Bare infinitival complement clauses selected by perception verbs like ziento see or the causative/permissive verb latento make/let exhibit an accusativus-cum-infinitivo effect: the subjects of the bare infinitival clauses do not appear as the phonetically empty element PRO, as would normally be the case in infinitival clauses, but as an accusative noun phrase. This is illustrated in (39), where the subject of the infinitival clause is underlined.

It is generally claimed that the subject of the infinitival complement is assigned accusative case by the matrix verb; we are dealing with so-called exceptional case marking across the boundary of an infinitival clause. However, the claim that the matrix verb assigns case to the subject of the embedded clause is not so easy to prove for Dutch, because the examples in (40) show that the matrix verb cannot be passivized in such contexts. Since the standard passivization test is not available, the claim can only be justified by appealing to the fact that there simply seems to be no other element that can be held responsible for the case assignment.

That the underlined phrases in (39) are not selected by the matrix verbs but function as subjects of the bare infinitival clauses can be tested by pronominalization of the infinitival clause; the fact that the accusative noun phrase cannot appear in (41a) strongly suggests that it is not selected by the matrix verb horento hear but must be part of the infinitival clause pronominalized by datthat. Unfortunately, (41b) shows that pronominalization cannot easily be used as a test in the case of the verb latento make/let; it is at best marginally acceptable with this verb in its permissive reading, and completely excluded in its causative reading, regardless of the presence of the noun phrase het meisje/haar. This means that especially the hypothesis that constructions with laten are AcI-constructions is based on rather weak evidence; we will ignore this for the moment, but refer the reader to Section 5.2.3.4 for further discussion of this issue and arguments for a completely different analysis.

Accusativus-cum-infinitivo constructions of the type discussed here exhibit monoclausal behavior. First, as indicated in italics in (39) above, the bare infinitival complements are normally split; while the bare infinitives follow the matrix verbs in clause-final position, their arguments must precede them. Second, the examples in (42) show that they exhibit the IPP-effect; the matrix verb cannot appear as a past participle in perfect-tense constructions, but must be realized as an infinitive.

A bare infinitival clause that is selected by a so-called root/deontic modal like kunnento be able, mogento be allowed or willento want, or by a verb such as lerento teach/learn, has its subject realized as the phonetically empty pronominal-like element PRO. As in the case of (om +) te-infinitivals, the PRO-subject of the bare infinitival can be controlled either by the subject or by the object of the matrix clause. The choice again depends on the matrix verb: for instance, while the deontic modals and the intransitive verb lerento learn require PRO to be controlled by their subjects, the transitive verb lerento teach requires PRO to be controlled by its object. Again, we have italicized the bare infinitival clause and underlined its subject.

Control constructions of the kind discussed here exhibit monoclausal behavior. First, the constructions in (43) show again that the bare infinitival can be split; as indicated in italics, the arguments of the bare infinitival precede the matrix verb in clause-final position, while the bare infinitive normally follows it. Second, the examples in (44) show that the construction exhibits the IPP-effect; the matrix verbs cannot appear as past participles in perfect-tense constructions, but must appear as infinitives.

That the noun phrases Jan in (43a) and zijn dochtertje in (43b&b') do not function as subjects of the bare infinitivals is clear from the fact, illustrated in (45), that they must be present when the infinitival clauses are pronominalized; this shows that these noun phrases are assigned thematic roles by the matrix verbs. The agent role of the bare infinitive must therefore be assigned to some independent argument, which motivates the stipulation of a PRO-subject in these examples.

Note that if we accept the conclusion from Section 4.6 that being an argument taking predicate is a defining property of main verbs, the fact that the root modal kunnento be able in the (a)-examples above licenses the subject noun phrase Jan independently of the embedded infinitival shows that the traditional assumption that root modals are non-main verbs cannot be maintained, and that they must instead be seen as regular transitive verbs; cf. Klooster (1984/2001). We will return to this issue in Section 4.5.

Subsection B has put on hold the fact that examples such as (43a) are actually ambiguous: the modal matrix verb can receive not only a deontic/root reading, but also a so-called epistemic reading. Although the most prominent reading of (43a) is the deontic one, the ambiguity can be brought out by putting this example in the perfect tense; if the modal verb is realized as a non-finite verb, it receives the deontic interpretation “to be able to”, but if it is realized as a finite verb, it receives the epistemic interpretation epistemic “may”.

That constructions with epistemic modals exhibit monoclausal behavior cannot be demonstrated by the IPP-effect, since the perfect auxiliary is now part of the infinitival complement of the modal verb (see below), but it is still clear from the fact that the bare infinitival can be split: the arguments of the infinitival clause precede the modal verb in clause-final position while the bare infinitive normally follows it. The underlining in (46) suggests completely different structures for the two constructions: if the modal verb has a deontic interpretation, the subject of the infinitival clause is realized as the phonetically empty pronominal element PRO; if the modal verb has an epistemic interpretation, the subject appears as the nominative subject of the whole sentence; it is promoted to subject (“raised to the subject position”) of the matrix clause. The reasons for this assumption are again related to pronominalization of the infinitival clause; example (47a) again illustrates that the nominative subject is not affected by pronominalization when the modal verb is deontic, whereas (47b) shows that the nominative argument cannot be realized when the modal is epistemic and should therefore be assumed to belong to the pronominalized infinitival clause. We have added example (47b') to support the earlier claim that the perfect auxiliary in the epistemic constructions in (46b) is part of the infinitival complement, pronominalized by dat in (47b&b').

Another good reason for assuming that the nominative subject in the epistemic example in (46b) originates as the subject of the infinitival complement clause is that this immediately explains why in passive constructions such as (48b) the internal argument of the bare infinitive stelento steal appears as the nominative subject of the sentence; passivization of the bare infinitive first promotes the noun phrase die autothat car to subject of the infinitival clause, and subject raising subsequently promotes it to subject of the matrix clause. On the alternative hypothesis that the nominative noun phrases in (48) originate as arguments of the epistemic modal kunnen, we can account for the pattern in (48) only by making the highly unlikely assumption that passivization of the embedded verb also affects the selection properties of the matrix verb.

A final argument to be mentioned here is that example (49a), with a subject clause introduced by the anticipatory pronoun hetit, is semantically more or less equivalent to (49b), at least with respect to the thematic relations between the italicized elements. If we assume that the nominative subject in (49b) originates as the subject of the infinitival clause and is subsequently promoted to subject of the matrix clause, the observed semantic equivalence follows straightforwardly.

Each of the examples in (47) to (49) strongly suggests that nominative subjects in epistemic constructions such as (46b) originate as the subject of the infinitival clause and are subsequently “raised” to the subject position of the matrix clause. The standard analysis formally derives this by assuming that the subject of the infinitival clause cannot be assigned accusative case and must therefore be assigned nominative case by being promoted to subject of the matrix clause in a manner similar to objects in passive constructions; this analysis is sketched in (50).

The analysis in (50) implies that epistemic modals do not assign an external thematic role. However, they must be able to assign an internal thematic role, which is clear from the fact that the finite complement clause in (49a), or the anticipatory pronoun in subject position introducing it, must be semantically licensed. Given the similarity in meaning between the two constructions in (49), we can also assume that the infinitival clause in (49b) must also be assigned an internal thematic role. If we accept the conclusion from Section 4.6 that being an argument taking predicate is a defining property of main verbs, the conclusion that epistemic modal verbs assign an internal thematic role implies that the traditional view that epistemic modal verbs are non-main verbs cannot be maintained; we should instead consider them as unaccusative main verbs.

This subsection has shown that bare infinitival clauses can occur in at least three types of syntactic environment that affect the way their subject is realized: the subject can be realized as an accusative noun phrase, as the phonetically empty element PRO, or it can be “raised” (i.e. promoted) to subject of the matrix clause and assigned nominative case. What we have not discussed, and what remains a largely unresolved issue, are the syntactic mechanisms that determine the form of the subject of the infinitival clause. For example, why does the modal verb willenwant in (51a) lack the ability of perception verbs to assign accusative case to the subject of their infinitival complement? Is this simply a lexical property of the verbs involved, or are we dealing with different syntactic structures? And why is the subject of the infinitival clause realized as PRO when the modal verb moeten in (51b&c) is deontic, but not when it is epistemic; cf. Klooster (1986)?

Since we have nothing illuminating to say about the first question, we leave it as an unsolved problem for future research. The second question poses a serious problem for the traditional formulation of control theory in Chomsky (1981), which in effect states that traces of movement and PRO cannot occur in the same syntactic configuration. The answer may lie in an appeal to the alternative proposal in Koster (1978: §2), and more specifically in Koster (1984a/1984b), that the difference is a property of the antecedent of the empty category; we will return to this issue briefly in the conclusion of Section 5.2.2.1.