Landing-site restriction (559a) correctly predicts that wh-moved phrases can serve as antecedents of parasitic gaps. This is illustrated in (567) for a wh-question, a topicalization construction, and a relative clause.

The overt movement restriction in (559b), on the other hand, does not seem to hold for Dutch as the multiple wh-question in (568a) is perfectly acceptable. However, the situation is more complex than it seems at first glance, since (568b) is unacceptable.

Since the position of the object in (568b) is taken to be its base position within the VP, we may assume that this is the construction that resembles the unacceptable English multiple wh-question in (560) most closely. It seems that (568a) is derived from this structure by leftward movement of the object into some structurally higher position; more precise representations of the examples in (568) are thus as indicated in (569).

The leftward movement of the object in (569a) is known as scrambling, and the non-interrogative counterparts of the examples in (569) in (570) show that scrambling is indeed capable of licensing parasitic gaps; cf. Bennis & Hoekstra (1984).

The contrast between the (a) and (b)-examples of (569) and (570) would follow from the landing site and overt movement restriction in (559a&b) if scrambling were an instance of A'-movement. The following subsection will show, however, that this is not easy to determine and that much depends on the specific version of the overall theory adopted.

But first, consider the examples in (571), which show that parasitic gaps easily alternate with overt referential personal pronouns when their antecedent is a scrambled phrase, as shown in Bennis & Hoekstra (1984) and Huybregts & Van Riemsdijk (1985), but that this is more difficult when the antecedent is interrogative; this is especially true when the wh-phrase is not D-linked, which is prototypically the case for the interrogative pronoun watwhat (although it can sometimes get a D-linked reading in specific contexts not discussed here).

To our knowledge, the contrasts in acceptability between the three types of example in (571) have not been noted before. It seems plausible to relate the differences to the degree of referentiality of the antecedents of the parasitic gap; referential noun phrases and pronouns obviously have a high degree of referentiality, while D-linked wh-phrases such as welk boekwhich book and non-D-linked wh-pronouns such as watwhat have an intermediate and low degree of referentiality, respectively.

The term scrambling refers to the fact that in certain languages the word order of the sentence can vary; for Dutch and German it is normally used to refer to certain variations in the word order of the middle field of the clause. The term is somewhat misleading, however, because it suggests that it refers to a single operation with well-defined properties. Chapter 13 will show, however, that there are different types of operations with quite different properties that can affect the word order of the middle field of the clause: some have properties of A-movement, while others have properties of A'-movement. But even if we restrict the notion of scrambling to the leftward movement of nominal arguments (i.e. subjects and objects), it is very difficult to determine definitively which type of movement we are dealing with, since this is closely related to the overall theory one adopts. This subsection contains a brief theoretical digression to illustrate this.

The term argument position (A-position) denotes positions in the clause that can only be occupied by arguments of the verb. Such positions are characterized by the fact that they can be assigned specific syntactic features, the three main types of which are: thematic roles, structural case, and nominal agreement features (person, number, and gender). Prototypical A-positions are the subject and the object position. The term non-argument position (A'-position) denotes positions that can also be occupied by non-arguments (adverbial phrases, etc.). Such positions function as landing sites for elements with a specific logico-semantic role (such as operator or negation) or an information-structural function (topic, focus, etc.); a prototypical A'-position is the clause-initial position, which can be occupied by any clausal constituent as a result of wh-movement.

The number of A and A'-positions proposed in generative grammar has increased considerably over the years. As for A-positions for nominal arguments of verbs, in the early 1980s only two positions were available: the object and the subject position in the simplified structure in (572a). The object position within the VP is the position to which the thematic role of theme, accusative case, and (for languages with object agreement) object agreement features can be assigned; the subject position is the position to which the thematic role of agent, nominative case, and subject agreement features can be assigned. Arguments can also sometimes pick up their features in different places; in the unaccusative construction in (572b) the subject John is base-generated in the object position, where it is assigned the thematic role of theme, and subsequently moved into the subject position, where it is assigned nominative case and subject agreement features.

Given that the object and subject positions exhaust the A-positions postulated for nominal arguments at this stage, it is a virtual necessity to assume that scrambling is A'-movement, i.e. movement targeting some A'-position in the middle field of the clause. It is therefore not surprising that an early article such as Bennis & Hoekstra (1984) comes to this conclusion.

The fact, illustrated in (572b), that the syntactic features of a given argument can be distributed over more than one position within the clause has led to the hypothesis that there is a one-to-one relation between features and positions. For example, instead of assuming that all features for the direct object are generated in a single position, it is now generally assumed that they are assigned by different functional heads, like those indicated in capitals in (573), to their complement or specifier: the main verb assigns the role of theme to its complement, the AGR-head assigns the agreement features to its specifier, and the CASE-head assigns accusative case. Something similar is assumed for subjects. Note that the names used in (573) for these functional heads are arbitrary; since Pollock’s (1989) seminal work on the subject, a large number of implementations of the main idea can be found in the literature.

Because all A-positions in (573) are considered potential landing sites for the theme argument, it will be clear that the number of potential A-movements in the derivation of sentences has increased enormously compared to the earlier proposal in (572); the same holds for verb movement, because all functional heads in (573) are assumed to be potential landing sites for the verb. This makes it possible to analyze scrambling of nominal arguments as A-movement; this position is taken in Broekhuis (2008/2011), where it is argued that the theme position in (573) is cross-linguistically the base position of the object (and where object clauses usually appear), that agreement features are located in the object position preceding the clause-final verb (which in earlier versions of the theory was considered the base position of the object in Dutch), and that scrambling of the object targets the accusative position.

Since the seminal work of Haegeman & Zanuttini (1991), Haegeman (1995), and Rizzi (1996/1997), there has also been a proliferation of A'-positions; whereas in the early 1980s there was only one clearly defined A'-position, the landing site of wh-movement, more recent research claims to have identified a large number of additional A'-positions in structurally lower positions, which can be targeted by negative, focused, topical, quantified phrases, etc. Again, this makes it possible to analyze certain forms of scrambling, including those involving leftward movement of nominal arguments, as A'-movement. All of this means that we can no longer rely, as in the early 1980s, on theory-internal considerations, but must develop empirical tests for determining the A or A'-status of a particular form of scrambling.

We will use binding as a diagnostic tool to determine whether the object movement found in the scrambling variant of the parasitic gap construction should be considered A or A'-movement, as these movement types can be shown to differ in whether or not they affect binding relations. We illustrate this with English data so as not to bias our discussion of Dutch.

A'-movement does not change the binding possibilities, as is clear from the examples in (574): the (a)-examples show that topicalization of the reflexive pronoun does not change its binding potential, and the (b)-examples show that topicalization of a potential antecedent does not create new binding possibilities; cf. Section 11.3.7 on reconstruction for a more detailed discussion as well as the relevant Dutch data.

A-movement differs from A'-movement in that it does affect binding. This can be seen from the subject-raising examples in (575), taken from Den Dikken (1995b); cf. Section 5.2.2.2 for an introduction to subject raising. The traces indicate the current standard analysis of examples of this kind: in (575a) the expletive there is raised from the subject position of the infinitival clause into the subject position of the matrix clause, while in (575b) it is the noun phrase some applicants that is ultimately raised into the subject position of the clause. The crucial point is that in (575a) the noun phrase some applicants is clearly within the infinitival clause and therefore does not c-command the nominal complement of the to-PP, the reciprocal each other, while in (575b) the noun phrase some applicants is moved into the subject position of the matrix clause, from where it c-commands the reciprocal each other. The acceptability contrast between the two examples thus proves that A-movement differs from A'-movement in that it affects binding.

The examples in (576) show essentially the same for the bound-variable reading of referential pronouns: the quantifier in (576a) is embedded in the infinitival clause and therefore does not c-command the pronoun embedded in the complement of the to-PP, while the quantifier in (576b) is in the subject position of the matrix clause, from where it does c-command the pronoun. This explains why the bound-variable reading is only available in the latter case.

The previous subsection has shown that A but not A'-movement feeds binding. This finding has played an important role in the discussion of whether scrambling of nominal arguments should be seen as A or A'-movement, or perhaps as not involving movement at all; a representative sample of such approaches can be found in Corver & Van Riemsdijk (1994).

Webelhuth (1989/1992) has argued that Dutch/German object scrambling exhibits properties of both A and A'-movement in that it not only licenses parasitic gaps but also feeds binding, a fact known as Webelhuth’s paradox. That object scrambling can license parasitic gaps was already illustrated in (570), and that it can also feed anaphor binding is illustrated in (577); cf. Vanden Wyngaerd (1988/1989b). Note in passing that we only consider cases with a neutral (non-contrastive) intonation pattern; example (577a) seems to improve somewhat when the adverbial phrase namens elkaaron behalf of each other is given a contrastive accent, which may be because in that case example (577a) would be derived from (577b) by reconstructible focus movement.

Webelhuth’s crucial observation, illustrated in the German example in (578), is that scrambling can simultaneously feed binding and license a parasitic gap. The structure shown is that given in Webelhuth (1992:207): the scrambled quantified direct/accusative object jeden gast binds the possessive pronoun embedded in the indirect/dative object seinem Nachbarnhis neighbor (resulting in a bound-variable reading), while simultaneously licensing a parasitic gap. Such examples cannot be reproduced in Dutch because it does not easily allow the inversion of indirect and direct objects in double object constructions.

Webelhuth gives a question mark to examples such as (578), noting that they are “as good or as bad as” other parasitic gap constructions. He concludes from these examples that the dichotomy between A and A'-positions is too coarse, and that we must assume a third, Janus-faced position with properties of both A and A'-positions. This reasoning was sound in the early 1990s, but the subsequent proliferation of A and A'-positions allows a somewhat different view on examples of this kind: instead of assuming that the scrambled phrase is moved to its surface position in one fell swoop, we can now claim that it arrives there in a step-by-step fashion; cf. Mahajan (1990/1994) for early proposals of this kind. This leads to structures such as the one given in (579) with an additional trace t': if the first movement step is of the A-movement type, the trace t' is in an A-position and thus able to bind the possessive pronoun seinem; if the second step is of the A'-movement type, the scrambled phrase ends up in an A'-position, from where it can license the parasitic gap.

Since it is generally assumed since Chomsky (1986a) that A'-movement cannot precede A-movement, a restriction known as the ban on improper movement, the proposed solution to Webelhuth’s paradox makes a very strong prediction: the phrase containing the parasitic gap must be in a structurally higher position than the phrase containing the A-bound pronoun. However, this does not seem easy to test. At first glance, the German example in (580a), taken from Mahajan (1990:60), seems to confirm this prediction: since the direct object binds a parasitic gap, it must be in an A'-position and therefore cannot bind the possessive pronoun. However, Müller & Sternefeld (1994) and Lee & Santorini (1994) have shown that replacing the indirect object seinem Nachbarn with an indirect object without a pronoun, such as der Maria in (580b), does not improve the result. This suggests that example (580a) is excluded for independent reasons and therefore has no bearing on the issue under discussion; cf. Kathol (2001) for an argument based on these examples that we are not dealing here with parasitic but pseudo-parasitic gaps in the sense of Postal (1994), according to which the construction is derived from a coordination-like structure by across-the-board movement (a proposal that may raise more questions than it solves).

We cannot replicate the German data for Dutch double object constructions, because indirect objects must precede direct objects in non-contrastive contexts. But perhaps the examples in (581), in which the bound pronoun and the parasitic gap are both embedded in an adjunct, can be used to illustrate the same thing; note that the (b)-examples should be read as continuations of the (a)-example.

Unfortunately, the judgments on these examples are also somewhat problematic. First, we should note that Neeleman (1994a) gives the continuation in (581b') as acceptable, which means that the judgment given here is not uncontroversial. Second, we tend to think that this continuation leads to a marginally acceptable result only if the adverbial PP namens elkaar is followed by an intonation break. If so, the infinitival clause containing the parasitic gap might be epenthetic, which would complicate the analysis considerably, as epenthetic phrases are often assumed to be clause-external. All in all, this means that the status of the continuation in (581b') is simply not clear enough for us to draw any firm conclusions. We therefore tentatively assume that the predictions that follow from the ban on improper movement are essentially correct, until more conclusive counterevidence is provided.

Note in passing that Neeleman provides example (581b') to show that object scrambling is not in fact a movement operation, but that scrambled objects are base-generated in their surface position, as in representation (582a): if true, this would imply that movement is not necessary for the licensing of parasitic gaps, and consequently that both the landing-site restriction and the overt movement restriction should be rejected. Neeleman further claims that nominalizations such as (582b) support this hypothesis that parasitic gaps can be licensed by noun phrases in their base position: the theme argument boeken is able to license the parasitic gap, despite the fact that it is putatively base-generated as the complement of the preposition van.

This argument may have been theoretically sound in the early 1990s, but in more recent years it has been argued that there is much more movement within noun phrases than meets the eye; cf. Hoekstra (1999) for an analysis of this example that takes the traditional (A'-movement) approach to parasitic gaps. It is therefore no longer obvious that example (582b) can be used as evidence for the hypothesis that scrambled phrases are base-generated in their surface position, from where they can license parasitic gaps; we will return to this hypothesis in the next subsection, where it will be shown to have a serious empirical inadequacy.

The previous subsection has shown that Webelhuth’s paradox finds a more or less natural explanation in more recent versions of generative grammar, which make more clause-internal A and A'-positions available. However, there is still a serious problem with the landing-site restriction (559a), since several linguists have independently claimed that parasitic gaps can occur in Dutch passive constructions. Broekhuis (1987/1992) has claimed that the result is somewhat less acceptable than in other cases, but attributes this to the fact that the implied PRO-subject of the infinitival clause requires a controller (Van Haaften 1991), as is clear from the fact, illustrated in (583a), that the construction is also marked when the parasitic gap is replaced by an overt pronoun; cf. Kathol (2001: §9.6), where the same account is given for similar German examples, which are deemed ungrammatical. Neeleman (1994a) and De Hoop & Kosmeijer (1995) give their examples as straightforwardly acceptable, which may have to do with the fact that they include an agentive door-PP, which may help to identify the implied PRO-subject; example (583b) shows that the addition of a door-phase indeed improves the parasitic gap construction in (583a).

To our knowledge, the consequences of the (relative) acceptability of the passive constructions in (583) have not yet been sufficiently studied. Broekhuis (1987/1992) suggests that the subject position in Dutch is in fact not an A but an A'-position, which he supports by claiming that subjects of subject-raising constructions such as (584a) are not able to bind (into) an indirect object of the matrix clause; cf. the discussion of the English examples in (575) and (576). Much rests on the claim that examples like those in (584) are ungrammatical, but this may be overstating the case; the judgments may simply not be clear enough to draw firm conclusions.

Another possibility, which has not yet been explored, is that the nominative noun phrase die boeken does not occupy the subject position at all in the examples in (583). This is a plausible option, because it can easily be shown that definite subjects do not have to occupy the regular subject position when they are part of the (new information) focus of the clause. This is illustrated in (585a), where the definite subject does not have to be right-adjacent to the complementizer datthat, but can also occur further to the right. That information structure is involved is clear from the fact that (phonetically reduced) referential subject pronouns, which are intrinsically part of the presupposition of the clause, do not have this option; cf. Section 13.2.

This would predict that the examples in (583) would become unacceptable if we replace the noun phrase die boekenthose books with a referential pronoun for. It is not clear to us whether this prediction turns out to be true; although the examples in (586) may indeed be somewhat harder to interpret, this may simply be a side effect of the fact that they are given without an appropriate context.

De Hoop & Kosmeijer (1995) and Neeleman (1994a) claim that parasitic gaps can be licensed by an antecedent in an A-position, which amounts to saying that the landing-site restriction does not apply to Dutch. Their claim implies that the standard assumption that parasitic gaps are subject to binding condition C should be replaced by the assumption that they are subject to binding condition A or B. The fact that the antecedent of a parasitic gap is external to the infinitival clause in (587) suggests that the parasitic gap is free in its local domain, and therefore not subject to binding condition A.

The claim that the antecedent of parasitic gaps can be in an A-position thus inevitably leads to the conclusion that parasitic gaps are subject to binding condition B. This, in turn, predicts that the antecedent of the parasitic gap can be bound by the subject of the matrix clause in the representation in (588a). We have not been able to construct such cases, but this may be for the independent reason mentioned above that the implicit PRO-subject of the infinitival adjunct clause containing the parasitic gap is usually controlled by the subject of the matrix clause: if this subject controls PRO and binds the parasitic gap, this results in a violation of binding condition B, because the parasitic gap would then also be bound within its local domain by the PRO-subject. A concrete example of this is given in (588b), with indices indicating coreferentiality.

However, the claim that parasitic gaps are subject to binding condition B also predicts that they behave like referential personal pronouns in that they should be able to be bound by a nominal argument in some higher clause, but this is at odds with the contrast found in (589), which shows that while the referential personal pronoun haarher can be bound by the subject of the matrix clause Els, the parasitic gap cannot; cf. Bennis (1986:55).

If we want to maintain that parasitic gaps are subject to binding condition B, we can only account for this contrast in a principled way by appealing to one of the other restrictions in (559). If we follow De Hoop & Kosmeijer (1995) and claim that the gap of the infinitival clause is still parasitic on some movement operation in the matrix clause, we can appeal to the anti-c-command restriction in (559c), which will be discussed in the next subsection. This would not be possible if we follow the hypothesis in Neeleman (1994a) that parasitic gaps can be licensed by noun phrases from their base position, as this would invalidate the overt movement and anti-c-command restrictions; since the categorial and island restrictions are both satisfied, this requires the introduction of some (as yet unknown) ad hoc stipulation.

This subsection has discussed a final problem for the landing-site restriction by showing that the subject of Dutch passives can function as the antecedent of a parasitic gap. We have shown that if the antecedent of parasitic gaps can indeed be located in an A-position, then the movement approach should be considered superior to a base-generation approach. However, we can also consider the possibility that Dutch parasitic gaps are not true parasitic gaps, as has been proposed for Dutch on other grounds in Huybregts & van Riemsdijk (1985) and for German in Kathol (2001), but this seems less attractive because Dutch parasitic gaps seem to be well-behaved with respect to the other restrictions in (559); cf. also Culicover (2001) and Chocano & Putnam (2013) for discussion. Another possibility is that there is simply something special about the infinitival clauses in the passive constructions in (583), given the surprising fact that Van Haaften’s (1991:108) comparable passive examples without a parasitic gap are all severely degraded regardless of the presence of a door-PP; this is illustrated in (590b).

If the PRO-subject of an adverbial zonder-clause must indeed be controlled by the subject of the matrix clause, the examples in (583) are surprising not only because they violate the landing-site restriction, but also because they exhibit exceptional control behavior. This should make us cautious about jumping to far-reaching conclusions on the basis of these examples alone. Since our discussion of parasitic gaps in passive constructions did not lead to a clear conclusion, we ended with a list of possible ways to approach such examples. Since we have no further insights to offer at the moment, we must leave this issue to future research.

All Dutch examples so far involve parasitic gaps embedded in an infinitival adjunct clause, and not without reason: as indicated in the accessibility hierarchy in (591), this is by far the easiest place to find parasitic gaps. Such infinitival clauses are usually introduced by a prepositional-like complementizer like zonder ‘without’, in plaats vaninstead, omin order to, and alvorensbefore; judgments about the examples in (592) vary not only from person to person, but also from case to case. The subject pronoun PRO of the infinitival clause is typically controlled by the subject of the matrix clause.

We have seen that the antecedent of a parasitic gap can be a scrambled or a wh-moved phrase; this is illustrated again in the examples in (593). Recall from Subsection A that parasitic gaps can alternate with overt pronouns in the case of scrambling, but that this is less common in wh-constructions, especially when the wh-phrase is a non-D-linked pronoun such as watwhat. Since the reader may also want information about the distribution of the pronominal counterparts of parasitic gaps, we will often prefer scrambling constructions for illustration in what follows.

If we follow the standard assumption that the object traces in (593) are embedded in the VP, while the adjunct clauses are external to the VP, the acceptability of the parasitic gap constructions is expected as far as the anti-c-command restriction is concerned. The examples in the literature are mostly cases where the adjunct clause precedes the clause-final verbs. This raises the question of what happens when such clauses are extraposed, i.e. when they follow the clause-final verbs. Although speakers disagree about the precise status of the parasitic gap constructions in (594), they generally agree that they are degraded compared to those in (593.); cf. while Kathol (2001) for similar judgments on German. However, Huybregts & van Riemsdijk (1985) gives a similar Dutch example as perfectly acceptable.

Assuming that the difference in word order corresponds to a difference in structure, it seems possible to account for the acceptability differences by appealing to the anti-c-command restriction. This may also explain why acceptability contrasts such as those in (593b) and (594b) have not been reported for English; the fact that parasitic gaps and referential pronouns are often assumed to alternate freely may be due to the fact that it is often not immediately obvious from the linear order of utterances in English whether or not extraposition has occurred. We leave the exploration of this analysis to future research; we will return to extraposition in Subsections 2 and 3.

The anti-c-command restriction also predicts that parasitic gaps in infinitival adjunct clauses cannot be licensed by the subject of the first higher matrix clause. Subsection A27 has already shown that it is very difficult to test this prediction because parasitic gaps are excluded in such constructions for independent reasons: the discussion of (588) has shown that PRO-subjects of infinitival adjunct clauses are normally controlled by the subject of the matrix clause and that parasitic gaps are consequently excluded because they would be locally bound by PRO. The anti-c-command restriction can, however, easily be demonstrated by examples in which the antecedent of the parasitic gap is a nominal argument in some higher clause; this was shown for a subject in (589) and illustrated again for an indirect object in (595).

Although parasitic gaps are virtually perfect in infinitival adjunct clauses (and even preferred to overt pronouns in wh-questions), they are less common in Dutch than in English. The reason for this is that Dutch differs from English in that it does not allow preposition stranding by extraction of a noun phrase or a pronoun, as shown in (596b); preposition stranding occurs only as a result of R-extraction from pronominalized PPs such as er/daar/waar/... + P P it/that/what, as illustrated in (596c). We refer the reader to Chapter P36 for a more detailed and careful discussion.

The ban on preposition stranding by extraction of noun phrases and pronouns severely limits the types of construction in which parasitic gaps can occur; cf. Bennis (1986). First, R-pronouns such as er/daar/waar/... cannot act as antecedents of parasitic gaps in nominal argument position, as shown in (597b).

Second, noun phrases and pronouns are not able to license parasitic gaps in PPs, as in example (598b). In short, a noun phrase or pronoun can only license parasitic gaps in a nominal argument position, as in all earlier examples.

This leaves us with the third case in which an R-pronoun licenses a parasitic gap in a PP; example (599b) shows that this leads to a perfectly acceptable result when the R-pronoun is part of a pronominal PP (here: daar ... uit).

Because the unacceptability of the parasitic gap constructions in (597b) and (598b) is not due to problems with the anti-c-command restriction, we may conclude from examples like (589) and (595) that the anti-c-command restriction applies to Dutch in full force, ... provided that it should be possible for a parasitic gap to have an antecedent external to its minimal finite argument clause; this is the topic of the next subsection.

This subsection discusses parasitic gap constructions in which the parasitic gap has an antecedent external to its own minimal finite argument clause. The examples in (589) and (595) have already shown that the anti-c-command restriction (binding condition C) does not allow the subject/object of a matrix clause to function as the antecedent of a parasitic gap within an infinitival adjunct clause embedded in a finite complement clause. The same is shown in (600) for the somewhat simpler abstract structures in which the parasitic gap functions as a nominal argument of the finite argument clause itself; the primed examples provide concrete instantiations of these structures. We do not give similar cases in which the parasitic gap is embedded in a PP, because the previous subsection has shown that noun phrases and pronouns cannot license such parasitic gaps.

In order to see whether an antecedent in a matrix clause can license a parasitic gap in a complement clause, we have to appeal to wh-moved complements of PPs (in order to avoid a violation of the anti-c-command restriction). Since nominal phrases cannot strand prepositions, we can confine our discussion to structures like (601a&b), in which some PP in the matrix clause has been split by R-extraction.

The parasitic gaps in the structures in (601) are embedded in a PP because the previous subsection has shown that R-pronouns cannot license parasitic gaps in nominal argument positions; it is therefore expected that example (602) is unacceptable with a parasitic gap as the object of the embedded clause.

This leaves us with the option that a wh-moved R-pronoun binds a parasitic gap embedded in a PP. Bennis (1986) claims that such cases are indeed grammatical. His example is given as (603a); although we are not aware of any objections to his judgment in the literature, we assigned a question mark to this example because we tend to think that the parasitic gap construction is marked in comparison to example (603b) with the overt R-pronoun er. Note that the structure assigned to the parasitic gap construction in (603a) is quite different from what Bennis suggests; we return to the reason for this presently.

Another example with the abstract structure in (601b), taken from Huybregts & Van Riemsdijk (1985), is given as (604a). Although this example is given as fully grammatical, we have again assigned a question mark to it in order to express that it is marked in comparison to the construction in (604b) with the overt R-pronoun er.

The contrasts we find between the (a) and (b)-continuations in (603) and (604) are not surprising in light of Engdahl’s accessibility hierarchy in (591), according to which finite argument clauses are less amenable to parasitic gaps than are infinitival adjunct clauses. It is important to note that we cannot account for these contrasts by appealing to the fact that finite argument clauses are generally extraposed; the fact that extraposition of infinitival adjunct clauses with a parasitic gap in (594) has a degrading effect was attributed to the anti-c-command restriction, but this restriction is satisfied in the parasitic gap constructions in (603a) and (604a).

If we follow Bennis (1986) and Huybregts & van Riemsdijk (1985) and assume that the structures marked with a percentage sign are grammatical, we still have to deal with the question of what the exact structure of these examples is. We have already indicated that the structures that we have assigned to the parasitic gap constructions in (603a) and (604a) differ from those assumed in Bennis (1986). For theory-internal reasons, Bennis claims that the trace is embedded in the finite clause, while the parasitic gap is embedded in the adjunct PP in the matrix clause, so that they should be swapped in the examples (601) to (604) above. Although this claim would be fully consistent with the anti-c-command restriction, we will show in the remainder of this subsection that there are compelling reasons to reject it.

First, it should be noted that wh-extraction of a relative pronoun from an embedded clause is possible but not much favored by many speakers; the percentage sign in (605a) is used to indicate that many speakers prefer to use the resumptive prolepsis construction in (605b), which was discussed in Sections 11.3.1.3, sub VII, and 11.3.2, sub III. The crucial observation is that (605c) is completely unacceptable, which shows that wh-extraction is excluded in the resumptive prolepsis construction. The judgments on (605b&c) therefore suggest that the corresponding parasitic gap construction should be analyzed as in (605d); cf. Huybregts & Van Riemsdijk (1985). Note that the percentage signs in (605d) is again used to indicate that many speakers prefer to use the resumptive prolepsis construction in (605b); this may also account for the fact that Kathol (2001) give German cases similar to (605d) as ungrammatical.

For completeness’ sake, it should be noted that this argument is not accepted in Bennis (1986), because the ungrammaticality of (605c) can plausibly be attributed to an intervention effect, according to which an R-pronoun (here: waar) cannot be moved across another c-commanding R-pronoun (here: er); cf. Section P36.5 for a detailed discussion of this restriction. This intervention effect cannot occur in the parasitic gap construction in (605d) because there is only one R-pronoun, the second interpretive gap being the parasitic one; it is therefore concluded that the ungrammaticality of (605c) should be set aside as irrelevant.

A second and possibly better argument for claiming that the interpretive gap in the embedded clause is the parasitic one (i.e. not the trace) is that the parasitic gap construction is possible when the finite complement clause constitutes an island for wh-movement. Example (606a) first shows that wh-extraction from an interrogative clause is impossible. Example (606b), on the other hand, is perfectly acceptable. The two competing analyses thus make very different predictions: the claim that the trace is in the embedded clause incorrectly predicts that (606c) has the same status as (606a), while the alternative analysis correctly predicts that it is as good or as bad as example (605d) above. To our ear, the latter prediction is the correct one, and we therefore conclude that the structure is as indicated in (606c); of course, this analysis is also favored because parasitic gaps are typically found in islands for wh-movement.

The final argument for the claim that the interpretive gap in the embedded clause is the parasitic one is that it does not seem possible to construct acceptable parasitic gap examples if the PP in the matrix clause cannot be pronominalized. First, note that example (607a) is less preferred than the resumptive prolepsis construction (i.e. de mani [[van wiei]j ik dacht tj [dat zij eri op zou wachten]]), but it is possible, which is why we have given it a percentage sign. Example (607b) is unacceptable because the volgens-PP does not allow R-pronominalization and R-extraction: the only option is pied piping, as in de mani [[volgens wiei]j ik tj dacht [dat zij op hemi zou wachten]] the man according to whom I believed that she would wait for him. The fact that the heads of such adjunct PPs cannot be stranded should not affect the acceptability of the parasitic gap construction if the parasitic gap were located in the PP. Thus, the crucial example is the parasitic gap construction in (607c). Since the claim that the trace is in the embedded clause incorrectly predicts that it has more or less the same status as (607a), we conclude that the structure given in (607c) is the correct one.

The relative acceptability of (606c) and the unacceptability of (607c) show that the proposal according to which the trace is in the embedded clause and the parasitic gap is in the adjunct PP in the matrix clause is problematic because it incorrectly predicts that these examples should have a status similar to that of the corresponding (a)-examples. We therefore adopt the alternative proposal that the PP in the matrix clause hosts the trace of the moved constituent, while the parasitic gap is in the finite argument clause; this correctly predicts that the (c)-examples should have a similar status to the corresponding (b)-examples. This proposal is also preferable because it is consistent with the generalization that parasitic gaps are typically embedded in islands for wh-movement.

Bennis (1986) claims that parasitic gaps cannot be embedded in finite adjunct clauses; two of his examples are given in a slightly adapted form in (608). This claim would be consistent with Engdahl’s accessibility hierarchy in (591) according to which parasitic gaps are less common in finite adjunct clauses than in finite argument clauses: because the previous subsection has shown that parasitic gaps are marked in Dutch infinitival adjunct clauses and finite argument clauses, it is to be expected that they give rise to even more degraded results in finite clauses.

Note, however, that examples such as (608) are not suitable for showing that parasitic gaps cannot occur in finite adjunct clauses. The reason is that they are in extraposed position and we have seen that this also has a degrading effect on the acceptability of infinitival adjunct clauses: as we attributed this to the anti-c-command restriction, the unacceptability of the parasitic gap constructions in (608) may simply provide additional support for this restriction. In order to conclusively show that parasitic gaps cannot occur in finite adjunct clauses, the adjunct clause must be in the middle field of the clause, as in (609). As this does not seem to improve the parasitic gap constructions, we may indeed safely conclude that parasitic gaps cannot be embedded in finite adjunct clauses.

Because the parasitic gap constructions in (609) do satisfy the anti-c-command restriction, their unacceptability must be due to some other restriction. Since the other restrictions in (559) are also satisfied, some additional constraint is needed; cf. Bennis (1986:48ff) for a proposal embedded in terms of Kayne’s (1984) path theory.

This subsection discusses cases where a parasitic gap is embedded in a postnominal PP. A prototypical English example is given in (610a). Its contrast in acceptability with example (610b) again illustrates the effect of the anti-c-command restriction: since the object trace in (610a) does not c-command the subject position, the parasitic gap embedded in the subject can be licensed by wh-movement of the object; since the subject c-commands the object position, the parasitic gap embedded in the object cannot be licensed by wh-movement of the subject. Assuming that direct objects function as external arguments of complementives, the anti-c-command restriction also correctly predicts that a parasitic gap embedded in an object can be licensed by wh-movement of the nominal complement of a complementive PP; this is illustrated in (610c), where the label SC stands for small clause, the phrase containing both the complementive and its logical subject. An interesting feature of the acceptable parasitic gap constructions in (610a&c) is that they do not allow a bound pronoun in the position of the parasitic gap; this is illustrated in the primed examples. For a more detailed discussion of the English data, see Engdahl (1983: §5).

Constructions like those in (610) are largely ignored in the literature on Dutch. Parasitic gap constructions of the form in (610a) are expected not to occur in Dutch because, as discussed in Subsection 1 above, noun phrases cannot bind parasitic gaps embedded in PPs; this correctly predicts that example (611a) is excluded. However, parasitic gaps are expected to be possible when the wh-moved phrase is an R-pronoun, as in example (611b); the result is clearly not perfect, but this example seems significantly better than (611a). The two examples in (611a&b) are also noteworthy because they differ in whether a bound pronoun can be used in the position of the parasitic gap while maintaining the same meaning: this is possible only if the parasitic gap construction is completely unacceptable. The linear string in (611b') is of course acceptable as such, but crucially not with the wh-moved R-pronoun waar construed as the antecedent of the pronoun haarher. As usual, coreferentiality is indicated in italics.

The anti-c-command restriction cannot be demonstrated by a Dutch version of (610b), because parasitic gaps embedded in a PP cannot be bound by a nominal argument in general. What we can show, however, is that parasitic gaps can at least marginally be embedded in a direct object when an R-pronoun is extracted from a complementive PP; cf. (612a). Example (612b) again shows that the parasitic gap cannot be replaced by a bound pronoun while maintaining the same meaning: this example is only acceptable if the pronoun haarher refers to some other person in the domain of discourse.

This subsection has shown that Dutch at least marginally allows parasitic gaps in postnominal PPs, provided that their antecedent is an R-pronoun. The restriction mentioned in the previous sentence makes it impossible to determine whether the anti-c-command restriction is applicable; we only have the weaker evidence that the marginally acceptable cases do not violate this restriction. It should also be noted that the marginally acceptable Dutch parasitic gap constructions are similar to their English counterparts in that the parasitic gaps cannot be replaced by bound pronouns.

Constructions in which a parasitic gap is embedded in a relative clause differ from those in which a parasitic gap is embedded in a postnominal PP in that they always lead to an unacceptable result. This is illustrated in example (613), taken from Bennis (1986); the reader should ignore the pseudo-intransitive reading of lezento read. The fact that these examples are not acceptable is consistent with Engdahl’s accessibility hierarchy in (591); cf. Bennis (1986) for an account of these examples, again in terms of Kayne’s (1984) path theory.

Of course, there is no prohibition on parasitic gaps within relative clauses per se, since they do occur when their antecedent is also part of the relative clause, as in (614), where the antecedent is a scrambled object. This shows that while relative clause cannot embed parasitic gaps, they can embed full parasitic gap constructions.

As expected, parasitic gaps can also be licensed when the antecedent is the wh-moved relative pronoun, as in (615a); the percentage sign is used because Den Dikken (2005:506) notes that speakers vary considerably in their judgments on this example. The main point, however, is that all speakers find it much better than examples in which the relative pronoun is extracted from an adjunct clause, i.e. when the trace ti is replaced by the pronoun ’mhim, as in (615b). Note that the unacceptability of (615b) is not the result of binding condition B, as can be seen from the fact that the pronoun can be coreferential with Jan in [Zodra je Jani ziet] ga je hemi meteen haten As soon as you see Jan, you come to hate him right away.

Den Dikken (2005) also gives the examples in (616) to show that the two clauses in comparative correlative constructions such as hoe vaker je Jan ziet, hoe meer je hem gaat hatenthe more you see Jan, the more you get to hate him are not paratactically related, but that the first clause is an adjunct, just like the adverbial clause in (615). We agree with the judgments on these parasitic gap constructions, but we give them here without further comment, because there are a number of properties of (616a) that we do not fully understand; this holds in particular for the fact that the correlative marker des te differs from its more common counterpart how in that it can remain in situ (as can be seen by the fact that it appears to the right of the subject je instead of to the left), which of course is necessary to allow the relative pronoun die to undergo wh-extraction from the second clause.

This subsection has shown that parasitic gaps embedded in a relative clause cannot be licensed by a c-commanding wh-moved antecedent external to the noun phrase containing the relative clause.