Not all reconstruction effects are syntactic

Abstract

This paper argues that not all reconstruction effects can be reduced to a syntactic mechanism that selectively interprets copies at LF. The argument is based on the novel observation that some but not all reconstruction effects induce Condition C connectivity in Hindi-Urdu. We contend that Hindi-Urdu requires the hybrid approach to reconstruction developed on independent grounds by Lechner (1998, 2013, 2019), where both copy neglect (a syntactic mechanism) and higher-type traces (a semantic mechanism) are available as independent interpretive mechanisms. We show that the interaction of these two modes of reconstruction derives the intricate reconstruction facts in Hindi-Urdu.

1 Introduction

Moved elements exhibit reconstruction effects—or more neutrally connectivity effects—with their premovement positions see Sportiche 2017 for a recent overview. That is, moved elements may display behavior that we would expect them to display if they had not undergone movement. For example, the A-moved subject in (1) may take scope either above or below the intensional operator . The narrow-scope interpretation in (1b) corresponds to the launching site of movement and thereby the scope that the element would have received if it had not moved.

There are two main lines of approach to reconstruction effects in the literature. The predominant approach since the advent of the Copy Theory of Movement (Chomsky 1993, 1995) is that reconstruction effects are the result of interpreting only the lower copy of the moved element at LF, as schematized in (2b). Assuming the relevant syntactic and semantic constraints apply at LF, then they will only apply to the lower copy. It will thus appear as if the element had not undergone movement, yielding reconstruction effects. Adopting the terminology in Sportiche (2016), we will refer to this procedure as higher-copy neglect.¹

Analyses of reconstruction effects in the copy-theoretic approach have been developed by May (1977, 1985), Cinque (1990), Chomsky (1993, 1995), Heycock (1995), Hornstein (1995), Romero (1997, 1998), Lechner (1998, 2013, 2019), Sauerland (1998, 2004), Fox (1999), Sportiche (2016), and Poole (2017, 2024), amongst many others.

The second line of approach is in terms of higher-type traces. This approach extends the standard analysis of the surface reading (which both approaches assume): interpret the moved element in its landing site and replace the launching site with a variable that is bound by a λ-operator inserted immediately below the landing site (Beck 1996; Heim and Kratzer 1998; Sauerland 1998). We will refer to these λ-bound variables as traces. (For the sake of simplicity, we set aside Trace Conversion and represent traces as simplex variables, though nothing crucial hinges on this choice; see Sect. 7.4 for discussion.) On the surface reading, the trace is of semantic type e, as sketched in the simplified derivation in (3).

On the higher-type-traces approach, traces may also be of higher semantic types, which can in turn be used to achieve various kinds of reconstruction effects. For example, a trace of the semantic type of a generalized quantifier (〈et,t〉) yields scope reconstruction (Cresti 1995; Rullmann 1995), as schematized in (4) (again, simplified). The crucial step of the derivation to note is when the moved element combines with the λ-abstraction created by movement (4c). With an ordinary type-e trace, the moved quantificational element takes as argument the λ-abstraction (3c). However, with a type-〈et,t〉 trace, it is vice versa: the λ-abstraction takes as argument the moved quantificational element (4c).

Higher-type traces have been extended to account for other types of reconstruction effects, such as pronominal-binding reconstruction e.g. Engdahl 1980, 1986, Jacobson 1999, 2004; though see Heim 2019 and referential-opacity reconstruction e.g. Sharvit 1998; though see Romero 1998: 108–114, though the specific semantic type of the trace will be different. Analyses of reconstruction effects in this line of approach have been developed by von Stechow (1991), Chierchia (1995), Cresti (1995), Rullmann (1995), Lechner (1998, 2013, 2019), Sharvit (1998), Sternefeld (2001), and Ruys (2015), amongst others.

In what follows, we will use the term ‘reconstruction’ in a purely descriptive manner. ‘α reconstructs for β’ should be understood as ‘α is evaluated for β in its premovement position’, without any implication about how that evaluation is achieved.

The two approaches can be characterized in terms of where in the grammar they situate reconstruction effects. Higher-copy neglect manipulates the LF, a syntactic level of representation, and thus is a syntactic procedure. Higher-type traces, on the other hand, manipulate the semantic type of the trace, not the LF, and thus constitute a purely semantic procedure.² However, to account for each type of reconstruction effect, the two approaches ultimately generate the same truth conditions. As such, they are mostly indistinguishable on these metrics alone. Thus, it is necessary to look elsewhere to empirically distinguish them. In their seminal work, Romero (1997, 1998) and Fox (1999) contend that the two approaches can be teased apart using Condition C. In particular, they argue that scope reconstruction correlates with Condition C connectivity: when a moved element reconstructs for scope, it is evaluated for Condition C in the position to which it scopally reconstructs (see also Heycock 1995). (This correlation will be discussed in greater detail in Sect. 2.) This state of affairs follows straightforwardly from higher-copy neglect: the moved element is evaluated for Condition C in its launching site because that is where it is located at LF. However, it does not immediately follow from the higher-type-traces approach, which places the moved element in the same position at LF regardless of whether it reconstructs for scope or not.

In this paper, we argue that not all reconstruction effects can be reduced to neglecting copies. The argument is based on a detailed empirical investigation of reconstruction in Hindi-Urdu (henceforth, Hindi), where we make the novel observation that not all reconstruction effects induce Condition C connectivity in Hindi. In particular, scope reconstruction does not correlate with Condition C, unlike what Romero and Fox claim for English. However, neither is it the case that all reconstruction effects are independent from Condition C in Hindi: Condition C connectivity does in fact correlate with reconstruction for referential opacity—that is, when the moved element is interpreted opaquely with respect to an intensional operator that it crosses (also claimed for English in the “lost” paper of Sharvit 1998).³ This state of affairs does not follow from an all-or-nothing approach to reconstruction, like the higher-copy-neglect approach. Rather, we argue that Hindi requires the hybrid approach to reconstruction developed on independent grounds by Lechner (1998, 2013, 2019), where both higher-copy neglect and higher-type traces are available as independent interpretive mechanisms. Crucially, higher-copy neglect induces Condition C connectivity, but higher-type traces do not. We show that the interaction of these two independent mechanisms derives the intricate reconstruction facts in Hindi. This more fine-grained approach to reconstruction importantly entails that some but not all reconstruction effects are syntactic.

The argumentation proceeds as follows: We begin in Sect. 2 by discussing the relationship between reconstruction effects and Condition C connectivity as it stands in the literature, which we frame in terms of two competing empirical generalizations. We then turn our attention to Hindi in Sect. 3. We argue that Hindi (long) scrambling adjudicates between these competing generalizations, in particular showing that reconstruction for referential opacity, but not reconstruction for scope correlates with Condition C connectivity. Section 4 develops an analysis of Hindi scrambling under Lechner’s (1998, 2013, 2019) hybrid model of reconstruction and discusses the necessary restrictions that must be imposed on higher-type traces under this model. This analysis is then extended in Sect. 5 to reconstruction for pronominal binding and weak crossover in Hindi. In Sect. 6, we zoom out to consider the typology of traces and reconstruction predicted under our proposal. Section 7 concludes by discussing several implications and open questions.

2 Scope, referential opacity, and Condition C

As mentioned in the introduction, higher-copy neglect and higher-type traces both generate the same truth conditions to account for each type of reconstruction effect. Thus, the two approaches cannot be distinguished merely on their basis to produce certain reconstruction effects. However, there is a key difference between the two approaches: where the moved element is located at LF. With higher-copy neglect, the moved element is evaluated in its launching site at LF, but with higher-type traces, it is evaluated in its landing site. The insight of the previous literature has been to use Condition C (i.e. R-expressions must be free; Chomsky 1981). Condition C is standardly taken to be evaluated at LF (Lebeaux 1988, 1990, 2000, 2009; Chomsky 1995), and so it can be used as a means to independently detect the location of the moved element at LF in the presence of reconstruction effects (Heycock 1995; Romero 1997, 1998; Sharvit 1998; Fox 1999; Sternefeld 2001; Lechner 2013, 2019; Ruys 2015).⁴ Higher-copy neglect predicts that reconstruction effects should cooccur with Condition C connectivity at the launching site of movement, because this is the position of the moved element at LF (5). Conversely, higher-type traces predict that reconstruction effects should not cooccur with Condition C connectivity (unless further assumptions are made), because the moved element instead occupies its landing site at LF (6).⁵

In short, under higher-copy neglect, reconstruction effects should be sensitive to Condition C, but under higher-type traces, they should not be.

The crucial configuration for testing these predictions involves “Lebeaux” effects. It is well-known that $\overline{\text{A}}$-movement may obviate Condition C violations incurred in the absence of movement if the offending R-expression is embedded inside a relative clause (van Riemsdijk and Williams 1981; Lebeaux 1988, 1990, 2000, 2009), as illustrated in (7).

The crucial property of (7b) is that the moved element would incur a Condition C violation in its base position—parallel to (7a)—, but not in its surface position. Thus, (7b) demonstrates that a moved DP can be evaluated in its landing site for Condition C; otherwise (7b) would be ungrammatical.⁶ For investigating the relationship between reconstruction effects and Condition C connectivity, the test configuration has the general form of (8), where a DP with a relative clause that contains an R-expression is moved over both a scope-bearing operator and a pronoun coindexed with the R-expression.

In this configuration, a reconstruction effect that correlates with Condition C connectivity should be blocked in (8) because it would yield a Condition C violation; that is, Op ≫ DP should be impossible. By contrast, a reconstruction effect that does not correlate with Condition C connectivity should be permitted in (8) because Condition C is not a factor; that is, Op ≫ DP should be possible.

Empirical investigations of these predictions have produced conflicting results. From the literature, we extrapolate two competing generalizations, which we discuss in Sects. 2.1 and 2.2 respectively. Section 2.3 summarizes the state of affairs, setting the stage for our own investigation of Hindi in Sect. 3.

2.1 Quantifier–Condition C correlation

The first proposed generalization comes from Romero (1997, 1998) and Fox (1999), who argue that if a moved element reconstructs into its launching site for quantificational scope, then it is evaluated in its launching site for Condition C (see also Heycock 1995).⁷ Thus, scope reconstruction is blocked in configurations like (8). This is encapsulated in the generalization in (9), which we will refer to as ‘Q→C’.

Their argument is based on sentences like (10). In (10), the wh-element contains an R-expression that is coindexed with the matrix subject; the movement step crosses the matrix subject, in addition to the attitude predicate . (10) thus instantiates the schema in (8). (For reasons of space, we assume familiarity with the ambiguity how many gives rise to when it moves over another scope-bearing element.⁸) Crucially, Romero and Fox report that (10) is unambiguous. It only has the surface-scope reading in (10a), where it is assumed that there is a set of particular pictures that John wants the editor to publish, and the question is asking how many such pictures there are. The sentence is claimed to lack the reconstructed-scope reading in (10b), where the question is asking about the quantity of pictures that John wants the editor to publish, without having any particular pictures in mind.

Compare (10) to (11), where the R-expression and the pronoun are swapped, so that binding connectivity would not induce a Condition C violation. In this case, scope reconstruction is possible, and the sentence is ambiguous.

Romero and Fox argue that the contrast between (10) and (11) shows that scope reconstruction correlates with Condition C connectivity. That is, reconstructing for scope is impossible in (10) because it would give rise to a Condition C violation and possible in (11) because Condition C is not at stake. This conclusion supports the empirical generalization Q→C. They present converging evidence for Q→C from A-movement and other $\overline{\text{A}}$-movement configurations.

Romero and Fox take Q→C as evidence in favor of higher-copy neglect and against the availability of higher-type traces. As discussed above, higher-copy neglect inherently predicts that reconstruction effects are sensitive to Condition C (see (5)), and thereby derives Q→C for free. All else being equal, on a higher-type-trace account, there is no expectation that Condition C should be able to influence the availability of a reconstructed-scope reading (see (6))—contrary to Q→C. To illustrate this point, the LFs for the reconstructed-scope reading of (10) under high-copy neglect and higher-type traces are included below in (12) and (13) respectively.⁹ Crucially, (12), but not (13), places the R-expression in the moved element inside the c-command domain of the coindexed pronoun, thereby yielding a Condition C violation.

Accordingly, Romero and Fox conclude that data like (10) support a purely syntactic approach to reconstruction, in which reconstruction effects result exclusively from neglecting a higher copy (see also Sportiche 2016; Poole 2017, 2024).

The conclusion that Q→C favors higher-copy neglect has been called into question by Sternefeld (2001) and Ruys (2015). They propose enriched versions of higher-type-trace accounts that are able to derive Q→C: Sternefeld (2001) by placing Condition C into the semantics and Ruys (2015) by imposing a general constraint on the availability of higher-type traces. We discuss these accounts in Sect. 4.1.

2.2 Intensionality–Condition C correlation

The second proposed generalization comes from Sharvit (1998) and Lechner (2013, 2019). They argue that scope reconstruction does not generally correlate with Condition C connectivity, contra Q→C. At the same time, they argue that reconstruction effects are also not entirely dissociated from Condition C. Instead, rather than scope reconstruction, what Condition C connectivity correlates with is reconstruction for referential opacity (i.e. the moved element being interpreted opaquely with respect to an intensional operator that it crosses over). We will refer to this competing generalization as ‘I→C’, given in (14).

The crucial piece of evidence for I→C comes from the paradigm in (15), which Romero (1998: 97) and Lechner (2013: 175, 2019: 120) attribute to Sharvit (1998) (see fn. 3). The sentence in (15) instantiates the test schema in (8) above, but in addition to quantifier scope, it manipulates referential opacity. The three a priori possible readings in (15) are conditioned by whether or not the moved DP reconstructs for quantifier scope and for referential opacity.¹⁰

Setting aside the first reading (surface-scope, transparent), which requires no reconstruction, let us step through the other two logically possible readings of (15). The second reading (15b) involves the quantificational force of how many taking scope below , but the restrictor NP being interpreted transparent to ; this is the so-called “third reading” in the de re/de dicto literature (Fodor 1970). On this reading, the question is asking about the quantity of individuals who Anton hopes will buy him a beer, without having any particular individuals in mind. These individuals are students who hate Anton in the actual world—potentially unbeknownst to him. This reading requires reconstruction for scope, but not for referential opacity. According to Q→C, this reading should be unavailable because it involves scope reconstruction and thus should induce Condition C connectivity. The fact that (15b) is a possible interpretation of (15) presents a challenge for Q→C.

Crucially absent is the reading in (15c), where the quantificational force of how many takes scope below and the restrictor NP is interpreted opaque to . This reading is like the second reading in that the question is asking about the quantity of individuals who Anton hopes will buy him a beer, without having any particular individuals in mind. However, unlike the second reading, these individuals are students who hate Anton in Anton’s bouletic alternatives. That is, they may not be students who hate Anton in the actual world. This reading would involve reconstruction for both scope and referential opacity.

Sharvit (1998) and Lechner (2013, 2019) conclude from (15) that what correlates with Condition C connectivity—and thus may be blocked by a Condition C violation—is not scope reconstruction, but referential-opacity reconstruction (as in (15c)). An example analogous to (15) is mentioned in passing by von Fintel and Heim (2011: 114–115), also attributed to Sharvit (1998). Lechner (2013, 2019) provides converging evidence from A-movement in (16). In (16), assuming that must be read transparent to (i.e. in the actual world), the A-moved element must be interpreted opaque to in order to avoid a contradiction. If both elements were interpreted transparently, then they would be the same value and so neither could exceed the other. Lechner claims that the noncontradictory reading is available in (16a), but not (16b), where Condition C is at stake.

This conclusion stands in opposition to Q→C. Sharvit (1998) and Lechner (2013, 2019) therefore reject Q→C and conclude that the correct generalization is I→C. Lechner (2013, 2019) further argues, based on I→C, that an account of reconstruction effects that is confined to higher-copy neglect (e.g. Romero 1998 and Fox 1999) is empirically insufficient. He instead proposes a hybrid theory of reconstruction; we will argue for and extend this hybrid theory of reconstruction in Sect. 4.

2.3 Interim summary

We are faced with a conundrum. The two generalizations, Q→C and I→C, are mutually incompatible, because they make contradictory statements about which types of reconstruction effects correlate with Condition C connectivity. Despite their incompatibility, these two generalizations have not, to our knowledge, been contrasted directly.¹¹ This state of affairs has repercussions for the proper analytical treatment of reconstruction effects. As noted above, Q→C has been taken to support higher-copy neglect (Romero 1997, 1998; Fox 1999) and I→C to support a hybrid model (Lechner 2013, 2019). In light of the uncertainties about whether Q→C or I→C is the correct empirical generalization, it thus stands to reason that a proper evaluation of the analytical consequences requires a better understanding of the empirical relationship between reconstruction effects and Condition C connectivity.

Admittedly, one might take the English paradigm in (15) as definitive evidence against Q→C and in favor of I→C. Although this paper will ultimately vindicate that conclusion, we believe that based on the already available data, it would be a premature conclusion. First, the data motivating I→C are scant: essentially three sentences, two of which are from an unavailable manuscript (see fn. 3). Second, the crucial contrasts—(15b) vs. (15c) and (16a) vs. (16b)—are subtle judgements. Thus, as von Fintel and Heim (2011: 115) and Ruys (2015: 479n27) also mention, the data behind I→C need further research. This stands in contrast to the clearer and more systematic data of Romero (1997, 1998) and Fox (1999).

In what follows, we attempt to develop a better understanding of the empirical patterns that are at stake with reconstruction effects. We take up this task by investigating the reconstruction profile of long scrambling in Hindi. For language-internal reasons (namely, the surface-scope reading being absent), the judgements in Hindi are sharper than the relevant judgements in English, and thus the predictions of Q→C and I→C come apart in a particularly clear and clean way. We will argue that this investigation provides striking support for I→C and against Q→C as a crosslinguistic generalization about reconstruction effects. In addition to contributing to our understanding of the empirical issues involved, this conclusion also has a number of general implications for theories of reconstruction. In particular, we will argue that the Hindi evidence provides novel support for Lechner’s (1998, 2013, 2019) hybrid theory of reconstruction, which encompasses both higher-copy neglect and higher-type traces as independent and complementary mechanisms of reconstruction.

3 Long-distance scrambling in Hindi

This section investigates the reconstruction profile of long-distance scrambling in Hindi. We show that long-distance scrambling obligatorily reconstructs for scope (Sect. 3.1). This property provides an exceptionally clear window into the relationship between Condition C and reconstruction effects, which we explore. We demonstrate that scope reconstruction is not constrained by Condition C connectivity in Hindi (Sect. 3.2), which indicates that Q→C is not a universally valid constraint on reconstruction. We then show that there is nonetheless a correlation between Condition C and reconstruction for referential opacity in Hindi (Sect. 3.3), supporting the validity of I→C.

Unless indicated otherwise, the data reported in this paper come from working with two primary consultants, with crucial judgements confirmed by one additional consultant. All three of the consultants are native speakers of Hindi and are linguists. Undoubtedly, our consultants being linguists simplified data collection. In particular, we relied on them to make judgements regarding (i) the scope relation between a universal quantifier and an existential quantifier and (ii) coreference between two DPs—without any context or specialized elicitation method. Both of these kinds of judgements, we take to be run-of-the-mill tasks for linguists.

It is worth explicitly pointing out that we are interested in the entailment relations that Q→C and I→C embody, and that several preconditions must be satisfied to test these entailments: (i) the language must have vanilla Condition C effects, (ii) the movement type must allow for scope reconstruction, and (iii) the movement type must allow for referential-opacity reconstruction. For example, a movement type that does not reconstruct for referential opacity does not bear on the relationship between referential opacity and Condition C. Furthermore, to control for potential interspeaker variation, it is important to ensure that these preconditions hold for individual speakers. For example, if Speaker A allows for scope reconstruction for a given movement type and Speaker B does not, then it is only Speaker A’s judgements that bear on whether scope reconstruction correlates with Condition C connectivity. (Technically, Speaker A and Speaker B have different movement types.) These preconditions hold for long-distance scrambling for our Hindi consultants, as will be shown in what follows.

3.1 Setting the stage: The scope of scrambling

It has been well-known since Gurtu (1985, 1992), Déprez (1989), and Mahajan (1990, 1994) that scrambling in Hindi is not a uniform phenomenon (see also Gambhir 1981; Dayal 1994a; Kidwai 2000; Keine 2016, 2019, 2020). We will distinguish between local scrambling, which does not cross a finite-clause boundary, and long-distance scrambling (LDS), which does cross a finite-clause boundary. One classical difference between the two scrambling types is with respect to weak crossover: LDS is subject to weak crossover, whereas local scrambling is not (Mahajan 1990, Gurtu 1992; see also Sect. 5.2). In this section, we use the terms “local scrambling” and “long(-distance) scrambling” as convenient descriptive labels without committing to an analysis of the distinction, which we defer until Sect. 4.

Keine (2016, 2019) notes that local scrambling and LDS in Hindi differ in their ability to extend quantifier scope. Like many other SOV languages with flexible word order (see e.g. Frey 1993 and Krifka 1998 for German), the scopal relations between two DPs are generally fixed in the base order, as shown in (17a) and (17c).^12,13 When a DP is locally scrambled over the subject, that DP may take scope in its landing site Mahajan 1997: 199–200, 2017: 426–428, as shown in (17b) and (17c). Scope reconstruction is also possible, as (18) illustrates.

This ability to extend scope is confined to local scrambling. Crucially, LDS (i.e. scrambling out of a finite clause) does not extend the scope options of the moved element.¹⁴ In other words, scope reconstruction is obligatory, or at least is strongly preferred. In (19), the embedded object is moved over the matrix subject, but it may not take scope over either the matrix subject (19a) or the matrix object (19b).^15,16

Note that local scrambling may feed LDS (Mahajan 1990: 38–47); only the movement step that crosses the finite-clause boundary constitutes LDS. This possibility predicts that a long-scrambled DP may take scope over elements within the embedded clause, because it may undergo local scrambling within the embedded clause before it undergoes LDS into the matrix clause. As (20) shows, this prediction is borne out.

In what follows, we are only interested in LDS itself. Therefore, we place all the relevant components (e.g. scope-bearing element, coindexed pronoun, intensional operator) in the matrix clause—except the moving element, of course—in order to rule out effects of local scrambling in the embedded clause. If the test configuration requires two argument DPs, the matrix predicate must then be ditransitive. (We leave further exploring local scrambling feeding LDS to future research.)

This restriction on LDS is also observed in how many-questions. Hindi does not have wh-movement, but wh-elements may be scrambled Mahajan 1990: 107–194, Dayal 1994b: 138–139, 1996: 35–38. Following the methodology in Fox (1999), we elicited scope judgements for how many-questions by setting up contexts in which the two interpretations yield distinct answers. In (21), the wh-element is moved into the matrix clause. The surface-scope reading ( ≫ ) is either impossible or severely degraded in (21). The reconstructed-scope reading ( ≫ ), by contrast, is readily available.

The lack of a wide-scope reading in (21) stands in direct contrast to otherwise analogous structures in English (cf. (10); see e.g. Cresti 1995 and Rullmann 1995). The reconstruction requirement is furthermore independent of the type of the embedding verb. It also holds for other attitude predicates like , as shown in (22).¹⁷

The relevant generalization in all of these cases is that LDS obligatorily reconstructs for quantifier scope.

3.2 Testing Condition C and quantifier scope

One might reasonably wonder at this point whether LDS in Hindi is simply semantically inert or, equivalently, whether it undergoes “radical reconstruction”, as has been claimed for long scrambling in Japanese (e.g. Bošković and Takahashi 1998, Bošković 2004, Saito 2004; see also Saito 1989). There is good indication that this is not the case. One interpretative aspect for which reconstruction is not obligatory is Condition C. As (23) shows, a Condition C violation between a matrix subject () and an R-expression inside the embedded object () is obviated by LDS of the embedded object over the matrix subject in (23b).¹⁸

The rescuing effect of LDS on Condition C violations demonstrates that LDS in Hindi is not simply semantically inert or the result of “radical reconstruction”. As a consequence, its properties are not amenable to a PF-movement account à la Aoun and Benmamoun (1998) and Sauerland and Elbourne (2002) or to an LF-lowering account such as Bošković and Takahashi (1998). This is consistent with the conclusion of Sect. 3.3, where we will show that LDS in Hindi does not have to reconstruct for referential opacity. Additionally, LDS does not need to reconstruct for wh-licensing (Mahajan 1990; Dayal 1994b, 1996), further suggesting that at least some of its effects are LF-visible.

We now turn to the relationship between scope reconstruction and Condition C. The reconstruction properties of LDS provide a particularly clear domain in which Q→C and I→C can be assessed. Because of the strong preference for scope reconstruction with LDS, Q→C predicts that scrambling a scope-bearing element out of a Condition C configuration (see (8)) should not only obligatorily reconstruct for scope, but also exhibit Condition C connectivity, thereby resulting in outright ungrammaticality. On the other hand, since I→C claims that scope reconstruction is independent of Condition C, I→C predicts LDS to be grammatical in a Condition C configuration and a reconstructed-scope reading to be possible.

First, notice that the observation that LDS strongly favors scope reconstruction (19)–(21), but does not require Condition C connectivity (23) provides a first indication that reconstruction for scope and Condition C connectivity do not correlate, contra Q→C.

We can sharpen this conclusion by manipulating Condition C and scope simultaneously, as in (24). In (24b), the moved element contains an R-expression coindexed with the matrix subject—which in situ would yield a Condition C violation (24a)—and moves across the quantificational element . The sentence is fully grammatical on a coreferential reading of the pronoun and with a reconstructed-scope interpretation of . Furthermore, in line with the scope observations in (19)–(21), surface scope of the moved element is impossible in (24b).

The crucial property of (24b) is that it readily allows a reconstructed-scope interpretation without incurring a Condition C violation. Thus, (24b) shows that scope reconstruction is possible even if Condition C connectivity with the launching site of movement would result in ungrammaticality. This provides an argument that reconstruction for quantificational scope does not generally entail Condition C connectivity, contra Q→C.

This pattern is general. It can be replicated with indefinites (25), -questions (26), and other embedding predicates (27). (Note that the indefinite in (25) must be interpreted de re, which we discuss in the next section.)

We conclude that scope reconstruction and Condition C connectivity do not necessarily correlate; that is, it is possible to reconstruct for quantifier scope without inducing Condition C connectivity. Because this is precisely what Q→C rules out, it strongly suggests that Q→C is not a universal constraint on reconstruction.

3.3 Testing Condition C and referential opacity

We have seen so far that reconstruction for scope in Hindi is independent of Condition C connectivity. This provides evidence against Q→C and is compatible with I→C. However, I→C makes a much stronger prediction: not only is scope reconstruction independent of Condition C, but Condition C is predicted to block reconstruction for referential opacity.

To investigate the empirical relationship between referential-opacity reconstruction and Condition C connectivity, let us begin by considering the paradigm in (28). The scenario in (28) is designed so that the description is true relative to Pratap’s doxastic alternatives, but false relative to the actual world, given that what Sangita saw was not actually a ghost. In the nonmovement baseline in (28a), the embedded object is embedded under the intensional predicate . As expected, the embedded object can be interpreted opaquely with respect to this predicate; (28a) is hence true in the given scenario. The examples in (28b, 28c) investigate how the availability of this reading interacts with movement. In (28b), the embedded object undergoes LDS into the matrix clause. While there is a bias for the transparent interpretation, (28b) allows for an opaque reading of the scrambled DP with respect to . Hence, reconstruction for referential opacity is possible in (28b). Against this backdrop, the crucial example is (28c). (28c) is identical to (28b), except that the R-expression and the pronoun have been swapped, so that the R-expression is now inside the scrambled DP. As such, if the scrambled DP were to be evaluated for Condition C in its launching site, it would incur a Condition C violation. Importantly, while the sentence in (28c) is grammatical, it is not judged as true in the given scenario. The only available interpretation is one where the moved DP is interpreted transparently with respect to . The opaque interpretation is not available at all. Thus, (28c) commits the speaker to the claim that Sangita saw an actual ghost and is thus infelicitous (under the assumption that ghosts do not exist in the actual world).

The impossibility of an opaque reading in (28c) demonstrates that reconstruction for referential opacity is impossible in a Condition C configuration. In light of the availability of such reconstruction in (28b), where Condition C is not at stake, this strongly suggests that reconstruction for referential opacity induces Condition C connectivity. Therefore, the paradigm in (28) indicates that reconstruction for an opaque reading is crucially not independent of Condition C, unlike reconstruction for quantifier scope. This finding aligns with the predictions of I→C.

Additional support comes from subjunctive relative clauses. Subjunctive relative clauses have the interesting property that they must be interpreted opaquely with respect to the intensional predicate that they are embedded under (Farkas 1985, 1997; Quer 1998; Romero 1998; Bhatt 2021). To illustrate, consider the DP in (29), which contains a subjunctive relative clause. The contexts in (30a) and (30b) are set up so that only an opaque reading or only a transparent reading of the DP holds respectively. The sentence in (29) is judged as true in (30a), but not in (30b). That is, the DP must be interpreted opaquely with respect to .

Crucially, the same pattern of judgements in (28) holds for long-scrambled DPs with subjunctive relative clauses, as shown in (31). When Condition C is not at stake, the scrambled DP can be interpreted opaquely with respect to the intensional predicate (31b)—in fact, given the subjunctive relative clause, it must be. The Condition C configuration in (31c) is outright ungrammatical: the subjunctive relative clause requires reconstruction, and this is blocked by Condition C.

Finally, a more complex example investigating the three-way relationship between Condition C, scope, and referential opacity is given in (32), modelled after Sharvit’s (1998) example in (15). The sentence only has the reconstructed-scope, transparent reading in (32b). It lacks the other two logically possible readings: a surface-scope, transparent reading (32a) and an (reconstructed-scope) opaque reading (32c).¹⁹

The observation that (32b) is the only available interpretation for (32) is fully consistent with the generalizations that we have reached thus far. First, because LDS in Hindi obligatorily reconstructs for quantifier scope, the surface-scope reading in (32a) is ruled out.²⁰ Second, we saw on the basis of (28) above that Condition C connectivity blocks reconstruction for referential opacity, thereby ruling out the opaque interpretation in (32c). Crucially, Condition C connectivity does not block reconstruction for quantifier scope. The reconstructed-scope, transparent reading in (32b) is therefore possible. As predicted, if the R-expression and the pronoun are swapped in (32) so that Condition C is no longer at stake, the opaque interpretation becomes available, as shown in (33).

In sum, the data in (28), (31), and (32) provide evidence that Condition C connectivity correlates with referential-opacity reconstruction, but not with scope reconstruction, converging with the previous evidence in this section. Taken together, the Hindi reconstruction data support I→C (repeated here as (34)) as an empirical generalization about the properties of reconstruction effects. The next section will explore the theoretical consequences of this conclusion and provide an analysis of Hindi scrambling in terms of Lechner’s (1998, 2013, 2019) hybrid model of reconstruction.

4 A hybrid account of reconstruction

The crucial takeaway from I→C and the Hindi reconstruction data supporting I→C (Sect. 3) is that some but not all reconstruction effects correlate with Condition C connectivity. Assuming that Condition C connectivity is indicative of a syntactic reconstruction mechanism, this means that some but not all reconstruction effects are syntactic; those that are not syntactic are semantic. This state of affairs represents a middle ground between the opposing views that have emerged in the reconstruction literature, where it is argued either that reconstruction is purely syntactic (e.g. Romero 1997, 1998; Fox 1999; Sportiche 2016; Poole 2017, 2024) or that reconstruction is purely semantic (e.g. Cresti 1995; Rullmann 1995; Sternefeld 2001; Ruys 2015). We contend instead that reconstruction is part syntactic and part semantic.

In this section, we argue that Hindi long scrambling provides novel support for the hybrid model of reconstruction developed by Lechner (1998, 2013, 2019). The core feature of Lechner’s system is that it uses both higher-copy neglect (a syntactic mechanism) and higher-type traces (a semantic mechanism). We show how such a model derives the intricate Hindi reconstruction facts via the interaction of these two modes of reconstruction.

4.1 The insufficiency of nonhybrid accounts

Before proceeding to the analysis, it is instructive to briefly consider the challenge that the Hindi reconstruction pattern poses for nonhybrid accounts of reconstruction, i.e. a purely syntactic account in terms of higher-copy neglect or a purely semantic account in terms of higher-type traces.

On one hand, an account limited to higher-copy neglect predicts that reconstruction effects should universally correlate with Condition C connectivity (Romero 1997, 1998; Fox 1999). As we have seen in Sect. 3.2, this is not the case. Thus, a purely syntactic account of reconstruction is too restrictive to accommodate the Hindi facts.²¹

On the other hand, an account limited to higher-type traces would dissociate Condition C from all reconstruction effects. This dissociation would fail to capture the empirical connection between referential-opacity reconstruction and Condition C connectivity from Sect. 3.3. Interestingly, Sternefeld (2001) and Ruys (2015) propose enriched versions of the higher-type-trace account that derive a strict correlation between Condition C and reconstruction effects. In this regard, these enriched accounts are empirically equivalent to higher-copy-neglect accounts (as Sternefeld 2001 himself emphasizes). As a consequence, these accounts are too restrictive for exactly the same reason.

Any account of reconstruction in which all reconstruction effects or no reconstruction effects correlate with Condition C connectivity is too coarse to capture the intricate empirical relationship between Condition C, scope, and referential opacity. A more nuanced account is therefore called for. In principle, either higher-copy neglect or higher-type traces might be augmented to variably induce Condition C connectivity, but we suspect that such an account will be stipulative. Instead, we pursue a hybrid theory of reconstruction, which makes available both reconstruction mechanisms and derives the limited correlation between reconstruction effects and Condition C connectivity from general syntactic and semantic principles.

4.2 Analysis

In this section, we show that the interpretation of Hindi scrambling receives a principled explanation under Lechner’s (1998, 2013, 2019) hybrid model of reconstruction. Let us begin by making some concrete assumptions about the nature of local and long-distance scrambling in Hindi. We adopt Mahajan’s (1990, 1994) account, according to which scrambling in Hindi is ambiguous between two distinct movement types (see also Bhatt 2016 and Keine 2016, 2020). One type of scrambling, which we will refer to as A-scrambling, exhibits A-properties and cannot cross a finite-clause boundary. The second type of scrambling, which we will call $\overline{\text{A}}$-scrambling, exhibits $\overline{\text{A}}$-properties and is able to leave a finite clause.²² Thus, LDS in Hindi is invariably $\overline{\text{A}}$-scrambling in this technical sense, whereas local scrambling is ambiguous between A-scrambling and $\overline{\text{A}}$-scrambling (35).

Turning now to the interpretation of the two scrambling types, we propose that $\overline{\text{A}}$-scrambling in Hindi can be interpreted either by neglecting the higher copy (36a) or by using a higher-type trace, in particular a generalized-quantifier trace (36b). By assumption, these are the only two options; in particular, translating the trace position of $\overline{\text{A}}$-scrambling into a type-e variable is impossible (see Sect. 6 for further discussion).²³ Because both procedures in (36) yield reconstructed scope, it follows that $\overline{\text{A}}$-scrambling never shifts the scope of the moved element. In turn, given that LDS is invariably $\overline{\text{A}}$-scrambling, LDS thus displays obligatory scope reconstruction.

The crucial component of (36) is that both higher-copy neglect and higher-type traces are in principle always available to interpret $\overline{\text{A}}$-scrambling. However, as we will show, they have slightly different effects, and higher-copy neglect is crucially blocked when it would induce a Condition C violation, leaving a higher-type trace as the only option in such cases.

Furthermore, we propose that A-scrambling in Hindi is interpreted with a type-e trace, as schematized in (37).

As local scrambling is ambiguous between A-scrambling and $\overline{\text{A}}$-scrambling, it (descriptively) has access to all three interpretive options in (36) and (37). It, therefore, follows that local scrambling can reconstruct (as in (18)), but that such reconstruction is optional (as in (17)), in contrast to LDS, whose two interpretive options both yield reconstruction.²⁴

This difference in how movement types are interpreted must be encoded somewhere in the grammar. Following van Urk (2015), we assume that such differences are encoded in the probe triggering the movement. Specifically, we propose that movement-triggering probes may either be interpreted as a λ-operator over a particular semantic type(s) (building on Kratzer 2002: 112–114) or as the identity function. Under this approach, the λ-operator binding the trace is not inserted at LF. Higher-copy neglect occurs whenever interpreting the higher copy in the landing site would result in a semantic-type mismatch. This state of affairs only arises when the probe is interpreted as the identity function—so that it is semantically vacuous—, because elements of type e and 〈et,t〉 cannot compose in this case. Applied to Hindi scrambling: the $\overline{\text{A}}$-scrambling probe is interpreted either as a λ-operator over type 〈et,t〉 (yielding a higher-type trace) or as the identity function (yielding neglect), and the A-scrambling probe is interpreted only as a λ-operator over type e.

We now proceed to demonstrating how the coexistence of the two reconstruction mechanisms for $\overline{\text{A}}$-scrambling in (36) and the division of labor between them enables a principled explanation of the Hindi reconstruction data from Sect. 3.

4.2.1 Higher-type traces

The principal motivation for higher-type traces comes from the observation in Sect. 3.2 that scope reconstruction is possible in Hindi even if evaluating the moved element in its launching site at LF would give rise to a Condition C violation. Consider the sentence in (24b), repeated here as (38). As discussed above, scope reconstruction is possible in (38)—and is in fact strongly preferred—despite the fact that the scrambled DP contains an R-expression and the movement crosses a pronoun coindexed with that R-expression.

Because (38) involves LDS, the scrambling must be an instance of $\overline{\text{A}}$-scrambling. According to (36), it can be interpreted either by neglecting the higher copy (39) or with a higher-type trace (40). (For the sake of simplicity, we abstract away from any movement that might happen within the embedded clause, such as type-driven movement or local scrambling.) As both options yield scope reconstruction, it immediately follows that (38) lacks a surface-scope reading. However, neglecting the higher copy would produce ungrammaticality in (38) because it would lead to a Condition C violation—but a higher-type trace would not.

It follows then that (38), where Condition C is at stake, cannot be interpreted via higher-copy neglect, and must be interpreted via higher-type traces. In cases where Condition C is not at stake, e.g. (19), both higher-copy neglect and higher-type traces are possible (i.e. there is an ambiguity). Higher-type traces hence derive the independence of scope reconstruction and Condition C connectivity in Hindi LDS, as documented in Sect. 3.2. This provides an argument that reconstruction effects may be the result of higher-type traces.

At the same time, we showed in Sect. 3.3 that Condition C and reconstruction effects are not entirely independent of each other: Condition C connectivity systematically bleeds reconstruction for referential opacity. The relevant example (28c) is repeated below in (41).

Given that higher-type traces do not induce Condition C connectivity, e.g. (40), the correlation between Condition C and referential-opacity reconstruction in (41) reveals that higher-type traces must be unable to produce referential-opacity reconstruction. Otherwise, if a higher-type trace could be used to derive an opaque reading of an $\overline{\text{A}}$-scrambled DP, then (41) would be felicitous, contrary to fact.

Following Lechner (2019), we propose that this restriction on higher-type traces is due to a general principle that the intensionality of a DP, specifically its NP restrictor, is evaluated in its position at LF. For concreteness, we adopt the theory of syntactically realized situation (or world) pronouns (Percus 2000), which has this principle as a consequence.²⁵ Under this theory, the opaque reading of a DP requires that its situation pronoun be bound by the λ-operator associated with the relevant intensional operator. When a moved DP is interpreted with a higher-type trace, its situation pronoun must be bound from the landing site of movement, as this is its LF position. Thus, an intensional operator that a DP has moved over cannot bind its situation pronoun, as schematized in (42).²⁶

In sum, higher-type traces yield reconstruction for scope, but not for referential opacity. Because a higher-type trace does not give rise to Condition C connectivity, scope reconstruction is independent of Condition C and thus is not constrained by it.

4.2.2 Higher-copy neglect

While we have seen evidence for higher-type traces in Hindi, such traces alone are insufficient; higher-copy neglect must be available as well. To illustrate why, consider again the sentence in (28b), repeated below as (43). (43) is a minimally different variant of (41) in which the positions of the R-expression and the pronoun have been swapped. Condition C is not at play in (43), and an opaque reading of the moved DP with respect to is possible.

As discussed above, reconstruction for referential opacity cannot be the result of a higher-type trace (see (42)). Therefore, the opaque interpretation in (43) must be the result of higher-copy neglect, as schematized in (44). By interpreting only the lower copy, the situation pronoun in the moved DP is in the scope of the λ-operator of , so that it may bind the situation pronoun, yielding an opaque interpretation. Crucially, because the moved DP contains a pronoun instead of an R-expression, interpreting only the lower copy in (44) does not result in a Condition C violation, unlike (39).

The option of neglecting the higher copy therefore explains why $\overline{\text{A}}$-scrambling allows for referential-opacity reconstruction. Crucially, because higher-copy neglect induces Condition C connectivity, referential-opacity reconstruction is only possible when it would not yield a Condition C violation. When Condition C is at stake, as in (41) above, reconstruction for referential opacity is impossible via higher-copy neglect, as schematized in (45).

In such cases where there would be a Condition C violation in the launching site of movement, as in (41), the only available interpretive option is thus a higher-type trace. In turn, because higher-type traces are unable to produce referential-opacity reconstruction, only a transparent interpretation of the scrambled DP is possible in such cases. This derives the observation that reconstruction for referential opacity, but not for scope, correlates with Condition C connectivity.

This account also extends to the more complex example in (32), repeated in (46). Here, Condition C connectivity blocks reconstruction for referential opacity, but allows it for scope.

The opaque reading in (46c) would require neglecting the higher copy. However, as this would give rise to a Condition C violation, this option is unavailable. Consequently, the only interpretive option in (46) is a higher-type trace. Because higher-type traces can produce reconstruction for scope but not for referential opacity, using a higher-type trace yields the reconstructed-scope, transparent reading in (46b), the only attested reading of (46). The surface-scope transparent reading in (46a) is ruled out because it would require a type-e trace, which is independently unavailable for $\overline{\text{A}}$-scrambling.

4.3 Interim summary

We have argued that Hindi provides evidence for the existence of both higher-copy neglect and higher-type traces as complementary mechanisms of reconstruction (Lechner 1998), because some but not all reconstruction effects in Hindi induce Condition C connectivity. The two reconstruction mechanisms have distinct empirical properties, which are summarized in (47) and (48). These conclusions converge with those reached independently by Lechner (2013, 2019).

This division of labor between the two reconstruction mechanisms derives the overarching empirical generalizations that we saw in Hindi. Because (i) reconstruction for referential opacity can only be achieved by neglecting the higher copy and (ii) such neglect induces Condition C connectivity in the launching site of movement, it follows that reconstruction for referential opacity correlates with Condition C, deriving the facts in Sect. 3.3. By contrast, scope reconstruction is not similarly restricted. It can be produced by either higher-copy neglect or higher-type traces. Because higher-type traces do not induce Condition C connectivity, scope reconstruction is not constrained by Condition C in the way that referential-opacity reconstruction is; this derives the facts in Sect. 3.2. Taken together, these consequences derive the empirical generalization I→C, repeated below in (49), from the interplay of the two mechanisms as complementary modes of reconstruction.

In the next section, we extend our account of Hindi reconstruction to two other semantic properties of $\overline{\text{A}}$-scrambling in Hindi: pronominal binding and weak crossover.

5 Extensions

The account developed in the previous section (Sect. 4) focused on the intricate relationships between reconstruction for scope, reconstruction for referential opacity, and Condition C connectivity. The claims that we made about the interpretation of scrambling in Hindi are general enough in nature to be assessed and applied in other domains as well, two of which we investigate in this section.²⁷ Sect. 5.1 assesses a prediction that emerges from our account with respect to reconstruction for pronominal binding. Section 5.2 extends the account to the classical weak crossover effects noted in Sect. 3.1.

5.1 Pronominal binding

LDS in Hindi is able to reconstruct for pronominal binding, as shown in (50), where the pronoun may be bound by the matrix subject , which it scrambles over. Because LDS invariably involves $\overline{\text{A}}$-scrambling, (50) thus demonstrates that $\overline{\text{A}}$-scrambling may reconstruct for pronominal binding.

Lechner (1998), Romero (1998), and Fox (1999) argue that higher-type traces do not allow for pronominal-binding reconstruction and that such reconstruction must therefore be the result of higher-copy neglect. In a nutshell, this restriction follows from the standard assumption that variables can only be bound by operators whose scope (i.e. c-command domain) they are in at LF. When using a higher-type trace, the scrambled element is in its landing site at LF, so a pronoun inside of it cannot be bound by an operator crossed by the movement, as schematized in (51).

This restriction entails that (50) must involve higher-copy neglect, as schematized in (52). By interpreting only the lower copy, the pronoun is within the scope of the quantificational matrix subject at LF, so that it may bind the pronoun. (We assume that the matrix subject undergoes a step of short movement, which binds both the trace and the pronoun, following Heim and Kratzer 1998.)

Against this backdrop, our account makes an immediate prediction: if reconstruction for pronominal binding requires higher-copy neglect, then it should induce Condition C connectivity. Fox (1999) shows that this prediction holds in English (data not included for reasons of space). As (53) demonstrates, this prediction is borne out in Hindi as well (Rajesh Bhatt, p.c.). In (53a), the $\overline{\text{A}}$-scrambled DP contains a bound pronoun () and an R-expression (). The $\overline{\text{A}}$-scrambling step crosses (i) a DP that binds the pronoun () and (ii) a pronoun that is coindexed with the R-expression (). The resulting structure is ill-formed. (53b) provides the relevant control structure, in which the positions of the R-expression and the coindexed pronoun have been swapped, so that Condition C is no longer at stake. The resulting structure is well-formed, demonstrating that the ill-formedness of (53a) is indeed the result of a Condition C violation.

The contrast in (53) is readily explained if (i) only higher-copy neglect may achieve pronominal-binding reconstruction and (ii) this procedure gives rise to Condition C connectivity, as argued in Sect. 4. Thus, in (53a), binding of the DP-internal pronoun requires interpreting the lower copy of the DP, which results in a Condition C violation. Interpreting the $\overline{\text{A}}$-scrambling via a higher-type trace is possible in (53a), but it is unable to produce a bound reading of the pronoun. Removing Condition C as a factor, as in (53b), permits higher-copy neglect and hence a bound reading of the pronoun.

5.2 Weak crossover

In this section, we briefly demonstrate that our analysis sheds light on another interpretive difference between A-scrambling and $\overline{\text{A}}$-scrambling. As Déprez (1989), Mahajan (1990, 1994), Gurtu (1992), and others have shown, local scrambling in Hindi is not subject to weak crossover and hence is able to feed pronominal binding from the landing site of movement (54a). By contrast, LDS displays weak crossover effects (54b).

While we are unable to do justice to the rich and varied literature on crossover phenomena within the scope of this paper, the proposal advanced in Sect. 4 provides a straightforward explanation for the contrast in (54), on the standard assumption that pronouns range over individuals (see e.g. Sauerland 1998; Ruys 2000). Let us first consider LDS, as in (54b). Because LDS in Hindi is invariably $\overline{\text{A}}$-scrambling, it must be interpreted via either higher-copy neglect or a higher-type trace (see (36)). If pronouns are of type e, then neither interpretive option allows binding of a pronoun from the landing site of movement. First, with higher-type traces, the λ-operator binding the trace is of type 〈et,t〉, but the pronoun is of type e. The λ-operator cannot bind the pronoun because their semantic types do not match (55). The resulting LF is wellformed, but lacks a bound reading of the pronoun.

Second, with higher-copy neglect, only the lower copy of the moved DP is interpreted. As this copy does not c-command the pronoun, binding is impossible (56).

The fact that neither interpretive mechanism allows the moved DP to bind a pronoun from its landing site derives the observation that $\overline{\text{A}}$-scrambling in Hindi—and hence LDS—is subject to weak crossover. Crucially, this follows from the independently motivated semantic interpretation of $\overline{\text{A}}$-scrambling from Sect. 4.

Compare LDS to local scrambling, which is not subject to weak crossover (54a). Because local scrambling can be A-scrambling, it is possible to interpret it with a trace of type e (see (37)), as motivated by the ability of a locally scrambled DP to take scope in its landing site. A second consequence of the type-e trace is that the λ-operator that binds this trace can additionally bind pronouns, as their semantic types match; this is shown in (57).

It follows then that A-scrambling allows binding of a pronoun, but $\overline{\text{A}}$-scrambling does not. Due to the clauseboundedness of A-scrambling, LDS is necessarily $\overline{\text{A}}$-scrambling, and hence is unable to bind pronouns. In this way, our account derives the crossover asymmetry between the two scrambling types from the independently observable scopal differences between them. To the extent that this extension to crossover is on the right track, it provides support for the view that crossover phenomena reduce to properties of quantificational scope (Ruys 2000).²⁸

6 The typology of traces and reconstruction

Despite the coexistence of two independent reconstruction mechanisms in our model, the model nonetheless imposes systematic restrictions on possible reconstruction profiles. For example, because reconstruction for referential opacity and pronominal binding can only be achieved through higher-copy neglect, the model predicts that such reconstruction effects should always induce Condition C connectivity, unlike reconstruction for quantifier scope. Furthermore, because higher-copy neglect gives rise to scope reconstruction, the model also predicts that any movement type that allows for reconstruction for referential opacity or pronominal binding—and therefore must have access to higher-copy neglect—should also permit scope reconstruction.

Within these restrictions imposed by the model, there is variability. Our model involves three mechanisms for interpreting movement dependencies: (i) type-e traces, (ii) type-〈et,t〉 traces, and (iii) higher-copy neglect. We have proposed that movement types have access to a subset of these mechanisms. If we assume that all combinatorial options are in fact possible, the model predicts 2³ interpretive profiles, which are listed in (58). (We focus here on movement of DPs; we discuss predicate movement below.)

Let us consider the reconstruction profiles in (58) in turn. Row (A) would correspond to a movement type that cannot be interpreted in any of the three ways. We assume that such a movement type is ruled out simply because the structure that it creates would be semantically uninterpretable. This leaves the seven combinations in rows (B)–(H).

Row (B) corresponds to a movement type with access to all three interpretive mechanisms. Such a movement type is instantiated by English wh-movement as characterized by Lechner (2013, 2019) (following Sharvit 1998). This movement type allows wide scope in the landing site of movement, implicating a type-e trace; it allows for referential-opacity reconstruction, implicating higher-copy neglect; and it dissociates scope reconstruction from Condition C, implicating a type-〈et,t〉 trace.²⁹

Row (C) is plausibly exemplified by what we will call “superlocal” scrambling in German: scrambling of a direct object over an indirect object but not over the subject (Lechner 1998, 2019). Lechner shows that superlocal scrambling may reconstruct for scope (59a), but not pronominal binding (59b) (though see Wurmbrand 2008 and Bobaljik and Wurmbrand 2012 for some qualifications). He proposes that this asymmetry follows if superlocal scrambling has access to a type-〈et,t〉 trace, but not to higher-copy neglect.

It is also clear that German superlocal scrambling has access to a type-e trace, because it may extend scope (59a) and feed pronominal binding (60), thereby obviating WCO.

Note that it is not possible to assess German superlocal scrambling for Condition C connectivity, because there is an independent constraint in German forcing pronouns to precede full DPs in the middlefield (Lechner 2019: 131). Moreover, referential-opacity reconstruction cannot be tested for superlocal scrambling more generally: because it is a very short movement step, there are no intensional operators that it may cross.

Row (D) corresponds to the properties of English wh-movement as characterized by Romero (1997, 1998) and Fox (1999): both scope extension and scope reconstruction are possible, but scope reconstruction induces Condition C connectivity. We return to the conflicting characterizations of English wh-movement below.

Row (E) is exemplified by superlocal scrambling in Hindi (i.e. scrambling of a direct object over an indirect object, but to the right of the subject; see Bhatt 2016 for an overview). Hindi superlocal scrambling extends scope (61), but it does not permit scope reconstruction (62).³⁰ Because both type-〈et,t〉 traces and higher-copy neglect result in scope reconstruction, both must be unavailable. Only type-e traces are available then—hence the scope extension.

If higher-copy neglect is unavailable, then pronominal-binding reconstruction should be unavailable as well. As (63) shows, this prediction is borne out.

According to Bhatt and Anagnostopoulou (1996: 16), Hindi superlocal scrambling also obviates Condition C violations, though their test configurations differ from ours (see (8)), which deserves followup in future research.

Row (F) corresponds to Hindi long scrambling, as argued at length in this paper.

The remaining two rows are not currently attested, as far as we know. Row (G)—only type-〈et,t〉 traces—would correspond to a movement type that obligatorily reconstructs for scope without incurring Condition C connectivity, but does not allow reconstruction for referential opacity or pronominal binding.³¹ Row (H)—only higher-copy neglect—would correspond to a movement type that obligatorily reconstructs for scope and allows reconstruction for pronominal binding and referential opacity (though does not require it). It would also always exhibit Condition C connectivity, and so in a Condition C configuration, it would be ungrammatical irrespective of the interpretation. Moreover, with no access to a type-e trace, both movement types described in (G) and (H) would not obviate WCO.

We note that the sample of languages and movement types that the existing reconstruction literature is based on is quite limited. It is striking though that even within the very limited sample size of English, German, and Hindi, there is such variability: five out of the seven patterns predicted by the model. At the same time, the current absence of movement types that instantiate rows (G) and (H) is difficult to interpret given this small sample size, and so we are hesitant to draw any conclusions from this absence. A more comprehensive evaluation of (58) thus calls for expanding the empirical basis, in particular for carrying out careful investigation of reconstruction profiles in a greater range of languages. We hope that our investigation of the reconstruction profile of Hindi long scrambling may serve as a model for such future work.

If the typology in (58) is on the right track, there are two immediate consequences. First, there is no default, always-available interpretive mechanism—contra Ruys’s (2015) Condition on Trace Typing, which always permits a type-e trace. Second, the attested variability demonstrates the need for an articulated theory of reconstruction mechanisms. It is difficult to see, for instance, how any theory that recognizes only a single reconstruction mechanism or a binary A/$\overline{\text{A}}$-distinction in reconstruction profiles would be able to accommodate the attested patterns. By postulating several interpretive mechanisms that may or may not be available for any given movement type, the theory developed here promises a more comprehensive analysis of reconstruction phenomena across a range of languages and movement types.

We have focused thus far on the combinatorial typology predicted by our model if otherwise left unconstrained. An important next question is whether there are further constraints on the availability of interpretive mechanisms for any given movement type in any given language. In other words, is it possible to predict the interpretive mechanisms available to a given movement type from other properties of the movement type or the language? This question, while important, falls outside the scope of this paper.³² It is also, we think, somewhat premature, given the empirical limitations discussed above, again highlighting the importance of establishing a broader empirical base of reconstruction effects.

The typology in (58) also invites us to reconsider the relationship between the conflicting generalizations about English wh-movement presented by Romero (1997, 1998) and Fox (1999) on the one hand (as obeying Q→C) and by Sharvit (1998) and Lechner (2013, 2019) on the other (as obeying I→C). Because our model can produce both patterns—rows (D) and (B) respectively—, it is possible, in principle, that both reconstruction profiles coexist and that the two generalizations are simply based on speakers with distinct grammars. This view would not affect the core argument in this paper: it is still the case that any theory of reconstruction that has Q→C as a necessary consequence is empirically too restrictive as a general theory of reconstruction (see Sect. 4.1).

Alternatively, it is possible that I→C is in fact the correct generalization for English wh-movement for all speakers, and that Romero’s and Fox’s failure to observe it stems from the fact that their examples did not control for intensionality. In this case, we predict that once intensionality is controlled for, a reconstructed-scope, transparent reading becomes possible. Consider again Romero’s (1998) original example in (10), repeated here as (64), which we have supplemented with a scenario that enforces a transparent interpretation of the moved DP. In this scenario, the English speakers we have consulted accept the sentence with a reconstructed-scope reading, though the judgement is subtle.

It is conceivable that reconstruction for scope preferentially coincides with reconstruction for referential opacity (plausibly as a parsing principle), and that as a result, scope reconstruction is degraded in cases where reconstruction for referential opacity is blocked, as in Condition C configurations. This is consistent with (64), where intensionality is controlled for, bringing to the fore the otherwise dispreferred reconstructed-scope interpretation. If this line of reasoning is on the right track, it resolves the apparent contradiction noted in Sect. 2, and it reconciles Romero’s (1997, 1998) and Fox’s (1999) evidence with I→C. We leave it up to future research to decide between these two approaches to the conflicting generalizations about English wh-movement.

Finally, we have focused on movement of nominals (i.e. DPs), but there are also movement types that target predicates (e.g. VPs, APs). Under the typology in (58), predicate movement would have to be interpreted via higher-copy neglect. Because predicates are of type 〈s,t〉 (or type 〈e,st〉 if not assuming predicate-internal subjects), they cannot be represented by traces of type e or 〈et,t〉 (or intensional variants thereof). Such an analysis of predicate movement is supported by the fact that predicate movement (in English) exhibits robust Condition C effects (Barss 1986; Huang 1993; Heycock 1995; Takano 1995). We think that the restriction on higher-type traces blocking referential-opacity reconstruction (see Sect. 4.2.1) would likewise block type-〈s,t〉 traces, though we leave exploring this issue to future research.

7 Summary and consequences

In this paper, we have offered an assessment of two longstanding but conflicting empirical generalizations about reconstruction effects, through the lens of scrambling in Hindi. One generalization, Q→C, claims that Condition C connectivity correlates with scope reconstruction (Romero 1997, 1998; Fox 1999). The other generalization, I→C, claims instead that Condition C connectivity correlates with referential-opacity reconstruction (Sharvit 1998; Lechner 2013, 2019). Based on novel evidence from Hindi, we have argued that Q→C does not represent a valid universal characterization of reconstruction effects, but that I→C plausibly does: Condition C correlates with reconstruction for referential opacity, not with reconstruction for scope.

We then explored the consequences of this finding for the mechanisms that underlie reconstruction. We argued that any account with Q→C as a consequence is empirically too restrictive. This conclusion challenges purely syntactic accounts of reconstruction that treat all reconstruction effects as the result of neglecting the higher copy (e.g. Romero 1997, 1998; Fox 1999; Sportiche 2016; Poole 2017, 2024). It also casts doubt on the purely semantic accounts of reconstruction in Sternefeld (2001) and Ruys (2015) that employ enriched higher-type traces to derive Q→C. We instead proposed that the Hindi reconstruction facts provide evidence that higher-copy neglect and higher-type traces coexist as complementary mechanisms of reconstruction, giving novel support for Lechner’s (1998, 2013, 2019) independently motivated hybrid model of reconstruction. We showed how together (i) the interaction of higher-copy neglect and higher-type traces and (ii) the restrictions on these two mechanisms derive the intricate Hindi reconstruction facts, viz. I→C.

The key consequence of our proposal is that some but not all reconstruction effects are syntactic—that is, amount to interpreting the lower copy at LF; other reconstruction effects are purely semantic. The remainder of this paper is devoted to discussing several consequences and issues that emerge from our proposal: the need for LF, reconstruction effects (counter)bleeding binding, the Trace Interpretation Constraint, and Trace Conversion.

7.1 The status of LF

The debate about whether reconstruction effects are syntactic or semantic in nature often features in the debate about whether or not it is necessary to posit Logical Form (LF), a level of syntactic representation distinct from surface structure that serves as the input to the semantic computation (e.g. Fox 1999; Jacobson 2002, 2004). Concerning this topic, we make two points: First, LF is necessary if we want to systematically account for how Condition C connectivity applies to (some but not all) reconstruction effects—some amount of copy neglect is needed. Approaches like Direct Compositionality, which do not posit LF, are thus inadequate. Second, at the same time, not all quantifier scope relations map onto c-command relations at LF. Higher-type traces allow a DP to scope below another scope-bearing element that it c-commands at LF. This state of affairs is, in some sense, a middle ground between what proponents and opponents of LF advocate for.

7.2 Reconstruction (counter)bleeding binding

The empirical focus of this paper has been on binding-theoretic connectivity bleeding and counterbleeding reconstruction effects. There is another side to this puzzle though, namely the inverse: reconstruction effects bleeding and counterbleeding (surface) binding possibilities.³³ Chomsky (1993, 1995) observes that in English, an anaphor in a moved wh-phrase can be evaluated for Condition A in the landing site of movement, so that it is bound by the matrix subject (65b). This interpretation is unavailable if the wh-phrase remains in situ (65a). However, as shown in (65c), when the wh-phrase must reconstruct for an idiomatic interpretation, binding by the matrix subject is no longer available. In other words, reconstruction for idiomatic interpretation bleeds anaphora binding in the landing site of movement.

Chomsky argues that the idiomatic interpretation in (65c) involves neglecting the higher copy, so that is not in a position at LF where it can be bound by . This analysis is compatible with the hybrid model of reconstruction from Sect. 4 if we assume that higher-type traces do not yield reconstruction for idiomatic interpretation. This predicts though, that reconstruction effects that can be achieved with higher-type traces—under our proposal, only scope reconstruction—should not bleed surface binding possibilities.³⁴

Crucially, to assess this prediction, the movement type in question must independently feed anaphora binding, like in (65b). Unfortunately, Hindi scrambling does not have this property; as shown in (66), scrambling a DP containing an anaphor does not change binding possibilities. Thus, Hindi does not bear on this prediction. (Note: The anaphor in (66) can marginally be bound by with a logophoric interpretation, but this reading is available irrespective of scrambling.)

7.3 Trace Interpretation Constraint

Poole (2017, 2024) proposes a general ban on higher-type traces (his Trace Interpretation Constraint): movement may either reconstruct (via higher-copy neglect) or be interpreted with an individual-type trace (see also Chierchia 1984; Landman 2006). This proposal is at odds with our arguments in favor of higher-type traces in Hindi. We leave reconciling these two proposals for future research. However, we would like to highlight what we believe to be a substantive difference between the kinds of evidence considered in this paper and in Poole (2017, 2024): the empirical arguments for Poole’s constraint do not involve Condition C connectivity but rather involve instances where reconstruction is blocked or is obligatory. This difference might represent a path towards reconciling these two conflicting proposals.

7.4 Trace Conversion

Throughout this paper, we have assumed that traces are simplex variables, but a substantial body of work has advanced the hypothesis that traces are in fact bound definite descriptions, commonly known as Trace Conversion (Engdahl 1980, 1986; Sauerland 1998, 2004; Fox 1999, 2002, 2003; Poole 2024). The issue of whether traces are simplex or definite descriptions is largely orthogonal to considerations of what semantic types a trace can be. It is, in principle, possible for bound definite descriptions to be of higher semantic types.

As a proof of concept, (67) sketches a variant of that Trace Conversion could substitute in for the lower copy’s determiner in order to produce a generalized-quantifier trace (〈et,t〉).³⁵ For the sake of simplicity, (67) encodes the variable in the definite determiner itself and sets aside the definite’s presupposition. A sample derivation is given in (68), where Op stands for some scope-bearing operator (e.g. a modal).

There are two general issues concerning Trace Conversion and higher-type traces that need to be addressed. The first is that Trace Conversion must be unable to produce reconstruction for referential opacity. In addition to the proposals in Sect. 4.2.1 (see (42)), we need to say something else to block an intensional operator occurring between the two copies from binding the situation pronoun in the lower copy but not the higher copy (since the operator does not c-command the higher one). This restriction presumably follows from the higher and lower copies needing to be evaluated at the same situation. Assuming a Lewisian theory of transworld identity between objects (Lewis 1968, 1986), where no individual exists in multiple possible worlds (i.e. maximal situations), if the higher and lower copies are evaluated with respect to different worlds, then the statement $\mathcal{Q}=g(n)$ in (an intensionalized version of) (67) cannot be satisfied. Thus, both the higher copy and the lower copy must be bound by an intensional operator that has both copies in its scope; that is, it must c-command both copies at LF.

The second issue is more general: Trace Conversion must obviate Condition C violations for relative clauses, irrespective of the semantic type of the trace. On standard assumptions, this property follows from the relative clause late-merging onto the moved element after movement has occurred so that the lower copy never contains the offending R-expression (Lebeaux 1988, 1990, 2000, 2009; Fox 2002). Late-merge blocks neglecting the higher copy because it would strand the relative clause without a host. As a result, neglecting the higher copy is only possible if the relative clause is first-merged in the lower copy. Therefore, the derivation that allows circumventing Condition C is possible with traces (as definite descriptions), but not with higher-copy neglect. However, alternative explanations of Lebeaux effects (e.g. Sportiche 2016; Poole 2017) are equally compatible with the claims made here.