Doen-support is common in cases of VP-topicalization, i.e. in cases where a verbal projection is topicalized. A typical example is given in (263). One possible account of the insertion of doen is to appeal to the verb-second constraint on main clauses; because the main verb is part of the fronted VP, there is no verb available to satisfy this constraint, and the verb doen must therefore be inserted to save the resulting structure from ungrammaticality. Another possibility is to say that VP-topicalization makes it impossible to express the tense features of the clause on the main verb, and that doen must be inserted to make the expression of these features possible. We prefer the latter option, as it correctly predicts that do-support is not restricted to main clauses, but can also be found in embedded clauses; note that we will show shortly that the markedness of example (263b) is not due to the presence of doen, but to VP-topicalization across the boundary of the embedded clause.

That we are dealing with “last resort” insertion can be supported in several ways. First, the examples in (264) show that VP-topicalization is an absolute prerequisite for doen-support; if the verb phrase is in clause-final position, doen-support is impossible in both main and embedded clauses. For completeness, note that pronouns usually precede the negation, but that this does not improve the result in (264a), and that (264b) is unacceptable regardless of the word order of the clause-final verb cluster in the embedded clause.

Second, the (a)-examples in (265) show that the insertion of doen is possible only if there is no other verb capable of satisfying the verb-second requirement and/or expressing the tense features; since the modal verb kunnencan can perform these functions, the insertion of doen is not needed and is therefore excluded by the last-resort nature of doen-support. The (b)-examples in (265) are interesting not because of the unacceptability of (265b), which is to be expected given the unacceptability of (265a), but because the markedness of (265b') shows that the markedness of (263b) is not due to the fact that doen is part of an embedded clause, but to the fact that VP-topicalization from an embedded clause leads to a somewhat marked result.

Doen-support is also possible in cases of contrastive left dislocation in examples such as (266). It is a matter of debate whether or not the VP-topicalization constructions in (263) are derived from these left-dislocation constructions by deletion of the deictic pronoun dat in main-clause initial position. If so, the constructions in (266) could be analyzed similarly to the examples in (263).

That we are dealing with “last resort” insertion of doen is supported by the fact that it is impossible if there is another verb that can satisfy the verb-second requirement and/or to express the tense features.

Example (268a) shows that Dutch VP- pronominalization involves the pronoun datthat or watwhat. The obligatory insertion of doento do in this example can perhaps be explained in the same way as in the case of topicalization and left dislocation of the verb phrase: since VP-pronominalization removes the main verb, some other verb is needed to satisfy the verb-second requirement and/or to express the tense features. However, the “last resort” nature of doen-support is less clear, since doen is also possible if there is another verb that can perform these functions (although some speakers may consider expressing doen as the less preferred option).

The first two examples in (269) show that the object of the main verb haarher need not be pied-piped by a topicalized/left-dislocated VP, but can also be stranded. VP-pronominalization, on the other hand, can must include the direct object, i.e. example (269c) is unacceptable if we add the direct object haar to the second conjunct; cf. Section 2.3.1, sub VII, for similar data with prepositional objects.

This suggests that pronominalization differs in a crucial way from left dislocation and topicalization. It is important to note that the difference is not located in the verb doen, since the acceptability judgments on (269) do not change when we substitute the modal verb willen for doen.

The fact that the object haarher cannot be expressed in (269c) suggests that doento do can be analyzed as a regular transitive main verb in examples such as Jan doet het graagJan does it with pleasure. The verb doen in (269b), on the other hand, cannot be analyzed as a main verb, because the main verb doen is not a ditransitive verb: cf. *Jan doet het haar graag (lit: Jan does itDO herIO with pleasure). For completeness, note that the contrast between the (b) and (c)-examples in (269) and (270) also shows that the two occurrences of dat have different functions: in the (c)-examples it clearly functions as a demonstrative pronoun with the function of direct object, whereas in the (b)-examples it does not.

We have argued that doen is used as a non-main verb only in VP-topicalization and left-dislocation constructions; in VP-pronominalization contexts it is a main verb. This conclusion can be further supported by considering the restrictions on VP-pronominalization in the contexts of the main verb doen. First, consider the examples in (271), which show that doen typically expresses an activity controlled by the subject of the clause; while the object pronoun dat in the second conjunct can easily refer to activities like reading something with an agentive subject, as in (271a), it is impossible for it to refer to non-controlled events like getting something or knowing something with a goal/experiencer subject, as in (271b&c). The fact that all these examples are perfectly acceptable without doen shows that it is not pronominalization as such that causes this deviation, but the use of doen.

The VP-pronominalization construction in (271) contrasts sharply with the VP-topicalization and left-dislocation constructions in (272), which allow doen with both controllable activities and uncontrollable states. This is true both when the direct object of the preposed verb phrase is pied-piped and when it is stranded.

The previous subsections have shown that there are at least two sets of facts supporting the claim that VP-topicalization and left-dislocation constructions involve non-main verb doen, whereas VP-pronominalization always involves main verb doen. What we have not discussed is whether VP-topicalization and left-dislocation constructions must involve non-main verb doen; after all, it may well be the case that the preposed phrases in the primeless examples in (272) are not VPs, but nominalizations comparable to those in (273).

It is not easy to find a conclusive answer to the question whether the preposed phrases in the primeless examples in (272) can also be nominalizations, but the fact that the examples in (274), in which the supposed nominalizations are clause-internal, are unacceptable seems to make this analysis very unlikely.

That the primed examples in (272) do not involve nominalizations seems uncontroversial, since nominalizations behave as a unit under movement and are therefore normally not split by topicalization.

We can tentatively conclude that the verb doen in VP-topicalization and left-dislocation constructions cannot be analyzed as a main verb.

This subsection has argued that VP-topicalization, VP-left-dislocation and VP-pronominalization constructions with the verb doen cannot all be treated on a par: while doen must be analyzed as a non-main verb in the first two cases, it clearly acts as a transitive main verb in the third case.