Obligatory control is usually given an operational definition; to speak of obligatory control, the antecedent of PRO must at least satisfy the four requirements in (395), repeated here as (398).

These properties of obligatory control are illustrated by the English examples in (399) to (401). The examples in (399) show that the antecedent must be overtly realized in the sentence containing PRO; cf. Bresnan (1982) and Manzini (1983). Example (399a') shows that passivization and the concomitant demotion of the subject is impossible in subject-control constructions, while example (399b'') shows that omission of the nominal object is impossible in object-control constructions. We use the index “?” to indicate that this is due to the lack of a suitable controller available in the syntactic structure.

That the antecedent of PRO must be a co-argument of the infinitival clause containing PRO can be illustrated by the examples in (400), which show that the unacceptable examples in (399) cannot be saved by embedding them in a larger sentence that does have a potential antecedent of PRO; since the antecedent must be within the clause headed by the subject/object-control verbs to promise and to ask, the subjects of the main clauses headed by to think cannot function as such.

That the antecedent of PRO must be a c-commanding nominal argument (subject or indirect object) is clear from the fact that the passive counterpart of (399a') does not improve when we add an agentive by-PP: *Billj was promised by Johni [PROi to shave himselfi] is unacceptable because the antecedent of PRO is not a nominal argument of the matrix verb but part of the adverbial agentive by-PP. Finally, the unacceptability of the examples in (401) shows that the antecedent of PRO must be unique in the sense that PRO cannot have a split antecedent.

It is usually assumed that obligatory control requires all four restrictions to be satisfied. The theoretical motivation is that obligatory control is comparable to binding of reflexive pronouns and NP-movement to subject position in passive, unaccusative, and raising constructions. All of these exhibit properties of locally restricted syntactic dependencies, i.e. they are characterized by being obligatory (398a), being local (398b), involving c-command (398c), and being unique (398d); cf. Koster (1984a/1984b) for a detailed discussion. This means that it is sufficient to show for just one of the restrictions in (398) that it does not hold in order to establish that we are dealing with non-obligatory control.

By definition, subject-control verbs must be minimally dyadic: they have an infinitival argument clause with an implied PRO-subject as well as a subject that can act as the antecedent of PRO. This is consistent with the fact that subject-control verbs are usually transitive or ditransitive verbs, or verbs that take a prepositional object. In (402) we first give a small sample of transitive subject-control verbs.

In (403) we give two concrete examples. Note that they may contain the anticipatory pronoun hetit to introduce the infinitival object clause. This pronoun is normally optional, but some verbs in (402) require its presence; this is especially true for particle verbs like aandurvento dare and afzwerento renounce, and some of the verbs prefixed with be- (bestaanto have the nerve), ont- (ontwennento break oneʼs habit), and ver- (vertikkento refuse, verlerento lose the hang of). At first glance, the primeless examples in (403) seem to be good candidates for obligatory control constructions: the antecedent of PRO is local, a nominal argument (subject), and unique. However, it turns out that the antecedent is not obligatory: the primed examples show that the matrix clause in examples such as (403) can be passivized.

Passive examples of the kind in (403) can be easily found on the internet by doing a Google search on the strings [er werd/is geprobeerd om] it was/has been tried and [er werd/is vergeten om] It was/has been forgotten. The other verbs in (402) also seem to allow passivization when they take an om + te-infinitival as direct object. If obligatory control indeed requires all four properties to be met, we have to conclude that the primeless examples in (403) are non-obligatory control constructions.

There are not many ditransitive subject-control verbs that take an om + te-infinitival clause as direct object. The set given in (404) seems to exhaust the possibilities. Note in passing that this set of subject-control verbs can be extended as a result of so-called control shift: we will ignore this issue here, but return to it in Subsection E.

At first glance, we seem to be dealing with obligatory control: the antecedent of PRO is local, a nominal argument (subject), and unique. But again it turns out that the antecedent is not obligatory, in the sense that the three verbs in (404) can all be passivized. The acceptability of passivization, illustrated in (405) for aanbiedento offer and belovento promise, shows that we are dealing with non-obligatory control constructions.

Again, more examples can easily be found on the internet by doing a Google search on the strings [er wordt/is aangeboden om], [er wordt/is beloofd om] and [er wordt/is toegezegd om]. For completeness, note that the verbs in (404) are also compatible with object control under certain conditions; we will ignore this for now, but return to it in Subsection E.

Subject control also occurs with verbs that take a prepositional object. Section 2.3 has shown that such PO-verbs can be intransitive, transitive, or unaccusative. A sample of each type is given in (406). Like regular ditransitive verbs, ditransitive PO-verbs are not very common as subject-control verbs; the only case we have found is dreigen (met)to threaten, and even this verb is very often (or perhaps normally) used without a nominal object. Note that the infinitival clause can optionally be introduced by an anticipatory pronominal PP; whether this PP is obligatory, optional, or preferably left implicit depends on the verb in question, and may also vary from person to person; cf. Section 2.3 for more discussion.

The examples in (407) show that intransitive PO-verbs that allow passivization also do so when they take an om + te-infinitival as prepositional object; more examples can be found with a Google search on the string [er wordt/is (naar) gestreefd om].

Since dreigen (met) is the only transitive PO-verb that triggers subject control, it is difficult to determine whether such verbs allow passivization, especially since examples such as (408b) are at best marginally acceptable. What we can conclude from this is unclear: since dreigen (met) is normally used without a nominal object, it is not surprising that we normally find the impersonal variant in (408b'). In the absence of sufficient evidence, we have to leave open the question of whether transitive PO-verbs involve obligatory or non-obligatory control.

Despite the unclear status of the (b)-example in (408), the acceptability of the impersonal passive examples in (403), (405), and (407) clearly shows that we are dealing with non-obligatory control constructions. Of course, there are also a reasonable number of unaccusative subject-control verbs, but these do not shed any light on the question as to whether we are dealing with obligatory or non-obligatory control, because they do not allow passivization anyway. The examples in (409) therefore only serve to illustrate the use of these verbs.

Note, however, that Section 2.3.2, sub IV, mentioned a number of possible unaccusative PO-verbs that are special in that they allow passivization. Some of these verbs (aanvangen/beginnen (met)to start with, ophouden/stoppen (met)to stop with, overgaan (tot)to proceed with) can be used as subject-control verbs and then retain their ability to undergo passivization. This again shows that subject-control structures of the kind discussed here do not involve obligatory control.

A final set of verbs that seem to trigger subject control are the inherently reflexive verbs in (411). Such verbs do not shed any light on the question as to whether subject control with verbs taking an om + te-infinitival as complement involves obligatory control, because inherently reflexive verbs never undergo passivization.

In fact, it is not entirely clear whether we are really dealing with subject control in such cases, as it depends on whether one is willing to assign argument status to the weak reflexive (which is not usually done). If so, one might as well assume that the subject-control reading is mediated by the reflexive, in which case one might claim that we are dealing with object control in the examples in (412).

For completeness, note that undative verbs are not used as subject-control verbs for the simple reason that verbs like hebbento have, krijgento get and houdento keep do not take infinitival complements.

Object-control verbs must be at least dyadic by definition: they must have an infinitival argument clause with an implied PRO-subject, as well as an indirect (dative) object that can act as the antecedent of the implied PRO-subject. Table 1 shows that object-control verbs with an external argument are typically ditransitive, unless they take an additional prepositional object, in which case they can also be transitive; we give a small sample of such object-control verbs in (413).

Note that some of the verbs in (413a), such as verzoeken/vragento request and overredento persuade, are also compatible with subject control under certain conditions; we defer discussion of this to Subsection E.

The examples in (414) provide a concrete case of object control with a ditransitive verb and show that passivization is easily allowed. However, it is not immediately clear whether this shows that we are dealing with non-obligatory control. One could argue that the promotion of the infinitival clause to subject destroys the c-command relation between the indirect object and the PRO-subject; indirect objects do not c-command subjects. However, one could also argue that we are dealing with a reconstruction effect, in that it is not the surface but the underlying representation that matters. Moreover, Section N22.1 will show that linear order can sometimes also affect the binding possibilities. Therefore, we will not digress on this issue here and tentatively assume that the pattern in (414) is expected.

A more conclusive reason for thinking that we are not dealing with obligatory control is that the indirect object is often omitted. The results of Google searches for [raad(t)/raden af om] and its passive counterpart [er wordt afgeraden om] show that strings without an indirect object are very common. However, not all verbs allow the omission of the direct object: it seems to give less felicitous results with the verbs belettento prevent, misgunnento envy, opdragento assign, toelatento allow and verhinderento prevent (although actual judgments vary from verb to verb and speaker to speaker).

The examples in (415) provide a similar concrete case of object control with a transitive PO-verb and shows that passivization is readily allowed. This does not provide evidence against an obligatory control analysis, but it does show that the notion of object control should be taken with a pinch of salt; it is clearly not the syntactic function of the antecedent that is at stake but its semantic function.

Again, the fact that the antecedent is often omitted shows that we are dealing with a non-obligatory control. The results of Google searches for [roep(t)/roepen op om] and its passive counterpart [er wordt opgeroepen om] show that strings without a noun phrase that could act as an antecedent for PRO are common. However, not all verbs allow the omission of the direct object: it seems to give less felicitous results with the verbs activeren (tot)to activate, belemmeren (in), dwingen (tot)to force, machtigen (tot)to authorize, herinneren (aan)to remind of, overhalen (tot)to persuade, overreden (tot)to persuade, overtuigen vanto convince of, stijven *(in ...)to confirm someone in, uitdagen (tot)to challenge, uitnodigen (tot)to invite to, verleiden (tot)to tempt to, verplichten (tot)to oblige to (although the actual judgments again vary from verb to verb and speaker to speaker).

Since the ditransitive verbs in (413) have the possibility of passivization when they take an om + te-infinitival as direct object, it is not surprising that there are also dyadic unaccusative (nom-dat) verbs that select om + te-infinitivals; in both cases the infinitival clause is an internal argument of the verb that is promoted to subject. Section 2.1.3, sub II, has shown that there are two types of nom-dat; both types do indeed include object-control verbs.

Two concrete examples of the object-control version of these verb types are given in (417). Of course, passivization cannot be used to show that we are dealing with non-obligatory control, because passivization of unaccusative verbs is impossible anyway.

However, it should be clear from the fact that the antecedent can often be omitted that we are dealing with non-obligatory control. Our Google search has shown that strings like [het bevalt goed om] it pleases and [het spreekt aan om] it appeals occur regularly. However, there seems to be a preference to interpret PRO as referring to the speaker in these cases. Furthermore, the omission of the indirect object seems to give less felicitous results with the verbs (gemakkelijk) afgaanto come easy to, (e.g. goed) bekomento agree with, ontgaanto escape, ontschietento slip oneʼs mind, aanstaanto please, behagento please, berouwento regret, betamento befit, tegenstaanto pall on, (niet) zinnento (not) please (where the actual judgments may again vary from verb to verb and speaker to speaker).

Finally, note that the vast majority of the causative object-experiencer psych-verbs discussed in Section 2.5.1.3, sub II, i.e. verbs like amuserento amuse, bemoedigento encourage, boeiento fascinate, ergerento annoy, fascinerento fascinate, and grievento hurt, can be used as object-experiencer verbs with the om + te-infinitival functioning as a cause. Again, passivization cannot be used to demonstrate that we are dealing with non-obligatory control, because passivization of causative object-experiencer psych-verbs is impossible anyway (cf. Section 2.5.1.3, sub IID), but it is supported by the fact, illustrated in (418), that the object can be omitted in various cases (with the actual judgments again varying from verb to verb and speaker to speaker).

This section has shown that there is reason to think that object control verbs do not involve obligatory control, because there are many cases in which the antecedent of PRO need not be expressed overtly; this violates requirement (398a) for obligatory control constructions.

This subsection discusses a number of cases involving non-obligatory control. In particular, we will discuss cases that violate the uniqueness requirement for obligatory control in (398d) by allowing PRO to take a split antecedent, as well as cases that violate the overt antecedent requirement in (398a) by allowing PRO to receive an arbitrary interpretation.

The previous subsections have shown that there is actually no reason to claim that PRO-subjects of om + te-infinitivals used as arguments involve obligatory control in the sense defined in (398), repeated here as (438).

This does not mean that the interpretation of PRO-subjects is completely free, only that it is not subject to syntactic restrictions; there is good reason to think that the meaning of the matrix verb imposes constraints on the interpretation of the PRO-subject. For example, the ditransitive verb belovento promise in (439a) can be characterized as a verb requiring control by its agent, which would also explain why the controller of PRO must be the nominal complement in the optional agentive door-PP (and not the c-commanding indirect object pronoun ons) in the corresponding passive construction in (439a'). In fact, we can show the same with the nominal constructions in the (b)-examples: the controller of the PRO-subject must be bound to the agent of the nominalization beloftepromise, regardless of whether it c-commands PRO or not: this is clear from the fact that it can be expressed not only by a c-commanding prenominal possessor, but also by a postnominal van-PP. The doubly-primed examples are added to show that the agent can also be left implicit in both the verbal and the nominal construction, in violation of the condition on obligatory control in (438a).

The primeless examples in (440) show that bare indirect objects in object-control constructions must be realized as prepositional indirect objects in the corresponding nominalizations. Since the controller hemhim is part of the prepositional indirect object, it does not c-command the PRO-subject of the infinitival direct object clause in (440b); again, this shows that the c-command restriction in (438c) can be violated in the case of om + te-infinitivals. The primed examples are added to show that the indirect object can also be omitted in both the verbal and the nominal construction, in violation of the condition on obligatory control in (438a).