Section 4.2 has already shown that there are different restrictions on the interpretation of referential subject pronouns of finite clauses on the one hand, and PRO-subjects of infinitival complement clauses on the other. The former can freely take some antecedent from the matrix clause or refer to some entity that is part of the domain of discourse, whereas the latter must be coreferential with some noun phrase in the matrix clause. We illustrate this again by showing that passivization of the primeless examples in (26) gives rise to different results in acceptability: the subject pronoun hij'he' of the embedded finite clause in (26a') can readily take some antecedent from the discourse domain, whereas the PRO-subject of the infinitival clause in (26b') cannot.

The restrictions on the interpretation of PRO-subjects of infinitival complement clauses have become known as control theory. In many cases, it is required that PRO should be controlled, that is, bound by some antecedent in the matrix clause. The examples in (27) show, however, that PRO cannot take just any antecedent; in (27a) PRO can only be controlled by the subject and in (27b) it can only be controlled by the object of the matrix clause. The available readings are indicated by means of referential indices.

The examples in (27) suggest that the interpretation of PRO is determined by the matrix verb: accordingly, verbs like beloven'to promise' have become known as subject control verbs, and verbs like vragen'to ask' as object control verbs. However, the situation is more complex given that the contents of the embedded clause may also affect the control options; adding a deontic modal verb like mogen'to be allowed' to the infinitival clauses in (27), for example, reverses the interpretation possibilities of PRO, a phenomenon known as control shift.

The examples in (27) and (28) show that the interpretation of PRO can be affected by properties of both the matrix verb and the infinitival clause. Moreover, it would seem that these restrictions are not syntactic in nature but related to our knowledge of the world; the interpretation of example (27a), for instance, is related to the fact that the speaker has the ability to promise that he will perform a certain action himself but he cannot promise that the addressee will perform that action; the interpretation of example (28a), on the other hand, is based on the fact that the speaker may grant permission to the addressee to do something, whereas it is much less likely that he will or needs to grant such permission to himself. Consequently, it is not at all surprising that we find similar shifts when the verbs beloven and vragen take finite clauses as their complement.

For the moment we provisionally conclude that the PRO-subject of infinitival argument clauses must be controlled by some antecedent in the matrix clause, but that the actual choice of the antecedent must be compatible with our knowledge of the world. Section 5.2 will show, however, that there are circumstances under which the PRO-subject may be exempt from the requirement that it should be bound.