Example (453) provides a sample of three subtypes of verbs with te-infinitival argument clauses that normally trigger subject control. The transitive and ditransitive verbs in (453a&b) are propositional verbs with a factive or non-factive complement clause; cf. Cremers (1983) and Van Haaften (1991: §4). The prepositional object verbs in (453c) also trigger subject control.

We have already shown in (452b) with the transitive verb bewerento claim that passivization of subject-control verbs is impossible because it demotes their subjects to adjunct status. This is illustrated again in (454); the passivization of the transitive verb gelovento believe results in unacceptability, regardless of whether the demoted subject is expressed by a door-PP or not. This supports the claim that we are dealing with obligatory control.

The examples in (455) show the same for the ditransitive verb garanderento guarantee; passivization is blocked regardless of whether the demoted subject is expressed by a door-PP or not. Again, this supports the claim that we are dealing with obligatory control.

The ditransitive verb schrijvento write is special in that it allows not only subject but also object control. Consider first the primeless examples in (456), which show that the actual interpretation of PRO depends on the pronoun hem/haarhim/her in the infinitival clause. On the reading that the pronoun hem is coreferential with the object of the matrix clause, example (456a) can only be interpreted as PRO being controlled by the subject of the matrix clause: object control would violate the requirement that the pronoun be free (= not bound) within its own clause; cf. Section N22.3. Similarly, on the reading that the pronoun haar is coreferential with the subject of the matrix clause, example (456b) must be interpreted as PRO being controlled by the object of the matrix clause: subject control would again violate the requirement that the pronoun be free in its own clause. The crucial point is that the contrast in acceptability between the two primed examples in (456) shows that subject control blocks passivization, whereas object control allows it. These passivization facts again show that we are dealing with obligatory control: example (456a') is unacceptable in the reading that Jan will be sent the book, because passivization demotes the subject controller to adjunct status; example (456b') is acceptable, because passivization does not affect the status of the object controller.

That the pattern in (456) is not accidental is clear from the fact that we find essentially the same in the examples in (457), where we see that the actual interpretation of PRO is restricted by the fact that the simplex reflexive zich/me of the inherently reflexive verb zich haastento hurry must have an antecedent in its own clause; in (457a) the third-person reflexive zich requires PRO to be controlled by the third-person subject, and in (457b) the first-person reflexive me requires PRO to be controlled by the first-person object of the clause. The contrast in acceptability between the primed examples again bears out that the subject-control constructions cannot be passivized.

The examples in (458) show that PO-verbs such as rekenen opto count on are perfectly compatible with passivization when they take a finite complement clause but not when they take a te-infinitival clause; the two primed (b)-examples are clearly degraded. This again shows that PRO-subjects of te-infinitivals are obligatorily controlled; note that this cannot be illustrated for the inherently reflexive PO-verbs in (453c), because they cannot be passivized anyway.

For completeness’ sake, note that some adjective phrases also take te-infinitival clauses as prepositional objects; examples are doordrongen (van)convinced of the necessity of and zeker (van)certain of. In such cases, the PRO-subject is controlled by the logical subject of the adjective (which appears as the subject of a copular sentence): cf. Jani is ervan doordrongen [PROi dat boek te moeten lezen] Jan is imbued with the need to read that book. It is difficult to say whether we are dealing with obligatory control here, because subjects of predicatively used adjectival phrases usually cannot be omitted for independent reasons.

The discussion above has shown that there are good reasons to assume that PRO-subjects of te-infinitivals differ from PRO-subjects of om + te-infinitivals in that they are obligatorily controlled. This might also be supported by the nominalizations in (459a&b), as Haaften (1991:100) deems (459b) to be unacceptable due to the lack of an overt controller for PRO. A possible problem with this conclusion, however, is that (459c) is acceptable, which would then be unexpected given the c-command restriction on obligatory control in (450c). We refer to Hoekstra (1999) for a possible solution to the c-command problem posed by (459c) based on the claim that the preposition van is not a preposition in the traditional sense of the word but a complementizer-like element; cf. also Kayne (2000: Part III) and Den Dikken (2006a).

We did not mark example (459b) with an asterisk because some speakers at least marginally accept such examples. Koster (1984b: §5), for example, claims that “it is almost always possible to replace the subject controller of an NP by an article”, and he concludes that obligatory control requires that the te-infinitival should be a complement of a verb. If we are indeed dealing with non-obligatory control in (459a), this would not only account for the fact that some speakers accept (459b), but also explain that the controller in (459c) can be expressed by a van-PP. Although it is not obvious that the nominalization facts in (459) support the claim that PRO-subjects of all te-infinitivals are obligatorily controlled, we can still maintain that PRO-subjects of te-infinitivals selected by verbs cannot receive an arbitrary interpretation, but must be controlled by a nominal argument of the matrix verb.

There are not many object-control verbs that take te-infinitivals as arguments, and for this reason we have grouped the ditransitive verbs and the transitive PO-verbs together. Although causative psych-verbs acting as object-control verbs usually select om + te-infinitivals, a limited number of them take a te-infinitival.

The verbs in (460a) normally require object control, as shown by example (461a). It is difficult to establish, however, whether we are dealing with obligatory control because passivization does not affect the syntactic status of the indirect object of a ditransitive verb like verwijten in (461a). And although the object of a transitive PO-verb like verdenken (van)to suspect (of) is promoted to subject, the acceptability of (461b') is still in full accordance with the characterization of obligatory control in (450): the derived subject can function as a unique, local and c-commanding controller of the PRO-subject.

The hypothesis that we are dealing with obligatory control predicts that the indirect object cannot be omitted in examples such as (461). Example (462a) shows that this prediction is correct, but this is does not help much, as the indirect object must also be present in examples such as (462b), where the infinitival is replaced by a finite clause; it is therefore likely that the degraded status of (462a) is due to independent factors.

Nevertheless, there is some indirect evidence that the verbs in (460a) involve obligatory control, based on the fact that some of these verbs allow control shift by manipulating the content of the infinitival clause, e.g. by adding a deontic modal like mogento be allowed. This is illustrated for the verb verwijtento reproach in (463); the fact that (463a) cannot be passivized supports the claim that the PRO-subject of the te-infinitival is obligatorily controlled.

Another possible argument can be built on the nominalizations of the (a)-examples in (462) and (463). Since the indirect object must be realized as an aan-PP in nominalizations, we expect object control to be blocked by the c-command restriction on obligatory control in (450c). However, the result is ambiguous: although many speakers consider example (464a) to be marked compared to (464b), some speakers tend to accept it. The primed examples show that omitting the controller altogether leads to a degraded result, and this supports the idea that we are dealing with obligatory control. However, some speakers report that they do accept example (464c), where both arguments are implicit, which contradicts this idea. The examples in (464) again show that it is not obvious that PRO-subjects of te-infinitivals are obligatorily controlled in nominalizations, which can perhaps be seen as evidence for Koster’s (1984b) claim that obligatory control occurs only in te-infinitival complements of verbs.

Leaving aside the problematic status of the examples in (464), we can again conclude that the verbal constructions discussed in this subsection confirm the prediction that PRO-subjects of te-infinitival argument clauses cannot receive an arbitrary interpretation, but must be controlled by a nominal argument of the matrix verb.

There are good reasons to assume that the verbs in (453) and (460) trigger obligatory control when they select a te-infinitival clause. First, the restrictions on obligatory control in (450a-c) predict that PRO cannot have arbitrary reference, but must have an overt controller functioning as a nominal argument of the matrix verb; the previous two subsections have shown that this prediction is essentially correct. Second, the uniqueness restriction on obligatory control in (450d) predicts that PRO cannot have a split antecedent. This subsection will show that this prediction is also correct: the core data are provided in Subsection A, while Subsection B discusses a possible counterexample.

The comparison of control in te-infinitival complements with control in om + te-infinitivals in Section 5.2.1.3 shows that the two types of infinitival clauses differ systematically in that the PRO-subjects of te-infinitivals, but not those of om + te-infinitivals, are obligatorily controlled in the sense defined in (450); cf. Van Haaften (1991) and Everaert (1991) for the same conclusion. Van Haaften claimed that this distinction is related to the semantic interpretation of the infinitival clauses; while te-infinitivals are propositional in nature, i.e. can be assigned a truth value, om + te-infinitivals are not. This difference in meaning is illustrated in the primeless examples in (472); the English renderings in the primeless examples show that the te-infinitivals in the (b)-examples, but not the om + te-infinitivals in the (a)-examples, entail that Jan actually reads the book at speech time (the number signs in the translations indicate impossible readings). This shows that om + te-infinitivals refer to potential states of affairs in the non-actualized part of the tense interval; cf. Section 1.5.1, sub I, for this notion. This is also clear from the fact, illustrated in the primed examples, that om + te-infinitivals differ from te-infinitivals in that they cannot contain the adverbial element alalready in present-tense constructions; cf. Janssen (1992) for further discussion.

Van Haaften proposes the following diagnostics for distinguishing the two semantic types: propositional infinitivals allow epistemic modals, whereas non-propositional infinitivals do not (and the same holds in fact for deontic modals). Van Haaften further notes that propositional infinitivals can always be replaced by finite clauses, whereas this is often impossible with non-propositional clauses. See also Cremers (1983), where the distinction between propositional and non-propositional infinitival clauses was first proposed to account for a number of other differences (e.g. concerning tense, gapping and topicalization).

Van Haaften also claims that propositional infinitival clauses require obligatory control because they can only be assigned a truth value if their subject is assigned a referential value.

Although this semantic approach seems to provide a more or less descriptively adequate description of the control facts, it still does not explain why the locally restricted syntactic dependency relation of obligatory control applies only to te-infinitivals. The remainder of this subsection will attempt to formulate an explanation in terms of the CP/TP distinction introduced in Section 9.1. If we follow Bennis & Hoekstra’s (1985) claim, discussed in Section 5.2.1.2, that om is a complementizer-like element situated in CP, we can hypothesize that the impossibility of having om in te-infinitivals marks the absence of the CP-projection; they are TPs.

Note that the hypotheses in (474) are not uncontroversial; Bennis & Hoekstra (1989c), for example, assume that all control infinitivals are CPs, which is convenient for them, since they do not make the distinction between obligatory and non-obligatory control as defined in (450). We will try to provide a more solid basis for these hypotheses by considering in more detail the problematic verb zeggen discussed in Subsection IIIB. In line with the hypotheses in (474), we can assign the TP-structures in the (a)-examples in (475) to the non-directive subject-control examples in (469), since they cannot be introduced by the complementizer om; note that we added the epistemic modal zullen to block the non-propositional, directive reading. The directive object-control examples in (470), on the other hand, must be assigned the CP-structures in the (b)-examples to account for the fact that they can be introduced by om. The primed examples show again that non-directive zeggen triggers obligatory control, while directive zeggen does not, as is clear from the fact that it allows a non-c-commanding or implicit controller for PRO.

The proposed structural difference between non-directive and directive zeggen receives independent support from the fact that example (476a) only allows a directive reading of the verb. The reason for this is that embedded wh-movement requires there to be a CP-projection within the embedded clause, since it targets the position left-adjacent to the phonetically empty complementizer (indicated by Ø); compare Ik weet niet wat of hij doet I do not know what he is doing, where the wh-phrase is placed to the left of the interrogative complementizer ofwhether. This leads to a conflict in the case of the non-directive subject-control verb zeggen: this verb selects a TP, whereas wh-movement requires there to be a CP, with the result that structure (476b) is rejected. In the case of the directive object-control verb zeggen in (476b') there is no problem: both wh-movement and the selection restrictions of the verb require there to be a CP. Note in passing that we have marked wat in (476a) with the label [+wh] to exclude its indefinite referential interpretation “something”, as this interpretation would be compatible with a directive reading.

The hypotheses in (474) allow us to state the robust difference in control properties between om + te-infinitivals and te-infinitivals in more general terms: PRO-subjects of infinitival TPs, but not of infinitival CPs, are obligatorily controlled. This, in turn, can be formulated even more generally by using the term island for locally restricted syntactic dependencies; cf. Hornstein (2001:56ff) for a similar conclusion on the basis of English. The hypotheses in (477) express that such syntactic dependencies (like NP-movement and anaphor binding) can in principle be established across a TP-boundary, but not across a CP-boundary.

That obligatory control is also covered by the generalizations in (477) is exactly what one would expect on the basis of its definition in (450), since Section 5.2.1.3, sub IIIA, has shown that it is molded on the more general definition of locally restricted syntactic dependency found in Koster (1984a/1987). For example, Section 5.2.2.2 will show that CPs, but not TPs, are syntactic islands for subject raising (which is a subtype of NP-movement). It may be interesting to note here that Koster’s (1984b) claim that te-infinitival complements of nouns do not involve obligatory control suggests that TP-status may not be sufficient for transparency, and that (477b) may therefore be restricted to TPs governed by a verb. This would predict that subject raising is like obligatory control in that it is restricted to TP-complements of verbs, and this is indeed what Koster claims to be the case: cf. Section 5.2.2.2, sub C, for an illustration of this.

What remains to be determined is what kind of syntactic dependency is obligatory control: Does control involve a different kind of syntactic dependency than movement, as the more traditional generative approaches (such as Chomsky 1981) claim, or are control and movement essentially identical syntactic dependencies? If the latter, then their apparent differences cannot be explained by assuming some inherent difference between PRO and trace, but must be due to some other difference. For example, it has been argued that these differences are due to whether or not the antecedent of the empty element (PRO/trace) has an independent thematic role. We will not discuss this proposal here, but refer the reader to Koster (1978a: §2.1.1) and especially Koster (1984a/1984b) for this line of investigation, which has more recently been revived in a somewhat different form in Hornstein (2001) and in the articles collected in Hornstein & Polinsky (2010).