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. We have seen cases in which the pronoun can freely take an antecedent from the matrix clause or refer to an entity that is part of the discourse domain, whereas PRO 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) leads to different results in terms of acceptability: the subject pronoun hijhe of the embedded finite clause in (26a') can easily take some antecedent from the discourse domain, whereas the PRO-subject of the infinitival clause in (26b') cannot.

The set of restrictions on the interpretation of PRO-subjects of infinitival complement clauses has become known as control theory. In many cases, it is required that PRO be controlled, i.e. bound by some antecedent in the matrix clause. However, the examples in (27) show 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 referential indices.

The examples in (27) suggest that the interpretation of PRO is determined by the matrix verb: accordingly, verbs such as belovento promise have come to be known as subject-control verbs, and verbs such as vragento ask as object-control verbs. However, the situation is more complex because the content of the embedded clause can also affect the control options; for example, adding a deontic modal verb such as mogento be allowed to the infinitival clauses in (27) reverses the interpretation possibilities of PRO, a phenomenon known as control shift; cf. Section 5.2.1.3, sub IIIE.

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 seems that these restrictions are not syntactic but related to our knowledge of the world; for instance, the interpretation of example (27a) is related to the fact that the speaker has the ability to promise that he himself will perform a certain action, 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 can give the addressee permission to do something, whereas it is much less likely that he will or has to give himself such permission. Consequently, it is not at all surprising that we find similar shifts when the verbs beloven and vragen take finite clauses as their complement. Note that the discussion has ignored the perfectly acceptable reading of (29b'), in which Jan asks Peter if he (= Peter) has received permission from someone else (e.g. his parents) to come.

For the moment, we tentatively 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 antecedent must be compatible with our knowledge of the world. However, Section 5.2 will show that there are circumstances under which the PRO-subject can be exempted from the requirement that it should be bound.