP has the consequence that the copy of that DP at the phase edge is devoid of φ-features, which makes it invisible to a higher A-probe. Applied to the structures in (28), this approach does not make the desired cut, again because case assignment should have the same effect in finite and nonfinite clauses.

Another recent approach that ties together A-movement and case indirectly is Safir (to appear). In the context of a novel account of the A/Ā-distinction, Safir (to appear) proposes a case-based account of scrambling in Japanese. He focuses on the fact that, like in Hindi, local scrambling in Japanese can be A-movement, but scrambling out of finite clauses must be Ā-movement. His account is based on the assumption that a DP scrambled out of a lower clause must be “insulated” in order to shield it from double case assignment from the matrix v head. Insulated DPs are then unable to move to A-positions. While Safir (to appear) does not discuss scrambling out of nonfinite clauses, this account does not seem to carry over to the Hindi data presented here, because shielding from matrix v—and hence insulation—would also be required if the embedded clause is nonfinite. This would block A-movement out of nonfinite clauses as well. The crucial contrast in (28) would therefore remain underived.

In sum, I have argued that the locality contrast between A- and Ā-movement with respect to finite and nonfinite clauses in Hindi does not correlate with the distribution of case in the language. As such, a case-based account of the locality of A-movement—be it direct or indirect—does not generalize to the Hindi facts. Note that this conclusion of course does not, in and of itself, invalidate case-based accounts of hyperraising in English. But it does show that there are instances of A-movement that display the same locality as in English—hence suggesting a unified treatment—, but which nonetheless fall outside the scope of a case-based approach. This indicates that a generalization is being missed on such an account.

The next section will argue that a position-based account does not encounter similar challenges to a case-based and that it makes available a principled account of the Hindi locality facts. A position-based account furthermore has the attractive consequence that it affords a unified analysis of Hindi and English. As such, it provides an explanation for why the locality properties of A- and Ā-movement are identical.

4.1 The size of embedded clauses

There is good evidence that in Hindi nonfinite clauses are structurally smaller than finite clauses in that finite clauses are CPs whereas nonfinite clauses lack a CP layer, and this is indeed a common assumption in the relevant literature (see, e.g., Dayal 1996; Bhatt 2005; Chandra 2007; and Keine to appear). First, finite embedded clauses may contain the complementizer ki (30), but nonfinite clauses may not (31). Hindi has no equivalent of the English nonfinite complementizer for.

Second, finite clauses in Hindi provide an interrogative scope position, but nonfinite clauses do not (Mahajan 1990; Dayal 1994b; 1996). In (32), the wh-element kyaa ‘what’ must take wh-scope within the embedded finite clause.¹³ By contrast, in (33), a wh-element inside a nonfinite clause can only take matrix scope, and therefore an embedded-question reading is not available.

On the standard assumption that interrogative scope is associated with C, its obligatory absence in nonfinite clauses like (33) follows straightforwardly on the assumption that these clauses lack a CP projection.

Evidence from the distribution of complementizers and wh-scope thus converges on the conclusion in (34). For the sake of concreteness, I will treat nonfinite clauses as TPs, but their exact structural size is not relevant for our present purposes as long as they are structurally smaller than finite clauses and lack a CP layer.¹⁴

4.2 The landing sites of A- and Ā-movement

Due to the head-final phrase structure and the optionality of movement in Hindi, determining the landing sites of A- and Ā-movement is difficult. Mahajan (1990) assumes that A- and Ā-movement in Hindi target different positions (L-related and non-L-related, respectively), but for theory-internal reasons. Keine (to appear) is, to my knowledge, the first attempt to provide empirical evidence for the positions involved. His arguments are indirect and based on the size of finite and nonfinite clauses in (34). Keine (to appear) argues that Ā-movement in Hindi lands in Spec,CP, whereas A-movement lands in a TP-internal position. I will present these arguments here and provide novel support for their conclusion.

Let us first consider the landing site of A-movement. Keine (to appear: (27)) presents the example in (35), in which the embedded nonfinite clause is extraposed to the right of the matrix verb in order to demarcate its left edge (see Bhatt & Dayal 2007 for evidence that auxiliaries in Hindi do not move leftward, i.e., all elements following an auxiliary are extraposed). The embedded direct object har laṛkii-ko ‘every girl-ACC’ undergoes movement over the adjunct us-kii shaadii ke dauraan ‘during her wedding’. Crucially, the moved element is able to bind the pronoun us-kii ‘her’ inside the adjunct from its landing site, making it clear that this movement step is A-movement. The extraposition of the nonfinite clause ensures that har laṛkii-ko is not extracted out of the nonfinite clause itself: if it were, it would surface to the left of the matrix verb.¹⁵ Consequently, the landing site of har laṛkii-ko in (35) must be inside the nonfinite clause. What (35) shows, then, is that it is possible for A-movement to target a position inside a nonfinite clause. The sentence in (36) provides a parallel example.

In light of the evidence that nonfinite clauses in Hindi invariably lack a CP layer (see section 4.1), Keine (to appear) concludes from these considerations that Hindi A-movement does not land in Spec,CP (as the nonfinite clauses in (35) and (36) lack this position). Rather, A-movement must target a TP-internal position.

Converging evidence that is not considered by Keine (to appear) comes from nonfinite clauses in subject position. Such clauses are islands for extraction out of them, as shown in (37), where extraction of phal ‘fruit’ out of the subject clause in (37b) creates ungrammaticality.¹⁶

Against this background, A-movement within the subject clause (diagnosed again by the absence of weak crossover) is well-formed, as shown by (38) and (39). Because extraction of har laṛkii-ko ‘every girl-ACC’ and sab-se ‘everyone-INSTR’ out of the nonfinite subject clauses in (38) and (39), respectively, would induce an island violation, the A-movement step in the two structures must target a position inside the nonfinite subject clause.

This observation provides clear support for the conclusion that A-movement must be able to land inside a TP clause. As a result, A-movement must target a TP-internal position.

Let us now turn to the landing site of Ā-movement. Keine (to appear) argues that Ā-movement lands in a TP-external position, in direct contrast to A-movement. The argument is based on paradigms like (40). All sentences in (40) involve a double embedding structure in which a finite clause is embedded inside a nonfinite clause that is in turn embedded inside a finite matrix clause. As in (35), the nonfinite clause is extraposed to demarcate its left edge. (40a) is the baseline structure. (40b) is then derived by moving the DP kitaab ‘book’ from the lowermost clause into the intermediate nonfinite clause. As shown, the result is ungrammatical. In (40c), on the other hand, the same DP kitaab is moved into the matrix clause and the resulting sentence is grammatical.

Keine (to appear) reasons as follows: The movement of kitaab in both (40b) and (40c) proceeds out of a finite clause and hence must be Ā-movement in both cases (given that finite clauses allow only Ā-movement out of them). The ungrammaticality of (40b) then demonstrates that Ā-movement cannot land inside a nonfinite clause. (40c) serves as a control: here kitaab is moved into the highest finite clause and the result is grammatical. This demonstrates that it is not Ā-movement of kitaab itself that underlies the problem in (40b). Rather, the ungrammaticality of (40b) must be due to the fact that here the Ā-movement lands in a nonfinite clause.

Keine (to appear) proposes that the paradigm in (40) receives a principled explanation on the assumption that Ā-movement targets Spec,CP. Because nonfinite clauses obligatorily lack a CP layer in Hindi (recall (34)), they simply lack the functional structure necessary to provide a landing site for Ā-movement. The otherwise puzzling ungrammaticality of (40b) then follows, as does the contrast to (35) and (36).

I will now present novel evidence that corroborates this conclusion. First, the restriction exemplified by (40) holds more generally, e.g., independently of the case marking of the moving element. (41) shows that the same pattern obtains for overtly case-marked objects like raam-ko ‘Ram-ACC’:

Similarly, (42) shows that the same restriction also holds for movement of locative phrases like dillii-mẽ ‘in Delhi’:¹⁷

The generality of the restriction is precisely what is expected if it is a manifestation of a general constraint on where Ā-movement may land in Hindi.

Furthermore, there is no general ban against moving into intermediate clauses. If the intermediate clause is finite, hence a CP, Ā-movement into it is well-formed:

The contrast between (42) and (43) receives an immediate account if Ā-movement lands in Spec,CP.

Subject clauses provide further novel evidence for this conclusion. In (44), a nonfinite subject clause contains a finite complement clause and the embedded object drugs is Ā-extracted out of the finite clause. In light of the fact that subject clauses are islands in Hindi (recall (37)), it follows that the landing site of drugs in (44) must be located inside the nonfinite clause. The ungrammaticality of the resulting structure can be given the same explanation as the ungrammaticality of (40b) above: Ā-movement must land in Spec,CP, a position that nonfinite clauses lack.

In sum, these considerations provide strong evidence that Ā-movement in Hindi lands in a position higher than the position targeted by A-movement in that only A-movement targets a position low enough to land in a nonfinite clause:

The structural conclusions in (45) make an immediate prediction not noted by Keine (to appear). Recall from the discussion in section 2 that unlike finite clauses, nonfinite clauses do allow A-movement out of them in Hindi. If A-movement lands in a TP-internal position, we expect movement out of a nonfinite clause to be able to land inside another nonfinite clause, in direct contrast to what we saw for movement out of finite clauses in (40b), (41b), and (42a). The paradigm in (46) shows that this expectation is indeed borne out.¹⁸ This paradigm is analogous to the one in (40), the only relevant difference being that the innermost clause is nonfinite in (46). This nonfinite clause is embedded inside another nonfinite clause, which is itself embedded in a matrix clause. (46a) constitutes the baseline. In (46b), the object daal ‘lentils’ of the innermost finite clause is moved into the intermediate nonfinite clause and the resulting sentence is grammatical. In (46c), daal is moved into the matrix clause and the result is likewise grammatical.

The crucial contrast is between (40b)/(41b)/(42a) on the one hand and (46b) on the other. The former groups of examples demonstrates that movement out of a finite clause cannot land inside a nonfinite clause; (46b) shows that movement out of a nonfinite clause is not restricted in this way. This contrast is accounted for without further ado on the positional conclusions in (45). Because movement out of a finite clause is invariably Ā-movement, it must target Spec,CP, a position that nonfinite clauses lack. By contrast, movement out of a nonfinite clause may be A-movement and it may hence target a TP-internal position, which nonfinite clauses provide. The contrast between (40b)/(41b)/(42a) and (46b) therefore neatly converges with the conclusions in (45).

Note, incidentally, that a case-based approach does not lend itself to an account of the contrast between (40b)/(41b)/(42a) and (46b). This is because, as discussed in section 3, movement out finite and nonfinite clauses do not differ with respect to case. The differences in the availability of movement into a nonfinite clause in (40b)/(41b) and (46b) therefore cannot be due to case, further supporting the conclusion that the locality facts are conditioned by something other than case (viz. positions).

The conclusion that Ā-movement lands in Spec,CP in Hindi raises an interesting question with respect to the landing site of movement out of a finite clause with respect to the matrix subject.¹⁹ In many cases, such movement can land either above or below the subject (Déprez 1989: 134; Mahajan 1990: 38; Bhatt 2003: 18; Dayal 2017: 161), as illustrated in (47):

Because the movement in (47) leaves a finite clause, it must be an instance of Ā-movement. If Ā-movement lands in Spec,CP, then this poses the question of how it is able to target a position below the subject in (47b). I would like to suggest that (47b) is produced by movement of is kitaab-ko ‘this book-ACC’ to Spec,CP—hence above the matrix subject position—, followed by Ā-movement of siitaa-ne ‘Sita-ERG’ to a second, outer Spec,CP, as shown in (48).

On this view, (47b) involves Ā-movement of both the embedded object and the matrix subject, and hence a more complex syntactic derivation than (47a). Significantly, Bhatt (2003: 18) and Dayal (2017: 167–168) point out that crossclausal movement that lands below the subject, such as (47b), is usually degraded compared to movement that lands above the subject, and sometimes altogether ungrammatical.²⁰ This contrast is consistent with the claim that these structures involve a more complicated syntax.

Having thus narrowed down the positions involved in A- and Ā-movement, the next section will show how a position-based account directly extends to Hindi.

4.3 A position-based account of A-movement locality in Hindi

Recall from section 2 the crucial locality contrast between finite and nonfinite clauses in Hindi that we are seeking to account for: Finite clauses allow Ā-movement out of them, but block A-movement; nonfinite clauses are transparent to both (see (17)). The previous sections have presented independent evidence that allows us to pinpoint the positions that are involved in the relevant movement dependencies. I argued on the basis of this evidence that (i) finite clauses in Hindi are CPs, whereas nonfinite clauses lack a CP layer, i.e., they are TPs (see (34)), and (ii) A-movement in Hindi lands in a TP-internal position, whereas Ā-movement lands in Spec,CP (see (45)).

Viewed through the lens of the positions involved, the locality facts can now be reformulated as follows: Movement out of a nonfinite (i.e., TP) clause can land in an A-position (i.e., Spec,TP or lower) or in an Ā-position (i.e., Spec,CP), as in (49a); movement out of a finite (i.e., CP) clause may only target an Ā-position (i.e., Spec,CP), but not a TP-internal position, as in (49b). Assuming, as is standard, that extraction out of CP requires movement through the edge of that CP, we obtain:

This situation is of course precisely what the Ban on Improper Movement, repeated in (50), predicts. Due to CP being a phase, illicit hyperraising in Hindi would involve movement from Spec,CP to a TP-internal A-position, which is blocked by (50), just as it is in English. By contrast, Ā-extraction out of a CP is possible precisely because such movement lands in Spec,CP of the higher clause. Nonfinite clauses permit A-movement out of them because they lack a CP layer.

The improper-movement account for English thus straightforwardly generalizes to Hindi because it is stated in terms of the positions involved and, as we have seen, the positional properties in Hindi mirror those in English. Note also that (50) is directly supported for Hindi by (40), (41), and (42). The locality facts in section 2 are then derived from the positional facts in sections 4.1 and 4.2 in a systematic manner.

Let us now consider the set of facts that were problematic for a case-based account. In particular, we saw in section 3.2 that Hindi allows left-branch extraction of possessors and that such extraction may be A-movement, but only as long as it does not cross a finite clause boundary (in which case it must be Ā-movement; see (21) and (22)). As argued in section 3, because the genitive case of the possessor is invariably assigned within the container DP—hence before crossclausal movement takes place—, an account in terms of case fails to distinguish between the two configurations. (51) illustrates that a position-based account captures this restriction: once a CP is crossed, a possessor DP (Poss-DP) cannot subsequently reach an A-position, irrespective of considerations of case.²¹

Configurations like (51) provide the clearest argument for a position-based account because they demonstrate that it is movement to Spec,CP, not case assignment, that blocks subsequent A-movement.

The crucial property of (50) is that it draws a distinction between different types of positions. As noted in section 1, there are a number of proposal as to how types of positions and (50) can be characterized theoretically, and the matter is still under active investigation. Generally speaking, any characterization that has (50) as a result yields the desired outcome. But it is nonetheless instructive to briefly review some of the analytical options that are consistent with (50).

The classical way of differentiating between A- and Ā-positions for the purposes of (50) is to distinguish between the type of trace left behind by movement into them (May 1979; Chomsky 1981; Müller & Sternefeld 1993). Another possibility is to invoke an ordering relationship between movement into them (Abels 2007; 2009). Yet another possibility is to distinguish them in terms of their structural height (Williams 2003; 2013; Müller 2014a; b). Williams (2003; 2013) develops an intricate theory of locality that derives the condition in (52) as one of its consequences. In Williams’ system, this condition is not stipulated, but derives from a novel way of building syntactic structure. In the interest of space, I will not lay out the framework that gives rise to it here; see Hornstein & Nevins (2005) for an overview.

(52) blocks movement from a position α to a position β in another clause if α is “higher” in the abstract clausal spine than β. This is illustrated in (53). For example, movement from Spec,CP is not allowed to land in a projection lower than CP in the clausal spine, i.e., it is blocked from targeting Spec,VP, Spec,vP, or Spec,TP, whereas movement from Spec,TP may target Spec,TP or Spec,CP in a higher clause. On this account, the crucial distinguishing property of A- and Ā-positions is that Ā-positions are structurally higher than A-positions.

(53)	Effects of the Generalized Ban on Improper Movement (52)
	a.		b.

The effects of (52) are more stringent than those of the traditional Ban on Improper Movement in (50), which they comprise as a special case.²² I will not discuss here the relative merits and challenges of (52), for which see, e.g., Abels (2007), Müller (2014a), and Keine (2016). What matters for our concerns is that (52) likewise generalizes to the Hindi facts because A-movement demonstrably lands in a position lower than CP. Assuming, as before, that extraction out of CP proceeds through Spec,CP, A-movement out of a finite clause would therefore require a movement step that (52) blocks. Again, the positional properties discussed in the previous section immediately entail the locality facts.

One potentially problematic property shared by both the traditional and the Generalized Ban on Improper Movement is that they not only rule out movement from CP to TP, but also movement from CP to vP. This restriction is at odds with the widely held view that vP is a phase and thus it deserves consideration.²³ In recent work, Müller (2014a; b) suggests that it is possible to reconcile the desired locality consequences of these two principles with vP phases if these principles are restricted to criterial movement (in the sense of Rizzi 2006). Müller’s (2014a; b) proposal is couched within a framework in which every phrase is a phase, but this assumption is not crucial. As a result of every phrase being a phase, an element moves successive-cyclically through the specifier of every projection on its way to its criterial position. Müller (2014a; b) proposes that a moving item has a buffer, which keeps a record of the projections that this item has traversed. The Generalized Ban on Improper Movement (52) is then implemented as the requirement that an item’s buffer conform to the functional sequence when this item reaches a criterial position. As a result, this account delivers the same empirical effects as (52), but it allows for successive-cyclic movement through clause-internal phase edges. Like Williams’ (2003; 2013) approach, Müller’s (2014a; b) account distinguishes between A- and Ā-positions in terms of their height in the clausal spine. Because this account focuses on the positions involved, it likewise rules out CP-to-TP movement.²⁴

A further possibility might be to characterize the A/Ā-distinction in the domain of interest here in terms of phases. For example, Miyagawa (2010) suggests that A-movement does not cross a phase boundary, whereas Ā-movement may. Combined with the common assumption that Ā-positions are phase-edge positions and A-positions are phase-internal (e.g., Charnavel & Sportiche 2016), a version of the Ban on Improper Movement may then be stated as in (54), which among other things prohibits movement from Spec,CP to Spec,TP in the higher clause.

On the assumption that A-movement does not cross phase boundaries, its application is consistent with (54). The position-based nature of (54) allows it to make the right cut with respect to the Hindi evidence presented here.²⁵

Lastly, one might characterize the distinction between different positions not by making direct reference to different types of positions per se, but rather in terms of their syntactic context. For example, elements in Spec,CP are necessarily separated from higher syntactic material by the CP node that dominates them. If this CP node causes an intervention effect for syntactic operations initiated in the higher clause, it follows that such operations are unable to reach elements in Spec,CP. Versions of this general line of approach have been proposed, in varying forms, by van Urk (2015), Keine (2016; to appear), and Halpert (to appear).

In sum, position-based accounts of the locality of A-movement afford an immediate explanation of the Hindi facts, and this renders them empirically preferable to case-based accounts. Furthermore, because a position-based approach offers a unified account of Hindi and English, it thereby captures the generalization that the locality facts are identical in the two languages because the positions are. I have reviewed various ways in which the critical distinction between the different types of positions may be characterized analytically. To the extent that this conclusion is correct, it provides an argument that a position-based distinction between A- and Ā-movement is empirically necessary.

5.1 Proper improper movement? A look at licit hyperraising

The main line of argumentation here has focused on Hindi, which bans A-movement out of finite clauses. However, there is evidence that this ban is not universal, and that there are languages that do allow hyperraising out of finite or subjunctive clauses. Such instances of licit hyperraising have frequently been analyzed in terms of a case, which raises the questions of whether they are compatible with the key conclusion reached here.²⁶ Examples of languages that allow hyperraising include Greek (Alexiadou & Anagnostopoulou 2002), Brazilian Portuguese (Martins & Nunes 2005; 2010; Nunes 2008; 2010; Ferreira 2009), and several Bantu languages (Zeller 2006; Carstens 2011; Diercks 2012; Carstens & Diercks 2013; Halpert 2015; to appear; Mountjoy-Venning & Diercks 2016). The example in (56) provides an illustrative example from Lubukusu. Here the embedded subject babaandu ‘people’ raises out of the embedded clause into the matrix subject position, where it triggers noun-class agreement on the matrix verb (see the references just cited for extensive discussion and motivation that this dependency indeed involves movement).

As mentioned, the literature that investigates such licit instances of hyperraising commonly adopts a case-based approach and attempts to derive structures like (56) from differences in case and/or Activity. For example, Carstens (2011), Diercks (2012), and Carstens & Diercks (2013) develop a line of analysis according to which the relevant Bantu languages lack syntactic case, and DPs are rendered active through the presence of uninterpretable gender features. Because gender features on a DP are intrinsic and hence not subject to valuation, they render a DP continuously active. As a result, a DP may enter into more than one A-relation in these languages, yielding (among other things) hyperraising as in (56). In a similar vein, licit instances of hyperraising have been attributed to the absence of case assignment in the lower clause by Zeller (2006) for the Bantu language Nguni, by Martins & Nunes (2005; 2010), Ferreira (2009), and Nunes (2008; 2010) for Brazilian Portuguese, and by Alexiadou & Anagnostopoulou (2002) for Greek, among others.

It might thus appear that accounts along these lines provide motivation for a case-based approach to (the ban on) hyperraising. However, closer scrutiny suggests otherwise. Importantly, there is evidence that a position-based account is necessary even for languages that do allow hyperraising. As, e.g., Carstens & Diercks (2013) emphasize for Bantu, not all types of embedded clauses allow hyperraising out of them. To give an example, we saw on the basis of (56) above that Lubukusu allows hyperraising out of clauses headed by the complementizer mbo. As it turns out, not all complementizers pattern in this way. If the embedded clause is headed by the agreeing complementizer -li, hyperraising is impossible, as (57) shows.

A second example comes from Luusamia, where hyperraising is possible only if the embedded clause does not contain a complementizer (58a).²⁷ (58b) shows that in the presence of a complementizer, a reconstructed-scope reading of the matrix subject is impossible, suggesting that the subject did not raise out of the embedded clause. Carstens & Diercks (2013) conclude from this contrast that (58b) involves a structure analogous to copy raising. Hyperraising is hence a possibility only in (58a).

Significantly, there seems to be no indication that clauses that allow hyperraising out of them and those that do not differ internally with respect to case or Activity. A case-based account by itself therefore does not distinguish between them. Carstens & Diercks (2013) suggest that a factor other than case underlies these contrasts, and I adopt this conclusion. They propose that it is not possible to A-move out of a phasal CP, and I will assume that this follows from a position-based constraint that prohibits movement from an Ā- to an A-position or the restriction in (54). For the Luusamia contrast in (58), they propose that complementizer-less finite clauses are in fact just TPs in the language. Extraction out of them therefore does not have to proceed through Spec,CP, and it may thus target an A-position in the higher clause, as schematized in (59a). For the Lubukusu contrast in (56) and (57), Carstens & Diercks (2013) propose an analysis in terms of the extended left periphery: by assumption, the complementizer mbo in (56) realizes a low head in the left periphery, which is not a phase; the complementizer -li in (57), on the other hand, realizes a higher, phasal C head. Extraction over mbo can therefore proceed in one fell swoop, as in (59a), and licitly target an A-position in the higher clause.²⁸ By contrast, extraction over -li in (57) needs to proceed through the CP specifier. Assuming that this is an Ā-position, a position-based account then rules out subsequent movement to an A-position in the higher clause, see (59b).

What the Bantu data above suggest, then, is that a case-based account alone is insufficient and that a position-based account is required even for languages in which hyperraising is in principle able to leave a finite clause. Given this need for a position-based account, we can now ask whether the evidence additionally requires a case-based constraint. As far as I can see, the answer is no. For Bantu, the literature cited above has argued that case or Activity does not limit A-movement in these constructions, which entails that case and Activity only wield their influence in English and other languages that do not allow hyperraising. But as argued above, in these languages, an independently needed position-based account is sufficient to rule out hyperraising structures, rendering superfluous a case-based restriction to the same effect.

If these considerations are on the right track, they suggest a perspective on the contrast between languages that allow hyperraising and those that do not that is rather different than what the previous literature on the topic has assumed. Rather than attributing the contrast to differences in how case assignment and/or Activity of DPs work in these languages, it is possible that what distinguishes languages that allow hyperraising from those that do not is that the former have access to embedded finite clauses that are non-phasal and hence do not require movement through their left edge, as in (59a). Extraction out of such clauses may then target an A-position without violating the Ban on Improper Movement. For the purposes of this ban, these clauses are thus equivalent to TP clauses. By contrast, in languages that do not allow hyperraising—like English and Hindi—, finite clauses are by assumption always phasal and hence require movement through their Ā-edge position, which rules out hyperraising, as in (59b).

If this line of analysis is tenable, it has a number of broader consequences. First, it provides further support for the need for a position-based account, which then renders unnecessary an additional case-based account of hyperraising, supporting the key conclusions I have argued for on the basis of Hindi and English above. Second, it raises the possibility that the crucial feature that distinguishes languages that allow hyperraising from those that do not is unrelated to case after all (contra the literature cited above). Third, this line of analysis has the interesting consequence that it would potentially allow us to maintain the Ban on Improper Movement as a universal principle, in spite of the initial appearance to the contrary.

5.2 Implications for case

In addition to the central conclusion that a position-based account of possible and impossible instances of hyperraising is empirically necessary and that such an account makes a case-based approach unnecessary, the findings here also hold broader consequences for the role of case in syntax. First, they suggest that the A/Ā-distinction cannot be (at least entirely) reduced to case, because constraints on the locality of A-movement cannot exclusively refer to case. A second consequence concerns the relationship between case and movement. While in GB case played a central role in triggering and prohibiting movement, case has been largely abandoned as a trigger for movement, following Chomsky (2000; 2001). At least in the domain of hyperraising, case also does not seem to play a role in prohibiting movement. With the link between case and movement becoming more tenuous, an emerging issue is whether case ever prevents movement. More generally, while case has historically played a fundamental role in constraining syntactic structures and derivations (particularly through nominal licensing), the more recent literature has argued that case and licensing should be severed and that case plays much less of a role in constraining syntax (e.g., Preminger 2014) or even no role at all (e.g., McFadden 2004). The present paper has shown that in one domain in which case has been argued to play a crucial role (namely, constraining A-movement), case does not seem to be the determining factor after all. This conclusion is fully in line with the recent shift in the literature away from case-based explanations in syntax and it removes one potential obstacle to such a shift.

Lastly, the conclusion that A-movement is not constrained by case converges with the conclusion reached by Bhatt (2005) for φ-agreement in Hindi. Based on long-distance agreement across nonfinite clause boundaries in the language, Bhatt (2005) argues that DPs in Hindi can trigger φ-agreement in the matrix clause after they receive case in the embedded clause. Bhatt (2005) concludes that Agree for φ-features is not limited to goals with unvalued case features and therefore is not subject to the Activity Condition. This suggests that the Activity Condition is too strong on multiple grounds, and that the conclusion that the locality of A-movement is not determined by case extends to other A-processes as well.