Subsection II begins by briefly reiterating one of the main results from our discussion of (om +) te-infinitivals in Sections 5.2.1 and 5.2.2, which we will take as our starting point: while constructions with om + te-infinitival argument clauses such as (551a) are non-obligatory control constructions categorically, constructions with te-infinitival argument clauses like (551b&c) involve either obligatory control or subject raising, depending on whether or not the verb selecting the infinitival clause also selects an external argument.

Examples like (551a&b) can be distinguished by impersonal passivization of the matrix clause, which is possible in the case of non-obligatory control, but excluded in the case of obligatory control. Examples like (551b&c) can be distinguished by pronominalization, which also affects the nominative subject of the whole construction in the case of subject raising, but not in the case of control; cf. the examples in (552).

The generalization given above suggests that there are two main syntactic types of infinitival complement clauses: om + te-infinitivals, which are CPs and constitute islands for locally restricted syntactic dependencies like obligatory control and subject raising, and te-infinitivals, which are TPs and are transparent to such dependencies. The examples in (550) have further shown that there are reasons to divide the set of te-infinitivals into two subclasses; one type that behaves like om + te-infinitivals, in that they are in extraposed position and do not trigger the IPP-effect in perfect-tense constructions, and a second type that behaves more like bare infinitivals, in that they require verb clustering and do exhibit the IPP-effect. In more traditional terms, we can conclude that the first type is opaque to the movements that derive verb clustering, whereas the second type is transparent to such movements. For convenience, we have again italicized the infinitival clauses and underlined the verbs in our examples in (553).

If we also include the distinction between control and subject-raising constructions discussed in Sections 5.2.2.1 and 5.2.2.2, we arrive at the somewhat unexpected classification in Table (554), in which the split pattern and the IPP-effect are taken as diagnostics for transparency. The problem with this classification is that it does not take into account the previously established fact that control te-infinitivals always involve obligatory control and are therefore expected to be part of the set of transparent infinitival clauses.

		split pattern	ipp-effect
opaque	om + te-infinitivals	—	—
	control te-infinitivals (type A)	—	—
transparent	control te-infinitivals (type B)	+	+
	subject-raising te-infinitivals	+	+
	bare infinitivals	+	+

It has long been assumed that the distinction between transparent (= verb clustering) and opaque (= extraposition) infinitival clauses is exhaustive. Subsection IV will show, however, that there seems to be a third possibility: many (but not all) obligatory control constructions involve what we will call semi-transparent te-infinitivals. The label “semi-transparent” is chosen to express that such infinitivals seem to constitute a hybrid category in that they do not exhibit the IPP-effect, but still allow the split pattern; example (555a) illustrates this for the verb bewerento claim.

Note that a percentage sign has been added to (555a) to indicate that speakers tend to vary in their judgments about the split version; this observation is important because we will see in Subsection VII that it may provide us with a better understanding of the still unexplained fact, noted in Section 5.2.2.2, sub III, that passive subject-raising constructions such as (555b) are also considered marked by many speakers.

The discovery of the third type of semi-transparent te-infinitivals implies that we have not two but three subcategories: opaque, transparent and semi-transparent, respectively. This leads to the revised table in (556), which, however, still does not solve the problem of having to postulate a set of opaque te-infinitivals, even though they clearly involve obligatory control.

		split pattern	ipp-effect
opaque	om + te-infinitivals	—	—
	control te-infinitivals (type A)	—	—
semi-transparent	control te-infinitivals (type B)	+	—
transparent	control te-infinitivals (type C)	+	+
	subject-raising te-infinitivals	+	+
	bare infinitivals	+	+

Subsection V continues by examining the split patterns we find with transparent and semi-transparent te-infinitivals; we will see that these differ in a number of ways, from which we conclude that these patterns are not of the same type. In fact, the split patterns we see with semi-transparent te-infinitivals seem to have more in common with extraposed/opaque te-infinitivals. This raises the question as to whether it is really justified to distinguish semi-transparent from opaque te-infinitivals. This issue will be the topic of Subsection VI, where it will be argued that there is no reason to postulate opaque te-infinitivals: semi-transparent te-infinitivals are arguably derived from the allegedly opaque ones by optional leftward movement of one or more constituents of the te-infinitival into a position preceding the clause-final verbs; this is indicated by the analyses of the two versions of (555a) in (557). In short, the allegedly opaque te-infinitivals simply arise when the optional movement does not take place.

This allows us to draw the tentative conclusion that we can maintain that the transparency of infinitival clauses is closely related to the independently motivated categorial distinction between CP, TP, and VP. This implies that we should replace Table (556) with the simpler one in (558), which is consistent with our earlier conclusion that te-infinitivals are in principle transparent to locally restricted syntactic dependencies; they differ only in that their biclausal structure is still reflected by their ability to be in extraposed position. Subsection VII will provide independent evidence in support of the movement analysis in (557), based on a comparison of the examples in (555a&b).

		split pattern	ipp-effect
opaque (cp)	om + te-infinitivals	excluded	—
semi-transparent (tp)	control te-infinitivals (type A)	optional	—
transparent (tp or vp)	control te-infinitivals (type B)	obligatory	+
	subject-raising te-infinitivals	obligatory	+
	bare infinitivals	obligatory	+

The hypothesis in (558) that te-infinitivals are never opaque may seem at odds with the fact that some of them resist the split pattern. Subsection VIII addresses this problem and shows that this follows from the independently established fact that (semi-)transparency is not an absolute property of clauses, but arises only when a number of additional syntactic conditions are met: for example, they must be internal arguments of the matrix verb and occur as direct objects. This leaves us with one question, which will be briefly addressed in Subsection IX: in what way are control te-infinitivals of type A and type B different? The answer to this question will be somewhat sketchy and certainly needs to be elaborated by future research. Subsection X concludes the discussion by pointing out a more general complication for all research on complement clauses, namely that verbs do not seem to be very particular in their choice of complement clause: some verbs can combine with finite or infinitival clauses, om + te-infinitival or te-infinitivals, transparent or semi-transparent te-infinitivals, and so on. We will discuss the available options for a small sample of verbs.

The discussion of (om +) te-infinitivals in Sections 5.2.1 and 5.2.2 ultimately led to the four hypotheses in (559), in which the notion of syntactic dependency refers to locally restricted phenomena, including NP-movement (such as subject raising), binding of anaphors (such as the reflexive zichzelfhimself), and obligatory control; cf. Section 5.2.2.1, sub IV. Together, these hypotheses express that such dependencies can be established across the boundary of a te-infinitival, but not across the boundary of an om + te-infinitival.

Although the proper formulation of the restrictions on obligatory control are still under debate, we have assumed that they are as given in (560); failure to meet any of the clauses in (560) is sufficient to conclude that we are dealing with non-obligatory control; cf. Section 5.2.1.3, sub III, for discussion.

Infinitival clauses in subject-raising constructions do not have implied PRO-subjects, but involve movement of their subject into the subject position of the matrix clause, where it is realized as a nominative noun phrase. The choice between obligatory control and subject depends on the thematic properties of the matrix verb. Transitive verbs such as bewerento claim in (561a) are only compatible with a control analysis; subject raising is excluded because the regular subject position of the matrix verb is already occupied by the external argument of this verb. Unaccusative verbs such as blijkento turn out, on the other hand, do allow subject raising because the landing site of subject raising is free; on the other hand, PRO is excluded because there is no suitable controller for it and the resulting construction would thus violate restriction (560a) on obligatory control.

The hypotheses in (559) correctly predict that obligatory control and subject-raising constructions cannot be om + te-infinitivals, as is clear from the impossibility of adding the complementizer om to the examples in (561). Conversely, they predict that om + te-infinitivals cannot be used in obligatory control or subject-raising constructions. That om + te-infinitivals do not involve obligatory control is clear from the acceptability contrast between the impersonal passive constructions in (562): example (562a) is allowed because the PRO-subject of the om + te-infinitival is not obligatorily controlled, and consequently restriction (560a) is irrelevant; example (562b), on the other hand, is excluded because the PRO-subject of the te-infinitival should be obligatorily controlled, but cannot find an overt antecedent, which results in a violation of (560a). That the contrast is indeed due to control is supported by the fact that obligatory subject-control verbs such as beweren can be passivized in non-control contexts: cf. Dat wordt vaak beweerdThat is often claimed.

That om + te and te-infinitivals differ in that only the latter can be used in subject-raising constructions is illustrated in (563); example (563a) is excluded because the CP-boundary turns the infinitival clause into an island for movement and thus blocks subject raising of the noun phrase Jan; example (563b), on the other hand, is acceptable because the TP-boundary is not an island for movement and thus allows subject raising.

Note that the four hypotheses in (559) do not yet explain why speakers give varying judgments on an example such as (563b); we return to this issue in Subsection VII.

The discussion of (om +) te-infinitivals in Section 5.2.2.1 was simplified in that it abstracted from a number of issues. For example, the discussion of control constructions was strictly limited to cases with infinitival clauses in extraposed position, i.e. after the matrix verb in clause-final position. Such constructions are characterized by the fact that they do not allow the infinitivus-pro-participio (IPP) effect in the perfect tense; the matrix verb must be realized as a past participle. This is shown in (564a) for a non-obligatory control construction in which the verb proberento try selects an om + te-infinitival, and in (564b) for an obligatory control construction in which the verb bewerento claim selects a te-infinitival; we italicize the infinitival clause and do not indicate the implied PRO-subject for ease of presentation.

Section 5.2.2.2 has shown that subject-raising constructions normally do not allow extraposition, but require that the infinitival clause be split by the matrix verb in clause-final position, i.e. they exhibit verb clustering. Furthermore, subject-raising constructions normally exhibit the IPP-effect for those speakers who allow subject-raising constructions in the perfect tense (which is not the case for all speakers).

Note that the qualification “normally” in the two sentences preceding (565) is necessary because Section 5.2.2.2 has shown that the formal register allows subject-raising constructions with te-infinitivals in extraposed position; such constructions do not exhibit the IPP-effect either. We will return to this issue in Subsection VII.

The differences between examples like (564b) and (565) show that from a syntactic point of view it is not sufficient to distinguish between om + te and te-infinitivals, but that the latter should at least be divided into the two subtypes in (566).

This may seem like a nice result, given that (566) allows us to describe the data discussed so far with the help of the two independently motivated binary parameters in Table (567): the distinction between CP and TP can be motivated by the distribution of the complementizer om, and the distinction between transparent and opaque infinitivals by the behavior of te-infinitivals with respect to clause splitting and IPP. The empty cell may follow from the general claim from the earlier theoretical literature that CP-boundaries block the movements required for deriving the split pattern.

	transparent	opaque
CP (om + te-infinitival)	—	non-obligatory control
TP (te-infinitival)	subject raising	obligatory control

Unfortunately, the hypothesis in Table (567) that obligatory control holds only in opaque te-infinitivals is obviously incorrect. Consider the examples in (568) with the verb durvento dare. The (a)-example shows that durven requires clause splitting; the object of the te-infinitival te vertellen must precede the finite verb durftdares in clause-final position. The (b)-example shows that durven also triggers the IPP-effect; Evers (1975) and Den Besten & Edmondson (1983) both claim that perfect-tense constructions do not allow the past participle gedurfd, and a Google search (October 18, 2023) for the string [heeft gedurfd te] indeed yielded relatively few examples. The (c)-example is added to show that we are indeed dealing with a control structure: pronominalization of the te-infinitival does not affect the nominative subject of the construction as a whole, which therefore cannot originate as part of the infinitival clause, but must originate as a separate argument of the matrix verb durven. Consequently, the external argument of the verb vertellento tell must be realized as PRO.

The examples in (568) thus show that the situation presented in Table (567) is incorrect, since there are also transparent te-infinitivals with obligatory control. We therefore get the more complex situation in Table (554), repeated here as (569), in which the split pattern and the IPP-effect are taken as diagnostics for transparency. The problem with this classification is that it does not take into account the previously established fact that control te-infinitivals always involve obligatory control, and are therefore expected to be part of the set of transparent infinitival clauses.

		split pattern	ipp-effect
opaque	om + te-infinitivals	—	—
	control te-infinitivals (type A)	—	—
transparent	control te-infinitivals (type B)	+	+
	subject-raising te-infinitivals	+	+
	bare infinitivals	+	+

The following subsections will try to resolve this paradox, but before we get to that, it is important to emphasize that the conclusion that te-infinitivals in extraposed position involve obligatory control is fully consistent with the four hypotheses in (559). The fact that control constructions with verbs such as durven in (568) involve clause splitting and IPP shows that they are TPs, which in turn predicts that we are dealing with obligatory control constructions. That this prediction is correct seems to be supported by the fact that (568a) has no passive counterpart; a Google search (October 18, 2023) on the (passive strings) [<gedurfd> werd <gedurfd> te] yielded only a handful of possible cases with intransitive infinitives. In fact, there is reason to dismiss these as irrelevant, since the search strings [<gedurfd> werd <gedurfd> * te] resulted in a small number of cases with an extraposed infinitival clause preceded by the complementizer om; this makes it plausible that the passive cases with intransitive infinitives involve om + te-infinitivals with a phonetically empty complementizer. We therefore tentatively conclude that, as predicted, control constructions of the type in (568a) cannot be passivized.

Subsection III has shown that obligatory control te-infinitivals are traditionally divided into two categories: opaque and transparent infinitivals. Opaque infinitival clauses are extraposed, i.e. they follow the clause-final verbs, while transparent infinitival clauses participate in verb clustering, i.e. they are split by the clause-final verbs. This is again illustrated by the primeless examples in (570), where we have italicized the infinitival clause and underlined the relevant verbs; the primed examples illustrate the concomitant (lack of) IPP in the corresponding perfect-tense constructions.

It has long been assumed that the choice between extraposition and verb clustering is absolute, despite the fact that a large set of verbs seems to allow both options; cf. Bech (1955), Evers (1975), and much subsequent work. The primeless examples in (571) show that bewerento claim is such a verb: it seems compatible with both extraposition and verb clustering. Later research has shown, however, that (570b) and (571b) cannot be treated on a par, since the corresponding primed examples show that durven exhibits the IPP-effect in the perfect tense, whereas beweren does not; cf. Den Besten et al. (1988), Den Besten & Rutten (1989), Rutten (1991), Broekhuis et al. (1995), and the references therein. For simplicity, we refer to this collection of work as the DB/R research group, since much of it was initiated or directed by Hans den Besten and Jean Rutten.

This array of facts leads to the conclusion that besides the transparent and opaque te-infinitivals, a third type of semi-transparent te-infinitivals must be recognized. The postulation of a more hybrid set of te-infinitivals requires a revision of the classification of infinitival clauses in (569) as in (572). However, it still does not solve the problem that we have to postulate a set of opaque te-infinitivals despite the fact that they clearly involve obligatory control.

		split pattern	ipp-effect
opaque	om + te-infinitivals	—	—
	control te-infinitivals (type A)	—	—
semi-transparent	control te-infinitivals (type B)	+	—
transparent	control te-infinitivals (type C)	+	+
	subject-raising te-infinitivals	+	+
	bare infinitivals	+	+

Subsection IV has shown that, according to the traditional diagnostics of clause splitting and IPP, there are three types of te-infinitivals: opaque, semi-transparent, and transparent. Note, however, that it is not the case that verbs always select one particular type of te-infinitival. For instance, the examples in (573) suggest that the verb proberento try is compatible with all three types.

At this point we would like to make two brief remarks. The first is that it has not been established in the literature that the three variants in (573) differ in meaning or information-structural properties (but see Subsection IX); the translation given in (573a) seems adequate for all cases. The second is that the previous subsections have analyzed (573c) as an om + te-infinitival, since it seems to allow the addition of the complementizer om; whether this analysis is indeed an option is not directly relevant for the discussion in this subsection, but we return to this issue in Subsection VII.

At first glance, the examples in (573a&b) seem similar in that clause splitting leads to verb clustering. However, this similarity may only be apparent because the infinitival clauses are small; they contain just one phonetically realized constituent apart from the te-infinitive, the direct object het boekthe book, and clause splitting therefore inevitably leads to clustering, i.e. adjacency of the verbs. To see whether clause splitting leads to verb clustering in the technical sense of an impermeable series of verbs, we should consider te-infinitivals that are larger in size; (574) shows the result for infinitival clauses headed by the ditransitive verb geven. The two (a)-examples show that in the case of transparent te-infinitivals, splitting the te-infinitival inevitably leads to verb clustering: it is impossible to place the direct object een kusa kiss between the verbs. The two (b)-examples, both of which may be considered slightly marked by some Dutch speakers, show that in the case of semi-transparent te-infinitivals the verbs may group together, but that it is also possible to interrupt the sequence of verbs by placing the direct object left-adjacent to the te-infinitive. The contrast in acceptability between the two primed examples thus suggests that transparent and semi-transparent te-infinitivals differ in that only the former involve verb clustering in the technical sense given above.

The same is shown by the examples in (575), where the te-infinitivals contain the phrasal expression in ontvangst nemento take delivery, in which the PP in ontvangst probably functions as a complementive. The (a)-examples show that this PP must precede the matrix verb proberen in the case of transparent te-infinitivals, while the (b)-examples show that it can be placed between the verbs in the case of semi-transparent te-infinitivals; the marked status of (575b) shows that placing the complementive PP before the clause-final verbs is actually disfavored.

The contrast between the primed examples in (574) and (575) shows that clause splitting of transparent and semi-transparent te-infinitivals cannot be considered the result of the same operation. Since clause splitting of transparent te-infinitivals invariably leads to verb clusters in the technical sense, the DB/R research group concluded that clause splitting in this case is the result of a head movement operation traditionally called verb raising. Since clause splitting of semi-transparent te-infinitivals does not necessarily lead to verb clustering, the group concluded that we are dealing with a kind of extraposition.

The DB/R research group argued that semi-transparent te-infinitivals are just like opaque te-infinitivals in that they are in extraposed position, i.e. after the clause-final verbs. Semi-infinitival constructions, however, are special in that at least one of the constituents of the te-infinitival is extracted and moved to a position before the clause-final verbs. The structures in (576) show that on this analysis the only difference between the semi-transparent and opaque te-infinitivals in (573b&c) is whether or not the object dat boekthat book has been extracted from the extraposed clause and placed before the matrix verb geprobeerd.

It is crucial to note that extraction of the direct object is not obligatory in the case of semi-transparent te-infinitivals, as can be seen from the fact illustrated in (574) that verb clustering is not obligatory when the te-infinitive is ditransitive; the analyses of the relevant examples are given in (577a&b). This suggests that “opaque” te-infinitivals as in (577c) are essentially the same as the semi-transparent ones in (577a&b), except that both objects remain inside the te-infinitival clause.

The same applies to other cases where te-infinitival contains a larger number of constituents, as in (575), the analyses of which are given in (578a&b). This again suggests that the “opaque” te-infinitival in (577c) is essentially the same as the semi-transparent ones in (578a&b), except that the direct object also remains within the te-infinitival.

The DB/R research group suggests that the fact that (578a) is often judged as marked compared to (578b&c) can be used to support the movement analysis, since movement of complementives such as in ontvangst usually targets the main-clause initial position (unless they are assigned an emphatic accent); cf. the discussion of (583) for an alternative proposal.

The analysis of the DB/R research group thus suggests that the distinction between opaque and semi-transparent te-infinitivals can be abandoned in favor of the claim that the alleged opacity of te-infinitivals in extraposed position simply follows from the optionality of the movement that derives the split pattern. This is actually predicted by the hypotheses I-IV, repeated here as (579), since these explicitly state that there are no opaque te-infinitivals; only om + te-infinitivals are of this type.

In Subsection VIII we will look at some possible counterexamples to the claim that there are no opaque te-infinitivals, but for the moment we will simply adopt it as an idealization. This makes it possible to replace Table (572) by the simpler one in (580), in which we have also indicated whether the split pattern arises as a result of leftward movement of one or more constituents of the te-infinitival, or as a result of whatever operation is responsible for verb clustering.

		split pattern	ipp-effect
opaque (cp)	om + te-infinitivals	—	—
semi-transparent (tp)	control te-infinitivals (type A)	leftward mvt	—
transparent (tp or vp)	control te-infinitivals (type B)	verb cluster	+
	subject-raising te-infinitivals	verb cluster	+
	bare infinitivals	verb cluster	+

The analysis of the DB/R research group discussed above crucially claims that te-infinitivals are transparent to movement. Since this movement is not possible from om + te-infinitivals, it seems reasonable to conclude that the movement involved is of a locally restricted type (although the discussion below will show that it differs from A-movement of the kind we find in passive and subject-raising constructions in that it can also affect non-nominal phrases). This means that we can now also explain why te and om + te-infinitivals differ in their control properties; the former allow but the latter block the locally restricted syntactic dependency of obligatory control.

The term semi-transparent te-infinitival can be justified by pointing out that such infinitival clauses are transparent to certain locally restricted dependencies, like obligatory control and the movements that derive the split pattern, but not to others, such as the movement(s) involved in deriving verb clustering in the technical sense mentioned above. Note, however, that the DB/R research group also found that extraction from extraposed te-infinitivals is considered marked by some speakers, and that acceptability judgments on the resulting surface forms depend on several factors. Many speakers consider present-perfect examples such as (581b) to be marked, compared to simple-present examples such as (581a), which we have indicated by the percentage sign; note, incidentally, that the placement of the participle before or after the finite verb in clause-final position does not seem to affect the acceptability of (581b). Example (581c) is added to show that this contrast in acceptability cannot be explained by assuming that besluitento decide selects a transparent te-infinitival as its complement, since this would incorrectly predict that substituting an infinitive for the participle in (581b) (i.e. the IPP-effect) would yield a perfectly acceptable result.

The relative markedness of (581b) shows that the term semi-transparent te-infinitival is also quite apt because the boundaries of such clauses seem to be less easily crossed than those of transparent te-infinitivals. Further support for this conclusion comes from the fact that acceptability judgments also depend on the nature of the element being moved. Examples (582a&b) first show again that it is easily possible for most speakers to extract nominal objects from te-infinitivals in extraposed position; while the primeless (b)-example illustrates that extraction of the indirect object is independent of extraction of the direct object, the primed (b)-examples show that (as usual) the direct object cannot be moved across the indirect object. Example (582c) shows that extraction of PP-complements is also possible. The examples in (582) are of course also acceptable if the objects occur after the verb beslootdecided, which we have indicated by angled brackets.

Examples like (583a&b) with an extracted complementive or a particle, on the other hand, are often considered marked, and examples such as (583c) with an adverbial phrase even seem to be excluded for many speakers. The primed examples show that the markedness of the split patterns is even more conspicuous in the corresponding perfect-tense examples. All cases become perfectly acceptable when the complementive, particle or adverbial phrase is placed immediately to the left of the te-infinitive.

The contrast between complements and non-complements in the examples in (582) and (583) is familiar from other island contexts, and again justifies the conclusion that te-infinitivals in extraposed position are not fully transparent to movement. However, since om + te-infinitivals completely block the proposed movements, we conclude that te-infinitivals are not completely opaque either. Therefore, the term semi-transparency seems to be the most appropriate to describe the situation.

Subsection VI has shown that extraction from semi-transparent (i.e. extraposed) te-infinitivals often leads to a slightly marked result. This subsection shows that this is true not only for control constructions, which have been the focus of our attention so far, but also for subject-raising constructions. Recall that Section 5.2.2.2, sub II, has shown that subject-raising constructions are usually transparent in the sense that they involve verb clustering and IPP (for those speakers who allow subject raising in perfect-tense constructions). This is illustrated again in (584).

In the formal register, however, subject-raising constructions often do not have this format; extraposition of te-infinitivals seems to be quite common. We illustrate this in (585a) with the verb blijken, which seems to occur relatively often with extraposed infinitival copular clauses; (585b) shows that such subject-raising constructions do not exhibit the IPP-effect. Of course, the fact that these constructions do not occur in informal speech may be a reason to simply set them aside, but it is still useful to note that they are not unexpected given the claim that semi-transparent te-infinitivals are transparent to certain kinds of movement.

The claim that te-infinitivals in extraposed position are transparent for subject raising also accounts for the existence of the passive subject-raising construction discussed in Section 5.2.2.2, sub III; we repeat the core data in (586). Example (586b) first shows that the PRO-subject of the infinitival complement of beweren must satisfy restriction (560a) on obligatory control; impersonal passivization is excluded, since PRO must have an overtly realized controller. However, example (586c) shows that passivization is possible when the subject of the infinitival clause is raised to the subject position of the matrix clause, where it is assigned nominative case. It is important to note that the markedness of the resulting structure fits well with the observation from Subsection VI that extraction from extraposed te-infinitivals usually leads to a marked result; judgments on examples such as (586c) seem close to those on the perfect-tense constructions in (581b).

The analysis proposed for the subject-raising constructions in (585) and (586c) seems to be consistent with the analysis proposed by the DB/R research group for semi-transparent control constructions discussed in the previous subsections. Unfortunately, there is a complication that needs to be addressed. Consider the set of examples in (587); the question marks indicate that it is not clear a priori whether we are dealing with a CP or a TP.

In our earlier discussion, we accounted for the paradigm by claiming that the complement of proberen is an om + te-infinitival with an empty complementizer. If so, impersonal passivization of the matrix clause, as in (587b), is predicted to be possible because PRO is not obligatorily controlled and can therefore take the implicit agent of the matrix verb as its antecedent. The subject-raising construction in (587c), on the other hand, is ruled out because om + te-infinitivals block extraction.

However, we have seen that proberen may also occur with a semi-transparent te-infinitival. This predicts that PRO is obligatorily controlled, as a result of which impersonal passivization is excluded because PRO cannot take the implicit agent of the matrix verb as its antecedent. Subject raising, on the other hand, is wrongly predicted to be possible given that te-infinitivals do allow extraction.

Of course, it may be the case that proberen can select either a CP or a TP, which would predict that both (588a) and (589b) are grammatical. This raises the question as to why only the former is acceptable. It is clear that this cannot be due to the syntax proper, since both structures can be generated without any problem; for this reason we have not marked the unacceptability of (589b) with an asterisk, but with a dollar sign. Consequently, we have to assume that we are dealing with a kind of syntactic blocking; speakers do not opt for a marked structure when an unmarked one is available. We leave the question as to whether this line of reasoning is tenable to future research. If not, it is clear that the pattern in (587), which is typical of verbs selecting either an om + te or a te-infinitival, poses a problem for the attempt to extend the line of inquiry initiated by the DB/R research group to the patterns discussed in this subsection.

The previous subsections have discussed only direct object clauses, and not without reason, since the syntactic function of te-infinitivals seems to have an effect on their transparency to movement. Evers (1975:39ff) and Den Besten et al. (1988), for example, have found that te-infinitivals functioning as subject, nominal part of a PP-complement, or logical subject of a particle (i.e. a complementive) categorically resist clause splitting; cf. the primeless examples in (590). The corresponding perfect-tense constructions in the primed examples also show that constructions with such argument clauses do not exhibit the IPP-effect.

However, it seems unwise to take the examples in (590) as counterexamples to our classification in Table (580). To show this, we will compare the examples in (590) with similar examples involving wh-movement, but first note that wh-extraction is possible from the direct object clauses in (591).

This shows that finite direct object clauses are transparent to wh-movement (though opaque to other types of movement); cf. Section 11.3.1. However, the examples in (592) show that not all finite argument clauses allow this kind of extraction. Note that we used the same type of verbs as in (590) in order to keep the examples as parallel as possible.

The fact that we find the same effects in (590) and (592) for different kinds of movement suggests that “transparency of a phrase of type P for movement type M” is not an absolute property of P; there may be other factors involved. This implies, for example, that being a te-infinitival is a necessary but not a sufficient property for exhibiting clause splitting. The opacity of the clause types in (590) and (592) is a well-known fact from the formal linguistic literature, and it is often assumed that transparency of a phrase P requires (at least) the additional conditions in (593a&b) to be satisfied.

Let us now consider how (593) accounts for the opacity of the embedded clauses in (590) and (592). The (a)-examples, in which the clauses perhaps function as the subject of a nom-dat verb, satisfy (593a) if we analyze them as so-called nom-dat verbs (cf. Section 2.1.3), but they clearly violate (593b). Note that we used the modifier perhaps in the previous sentence because the examples contain the anticipatory pronoun hetit, which may function as the true syntactic argument of the verb, the extraposed “subject” clause being merely a right-dislocated adjunct that provides further information as a kind of afterthought (cf. Section 12.1, sub IB). That this may be the correct way to look at these examples is suggested by the fact that direct object clauses also become opaque when they are introduced by an anticipatory pronoun. The examples in (594) illustrate this for clause splitting and wh-movement from a complement clause of the verb bewerento claim. Although the use of the anticipatory pronoun het is marked in the sense that it is restricted to specific (e.g. factive) contexts, it is clear that the insertion of the pronoun has a dramatically negative effect on the acceptability of clause splitting and wh-extraction. Note that alternative placements of the anticipatory pronoun (e.g. in a position following the moved phrase dat boek) do not improve clause splitting in (594a).

In (595) we show essentially the same for the verb proberen, but since this verb does not take finite complement clauses, we illustrate the restriction on wh-movement by extraction from an infinitival clause.

The embedded clauses in the (b)-examples in (590) and (592) clearly violate both clauses in (593), and in addition they involve the anticipatory pronouns erop/ertoe; as expected, both clause splitting and wh-movement are impossible. The embedded clauses in the (c)-examples in (590) and (592) seem to satisfy clause (593b), but they are not internal arguments of the matrix verbs because they are introduced as logical subjects of the particles; both clause splitting and wh-movement do indeed produce degraded results, although many speakers report that they consider the deviation to be less severe than in the (a) and (b)-examples.

Finally, we take another look at the formulation of the additional restrictions on the transparency of argument clauses, in particular at clause (593b), which claims that the argument clause must not only be an internal argument of the verb, but must also function as a direct object. It is easy to check the validity of this claim by passivization of obligatory object-control constructions. The examples in (596) first show that such constructions can be passivized, since passivization does not affect the control relation; PRO has an overt controller in both constructions, the indirect object Marie/haarher.

Example (597a) further shows that obligatory object-control constructions allow clause splitting: that we are dealing with a semi-transparent construction can be seen from the fact that the perfect-tense construction does not exhibit the IPP-effect. The question mark between parentheses indicates that speakers tend to evaluate clause splitting as less acceptable with double object verbs.

The crucial question is whether clause splitting is possible in the corresponding passive construction in (598a). Broekhuis et al. (1995:113) has claimed that passive constructions of this kind are grammatical, and it is certainly true that they are much better than examples such as (590a). Regardless of whether (598a) is grammatical or whether we should assign it some intermediate status, we need to answer the question of what causes the observed difference in acceptability between the two examples. The most striking difference is that while (590a) is introduced by the anticipatory pronoun hetit, (598a) is not; in fact, (598b) shows that adding an anticipatory pronoun to (598a) also makes clause splitting impossible.

This suggests that the difference in acceptability between (590a) and (598a) reflects the so-called argument/adjunct asymmetry; adjunct clauses are always islands for syntactic dependencies, whereas argument clauses are not. On the crucial assumption that “argument” clauses introduced by an anticipatory pronoun are a special kind of adjunct clause, we can simplify (593) for te-infinitivals by eliminating the clause in (593b) about syntactic function, as in (599); presumably the same simplification can be made for other clause types.

It is important to note that we do not want (599) to imply that there are no additional restrictions on the (semi-)transparency of te-infinitivals; for example, it may be necessary for the infinitival clauses to be in their base position in order to avoid the so-called freezing effect. We leave the identification of such additional restrictions to future research.

The discussion in Subsection VIII has made it clear that the distinction between non-transparent and semi-transparent te-infinitivals not only reflects the categorial distinction between CP and TP, but also depends on properties of the wider syntactic context in which these clauses are used, such as their syntactic function or the presence of an anticipatory pronominal element. Thus, we are left with the distinction between semi-transparent and transparent te-infinitivals, which we called Type A and B in Table (580). If we want to maintain our earlier conclusion that these are of the same category, TP, we have to ask in what respect they are different. Bennis & Hoekstra (1989c:141ff) make the more general claim that infinitival clauses allowing verb raising (in traditional terms) differ from extraposed infinitival clauses in that the temporal interpretation of the former, but not the latter, depends on the temporal interpretation of the matrix clause. Applied more specifically to te-infinitivals of type A and B, this implies that the former, but not the latter, constitute an independent temporal domain. That something like this is indeed the case is shown by the examples in (600a&b), adapted from Pardoen (1986), which show that independent temporal modification of te-infinitivals is only possible when the infinitival clause is in extraposed position. Whether independent temporal modification is also possible with split te-infinitivals is not discussed in the literature and is also difficult to determine with certainty, since examples such as (600a') are clearly marked in comparison to (600a).

However, since the marked status of (600a') may be due to the fact that splitting the nonverbal constituents of the infinitival clause always yields a somewhat marked result, we can perhaps assume that (600a') is grammatical. If so, we must conclude that split te-infinitivals can receive an independent temporal interpretation, just like their non-split counterparts.

We provide a second set of examples with the verb weigerento refuse in (601), which produce more or less the same acceptability judgments. Note in passing that some speakers may also feel less comfortable with example (601b) even without the adverbial phrase morgen; a Google search on [heeft weigeren/geweigerd te] reveals that such examples with the infinitive weigeren (i.e. with the IPP-effect) are far less common than examples the past participle geweigerd (i.e. without the IPP-effect). However, since a similar minimal pair is given independently in Haeseryn et al. (1997:1037), we assume that such examples are acceptable to many speakers.

We should keep in mind, however, that our evaluation of the acceptability of the primed examples in (600) and (601) is not very sharp, and that our tentative claim regarding their grammaticality must therefore be treated with caution; it is clearly necessary to examine more closely the similarities and differences between the transparent and semi-transparent te-infinitivals before drawing any firm conclusions about their syntactic status and/or semantic interpretation.

The previous subsections have shown that we need to distinguish between at least the three types of (om +) te-infinitivals in Table (580), repeated here as (602). This table omits semi-transparent infinitivals in subject-raising constructions, since these seem to be restricted to the formal register (as relics of earlier stages of the language).

		split pattern	ipp-effect
opaque (cp)	om + te-infinitivals	—	—
semi-transparent (tp)	control te-infinitivals (type A)	leftward mvt	—
transparent (tp or vp)	control te-infinitivals (type B)	verb cluster	+
	subject-raising te-infinitivals	verb cluster	+
	bare infinitivals	verb cluster	+

Regarding the opposition between extraposition and verb clustering, it seems that the opaque and semi-transparent infinitival clauses must be opposed to the transparent ones in that they must be in extraposed position. The IPP-effect can be used as a diagnostic in the sense that the IPP-effect is both necessary and sufficient to conclude that we are dealing with verb clustering; it is never found in extraposition constructions. The presence of the complementizer om can be used to distinguish between opaque and (semi-)transparent infinitivals: the presence of om is sufficient but not necessary to conclude that we are dealing with opaque infinitival clauses. Finally, recall from Subsection VIII that semi-transparent te-infinitivals allow the split pattern only if they satisfy a number of additional conditions: for example, they should be at least internal arguments of the matrix verb. The same condition also applies to verb clustering.

Many verbs are compatible with more than one type of infinitival clause. A good example is the verb proberento try, which has featured prominently in the previous subsections. The examples in (603) show that this verb is compatible with opaque, semi-transparent and transparent te-infinitivals.

The fact that these three examples are virtually equivalent (apart from the differences discussed in Subsection IX) raises the question as to why the examples in (603) can occur side by side in the current language. One answer might be that this is due to the diachronic development of the language, to the extent that there is a tendency to make infinitival complements smaller, perhaps as a side effect of a tendency for matrix verbs to lose their independent status as main verbs. This has been argued explicitly for the modal (subject-raising) verbs blijkento turn out, lijkento appear and schijnento seem, which can still take extraposed (hence semi-transparent) clauses in the formal register, but not in colloquial speech; cf. Haegeman (2006) and Vliegen (2010) for discussion. The same line of thought might also be supported by the fact that some semi-aspectual verbs, such as zittento sit in (604), seem to take te-infinitivals in the simple present/past tense, but (preferably) bare infinitivals in the perfect tense; cf. Section 6.3.1 for more discussion. The infinitival clauses in (604) have been italicized, and the number sign # indicates that the use of te in (604b) is not impossible, but restricted to certain contexts; the relevant point here is the contrast between the two examples without te.

The diachronic development suggested above seems quite plausible for “semi-lexical” verbs like modal blijken, lijken and schijnen as well as aspectual zitten, but we leave it open whether it can also be successfully applied to lexical verbs such as proberento try.

We conclude this section by providing the selection restrictions of the sample of verbs in Table 2. The data set has been garnered from Evers (1975), Den Besten et al. (1988), Den Besten & Rutten (1989), Rutten (1991), Van Haaften (1991), Broekhuis et al. (1995), Ter Beek (2008), and Zwart (2011). A number of things need to be said about the information in this table. First, the classifications in the above works were developed at different stages in the development of the theory and/or for different purposes, and therefore cannot always be directly applied to Table 2. Second, there seems to be considerable variation between speakers: it may therefore be the case that some Dutch readers find that they allow fewer or more options than indicated in the table. In this context, we should also point out that in some cases we did not strictly follow the judgments given in the publications, because the authors sometimes gave contradictory judgments or sharpened their views over time (the latter is especially true for the views expressed in the various publications of the DB/R research group). Occasionally, we had to supplement the data with our own observations or information from Google searches. Third, we restrict ourselves to (di)transitive verbs with object clauses, since we have seen that semi-transparent te-infinitivals usually do not exhibit the split pattern when they have the syntactic function of subject, correspond to complements of prepositional objects, or function as logical subjects of complementives (including verbal particles, which we have seen to give rise to an intermediate status). We also exclude verbs that obligatorily introduce their object clause with the anticipatory pronoun hetit, but include subject-raising verbs. Fourth, speakers often prefer one of the indicated options; we have not indicated this because the literature does not provide reliable information about it. Finally, we do not include bare infinitives in our inventory, since they are always transparent; we refer the reader to Section 5.2.3 for discussion.

Table 2 shows that it is possible to group the verbs into several classes according to whether they select opaque, semi-transparent and/or transparent te-infinitivals, but it is not clear whether it is also possible to find a semantic correlate of this formal classification. Classes I and II have in common that the verbs are able to select both opaque and semi-transparent infinitivals. It does not seem possible to distinguish the two classes semantically, since they both contain control verbs of different types, and they both contain implicative verbs (which assert or deny the proposition expressed by the te-infinitival) and irrealis verbs (which leave open whether the proposition expressed by the te-infinitival is/will be true). Note that it is difficult to classify vergetento forget, since the past participle and infinitive of this verb are homophonous. Class III consists of verbs that select only semi-transparent te-infinitivals and are often propositional in nature. When these verbs imply a specific mode of expression, such as fluisterento whisper, they do not allow the split pattern; this again shows that there may be additional restrictions on the transparency of te-infinitivals. Note also that the verb denken only allows a transparent complement when it is used with the same meaning as menento suppose. Verbs that select only transparent te-infinitivals fall into two groups, given here as IV and V. Class IV consists of subject-raising verbs, all of which are modal in nature. The members of class V are obligatory control verbs, but also seem to express some kind of modal meaning.

	Verb	translation	opaque	semi-transparent	transparent
I	beginnen	to start	+	+	+
	durven	to dare	+	+	+
	helpen	to help	+	+	+
	hopen	to hope	+	+	+
	leren	to learn/teach	+	+	+
	pogen	to try	+	+	+
	proberen	to try	+	+	+
	trachten	to try	+	+	+
	wagen	to dare	+	+	+
	weigeren	to refuse	+	+	+
	wensen	to wish	+	+	+
II	adviseren	to advise	+	+	—
	begeren	to desire	+	+	—
	beloven [+agens]	to promise	+	+	—
	beogen	to aim at	+	+	—
	besluiten	to decide	+	+	—
	bevelen	to order	+	+	—
	dreigen [+agens]	to threaten	+	+	—
	dwingen	to force	+	+	—
	eisen	to demand	+	+	—
	gebieden	to command	+	+	—
	verbieden	to forbid	+	+	—
	vergeten	to forget	+	+	?
	vermijden	to avoid	+	+	—
	verlangen	to desire	+	+	—
	verleren	to forget how to	+	+	—
	verplichten	to oblige	+	+	—
	verzoeken	to request	+	+	—
	verwachten	to expect	+	+	—
	verzuimen	to fail	+	+	—
	vragen	to ask/request	+	+	—
	vrezen	to fear	+	+	—
III	antwoorden	to answer	—	+ (no split)	—
	beweren	to claim	—	+	—
	denken	to think	—	+	—/+
	fluisteren	to whisper	—	+ (no split)	—
	garanderen	to guarantee	—	+	—
	geloven	to believe	—	+	—
	menen	to suppose	—	+	+
	schreeuwen	to yell	—	+ (no split)	—
	schrijven	to write	—	+ (no split)	—
	verklaren	to state	—	+	—
	vertellen	to tell	—	+	—
	verwijten	to blame	—	+	—
	verzekeren	to guarantee	—	+	—
	zeggen	to say	—	+	—
IV	beloven [-agens]	to promise	—	—	+
	blijken	to turn out	—	—	+
	dreigen [-agens]	to threaten	—	—	+
	lijken	to seem	—	—	+
	schijnen	to appear	—	—	+
V	behoren/horen	ought to	—	—	+
	dienen	to have to	—	—	+
	(niet) hoeven	need not	—	—	+
	plegen	to be used to	—	—	+
	weten	to know how	—	—	+