If the binding conditions were computed from the landing site of the topicalized phrase, one would expect topicalization of a reflexive pronominal direct object to bleed binding. However, example (513b) shows that the topicalized reflexive pronoun zichzelf behaves with respect to binding as if it were still in the position indicated by its trace; this shows that what is involved in binding is clearly not linear order. Note that coreferentiality is again indicated by italics.

That topicalization does not bleed binding is also illustrated in the examples in (514), in which a reciprocal possessive pronoun is embedded in a direct object. The fact that topicalization of this object does not affect the binding possibilities again shows that binding cannot be captured in terms of linear order. Note in passing that the complementarity distribution of referential and reciprocal pronouns does not hold for possessive pronouns, since elkaars can easily be replaced by huntheir without changing the binding relation (but of course a different meaning); cf. Section N22.1 for a detailed discussion.

Another case showing that topicalization does not bleed binding is illustrated by the examples in (515), which allow a bound-variable reading of the possessive pronoun zijnhis, according to which every person x admires his own parents: ∀x (x:person) admire (x, x’s parents). This reading can only arise if the quantifier binds (hence: c-commands) a referential pronoun, and we might therefore expect that topicalization in (515b) would make this reading impossible, but this expectation is not borne out.

If the binding conditions were computed from the landing site of wh-movement, one would expect that topicalization of a referential (pronominal) direct object would enable it to function as the antecedent of the subject of its clause, but example (516b) shows that this is not the case: with respect to binding, the objects hem and die jongen again behave as if they were still in the position indicated by their trace.

A plausible hypothesis would be that example (516b) is unacceptable because the subject Jan is bound by the topicalized phrase and thus violates binding condition C. However, this hypothesis is refuted by the fact that the matrix subject Jan in (517b) can be coreferential with the topicalized pronoun hemhim: this again leads to the conclusion that wh-movement does not affect binding possibilities. Note that the question mark in brackets indicates that (517b) may be slightly marked compared to example (509b), but this is a more general property of long topicalization; cf. Section 11.3.3, sub II.

Reconstruction is also sometimes illustrated in the literature by examples such as (518a), in which a bound nominal phrase is embedded in a complementive.

Some linguists do not accept (518b) as a convincing case of reconstruction because there is reason to assume that the subject originates as the external argument of the AP: assuming that the moved phrase is a small clause containing an NP-trace of the subject Jan, this trace can serve as an antecedent for the nominal phrase.

However, even if the representations in (519) are the correct ones, reconstruction may still be necessary, because it is generally assumed that NP-traces are also subject to binding condition A: like reflexive pronouns, they must be bound by their antecedent (i.e. the moved phrase) within their local domain. This is made explicit by the extended binding condition A in (520).

A similar configuration as in (519b) is found in VP-topicalization constructions, if we follow the now generally accepted hypothesis that subjects are base-generated within the lexical projection of the verb; we do not need to appeal to reconstruction to account for the acceptability of examples such as (521b), because the topicalized VP also contains the NP-trace of the subject Jan, which can serve as an antecedent of zichzelf. But then, according to binding condition A in (520), reconstruction is again necessary to satisfy this condition for the NP-trace ti.

VP-topicalization constructions of the type in (522) also provide evidence for reconstruction. Under the standard assumption that the clause-initial position can only be filled by phrases (and not by heads), the theme argument must have been extracted from the VP by NP-movement (i.e. nominal argument shift of the type discussed in Section 13.2) before VP-topicalization occurs. The topicalized VPs in the primed examples thus contain a trace of the theme argument and reconstruction is necessary in order for the NP-traces ti to be bound by mijn huismy house; cf. Section 11.3.3, sub VIC, for further discussion.

The examples so far all involve the topicalization of arguments, complementives, and VPs, and we have seen that such cases exhibit reconstruction effects: binding possibilities are computed from the base position of the moved phrase. However, this does not always seem to hold for adjuncts, as is clear from the contrast between the two examples in (523); if the adverbial clause in (523b) were interpreted in the same position as the adverbial clause in (523a), we would wrongly expect coreference between Jan and hij to be blocked by binding condition C in both cases; cf. Reinhardt (1983a: §3).

This contrast has given rise to the idea that examples such as (523b) are not derived by wh-movement, but involve base-generation of the adjunct in clause-initial position; that this is possible is then attributed to the fact that adjuncts are not selected by the verb and can thus be generated externally to the lexical projection of the verb. Note, however, that the lack of reconstruction cannot be demonstrated on the basis of binding condition B, since referential pronouns embedded in adverbial clauses can always be coreferential with the subject of a matrix clause; this is shown in (524).

A similar lack of reconstruction can be observed in the examples in (525); cf. Van Riemsdijk & Williams (1981). In this case an argument is topicalized, but the contrast between the two examples shows that the reconstruction effect does not occur: contrary to what would be expected if the topicalized phrase were interpreted in the position of its trace, the referential noun phrase Jan embedded in the relative clause can be coreferential with the pronoun hij in (525b). Of course, it is not possible to appeal to an argument-adjunct asymmetry in this case, but it has been suggested that the (optional) relative clause is an adjunct that can be generated after the object has undergone wh-movement; cf. Barss (2001) and Sportiche (2006/2017) for details.

The examples in (526) show again that the lack of reconstruction cannot be demonstrated on the basis of binding condition B, since referential pronouns embedded in a relative clause can be coreferential with the subject of a matrix clause.

The discussion of the data in this subsection has shown that a reconstruction effect is usually found when an argument, complementive or verbal projection is topicalized. Such effects may be absent when an adverbial clause occupies the clause-initial position or when the topicalized phrase is modified by a relative clause, but this can only be demonstrated for apparently pied-piped referential noun phrases, not for pronouns. We have given two early accounts of this lack of reconstruction, but it has also been suggested that it shows that binding condition C differs from binding conditions A and B in that it is not actually a syntactic rule; we will return to this issue in Section N22.1, sub IIA.

Because wh-movement has a clear effect on meaning, the standard (but not uncontroversial) assumption is that it precedes the semantic interpretation of the clause. The fact that, for the purpose of the binding theory formulated in (510), topicalized phrases behave as if they still occupy the position indicated by their traces has led to theories according to which wh-movement is at least partially undone before the semantic interpretation of the syntactic representation takes place; the technical term for this is Reconstruction. A more recent approach, which makes reconstruction superfluous, is Chomsky’s (1995a: §3) copy theory of movement, according to which movement is a copy-and-paste operation that leaves a phonetically empty copy (a copy that is not pronounced in the actual utterance) of the moved constituent in its original position. We will follow common practice and use the term reconstruction as a purely descriptive term. The core finding that all generative theories attempt to explain is that binding of nominal arguments should be formulated in terms of A-positions, i.e. argument positions to which thematic roles, agreement features, and/or case are assigned; movement into A'-positions (positions such as the clause-initial position, which can also be occupied by non-arguments) does not affect the binding possibilities. We refer the reader to Barrs (2001), Sportiche (2006/2017), and Salzmann (2006) for critical reviews and discussions of the various theoretical implementations of this insight.

The standard view seems to be that reconstruction effects are syntactic in nature, but there are reasons to doubt that these effects are part of syntax proper. To show this, we need to make a brief digression into hanging-topic and contrastive left-dislocation; cf. Section C37.2 for a detailed discussion. Left dislocation is characterized by the fact that there is a phrase preceding the clause-initial position that is associated with a resumptive element elsewhere in the clause. The two types of left-dislocation constructions differ in the form and position of the resumptive element: hanging-topic left-dislocation constructions have a resumptive pronoun in the form of a referential pronoun, such as hemhim in (527a), located in the middle field of the clause; contrastive left-dislocation constructions have a resumptive pronoun in the form of a demonstrative pronoun, such as diethat in (527b), located in the clause-initial position. Note that we have indicated the relation between the left-dislocated phrase and the resumptive pronoun by indices (i.e. in the same way as the relation between a moved phrase and its trace).

At first glance, the examples in (528) seem to show that left dislocation differs from topicalization in that it does affect the binding possibilities. However, Van Riemsdijk & Zwarts (1997) and Vat (1997) suggest that the unacceptability of the examples in (528) is due to the fact that resumptive pronouns are referential pronouns, which are themselves subject to binding condition B; in order to satisfy the binding conditions on the reflexive zichzelfhimself, the resumptive pronouns hemhim and diethat must take the subject Jan as a local antecedent, thus violating binding condition B. Note that the binding options for the resumptive pronoun die in (528b) should be computed from its original object position, indicated by its trace in object position, a case of reconstruction.

Violations of binding condition B induced by the resumptive pronouns themselves can be avoided when the reflexive/reciprocal pronoun is more deeply embedded in the topicalized phrase, as in the examples in (514). Their left-dislocation counterparts in (529) show that the two types of left dislocation behave differently in such cases; while the hanging-topic construction is given as ungrammatical in Van Riemsdijk & Zwarts (1997) and Vat (1997), the contrastive left-dislocation construction is perfectly acceptable. The fact that the left-dislocated phrase can be interpreted in the position of the trace of the wh-moved demonstrative die has become known as the connectivity effect.

Connectivity effects also occur in the left-dislocation counterparts of the topicalization construction in (515b) with a bound-variable reading. Van Riemsdijk & Zwarts (1997) and Vat (1997) show that there is again a contrast between hanging-topic and contrastive left-dislocation.

For completeness’ sake, consider the contrastive left-dislocation constructions in (531), which show again that the acceptability judgments on the contrastive left-dislocation constructions are more or less the same as in the corresponding topicalization constructions in (516b) and (517b).

The above discussion has shown that contrastive left-dislocation constructions exhibit connectivity effects that closely resemble the reconstruction effects found in topicalization constructions. Given this similarity, it is tempting to provide a single theoretical account of the two types of effect, e.g. by postulating that there is some kind of matching effect, in the sense that the demonstrative pronoun die simply adopts certain semantic properties of the left-dislocated phrase and transfers them to the position of its trace. However, this seems to run counter to the current leading idea that reconstruction effects follow from the copy theory of movement, according to which movement leaves a copy of the moved constituent in its original position.

Alternatively, one could try to show that left-dislocated phrases are base-generated within the clause to which they are attached, and find their surface position by (a series of movements including) wh-movement. If such an analysis is feasible, we could maintain that reconstruction effects result from the copy-and-paste operation proposed by the copy theory of movement; cf. Grohmann (2003: §4) and De Vries (2009a) for detailed proposals. This would immediately explain the differences in connectivity effects between hanging-topic and contrastive left-dislocation constructions found in this subsection: hanging-topic constructions have a resumptive pronoun in the middle field of the clause, and we can therefore safely conclude that they do not involve wh-movement, and consequently we expect no connectivity effects. However, there are two possible problems with this approach. First, there seems to be no independent evidence to support the assumption that left-dislocated phrases have ever occupied a clause-internal position. Second, this approach should provide an explanation for the fact that left-dislocated phrases can strand prepositions, while topicalized phrases (and wh-moved phrases in general) are usually unable to do so; cf. the contrast between the (a) and (b)-examples in (532).

We will return to the question of whether reconstruction and connectivity effects can be given a (more or less) unified treatment in the discussion of relativization in Subsection IV below.

As with wh-questions, reconstruction for binding condition A is difficult to detect because the reflexive/reciprocal pronoun must be embedded in a larger wh-moved noun phrase: relative pronouns are never themselves reflexive/reciprocal. Moreover, because the relative pronoun in such complex noun phrases is typically a possessive pronoun such as wienswhose, we expect it to be construed as the antecedent of the reflexive/reciprocal pronoun; cf. Section N22. The antecedent of the reflexive in examples such as (540a) must therefore be the relative possessive pronoun wiens, and the impossibility of construing the subject of the relative clause (i.e. the pronoun hijhe) as the antecedent of zichzelf, as in (540b), tells us nothing about reconstruction. As usual, the intended binding relations are indicated by italics.

Examples such as (541b) with a bound-variable reading do seem to provide evidence for reconstruction, although some speakers may find it difficult to give a judgment on this example due to its complexity.

Reconstruction for binding condition B is again difficult to establish, because referential pronouns embedded in a noun phrase containing a possessive pronoun can usually be coreferential with noun phrases external to that noun phrase. Moreover, the acceptability of (542b) tells us nothing about reconstruction, since referential pronouns do not require a c-commanding antecedent.

For binding condition C it is possible to show that reconstruction effects do occur: the intended coreference relation is excluded in both examples in (543). The fact that referential noun phrases can normally have a non-c-commanding antecedent suggests that reconstruction is obligatory.

Despite the difficulty of constructing relevant examples, the arguments based on the bound-variable reading of pronouns and binding condition C show conclusively that relative clauses exhibit similar reconstruction effects as wh-questions and topicalization constructions.

The discussion in the previous subsection has shown that reconstruction within relative clauses is indeed obligatory. However, the research on relative clauses that has attracted the most interest is not concerned with reconstruction effects of the kind discussed above, but with connectivity effects of the kind we have also found in contrastive left-dislocation constructions; cf. Subsection IIB.

The connectivity effect for binding condition A is illustrated in example (544); on the default interpretation that the rumors are not spread by Jan himself, the reflexive pronoun zichzelfhimself can only be bound by Jan if we assume that the antecedent of the relative pronoun datwhich is interpreted at the position of the relative pronoun.

Connectivity effects can also be illustrated by example (545) in its bound-variable reading. Since this reading arises only when a quantifier binds (and thus c-commands) a referential pronoun, we must again assume that the antecedent of the relative pronoun diewhich is interpreted at the position of the relative pronoun.

Establishing connectivity effects for binding condition B is again somewhat delicate, because referential personal pronouns embedded in a noun phrase can often be coreferential with the subject of their clause if they are phonetically reduced. However, example (537) has shown that phonetically non-reduced pronouns do not easily allow this. The fact that we do not find the same contrast in the relative construction in (546) may argue against the assumption of a connectivity effect, but this needs further investigation, as it is not clear whether we are really dealing with a syntactic restriction or with some other kind of restriction.

A more serious problem is that connectivity effects for binding condition C are not found in relative clauses: example (547) does readily allow an interpretation in which the noun phrase Jan and the subject pronoun of the relative clause are coreferential.

The examples in this subsection lead to a somewhat ambivalent result: connectivity effects can be established for examples like (544) and (545) involving binding condition A and the bound-variable reading of pronouns, but not for examples such as (547) involving binding condition C. This might lead to the conclusion that connectivity effects occur only in the case of local (i.e. clause-internal and NP-internal) syntactic dependencies. This may be consistent with the traditional view in generative grammar, currently embedded in Chomsky’s (2008) phase theory, that there are no syntactic constraints on non-local relations, but would be problematic in view of Salzmann’s (2006: §2.2) observation that connectivity effects differ crucially from reconstruction effects in that the latter effects also occur with non-local relations.

There has been a heated debate in the theoretical literature as to whether the connectivity effects in relative clauses can be reduced to reconstruction. This debate has its origin in Vergnaud (1974), where it was claimed that, descriptively speaking, the antecedent of the relative pronoun is base-generated within the relative clause, placed in the initial position of the relative clause by wh-movement, and subsequently raised to its surface position in the main clause; for updated versions of this so-called promotion/raising analysis, see Kayne (1994), Bianchi (1999), and De Vries (2002). Despite its popularity, the promotion/raising analysis is still not uncontroversial, as it raises a large number of technical and theory-internal problems; cf. Boef (2013) for a review. For example, it is still not clear why the antecedent is able to strand prepositions promotion/raising analysis, while this is usually impossible in cases of wh-movement like topicalization and question formation; cf. the contrast between the examples in (548a&b) and (548c).

Moreover, Salzmann (2006) points out that the differences between reconstruction and connectivity effects for binding conditions B and C discussed in this section are problematic for this analysis.