LD-constructions are characterized by the fact that left-dislocated phrases are associated with a resumptive element. If we restrict ourselves for the moment to cases such as (33) with a left-dislocated noun phrase, we observe that the resumptive element preferably takes the form of a referential personal pronoun such as hemhim when it is in clause-internal position, but takes the form of a distal demonstrative personal pronoun such as diethat when it is in clause-initial position. The main verbs in these constructions cannot semantically license both the left-dislocated and the resumptive element by assigning them a thematic role. Since the resumptive pronoun is clearly the recipient of the available thematic role, it is traditionally assumed that the left-dislocated constituent does not occupy a clause-internal position, but is instead base-generated in clause-external position, as indicated by the structures in (33); the left-dislocated constituent should then be semantically licensed by functioning as an antecedent of the resumptive element (indicated by co-indexing).

That the left-dislocated element must be licensed by functioning as the antecedent of a resumptive element can be demonstrated by the unacceptability of examples such as (34), in which no suitable resumptive pronoun is available. We refer the reader to Van Riemsdijk & Zwarts (1997:26) for discussion; note that there are a number of errors in the published version of that paper (such as missing asterisks), which we have tacitly corrected in the discussion below.

There are several empirical arguments in favor of the hypothesis that left-dislocated constituents are clause-external. First and foremost, it explains why the two types of LD-constructions in (32) are special in that they allow the finite verb to be preceded by two constituents: since left-dislocated elements are clause-external, they do not count for the verb-second restriction, and the representations in (33) are therefore perfectly consistent with this restriction. Second, the hypothesis is supported by the fact that the polar elements jayes and neeno can follow the left-dislocated constituent; under the standard assumption discussed in Section 37.1, sub III, that ja and nee do not occur clause-internally, the left-dislocated phrases in (35) must also be clause-external. We will return to such cases in Subsection VII.

Third, the hypothesis that left-dislocated constituents are clause-external provides a simple explanation for the fact that LD is a typical root phenomenon, i.e. cannot apply in embedded contexts: complement clauses cannot be preceded by a left-dislocated constituent. For completeness’ sake, note that the (b)-examples are unacceptable both with and without the complementizer datthat.

Salverda (2000:102) claims that embedded contrastive LD is acceptable in spoken Dutch when the left-dislocated element and the resumptive pronoun are placed after the complementizer datthat, as in (37b), but we agree with Van Riemsdijk & Zwarts (1997:13) that this pattern is not acceptable in the standard language; the same is true for the corresponding hanging-topic construction in (37a). The use of asterisks in (37) is not intended to express that the patterns in (37) cannot be found in other varieties of spoken Dutch. In fact, we expect them to occur in the regional variety of Dutch spoken in Friesland, since Frisian does allow (a kind of) embedded contrastive LD; cf. De Haan (2010: §5.3) for examples and discussion.

Example (36a) can be the input for LD if the left-dislocated element is to the left of the complete main clause, as shown by the examples in (38). That Jan can be construed as the object of the embedded clause in (38a) is not surprising, since it is normal for the (resumptive) referential pronoun hem to take a non-local antecedent, i.e. an antecedent that is not part of its own clause. That it can also be construed as the object of the embedded clause in (38b) can be accounted for by assuming that the resumptive demonstrative pronoun is extracted from the embedded clause by wh-movement, which we have indicated by the trace ti. Evidence that wh-movement is involved in contrastive (but not hanging topic) LD will be presented in Subsection IV.

Hanging-topic and contrastive LD do not allow stacking in Dutch; note that changing the order of the left-dislocated phrases in (39) does not affect the acceptability judgments. It is not clear how to account for this fact, given that some of the types of LD found in other languages do allow stacking; cf. Alexiadou (2006/2017) for discussion.

Combining hanging-topic LD and contrastive LD, on the other hand, is possible; cf. Zaenen (1997; §3). Note that in the examples in (40) the hanging topic must precede the contrastively left-dislocated phrase. This means that hanging topics can never separate a contrastively dislocated phrase from its wh-moved demonstrative correlate in main-clause initial position, despite the fact that other clause-external material may intervene between these elements; cf. example (35b).

Finally, observe that examples such as (41) are acceptable. Given the generalization that hanging topics precede contrastively left-dislocated phrases, example (41a) could perhaps be analyzed in the same way as (40a), with two independently left-dislocated phrases, a hanging topic followed by a contrastively left-dislocated phrase. However, a similar analysis is less likely for example (41b) because (39b) has shown that stacking of contrastively dislocated phrases is excluded.

This seems to support the analysis of (41b) in Haeseryn et al. (1997:1390), according to which Jan is left-dislocated to the object clause, as in the structure indicated in (42b). If this is correct, it is natural to assume a similar analysis for (41a), i.e. with the hanging-topic left-dislocated to the object clause, as indicated in (42a). However, this would be rather surprising given our earlier conclusion, based on the (b)-examples in (36), that complement clauses cannot be preceded by a left-dislocated constituent: the analysis implies that this is possible after all, but only if they themselves are left-dislocated.

A similar analysis seems plausible for the examples in (43), with a left-dislocated conditional clause; cf. Paardekooper (1986:417) for more examples of this kind.

Note that the resumptive demonstrative die is not in the initial position of the object clause in (42b)/(43b); this is not unexpected, since Subsection IV will show that the demonstrative can remain in situ when topicalization is impossible for independent reasons.

Semantically, the two types of LD-constructions can be characterized by saying that the sentence is “about” the left-dislocated complement, but they differ in that hanging-topic constructions are usually not contrastive. This difference can be seen in the coordinate structures with maarbut in (44), where the LD-construction in the first conjunct is contrasted with a fragment (i.e. partially elided) clause in the second conjunct: example (44a) is only acceptable if the resumptive object pronoun hemhim is given a contrastive accent, while the resumptive demonstrative die in (44b) does not need any special marking because, being topicalized, it is already accented.

The previous subsection has shown that noun phrases can occur in both hanging-topic and contrastive LD-constructions. The examples in (45) show that such left-dislocated nominal phrases can be associated with a resumptive pronoun with the function of subject, (in)direct object and the nominal part of a prepositional object. It may be that some speakers prefer the contrastive LD-construction in the case of a subject, but both constructions seem perfectly acceptable.

Left-dislocated nominal phrases can also be associated with resumptive pronouns originating in a more deeply embedded position. This is illustrated in (46) for a nominal complement and the nominal part of the PP-complement of a complementive AP, respectively; cf. Subsection V for a discussion of cases in which the resumptive pronouns originate in an embedded clause.

The primed examples in (47) show that, at least in contrastive LD-constructions, the left-dislocated element can also be an AP, a PP or a verbal projection. Zaenen (1997) claims that hanging-topic LD-constructions yield degraded results in these cases, but there seems to be speaker variation in this respect, which we have indicated by the percentage sign.

The examples in (45) and (47) also show that left-dislocated phrases can be antecedents of resumptive elements with different syntactic functions in the case of contrastive left-dislocation: the examples in the previous subsection have shown that they can be antecedents of resumptive demonstratives functioning as arguments, and the examples above show that the resumptive can also be a complementive (47a), an adverbial phrase (47b), and even replace part of the lexical projection of the main verb (47c). The markedness of the primeless examples in (47) suggests that the left-dislocated phrases in hanging-topic LD-constructions are usually antecedents of pronominal arguments. We will not discuss the examples (47) further here, but we will return to left-dislocated VPs such as in the (c)-examples in subsection VIII.

To say that left-dislocated phrases can be nominal is not very precise, since there are various additional restrictions on LD of noun phrases. Furthermore, hanging-topic and contrastive constructions seem to differ in that the left-dislocated element must be definite in the former, but not in the latter.

The use of the percentage sign in (48b) is motivated by the fact that Zaenen (1997) gives a similar example as marked. It seems to us that judgments may vary from case to case, perhaps depending on whether or not the indefinite noun phrase allows for a specific interpretation, i.e. whether the speaker is able to identify the referent of the noun phrase. Indeed, this would be consistent with Zaenen’s (1997:142) specificity requirement, according to which contrastive LD “can only be used felicitously if the speaker has a “recoverable” referent in mind for the initial constituent”. Accordingly, left-dislocated noun phrases are often introduced by a D-linked demonstrative like dit/datthis/that or a referential possessive noun phrase. As expected, there seems to be no contrast between hanging-topic and contrastive LD in such cases.

The acceptability of left-dislocated weak quantified noun phrases depends on the quantifier; Zaenen (1997:141) shows that negative articles such as geenno also block contrastive LD, while determiners like velemany and enkelesome at least marginally allow contrastive (but not hanging topic) LD.

Zaenen also claims that contrastive LD of strong quantified noun phrases introduced by alleall, elk(e)each and de meestemost is possible, but to our ear such cases seem somewhat marked, which we have expressed in (51) by a single question mark; cf. also Van Riemsdijk & Zwarts (1997:fn.5), in which examples such as (51a') are rejected. The hanging-topic LD-constructions all seem to be more degraded than the corresponding contrastive LD-constructions; cf. Vat (1997). Note that the judgments given here differ somewhat from those in Vat, which may be related to the fact that Vat idealizes the data somewhat for simplicity (cf. Vat’s remark on p.71).

Note that making judgments is tricky because the (a) and (b)-examples in (51) compete with the even more natural examples in (52). In the (a)-examples quantification is expressed by the floating quantifier allemaalall and in the (b)-examples by the definite article of the a nominal phrase de meeste pied-piped by the resumptive element er. So we want to emphasize that (as always) the diacritics on the examples in (48) to (52) express relative, not absolute, judgments.

The above discussion has shown that LD of noun phrases yields the best result when the left-dislocated noun phrase is referential: quantified noun phrases are always marked compared to definite noun phrases introduced by a definite article or a definite demonstrative/possessive pronoun. This conclusion is further supported by the fact that LD of non-referential noun phrases is not possible; cf. Van Riemsdijk & Zwarts (1997) and Vat (1997), although the latter provides a number of cases in note 5 that are said to allow contrastive LD. We illustrate this in (53) for the idiomatic expression ergens de ballen van geloven with the non-referential nominal phrase de ballen; note that we have added the asterisk that was inadvertently omitted in the published version.

In other cases, LD results in the loss of the idiomatic interpretation: the number signs in the (b)-examples in (54) indicate that only the literal transmission reading survives in LD-constructions.

The claim that left-dislocated phrases must be referential also accounts for the fact noted in Zaenen (1997) that wh-phrases cannot be left-dislocated. This is illustrated by the examples in (55), which show that interrogative phrases are very different from demonstrative phrases in this respect.

Another case that might show the same is LD of reflexive and reciprocal personal pronouns, since such pronouns are not inherently referential, but depend on an antecedent for their reference. However, Van Riemsdijk & Zwarts (1997) and Vat (1997) suggest that the unacceptability of the examples in (56) is due to problems with binding. An updated version of this proposal would attribute it to the fact that the resumptive pronouns henthem and die must take the subject zijthey as antecedent in order to satisfy the binding conditions on elkaareach other, which results in a violation of the binding conditions that they themselves must satisfy (i.e. that they are free in their minimal clause); cf. Chapter N22 for a detailed discussion of the binding conditions. Note that for the resumptive pronoun die in (56b), binding should be computed from its original object position indicated by its trace.

This binding approach to the unacceptability of the examples in (56) can be supported for the contrastive LD-construction in (56b) by the fact that example (57b) is perfectly acceptable, which can be made to follow from the fact that the resumptive pronoun die is not bound by the subject zij in this case. This leaves us with the question of why (57a) is still unacceptable. This may be explained by the connectivity hypothesis proposed in Vat (1997), according to which hanging-topic and contrastive LD-constructions differ in that only the latter allow left-dislocated phrases to be interpreted in the (trace) position of the resumptive pronoun; this means that the reciprocal elkaareach other is correctly bound in its local domain in (57b) (cf. Elkaarsi jassen dragen ziji niet graag), but not in (57a).

Further evidence for the connectivity hypothesis is provided by examples such as (58), where the indices indicate that pronouns embedded in left-dislocated phrases only allow a bound variable reading in contrastive LD-constructions: while the hanging topic in (58a) must refer to the mother of a certain person who is liked by everyone, the contrastive LD-construction in (58b) additionally allows an interpretation according to which everyone likes his own mother; cf. also Zaenen (1997).

See Vat (1997), Ott (2014/2015) and Dikken & Surányi (2017) for more discussion of the connectivity hypothesis, including its relevance for case agreement between the left-dislocated phrase and the resumptive pronoun in German, which is not visible in Dutch.

We conclude this subsection with a discussion of dislocated referential personal pronouns. Zwart (1997:249) has claimed that referential pronouns such as hemhim cannot be left-dislocated. If true, this would be surprising, since there seems to be no obvious reason why this should be the case. It seems to us, however, that with sufficient context (as in (59a)) such cases are quite acceptable, but we will leave it to future research to determine more precisely the acceptability status of such examples.

It has also been claimed more specifically that first-person and second-person pronouns cannot be left-dislocated; cf. De Wit (1997). Although such pronouns do indeed seem to yield a marked result when used as hanging topics, we think that they can be used comfortably in contrastive LD-constructions, which is why we have marked the primed examples in (60) with a percentage sign; we leave it to future research to determine more precisely their acceptability status.

Note that all examples (59) and (60) become completely impossible with a left-dislocated weak pronoun, which is of course not surprising for contrastive LD, but it also severely degrades hanging-topic LD cases (regardless of the form of the resumptive pronoun): cf. *ʼM, ik heb hem/ʼm niet gezien and *Je, Peter wil jou/je niet meer helpen. On the other hand, hanging-topic LD of pronouns is greatly improved when the left-dislocated phrase is a coordinate structure, as in (61); the presumed markedness of the primeless examples in (60) may therefore be related to the repetition of identical forms.

Contrastive LD also seems to improve when a focus marker like zelfseven or ookalso associated with the resumptive pronoun is added; this can be seen in (62).

The discussion of the examples in (56) to (62) shows that although LD of pronominal elements may lead to degraded results, this need not be due to a syntactic restriction on LD: the unacceptability of the examples in (56), for example, may be due to the independently motivated binding conditions, and the judgments on the examples in (59) and (60) are not clear enough to draw firm conclusions, and are also influenced by various (at least partly) non-syntactic factors that have not yet been sufficiently investigated. Finally, the examples in (62) show that LD of pronominal elements is sometimes perfectly acceptable.

The resumptive element in hanging-topic constructions is relatively straightforward: because the construction yields a perfectly acceptable result only if the left-dislocated element is nominal, it is typically a referential personal pronoun; examples can be found e.g. in the primeless examples in (45). The previous subsections have assumed, without discussion, that resumptive elements in contrastive LD-constructions with a nominal left-dislocated phrase are D-linked distal demonstrative pronouns (i.e. diethat/thoseand datthat). A good reason for adopting such an assumption would be that the resumptive pronouns behave like demonstratives in that they agree in gender and number with their antecedent, as shown in (63); cf. Section N18.2.3.1.1, sub I, for a discussion of the paradigm of demonstrative pronouns.

However, the agreement pattern in (63) would also allow for an analysis of the resumptive pronouns as relative pronouns (cf. Section N18.2.1.4) and, indeed, it has been suggested that this might be a reasonable line of inquiry; cf. Van Riemsdijk (1997: §4). Given that relative pronouns are obligatorily moved into the clause-initial position, one might expect that resumptive pronouns are also obligatorily moved into the clause-initial position in contrastive LD-constructions. Subsection IV will show, however, that this expectation is not borne out. Another problem for the analysis is that the resumptive pronouns can be replaced by epithets like die idiootthat idiot or dat wijfthat bitch in (64a&b) (cf. Van Riemsdijk & Zwarts 1997, Zaenen 1997, Ott 2014 and Dikken & Surányi 2017); it seems impossible to analyze such contrastive LD-constructions as relative constructions. Another problematic case, adapted from De Vries (2009), is (64c), where the left-dislocated coordinate structure is resumed by the encompassing phrase al die dingenall those things; again, analyzing such contrastive LD-constructions as relative constructions seems impossible.

A relative clause-like analysis is also unlikely in view of the contrastive LD-constructions in (65), which are repeated in a slightly different form from Subsection II. Although adjectival complementives and verbal projections can be used as antecedents of (non-restrictive) relative clauses, the relative pronouns would appear as wat, not as dat: cf. Jan is erg slim, wat/*dat ik niet benJan is very smart, which I am not and Jan heeft de boodschappen gedaan, wat/*dat ik vergeten wasJan has done the shopping, which I had forgotten. Moreover, the relative pronoun taking a phrase denoting a place as its antecedent is waar, not daar; cf. De stad waar/*daar hij woont is Amsterdamthe city where he lives is Amsterdam. The resumptive elements in (65) thus simply do not have the form expected of relative proforms; cf. Section N16.3.2.2 for a discussion of the form of relative proforms in Dutch.

The same holds for contrastive LD-constructions with left-dislocated temporal adverbial phrases: while the resumptive element toenthen in (66a) can indeed occasionally be found in relative clauses (cf. Section N16.3.2.2, sub IVC), this does not hold for the form danthen in (66a'). Locational proforms such as daarthere in example (66b) cannot be used in relative clauses either, and the same goes for the pronominal PP daarmee and the adverbial proform zo in (66c&d).

The discussion above suggests that it is safe to conclude that the resumptive elements in contrastive LD-constructions are distal demonstrative (and not relative) proforms; cf. Haeseryn et al. (1997:1389) for the same conclusion. That the resumptive demonstrative is distal may account for the fact, illustrated in (67), that left-dislocated time adverbials can refer to time spans either preceding or following speech time, but not including speech time; there is no way to express the intended meaning in (67c) by using a contrastive LD-construction (this case was brought to my attention by Liliane Haegeman p.c.).

Subsection III has argued that it is reasonable to assume that the verb assigns its thematic role to the resumptive pronoun and, consequently, that the left-dislocated element cannot be licensed within the clause. This implies that a left-dislocated noun phrase must be base-generated external to the clause and receive a semantic interpretation by acting as the antecedent of the resumptive pronoun. However, the examples in (68) and (69) show that the two types of resumptive pronouns seem to exhibit different behavior with respect to wh-movement. Although examples such as (68b) are generally not discussed in the literature, our informants indicate that wh-movement of the referential personal pronoun produces a highly marked result.

The reception of examples such as (69a) is mixed: Van Riemsdijk (1997) states without reservation that demonstratives are possible in hanging-topic LD-constructions, Vat (1997:70) claims that they are grammatical, although their acceptability depends on factors that are not well understood, and Odijk (1998:204). According to our own intuitions, examples such as (69a) are acceptable, but usually less preferred than examples such as (68a); cf. Den Dikken & Surányi (2017:544) for the same judgment.

Whatever the status of (69a), it is clear that we cannot categorically say that resumptive demonstratives are excluded in the middle field of the clause, since they can remain in situ in constructions that do not allow wh-movement, like the yes/no question in (70a) and the imperative construction in (70b); cf. Paardekooper (1986:39), Van Riemsdijk & Zwarts (1997:21), Zaenen (1997), Odijk (1998:205) and Ott (2014:274). The same applies to constructions in which the main-clause initial position is occupied by another element, such as wiewho in the embedded wh-question in (70a'); cf. Paardekooper (1986:39) and Van Riemsdijk (1997:4).

Note in passing that the distal demonstrative pronouns die and dat also occur in topic-shift constructions, where they usually undergo wh-movement as well; cf. Section N18.2.3.2, sub IIA. This suggests that there may be some relation between contrastive LD and topic-shift constructions. This idea is further supported by the fact that both seem to introduce a (new) topic in the domain of discourse. We leave this as a suggestion for future research.

Subsection IV has shown that resumptive personal pronouns prefer to remain in situ when they are referential, whereas resumptive demonstratives prefer to undergo wh-movement when the clause-initial position is available as a landing site. Although this formulation deliberately leaves some room for variation, it seems beyond doubt that the two LD-constructions differ with respect to wh-movement; wh-movement is not involved in the hanging-topic construction, while it is involved in the contrastive construction. This is also reflected in the island sensitivity of the two constructions. The examples in (38), repeated here as (71), show that the hang-topic and contrastive LD-constructions are equally acceptable when the (trace of the) resumptive pronoun is contained in a clausal complement of a bridge verb.

However, the acceptability of two constructions differs sharply when the embedded clause is an island for wh-movement. The examples in (72) first illustrate this for interrogative complement clauses (i.e. wh-islands): while the hanging-topic LD-constructions in (72a) is perfectly acceptable, the contrastive LD-constructions in (72b) is severely degraded.

The examples in (73) illustrate the same contrast for an adverbial clause (i.e. an adjunct island).

The examples in (74) show similar cases involving the coordinate structure constraint.

That contrastive LD is island-sensitive follows immediately from the fact that the resumptive demonstrative is subject to wh-movement. That hanging-topic LD is not island-sensitive follows from the fact that referential personal pronouns can generally take any (non-local) discourse-prominent element as their antecedent. For completeness, note that replacing the referential pronoun hem in the (a)-examples of (71) to (74) is not a preferred option, but at least marginally possible.

The fact that wh-movement of resumptive distal demonstratives is near-obligatory when the main-clause initial position is available, combined with the fact that this movement is island-sensitive, raises the expectation that contrastive LD can trigger pied piping, and this is indeed what we find. In (75a), LD requires movement of only the possessive demonstrative diens, but since a prenominal possessor cannot be extracted from its noun phrase, the entire noun phrase diens vader is moved into the clause-initial position. In (75b) we see a case where the demonstrative diens pied-pipes a PP.

Pied piping of PPs in contrastive LD-constructions is less common than in wh-questions, however, due to the fact that [-human] demonstrative pronouns are almost invariably subject to R-pronominalization, while [+human] interrogative pronouns cannot undergo R-pronominalization.

Since interrogative pronouns such as wie cannot be extracted from PP-complements, while the prepositional part of pronominal PP-complements is often stranded, we find the following contrast between the wh-questions in the (a)-examples in (77) and the contrastive LD-constructions in the (b)-examples.

There is a long tradition of analyzing topicalization as a form of contrastive LD followed by deletion of the wh-moved resumptive demonstrative; cf. Koster (1978), Weerman (1989), Kosmeijer (1993), and Zwart (1997:248). Others have argued against such a deletion analysis of topicalization: for instance, the examples in (78) show that certain topicalization constructions have no LD-counterpart, as illustrated for a topicalized modal adverb in the (a)-examples, taken from Odijk (1995:9), and for non-referential noun phrases in the (b)-examples, taken from Van Riemsdijk & Zwarts (1997) with the addition of an asterisk which was inadvertently omitted in print.

The same can be illustrated by quantified noun phrases, which were shown in Subsection II to produce degraded results in contrastive LD-constructions. We illustrate this in (79) by comparing examples with and without the demonstrative, which should be read with and without an intonation break, respectively. In example (79a), the noun phrase should be interpreted as specific because topicalization is like contrastive LD in that it does not easily affect non-specific indefinites.

The examples in (80) show the same for reflexive and reciprocal pronouns. While contrastive LD of the elements leads to degraded results, topicalization is perfectly acceptable; cf. Van Riemsdijk & Zwarts (1997).

Another problematic case for the deletion analysis of topicalization can be seen in (81). The (b)-examples show that while the contrastive LD-construction is perfectly acceptable, the corresponding topicalization construction is rejected by many speakers; cf. Bennis & Hoekstra (1983:8). The percentage sign again indicates that such topicalization constructions can be found in informal speech (Klooster 1989/2001), but the acceptability contrast is still unexpected if topicalization is derived by the omission of the demonstrative element (here the pronominal part of the discontinuous PP daar .. op); cf. the introduction to Chapter P36 for a more detailed discussion.

Note in passing that topicalization examples such as (81b') should not be confused with perfectly acceptable instances such as (82a), where the noun phrase bananen is followed by an intonation break. Barbiers (2007:101) derives such examples as indicated in (82b), i.e. by combining contrastive LD and topic drop, which was discussed in Section V11.2.2. The mere fact that the acceptability contrast exists between (81b') and (82a) can be seen as evidence against the deletion analysis of topicalization, although it will also be clear that the existence of the construction in (82a) has considerably obscured the discussion surrounding this hypothesis.

Odijk (1995:9) suggests that the unacceptability of example (83b') is also problematic for the deletion analysis of topicalization, because it is expected that this example can be derived from (83a') by deletion of the pronominal part in clause-initial position. Odijk’s original argument is actually more complicated, as he claims that (83b) should be unacceptable because it involves deletion of the preposition op in violation of the recoverability condition on deletion; cf. Chomsky & Lasnik (1977).

Section V11.3.3, sub VI, has shown that we also find acceptability contrasts between constructions with contrastively left-dislocated and topicalized VPs; in order to avoid unnecessary redundancy we simply repeat the table with the results of the discussion given there, and refer the reader to this section for detailed discussion; we return to left-dislocated and topicalized VPs in Subsection VIII.

		topicalization	left dislocation
finite clauses	argument (SU, DO) argument (PP)	✓ *	✓ ✓
	adverbial adverbial (conditional)	✓ ✓	* ✓
infinitival clause	om + te-infinitive	*	✓
	te-infinitives	?	✓
	bare infinitives	✓	✓
VP-topicalization	past/passive participle	✓	✓
	te-infinitival	?	?
	bare infinitival (full) bare infinitival (remnant)	✓ ✓	✓ ?

A final empirical argument against the deletion analysis of topicalization can be based on examples like (84) and (85) below. Example (84a) first shows that left-dislocated elements can easily be followed by clause-external pragmatic markers like vocatives, the polar element ja/neeyes/no or an interjection such as tjawell; cf. Haeseryn et al. (1997:1388). The acceptability of (84b) is less informative, since it can be analyzed either as a topicalization construction or as a contrastive LD construction plus topic drop, because the intonation break here is introduced by the intervening element nee/tja and thus cannot be used to distinguish between the construction types; cf. the discussion of (81b') and (82a). Since we have no other reliable tests at our disposal, the best we can do at this point is to appeal to our intuitions about this case, which point in the direction of a contrastive LD-construction with topic drop. Examples like (84b'), on the other hand, do not allow for a topic-drop interpretation without an intonation break after the noun phrase dat boek.

Topicalization constructions such as (78), which have no contrastive LD-counterpart, perhaps show more conclusively that topicalized phrases cannot be followed by pragmatic markers like neeno and tjawell. If these constructions were nonetheless derived from contrastive LD-constructions, we would predict that they could also precede nee/tja. The (a)-examples in (85) show that this expectation is wrong; topicalized modal adverbials must follow these elements. The (b)-examples illustrate the same for cases with a preposed non-referential part of an idiomatic expression.

The discussion in this subsection has shown that, despite its continuing appeal, the deletion approach faces a number of challenges; future research will need to show whether or not these can be overcome in a satisfactory way.