The examples in examples in (59) and (60) show that adding a perfect auxiliary to the superior main verbs in (57) and (58) seems to be easily possible, with the exception of the subject-raising verb schijnen: most people consider examples such as (59b) to be at least marked. All examples exhibit the infinitivus-pro-participio (IPP) effect, which is of course expected, since we have seen that this is a feature of verb clustering; cf. Section 7.1.1. For convenience, we have underlined the added non-main verbs in the following examples.

At first glance, it seems that control and subject-raising verbs differ in whether they can take a perfect te-infinitival as complement: while (61b) is impeccable, example (61a) seems unacceptable, or at least very marked.

However, there is reason to think that the unacceptability of (61a) is not due to some syntactic restriction imposed by proberen, but has to do with the fact that proberen triggers an irrealis reading of its complement: the eventuality expressed by the te-infinitival must be located in the non-actualized part of the time interval evoked by the present/past tense of the matrix clause; in the present tense, the eventuality expressed by the infinitival clause is located after speech time. This seems to clash with the default reading of the perfect, which locates the completed eventuality in the actualized part of the relevant tense interval. For instance, the present perfect example (62a) by default locates the eventuality before speech time; it normally expresses that Jan has finished the book at speech time. However, this default reading of the perfect is pragmatic in nature and can easily be overridden by the addition of an adverbial phrase such as morgentomorrow that refers to a time interval in the non-actualized part of the present-tense interval; example (62b) locates the completion of the eventuality of reading the book after speech time; cf. Section 1.5.4 for a detailed discussion.

This suggests that the default reading of the perfect tense in examples (61a) makes the assertion expressed incoherent. This is supported by the fact that this example becomes perfectly acceptable when we add the adverb morgentomorrow, as in (63): this provides additional temporal information that overrides the default reading of the perfect, as a result of which the eventuality expressed by the infinitival clause can be located in the non-actualized part of the present domain, and the message becomes fully coherent. We can conclude that the unacceptability of (61a) is not due to any syntactic or semantic (selection) restriction, but is simply an effect of pragmatics.

Note in passing that we cannot appeal to the IPP-effect to establish that we are indeed dealing with a verb cluster of three verbs in examples such as (63), since it is impossible to add a second perfect auxiliary associated with the superior verb proberen: cf. *dat Jan dat boek morgen gelezen heeft proberen/geprobeerd te hebben. However, it seems very unlikely that (63) can be analyzed as a remnant extraposition construction: under such an analysis, the fact that the participle gelezen precedes the verb proberen can only be derived if gelezen is extracted from the verb cluster gelezen te hebben of the extraposed te-infinitival clause, but such movements have not been attested (or even considered as an option) in the existing literature. Note, however, that we have found a small number of cases on the internet such as gehoord/gezien beweert te hebbenclaims to have heard/seen, despite the fact that there is strong evidence to assume that bewerento claim usually triggers (remnant) extraposition, not verb clustering; we will ignore this problem here, and leave the question of the status of these cases to future research.

We will now consider verb sequences with modal verbs, i.e. Modal1–Main2. Example (58a) in Subsection 1 has already shown that the perfect auxiliary can easily be added to the superior modal verb in such clusters; the relevant example is repeated here as (64a). Example (64b) shows that it is also possible to add a perfect auxiliary to the embedded main verb.

The two examples exhibit a striking difference in meaning: whereas the modal in (64a) receives a (directed) deontic “obligation” reading, the modal in (64b) receives an epistemic “necessity” reading; cf. Section 5.2.3.2, sub III, for a discussion of these types of modality. This contrast can also be demonstrated by the fact, illustrated in (65), that the hierarchical order Perf1–Modal2–Main3 has the property of deontic constructions that the subject of the sentence must be able to control (e.g. affect) the eventuality expressed by Main3, whereas the hierarchical order Modal1–Perf2–Main3 does not this property.

The contrast in meaning between the examples in (64) is not really surprising, since epistemic modal verbs usually occur as finite verbs, with two exceptions. First, epistemic modals can occur as infinitives when they are part of an infinitival complement clause; this is illustrated in (66a) by the te-infinitival counterpart of example (64b), adapted from Felbr (2022). This case would follow directly if we assume that epistemic verbs are usually the highest verb in the verbal sequence; then it appears as a finite verb in finite but not in infinitival clauses. Note in passing that this assumption correctly predicts that in example (66b), the two modals differ in that the higher (finite) modal must be epistemic while the lower is deontic.

Second, there is a case in which epistemic modals can be governed by a past-tense form of hebben (and, as we will see later, a limited number of other verbs such as zullen). This is shown in the examples in (67), which at first glance appear to be the past-perfect counterparts of the present-perfect examples in (64a) and (65a), but the auxiliaries here function as irrealis markers, which is always the most superior (i.e. finite) verb in the sequence. Given its distribution and semantic contribution, it seems better to classify this auxiliary as a category in its own right rather than as a perfect auxiliary. For further discussion, see Section 1.4.3, sub II.

Now that we have seen that examples such as (64), where the modals governed by a perfect auxiliary are always deontic, we will return to the interpretation of examples such as (64b), in which the modals govern the perfect auxiliary. Section 5.2.3.2, sub IIIC, has suggested that the epistemic reading of this example is favored by the default reading of the perfect tense, namely that the completed eventuality is located in the actualized part of the present-tense interval (i.e. before speech time). This would explain the otherwise surprising fact that a deontic interpretation of the modal is perfectly acceptable in (68), where the default reading of the perfect tense is overridden by the addition of the adverb morgentomorrow, since this adverb explicitly locates the eventuality in the non-actualized part of the present-tense interval.

Deontic modals can also be superior to the perfect auxiliary in irrealis contexts. Such contexts are found in the infinitival complement clauses of control verbs such as verwachtento expect, as in (69a), and also in examples with an irrealis marker such as the past tense form zou.

The reader is referred to Felbr (2022: §6) for examples showing that deontic modals can more generally be superior to perfect auxiliaries in infinitival clauses, but the above is sufficient for our main conclusion that for deontic modals there seems to be no syntactic restriction preventing the extension of the cluster Modal1–Main2 by a perfect auxiliary associated with either Modal1 or Main2; the hierarchical orders Aux1-Modal2-Main3 and Modal1-Aux2-Main3 are both possible with a deontic reading, although there seem to be additional non-syntactic constraints on the first order. Deontic modals differ in this respect from epistemic modals, which are typically superior to the perfect auxiliary (and all other verbs except irrealis markers like had and zou, which must be marked past tense); cf. Section 1.4.3, sub II.

We conclude with constructions containing a perception or causative verb. The examples (58b&c) in Subsection C1 have shown that perfect auxiliaries can easily be added to these verbs, but it seems impossible to add them to the embedded main verb; examples such as (70) are infelicitous.

The use of the dollar sign indicates that, again, it is not a priori clear whether the unacceptability of these examples is due to a syntactic, a semantic or a pragmatic constraint. We think that there is reason to suspect a pragmatic constraint. In the case of (70a), the reason is that examples such as dat Jan haar dat boek ziet lezenthat Jan sees her read that book express a notion of simultaneity: the eventuality of seeing occurs simultaneously with the eventuality expressed by the embedded bare infinitival, and the default reading of the simple present locates these eventualities at speech time. This seems to clash with the default reading of the perfect tense in examples such as (70a), which locates the completed eventuality expressed by the infinitival complement in the actualized part of the present-tense interval, i.e. before speech time.

In its causative interpretation, the construction in (70b) is an irrealis construction in the sense that the eventuality expressed by the embedded bare infinitival is located after speech time, which again clashes with the default interpretation of the perfect, which locates the completed eventuality before speech time. In its permissive interpretation, the eventuality expressed by the embedded bare infinitival is located either at or after speech time, and this again clashes with the default interpretation of the perfect. Note, however, that the addition of an adverb such as morgentomorrow does not seem to improve the result: ??dat Jan haar morgen dat boek laat hebben gelezen, perhaps because this construction is blocked by the simpler construction dat Jan haar morgen dat boek laat lezenthat Jan will make her read that book tomorrow. We will not pursue this issue any further.

The main conclusion of this subsection is that there is no reason to assume a syntactic restrictions prohibiting the selection of a perfect infinitival construction by the superior main verbs in (57) and (58). In some cases this leads to less felicitous results, but this seems to be due to semantic/pragmatic reasons. As expected, the matrix verb can usually also be combined with a perfect auxiliary (with the exception of the epistemic modal verb).

Passivization of the superior main verbs in the verb-clustering constructions in (57) and (58) from Section 7.0 is categorically impossible. We will first discuss the examples in (71) with the subject-control verb proberento try. The primeless examples show again that proberen can select either an opaque te-infinitival, which leads to extraposition, or a transparent te-infinitival, which leads to verb clustering. The primed examples show that passivization of the matrix verb is only possible when the infinitival complement is opaque/extraposed; replacing the participle geprobeerd with the infinitive proberen or changing the order of the verbs in the cluster (or a combination of the two) does not improve the status of (71b').

The fact that subject-control verbs such as proberento try resist passivization only when they are part of a verb cluster suggests that we are dealing with a syntactic constraint. Koster (1984b) proposed an account in terms of the notions of obligatory and optional control introduced in Section 5.2.1.3, sub III. Section 5.2.2.1 has shown that the implicit PRO-subjects of opaque infinitival clauses are optionally controlled; they do not require an antecedent in the matrix clause, so that passivization of the matrix clause (i.e. demotion of the subject to adjunct status) is possible. On the other hand, the implicit PRO-subjects of transparent infinitival clauses are obligatorily controlled: they require an antecedent in the matrix clause, so that passivization of the matrix clause leads to unacceptability. For completeness, note that (71b) can also be analyzed as a remnant extraposition construction, i.e. with a semi-transparent te-infinitival; this does not affect the argument, because Section 5.2.2.3 has shown that the PRO-subjects of such infinitival clauses are also obligatorily controlled.

The fact that subject-raising verbs such as schijnento seem cannot be passivized is expected; Section 5.2.2.2, sub IC, has shown that such verbs are unaccusative, and since unaccusative verbs always resist passivization, there is nothing special to discuss here.

The examples in (72) show that main verbs selecting a bare infinitival clause cannot be passivized, regardless of whether the passivized verb appears as an infinitive or a participle. We give the examples with clusters in the order Pass1–Main2–Main3; changing this order does not improve the acceptability judgments.

As constructions with perception and causative verbs are often analyzed as heads of AcI-constructions, the unacceptability of the (impersonal) passive constructions in (72b&c) is not surprising, since we expect passivization of the matrix verb to promote the subject of the bare infinitival clause (here the object pronoun haarher) to the subject of the matrix clause. However, the examples in (73) with the nominative pronoun zij instead of haar are also unacceptable.

The fact that perception and causative verbs are not usually analyzed as control verbs (but see Petter 1998: §4 for an alternative view) suggests that the unacceptability of passivization in (72) and (73) cannot be explained by an appeal to control theory. Since there is also no obvious semantic/pragmatic reason for the impossibility of passivization, it seems that we have to explain the unacceptability of these examples in terms of verb clustering, but how is not clear; cf. Bennis & Hoekstra (1989b) for an attempt of this kind.

This subsection discusses passivization of the more deeply embedded verbs in the examples in (57) and (58) from Section 7.0. We start with the case in (74), in which the superior verb is the subject-control verb proberento try; (74a') involves regular passivization and (74b') krijgen-passivization. The reader can easily identify the two main verbs by keeping in mind that the control verb is the finite verb while the embedded main verb (i.e. Main3) appears as a passive participle. For now, ignore the use of the dollar sign, which we will discuss shortly.

The fact that the krijgen-passive can easily be embedded under proberen in (74b') suggests that the unacceptability of (74a') has little to do with syntactic selection restrictions imposed by the verb proberen. Instead, it is probably due to the fact that we are dealing with an obligatory subject-control construction; the animate subject of proberen is simply not a suitable antecedent for the implied PRO-subject of the infinitival passive construction (which corresponds to the inanimate object het boek in the active construction in (74a)): #Jani probeert [PROi gelezen te worden]. Example (74b') is perfectly acceptable, of course, since the subject of the infinitival passive construction is a suitable antecedent for the implied PRO-subject of the infinitival passive construction (which corresponds to the animate goal/experiencer argument in (74b)): Jani probeert [PROi dat boek voorgelezen te krijgen].

This account of the unacceptability of (74a') in terms of obligatory subject control is supported by the fact that structurally parallel examples are perfectly acceptable when subject control leads to a result compatible with the selection restriction imposed by the passive verb on its internal arguments. This is illustrated by the primeless examples in (75), where the animate subject of proberen is a suitable antecedent for the PRO-subjects of the infinitival passive constructions: Jani probeert [PROi ontslagen/verkozen te worden]. The primed examples are added to show that we get the IPP-effect in the perfect tense, from which we can conclude that we are indeed dealing with verb clustering; the same is of course supported by the fact that the passive participles of the infinitival clauses precede the matrix verb proberen.

The acceptability of the examples in (75) is the reason why we have used a dollar sign in (74a'): some people might accept this example in a reading in which Jan is trying to get other people to read the work he has written; i.e. with the proper name referring to a body of work, as in Louis Couperus wordt nog steeds gelezenLouis Couperus is still being read.

The examples in (76) show that subject-raising verbs such as schijnen are quite capable of taking a passivized te-infinitival: (76a) involves a regular passive and (76b) a krijgen-passive. The most deeply embedded main verb (i.e. Main3) again appears as a participle, while the subject-raising verb is the finite verb.

The examples in (77) show that modal, perception and causative verbs are able to select a passivized bare infinitival. The acceptability of the results sometimes depends on the embedded main verb, which is why we did not use the verb lezen but the main verb slopento demolish in (77b&c).

A percentage sign is added to (77c) because some speakers object to this example, and the same seems to hold to a lesser extent for (77b). However, there is reason to think that this has nothing to do with a selection restriction imposed by the causative or perception verb. Instead, it seems to be related to the fact that there is an alternative way of expressing the passive meaning in the case of AcI-constructions; the subjects of the bare infinitival clause can simply be omitted or replaced by agentive door-PPs, as illustrated in (78).

One possible explanation for the markedness of (77c) would be an appeal to syntactic blocking; for one reason or another, speakers may value the structure in (78b') more highly (e.g. as more economical or simpler) than the one in (77c). If so, we can conclude our discussion by saying that the superior main verbs in (57) and (58) do not impose any restrictions on the voice of their infinitival complement.

The examples in (79) show that it is possible to add a (semi-)aspectual verb on top of control structures such as (57a). Some people may find example (79b) a little marked as given, but it becomes perfectly acceptable if we add an adverbial phrase of duration like al de hele dagalready the whole day: cf. dat Jan dat boek al de hele dag zit te proberen te lezen.

A problem with the examples in (79) is that we cannot prove that we are dealing with verb clusters of the type Asp1-Main2-Main3. The reason is that perfect auxiliaries must be superior to the (semi-)aspectual verbs; cf. example (56) in Subsection B. This means that we can make sequences of the form Perf1–Asp2–Control3-Main4, as in (80), which shows that the (semi-)aspectual and the control verb are indeed part of a verb cluster, because the IPP-effect applies to gaan and zitten. However, sequences of the form Asp1–Perf2–Control3–Main4 do not occur, which makes it impossible to show whether the IPP-effect applies to the control verb proberen, which would be necessary to determine that proberen and te lezen are also part of a verb cluster.

Since the examples in (81) must be analyzed as extraposition constructions (with the extraposed clause underlined), we must leave open the possibility that the examples in (79) are not verb-clustering but remnant-extraposition constructions.

Nevertheless, in the absence of solid reasons for claiming that the examples in (79) cannot be analyzed as verb-clustering constructions, we will provisionally assume that it is a possible analysis (besides the remnant extraposition analysis).

The examples in (82) show that the addition of a (semi-)aspectual verb on top of subject-raising constructions such as (57b) is impossible. The infelicity of these examples is clearly related to the fact noted earlier that subject-raising verbs do not easily appear as non-finite verbs, but a more insightful statement might be that we are dealing with a semantic restriction in that the evidential modality expressed by schijnen: the infinitival clause simply does not satisfy the semantic selection restriction of (semi-)aspectual verbs that their complements refer to an activity controlled by the subject of the clause (cf. Section 6.3.1, sub II).

The examples in (83) show that adding a (semi-)aspectual verb on top of modal structures such as (58a) leads to degraded results: examples such as (83a) do occasionally appear on the internet, but we think that these are incorrect forms, used instead of the perfectly acceptable form moet gaan lezenmust go read (although we cannot rule out that this is a non-standard usage). Again, it seems plausible that the unacceptability of the examples in (83) is due to the fact that the modal phrases do not satisfy the semantic selection restriction of (semi-)aspectual verbs that their complement refers to an activity controlled by an agent.

The results are different for AcI-constructions with the perception verb zien and the causative verb laten. The examples in (84) show that it is possible to add the aspectual verb gaan to the examples in (58b&c). We have added the particles nog and wel in (84a) in order to facilitate the intended posterior-to-speech-time reading; without these particles some speakers may have problems with this construction.

The examples in (85), on the other hand, show that it is not possible to add the semi-aspectual verb zitten to the examples in (58b&c).

The acceptability of the examples in (84) strongly suggests that the unacceptability of the examples in (85) is not due to some syntactic constraint, because the semi-aspectual verb zitten can usually be embedded under aspectual gaan. It is therefore more likely that the unacceptability of (85) is due to some semantic incompatibility between the semi-aspectual verbs and the verbs zien and laten This can be independently supported for the verb zien by the contrast between the two examples in (86), in which zien takes a nominal object.

The acceptability contrast indicated in (86) is confirmed by a Google search (November 28, 2023): while the (colloquial) question in (86a) occurs about ten times on the internet, the question in (86b) does not occur at all. We cannot provide similar evidence for causative laten as this verb does not allow nominal complements.

In general, it seems possible to add a (semi-)aspectual verb to the structurally lower main verbs of the examples in (57) and (58) from Section 7.0, although we will see that there are certain complications that deserve attention. The examples in (87) show that while the addition of an aspectual verb such as gaan is perfectly acceptable, the addition of a semi-aspectual verb such as zitten leads to a degraded result.

The acceptability of example (87a) strongly suggests that the unacceptability of example (87b) cannot be due to some syntactic constraint, because semi-aspectual verbs such as zitten can normally be embedded under aspectual verbs such as gaan, but that there must be some semantic or pragmatic reason for this. This is quite plausible: the fact that semi-aspectual zitten locates the eventuality expressed by the infinitival clause dat boek lezento read that book in a temporal interval that includes speech time clashes with the fact that the verb proberen triggers an irrealis reading, i.e. locates the eventuality expressed by the infinitival clause after speech time.

The acceptability of the examples in (88) shows that it is easily possible to add a (semi-)aspectual verb to the structurally lower main verbs in subject-raising contexts. Nothing special needs to be said here.

The examples in (89) show that the (semi-)aspectual verbs can also occur embedded under the modal verb moeten. The translations suggest that the modal can be interpreted either as an epistemic or as a directed deontic modal, but the judgments do not seem to be very sharp.

That the epistemic reading is possible seems clear, and it can also be supported by the fact that it is the most conspicuous reading of the examples in (90), in which the (semi-)aspectual verb is preceded by an additional perfect auxiliary. The representation of the clusters in (90) is Modal1–Perf2–AsP3/Semi-asp3–Main4.

The deontic reading of the examples in (89) seems less prominent, which is also suggested by the fact that the examples in (91), where the modal verb is preceded by a perfect auxiliary, are marked. We leave the status of clusters of the type Perf1–Modal2–Asp3/Semi-asp3–Main4 as an issue to future research.

The acceptability of the verb clusters in (92) with the perception verb ziento see seems to depend on tense marking. That example (92a) is somewhat strange may be related to the fact that AcI-constructions with perception verbs normally express a notion of simultaneity; an example such as dat Jan haar dat boek ziet lezenthat Jan sees her read that book expresses that the eventuality of seeing is simultaneous with the eventuality of her reading the book, expressed by the embedded bare infinitival. The problem with (92a) may therefore be that while the simple present on the verb zien locates the eventuality of seeing at speech time, the aspectual verb gaan locates the eventuality of her reading the book after speech time. The past-tense example in (92b') seems perfectly coherent, as it expresses that Jan witnessed the beginning of the eventuality of her reading the book.

The fact that examples (93a&b) are both acceptable shows that the semi-aspectual verb zitten does not raise similar problems as gaan, which is consistent with the fact that zitten locates the eventuality expressed by the bare infinitival in a temporal interval that includes speech time.

For completeness’ sake, note that the account of the markedness of example (92a) given above receives further support from the acceptability of present-tense ACI-constructions with an illusory reading in (94), which were briefly discussed at the end of Section 5.2.3.3, sub I; in such constructions simultaneity is not implied and the contradiction (or incoherence) does not arise.

The examples in (95) with latento make/let are also acceptable. However, the fact that the aspectual verb gaan locates the eventuality after speech time (i.e. in the non-actualized part of the present-tense interval) favors the causative interpretation of (95a). The fact that the semi-aspectual verbs locate the eventuality in a temporal interval that includes speech time makes the permissive reading of (95b) the most plausible one.

Finally, we would like to note that some speakers find AcI-constructions with (semi-)aspectual verbs somewhat harder to get with transitive verbs than with intransitive verbs. This also seems reflected in our Google searches; it is relatively easy to find examples with intransitive verbs like werkento work or slapento sleep, but more difficult to find examples with transitive verbs such as lezento read.