This subsection discusses verbal collocations with a noun, adjective, or verb as the first part. In general, we find two syntactically relevant types: inseparable and separable collocations. It seems that this distinction is weakly correlated with the semantic/syntactic status of the first part, as Ackema (1999) notes that in separable collocations the first part is usually an argument of the second, verbal part. This is illustrated in (35). The verb vechtento fight in (35a) is intransitive and the N-part bekmouth is interpreted as having the thematic role of instrument; cf. met de bek vechtento fight with the mouth. The verb halento get in (35b) is transitive and the N-part adembreath is interpreted as a theme argument. The primed examples show that only in the latter case can the N+V collocation be split.

The two lists in (36) are samples of the two types of N+V collocation, based on De Haas & Trommelen (1993) and Booij (2010). We do not include inseparable verbs such as voetballento play soccer that are (potentially) derived via conversion from complex nouns (here: voetbalfootball) or formations such as raadplegento consult with a non-transparent or non-compositional meaning for contemporary speakers, because these are expected to be inseparable anyway. Recall that we deviate from the orthographic convention of writing the N+V collocations in (36b) as a single word in order not to bias the discussion towards a compound analysis for these collocations.

Note that we have used the term “weak correlation” to characterize Ackema’s hypothesis. The reason is that it is not the case that N+V collocations are always separable when the N-part functions as a theme. This can be illustrated by the collocation stof zuigento vacuum, which can be used by many speakers as either a separable or an inseparable collocation. However, there is reason to assume that the N-part has lost its argument status in the inseparable form; cf. Ackema (1999) and the discussion of the examples in (48) in Subsection II.

We should also raise a warning flag and note that there are a number of cases of separable N+V collocations for which it is less clear that the N-part functions as a (direct) argument of the V-part. For example, this holds for piano spelento play the piano and televisie kijkento watch television, since spelen and kijken select a PP-complement in examples such as (38). In order to maintain the claim that the N-part is an argument of the V-part, we have to assume that the PP-complement is reduced in the separable N+V collocations piano spelen and televisie kijken; cf. Ackema (1999) and Booij (2010) for a discussion of these forms.

The examples in (39) show that the two main types can also be found in A+V collocations: (39a) is an example with the inseparable (compound) verb liefkozento caress and (39b) with the separable collocation bekend makento make known.

If we exclude examples such as blinddoekento blindfold, which is derived from the complex noun blinddoekblindfold, and cases such as dwarsbomento thwart with a meaning that is not transparent or compositional for the contemporary speaker, there are very few inseparable A+V collocations; the examples in (40a) are again taken from De Haas & Trommelen (1993). For the separable A+N collocations in (40b), Ackema’s hypothesis that the left part of the collocation is usually an argument of the verbal part of the collocation seems too strict, but we can easily repair this by loosening the statement slightly by requiring that the left part must be a complement of the verbal part, as this will include complementives. Again, we depart from orthographic convention by writing separable A+V collocations as separate words, so as not to bias the discussion toward a compound analysis for these collocations.

The proposed revision of Ackema’s hypothesis, which we will call Ackema’s generalization, also accounts for the fact that particle verbs (P+V collocations) like opbellento call (up) and overstromento run over in (41) are usually separable, since Section 2.2 has shown that verbal particles also function as complementives. Although there are a number of inseparable P+V collocations, we will not digress here, as this would simply repeat the discussion in Section P32.2.4.4. In fact, we will ignore P+V collocations altogether until we return to them in Subsection VI D.

There are very few inseparable V+V collocations like hoesteproestento cough and splutter in (42a); more transparent cases such as zweefvliegento glide (in a sailplane) belong to the set of immobile collocations, which will be discussed in Subsection IV. Separable V+V collocations are also rare and may not even exist: a possible case is laten vallento drop in (42b), but the fact that the dependent verb vallento fall does not precede but follows the causative verb latento make/let suggests that we are not dealing with a verbal collocation but with a regular causative laten-construction. We will not discuss such cases here, but in Section 5.2.3.4.

This subsection has shown that separable verbal collocations require their first member to function as a complement of the verbal part: the N-part in N+V collocations functions as a direct (and sometimes prepositional) object of the V-part, and the A-part in A+N collocations functions as a complementive, i.e. a predicative complement of the V-part. Since there are no clear cases of separable V+V collocations, and since particle verbs will be discussed separately in Subsection VID, the following subsections will deal only with N+V and A+V collocations.

Assuming that inseparable X+V collocations are true compounds, their syntactic behavior can be explained by appealing to the lexical integrity constraint, according to which syntactic operations cannot apply to subparts of words. If an inseparable N+V collocation such as bekvechtento squabble is analyzed as in (43a), where the label Vº stands for a word boundary, movement of the verb cannot strand the nominal part of the collocation. By the same logic, separable N+V collocations cannot be analyzed as compounds, but should be analyzed as phrasal: a separable N+V collocation such as adem halento breathe should be analyzed as in (43b), where the label V' stands for some phrasal projection of the verb containing a direct object; in this case, movement of the verb can strand the nominal part of the collocation.

There is morphological and syntactic evidence for this distinction in (43). First, we would expect inflectional material to be attached at the Vº rather than at the V'-level, and so we predict that the nominal part will follow preverbal inflectional material in the case of (inseparable) compound verbs such as bekvechten, but will precede such material in the case of (separable) phrasal collocations such as adem halen, as in (44).

The examples in (45) show that this is indeed correct: the preverbal part ge- of the participial circumfix ge-...-d/t and the infinitival prefix te must precede the nominal part in bekvechten, but must follow it in adem halen for most speakers.

Note in passing that there seems to be some variation among speakers, especially with regard to the infinitival marker te. For example, a Google search (May 2023, 2024) showed that the form bek te vechten is occasionally used on the internet (perhaps in jest), whereas we got no relevant hits for the strings [heb bekgevecht] and [heb * bekgevecht]. Similarly, the form te ademhalen is not difficult to find (though with a lower frequency than adem te halen), while we found only a handful of genuine cases with the form geademhaald. The judgments in (45) reflect our own acceptability judgments and may be an idealization of the actual situation in standard Dutch.

The form of the past participle gebekvecht in (45a) constitutes an additional argument for a compound analysis, given that the participle of the simplex verb vechten has the irregular form gevochten. The reason is that De Haas & Trommelen (1993:441) claims that a hallmark of compounds is that they have a regular declension; that this is a useful test is illustrated in (46), in which glimlachento smile is an inseparable N+V compound and paard rijden is separable phrasal N+V collocation; only in the former case does the collocation have the regular declension ge-...-d/t.

A rather spectacular illustration of De Haas & Trommelen’s claim is stof zuigen. The examples in (37) have shown that this collocation exhibits mixed behavior for many speakers: the N+V collocation can be split under verb-second, but it can also be moved as a whole. The simplex verb zuigento suck has an irregular declension: zuig-zoog-gezogen. The predictions of De Haas & Trommelen’s hypothesis are clear. First, we predict that stof zuigento vacuum is associated with two different past participial forms, depending on the position of the nominal part. The primeless examples in (47) show that this prediction is indeed correct. Second, we predict that the split under verb-second is only possible if the finite verb has the irregular declension; the primed examples show that this prediction is also correct.

Note that we have ignored the fact that the form stofzoog is sometimes found in verb-second position on the internet; this is not surprising, since speakers are quite uncertain about the “correct” form of the past tense, as is clear from the fact that it is a recurring topic of discussion on the internet. Moreover, there seems to be normative pressure to use the inseparable form, as can be seen from the fact that taaladvies.net/taal/advies/vraag/755 and the Van Dale Dictionary only give the regular declension.

The claim that stof zuigen allows two different analyses is also supported by the examples in (48), adapted from De Haas & Trommelen (1993:442). These examples show that this collocation can be used with the direct object de kamerthe room when it has a regular declension, but not when it has an irregular declension.

This contrast follows from the analysis proposed above: if stof zuigen is phrasal, the bare noun stof functions as a direct object, and thus blocks the addition of another direct object such as de kamerthe room: if, on the other hand, it is a compound, it can simply be stored in the lexicon as a transitive verb, and consequently the use of a direct object such as de kamer is fully licit. Other cases of such transitive, inseparable N + V collocations mentioned in Ackema (1999) are: beeldhouwento sculpt (lit: statue + chop) stand hersenspoelento brainwash.

In (49) we give similar examples for A+V collocations: liefkozento fondle (lit. sweet + caress) is a compound and the adjectival part lief must therefore follow the preverbal part of the participial circumfix ge-...-d/t and the infinitival prefix te; bekend makento make public, on the other hand, is phrasal and the adjectival part must therefore precede these elements.

Table 1 lists three morphological reasons for assuming that inseparable verbal collocations are compounds, whereas separable verbal collocations are phrasal. The first two reasons concern the placement of the inflectional affixes ge-...-d/t and te. The third reason concerns declension: inseparable verbal collocations always have a regular declension, which has been claimed to be a hallmark of compounds, whereas the declension of the verbal part of separable verbal collocations is completely determined by the verbal part.

Although the discussion in Subsection II strongly suggests that separable N+V collocations are phrasal and that the N-part usually functions as a direct (or perhaps prepositional) object of the V-part, the collocation has a number of properties not normally found in verb phrases consisting of a verb and an object. In terms of these peculiarities, separable N+V collocations behave more like N+V compounds. We will illustrate this compound-like behavior of separable N+V collocations by comparing the separable collocations adem halento breathe and piano spelento play the piano with the regular verb phrase iets halento fetch something and iets spelento play something (e.g. a sonata).

A first property is that the N-part of a separable N+V collocation is usually bare, i.e. not accompanied by a determiner, whereas singular regular direct objects are usually not bare, but require a determiner. Note that this difference cannot be observed when the N-part is plural, as in aardappels schillento peel potatoes, because indefinite plurals take a phonetically empty article.

Related to this difference concerning the determiner is the fact that the nominal part of the N+V collocation is not referential. This can be shown by comparing the examples in (51): example (51a) cannot be uttered out of the blue, since the reference of the deictic pronoun hij cannot be properly determined by the bare noun piano; example (51b) with the regular direct object de sonatethe sonata, on the other hand, is fine, since the pronoun can take this object as its antecedent.

For the same reason, it is normally impossible to modify the nominal part of an N+V collocation with an attributively used adjective, whereas this is of course possible with regular direct objects, as shown in (52).

Note in passing that attributive modification of the nominal part of a separable N+A collocation is marginally possible if the modifier-noun combination has a type reading: for example, Booij (2010) gives examples such as dat Jan klassieke piano speelt. However, the fact that Booij translates this example as “that John plays classical piano music” suggests that we may simply be dealing with a regular direct object in the form of a mass noun, comparable to Hij speelt klassieke muziek/jazzHe plays classical music/jazz. We will leave this issue to future research and simply conclude from the examples above that nominal parts of N+V collocations are not referential. In this respect they are similar to the first members of N+V compounds like beeldhouwento sculpt, N+A compounds like boterzachtsoft as butter, and N+N compounds like huissleutellatchkey, but unlike regular direct objects.

A second property of the N-part of separable N+V collocations is that speakers allow it to permeate clause-final verb clusters. In this respect, it resembles the nominal part of a N+V compound, for which this is obligatory, but differs from regular direct objects, which usually do not allow this (but see Sections 5.2.3 and 6.0 for some exceptions). The examples in (53) show this constructions with separable N+V collocations and constructions with a definite direct object.

A third property of the N-part of a separable N+V collocation is that it can be left-adjacent to the main verb in the progressive aan het + Vinfinitive construction; regular direct objects, on the other hand, must precede the aan het + Vinfinitive phrase.

A final property in which N-parts of separable N+V collocations differ from regular direct objects may be that they do not easily occur as part of a postnominal van-PP in nominalizations, as shown in (55). We marked these cases with a percentage sign, because they are not uncommon on the internet.

The discussion above has shown that the N-part of N+V collocations has several properties which are unexpected for regular direct objects, but which are similar to the properties of the N-part of an N+V compound: (i) it is not referential, (ii) it is allowed to interrupt clause-final verb clusters, and (iii) it can be left-adjacent to the main verb in the progressive aan het + Vinfinitive construction. These results are listed in Table 2.

The first property does not apply to A+V collocations for obvious reasons, but the examples in (56) with fijn malento grind show that the latter two properties also apply to such cases.

Table 2 shows that separable X+V collocations like (36b) and (40b) in Subsection I exhibit a certain degree of freedom in word order with respect to the placement of the X-part in constructions with a clause-final verb cluster or a progressive aan het-phrase. This raises the question whether it is justified to consider separable X+V collocations as a single class or whether we should distinguish two subtypes. Booij (2010: §4.3) argues for N+V collocations that there is reason to assume the latter, based on the morphological expression of sentence negation. Although sentence negation is typically expressed by the independent negative adverb nietnot, as in (57a), the examples in (57b&c) show that it usually obligatorily merges with existentially quantified elements in its clause; in (57b) the negation is expressed as part of a frequency adverb (neg + ooit → nooitnever) and in (57c) as part of an indefinite direct object (neg + een auto → geen autono car).

However, the examples in (58) show that in N+V collocations like auto rijdento drive a car and piano spelento play the piano, the merger of sentence negation is optional; it can either be expressed by the adverb nietnot or as part of the negative article geenno.

The examples in (59) show that the choice between the two options depends on the placement of the N-part of the collocation in clauses with a verb cluster: negation seems preferably to be expressed by the negative article geen, but if the N-part remains adjacent to the V-part, the negative adverb niet must be used. Although Booij considers the options marked with a number sign acceptable, there may be reasons to reject his claim; we will return to this issue in Subsection V.

The same seems to hold for the progressive aan het + Vinfinitive constructions in (60). Although such negated examples are extremely rare on the internet, they seem to be perfectly acceptable in contrastive contexts. The options marked with the number sign also occur on the internet in contrastive contexts, which is not surprising, since the merger of sentence negation is usually not obligatory in contrastive contexts: cf. Ik heb niet een auto, maar een fiets gekochtI did not buy a car, but a bicycle.

As such, the examples in (59) and (60) do not shed any light on the question of whether separable verbal collocations form a single class, or whether we should distinguish two subtypes: the merger of sentence negation may simply be subject to some adjacency restriction, which would effectively block the formation of geen in the primed examples. However, these examples are quite revealing in combination with the examples in (61), in which the N+V collocations are split by verb-second and sentence negation must be expressed by the negative article geen; the use of the adverb niet leads to ungrammaticality.

The fact that sentence negation cannot be expressed by the adverb niet, but must be expressed by the merged form geen, strongly suggests that these verb-second examples are more akin to the primeless than to the primed examples in (59) and (60); the merger of negation is restricted to those cases in which the N+V collocation can be split by syntactic operations like verb clustering and verb-second. This suggests that separable verbal collocations like (36b) and (40b) actually have two uses: they may be separable in all syntactic and morphological contexts, or they may be separable only in morphological contexts. It is the latter set of separable verbs that we have characterized as immobile in the sense that they resist verb-second. This line of reasoning would lead to the three groups of X+V collocations in Table 3; we have illustrated the clustering of properties on the basis of N+V collocations only, but it seems reasonable to assume that they also hold for A+V collocations.

The morphological properties given in the A-rows of Table 3 distinguish the inseparable verbal compounds from the separable verbal collocations; whereas the former require the X and V-part to be adjacent in past/passive participles and te-infinitives, the latter do not allow this. The properties in the B-rows are the crucial ones for distinguishing between the two types of separable verbal collocations. The C-row describes the verb-second behavior of the three types of verbal collocations we have distinguished on the basis of the properties in A and B.

This subsection attempts to outline a theoretical account of the clustering of the properties in Table 3. The first group of X+V collocations in Table 3 is the class of compounds, which differs from the other two groups in that the X+V collocations form an indissoluble morphological and syntactic unit. In short, they function as complex words of the form [Vº X V]: inflectional material is added externally to Vº, which accounts for their properties in the two A-rows in Table 3, and syntactic movements can only affect Vº as a whole, which accounts for their properties in the B and C-rows.

The original class of separable X+V collocations is now divided into two subclasses, which have in common that the X-part can be separated from the verbal part by inflectional morphemes. This strongly suggests that we are dealing with a regular verb phrase, [V' X Vº], in which N functions as a direct object and A functions as a complementive: inflectional material is thus added to Vº, which again accounts for the properties in the two A-rows in Table 3.

The claim that separable X+V collocations are regular verb phrases of the form [V' X Vº] also explains the properties of the first subclass of separable verbs given in the B and C-rows in Table 3: (i) the fact that the X-part is syntactically independent from the verbal part of the collocation predicts that these parts need not remain adjacent in syntax, but can be split by syntactic operations involved in the formation of verb clusters and the derivation of verb-second, and (ii) the fact that the N-part is in the regular object position accounts for the fact that the merger of sentence negation and the indefinite article (resulting in geenno) is obligatory.

This leaves us with the second group of separable verbs that do behave as a unit for syntactic purposes. It has been proposed that these involve incorporation, a syntactic operation that creates a syntactic unit by so-called head adjunction. This changes the phrase structure [V' X V] by head movement of X into [V' ti [V* Xi V]], where V* stands for a syntactically derived complex head and ti for the trace of the incorporated head X. In some languages noun incorporation is much more productive than in Dutch, and Baker (1988) has shown for such languages that incorporation is restricted to complements, providing a natural cross-linguistic rationale for Ackema’s generalization. The incorporation analysis also derives the properties in the (B)-columns of Table 3: (i) although the collocation can be split by morphological operations, this is impossible for syntactic operations involved in the formation of verb clusters or aan het-phrases, because they apply after incorporation has taken place, and (ii) the premise that the N-part is no longer in object position after incorporation can be held responsible for the impossibility of the merger of sentence negation. It remains puzzling, however, why this kind of separable X+V collocation cannot undergo verb-second. It has been suggested that the reason is that verb-second can only affect words, i.e. Vº’s: if Vº cannot be extracted from V* and V* itself cannot undergo verb-second, the impossibility of verb-second would follow. What such an approach still needs, of course, is a credible account of the observation that V*’s cannot undergo verb-second; cf. Koopman (1995), Vikner (2005), and Booij (2010) for various attempts at an explanation.

If the above proposal is on the right track, we can identify the three types of verbal collocation by their different types of verbal element: true compounds ([Vº X V]), phrasal projections ([V' X Vº]), and word-like V*-units derived by incorporation ([V* X Vº]). This makes it possible to replace Table 3 with Table 4.

To make the incorporation proposal watertight, we should say something about the negation data in the primeless examples in (59), repeated here as (62). Since the N+V collocation is split, we cannot assume that the N-part is incorporated into the V-part of the collocation. Therefore, we expect the N-part to be in the regular object position and, consequently, the merger of sentence negation to be obligatory; the use of niet should thus lead to ungrammaticality.

The position that the examples with niet are not (fully) acceptable may be supported by the fact that such examples are rare on the internet. Our Google searches (4/23/2024) on the strings [niet auto kan rijden] and [niet piano kan spelen] yielded a total of 35 relevant results (manual count). Moreover, the results include many cases in which niet is coordinated with or modified by another element (wel of nietwhether or not, al dan nietwhether or not, nog nietnot yet, helemaal nietnot at all, etc.), which may make the merger of the negation unnecessary for independent reasons; cf. the examples in (63).

Since the corresponding search strings [geen auto kan rijden] and [geen piano kan spelen] each yielded more than 100 hits, we might have to conclude that the use of niet should be dismissed as irrelevant for a variety of reasons. This would be in line with our own judgment that under neutral intonation the use of niet in examples such as (62) is quite marked compared to the use of geen. Recall that the restriction to neutral intonation is necessary because the discussion of (60) has already shown that the use of niet is possible in contrastive contexts.

Now take the opposite position and assume that the attested cases with niet are sufficient to claim that the use of niet leads to a grammatical result in (62). A coherent incorporation analysis would then be that the X-part can be incorporated not only into the verbal part of the collocation, but also into larger verb clusters, as proposed in Booij (2010). This result would be relatively easy to obtain under the traditional verb-raising analysis of verb clustering: verb raising is assumed to create an adjunction structure (similar to that resulting from noun incorporation) as a result of verb movement. If we assume that the base structure of an example such as (62a) is as given in (64a), verb raising can derive a structure such as (64b) with the syntactically derived complex head [V* kan rijden], and subsequent N-incorporation would lead to (64c). Since we have seen that N-incorporation is optional, we can now explain why both geen and niet are possible in (62): in structure (64b) the negation must merge with the indefinite N-part in direct object position, whereas in structure (64c) this is blocked by the fact that the N-part is part of an adjunction structure.

Independent support for the claim that N-incorporation into larger verb clusters is possible can be found in the fact that examples such as (65), in which the N-part in inside the cluster but not adjacent to the verbal part of the collocation, are at least marginally acceptable for some (but certainly not all) speakers of Dutch.

Under a verb-raising approach, example (65a) is derived as follows: starting from the structure in (66a), verb raising first produces the verb cluster [V* willen rijden] in (66b); subsequent N-incorporation in this cluster results in the structure [V* auto [V* willen rijden]] in (66c); finally, this complex is incorporated into the finite verb by verb raising, resulting in [V* zou [V* auto [V* willen rijden]]] in (66d): cf. Bennis (1992) for a similar derivation of verb clusters containing a particle verb in the order Vfinite–prt-Vinf–Vmain.

There are a number of possible problems with an approach based on noun incorporation and verb raising (including the theory-internal problem that the derivation is anticyclic, which we will ignore here). A minor problem is that the rules determining the word order of the complex V* are quite complicated: while noun and adjective incorporation involve left-adjunction, verb incorporation would (usually) involve right-adjunction. A more serious problem is related to the account of the primeless examples in (61), one of which is repeated here as (67a).

The fact that negation must be expressed by geen in verb-second structures such as (67a) is argued to follow from a constraint prohibiting the extraction of Vº from V*-units: since (67c) violates this constraint, (67a) must have the structure in (67b), which correctly predicts that the merger of negation with the direct object is obligatory. However, the claim that N-incorporation can also target verb clusters, which is needed to account for the examples in (62) marked by a number sign, presupposes that verb clusters are themselves V*-units ([V* V V]). This raises the question of how we can derive verb-second at all, since this would always involve extraction of Vº from a V*-unit (at least under the traditional standard assumption that verb raising is obligatory). A final problem is that the analysis rests on the traditional assumption that verb clustering involves V-movement, which has been challenged in much work since the early 1990s; cf. Section 7.5. We leave undecided the question whether the use of the adverb niet in examples such as (62) leads to a grammatical result or not, and consequently it also remains open whether the assumption that noun incorporation into verb clusters is possible is

needed.

Subsection IV discussed separable V+X collocations that are ambiguous between a mobile and an immobile form. This subsection discusses in more detail cases of verbal collocations that are (usually) of the immobile type. Subsections A-C focus on three different subtypes of immobile N+V collocations: we successively discuss immobile verbs of the type touwtje springento skip (lit: to rope skip), stijl dansento ballroom dance (lit. to style dance), and herinvoerento reintroduce. Subsection D examines inseparable complex particle verbs, and Subsection E concludes with a brief discussion of a type of immobile verb that has received relatively little attention in the literature.