Usually, main verbs are syntactically classified according to the number and type of arguments they take. These properties are sometimes formalized by assigning main verbs subcategorization frames, which specify the number of arguments they take as well as the categories (e.g. NP or PP) and the thematic roles of these arguments. An intransitive verb such as lachento laugh has one nominal argument with the thematic role of agent. A transitive verb such as lezento read has two nominal arguments with the thematic roles of agent and theme, respectively. A ditransitive verb such as gevento give has three nominal arguments, two noun phrases with the thematic roles of agent and theme, and a third noun phrase with the thematic role of recipient; we will return to the fact that the recipient of geven can also be expressed as a PP in Subsection D below.

At least some of the information in these subcategorization frames is systematically related to the meanings of the verbs in question. This is evident from the fact that the arguments mentioned in (23) fill slots in the semantic predicate frames implied by the verbs: lachen is a one-place predicate lachen (x) and the agent argument fills the single argument slot x; lezen is a two-place predicate, and the agent and theme arguments fill the x and y slots in the predicate frame lezen (x,y); geven is a three-place predicate, and again the three arguments fill the slots in the predicate frame geven (x,y,z).

The arguments that fill the slots in the predicate frames of two-place and three-place predicates are not all equal: filling the y and z slots in a sense creates one-place predicates that can be predicated of the arguments placed in the x slot. In syntactic terms, we can say that fillers of y and/or z correspond to objects of the clause, and fillers of x correspond to the subjects. Since the addition of the object(s) to the verb creates a predicate in the traditional, Aristotelian sense, objects are often referred to as complements or internal arguments of the verb. Subjects, on the other hand, are the arguments of which these one-place predicates are predicated, and are therefore also referred to as external arguments. In (24) the subcategorization frames in (23) are repeated, with the external arguments underlined to distinguish them from the internal arguments.

There are several complications that are not discussed here. Subsection II will show, for example, that so-called unaccusative and undative verbs do not have an external argument, but are predicated of an internal argument; cf. Table 2 below.

The fact that the three arguments selected by a verb such as gevento give function as agent, theme and recipient is often referred to as semantic selection. However, semantic selection can be more specific; for example, verbs like zich verzamelento gather, zich verspreidento spread and omsingelento surround in (25) usually require their subject to be plural when headed by a count noun, unless the noun denotes a collection of entities such as menigtecrowd.

There are also verbs like verzamelento collect and (op)stapelento pile up in (26) that impose similar selection restrictions on their objects: the object of such verbs can be a plural noun phrase, a non-count noun such as porseleinchina, or a singular noun phrase headed by a count noun denoting collections of entities such as serviesdinnerware, but not a singular noun phrase headed by a count noun like muntcoin or bordplate, which denote discrete entities.

The examples in (27) show that the information can be even more idiosyncratic: verbs of animal sound emission often select an external argument that refers to a specific or at least very small set of animal species, verbs that take an agentive external argument often require their subject to be animate, and verbs of consumption usually require their object to be edible, drinkable, and so on.

Since restrictions of the kind illustrated in (25) to (27) do not enter into the verb classifications that we will discuss below, we need not address the question as to whether such semantic selection restrictions are encoded in the subcategorization frames of the verbs, or whether they follow from our knowledge of the world and/or our understanding of the meaning of the verb in question; cf. Grimshaw (1979) and Pesetsky (1991) for related discussion.

Subcategorization frames usually provide information about the categories of the arguments, i.e. whether they should be realized as a noun phrase, a prepositional phrase, a clause, etc. That this is necessary can be motivated by the fact that languages may have different subcategorization frames for similar verbs; the fact that the Dutch verb houden requires a PP-complement while its English counterpart to like takes a direct object shows that the category of the internal argument(s) cannot be inferred directly from the meaning of the verb, but is a language-specific matter.

That the category of the internal argument(s) cannot be inferred directly from the meaning of the verb is also suggested by the fact that verbs like verafschuwento loathe and walgento loathe, which express more or less similar meanings, have different subcategorization frames.

The subcategorization frames also provide specific information about the prepositions heading PP-complements. This can be motivated by comparing the Dutch and English examples in (30): although the Dutch translation of the English preposition for provided by dictionaries is voor, in many (if not most) cases English PP-complements with for are not rendered as voor-PPs; vice versa, Dutch voor often has an English counterpart other than for. This again shows that the choice of preposition is an idiosyncratic property of the verb that cannot be inferred from its meaning.

Of course, the above does not imply that the choice between nominal and PP-complements is completely random. There are clearly a number of systematic correlations between the semantics of the verb and the category of its internal arguments; cf. Section 1.2.4. For instance, the examples in (31) show that incremental themes (themes referring to entities that gradually come into existence as a result of the event denoted by the verb) are typically realized as noun phrases, while themes that exist independently of the event denoted by the verb often appear as PP-complements.

Similarly, affected themes are usually realized as direct objects, while themes that are not (necessarily) affected by the event can often be realized as PP-complements. For instance, example (32a) implies that Jan hit the hare, while (32b) has no such implication; cf. Section 3.3.2, sub I.

The same is true for the choice between a nominal and a clausal complement. For instance, the examples in (33) show that verbs like zeggento say or denkento think, which semantically select a proposition as their complement, typically take declarative clauses and not noun phrases as their complement, since the former but not the latter are the canonical expression of propositions.

The examples in (34) show that something similar holds for verbs like vragen or zich afvragen, which semantically select a question; they typically take an interrogative clause as complement rather than a noun phrase.

For completeness’ sake, it should be noted that the choice of a particular preposition as the head of a PP-complement is not necessarily completely idiosyncratic: Loonen (2003) shows that there are several semantic subregularities, and Schermer-Vermeer (2006) shows that in some cases the locational meaning of the preposition used in PP-complements of the verb can still be recognized, as in volgen uitto follow from and zondigen tegento sin against; cf. Schermer-Vermeer &Vandeweghe (2023) for more cases and a more detailed discussion of this.

The examples in (31) and (32) have shown that some verbs can occur in more than one verb frame, i.e. either with a nominal or a prepositional/clausal complement. Another familiar example of such verb-frame alternations is illustrated in (35), which shows that ditransitive verbs such as schenkento give/present can realize their internal recipient argument either as a noun phrase or as an aan-PP.

In earlier generative grammar, this alternation was accounted for by assuming that the subcategorization frame of the verb schenken is as in (36), where the braces indicate that the NP and PP are alternative realizations of the recipient argument.

However, there are less stipulative ways to account for this alternation. One way is to derive example (35a) from (35b) by a transformation usually referred to as dative shift; cf. Emonds (1972/1976) and many others. Another way is to assume that there is only one underlying semantic representation, but that the syntactic mapping of the arguments may vary (possibly depending on other factors). We refer the reader to Levin & Rappaport Hovav (2005: §7) for an overview of different theoretical approaches to such verb-frame alternations. See also Chapter 3 for a discussion of various common verb-frame alternations in Dutch.

Traditional grammars usually classify main verbs on the basis of their adicity, i.e. the number of nominal arguments they take. For reasons that will become clear later, we will use the terms in (37) to refer to the three subclasses that are traditionally distinguished and reserve the traditional notions of intransitive, transitive, and ditransitive verbs to refer to specific subsets of these classes.

The classification of main verbs in (37) is crucially based on the notions of subject and object. This has been criticized by pointing out that in this way verbs with quite different properties are lumped together; cf. the discussion in Subsection B. This is due to the fact that whether an argument is realized as subject or object is determined by the syntactic properties of the construction as a whole and not by the semantic function of the arguments. The active/passive pair in (38) illustrates this: in (38a) the subject de bijthe bee is an external argument, as is clear from the fact that it has the prototypical subject role of agent, whereas in (38b) the subject de katthe cat is an internal argument, as is clear from the fact that it has the prototypical direct object role of patient.

In generative grammar, the semantic difference between the subjects of the examples in (38) is often expressed by saying that the subject de bijthe bee in (38a) is the logical subject of the verb, i.e. corresponds to x in the logical formula predicate (x,y), whereas the subject de kat in (38b) is a “derived” subject, corresponding to y in that logical formula. From now on we will refer to derived subjects as theme-subjects, since the discussion of the examples in (40) and (42) in Subsection B will show that such derived subjects originate in the same structural position in the clause as direct objects.

Perlmutter (1978) and Burzio (1986) have shown that the set of monadic verbs in (37a) can be divided into two distinct subclasses. Besides the usual intransitive verbs such as lachento laugh, there is a class of so-called unaccusative verbs, such as arriverento arrive, with a number of distinctive properties (which may differ from language to language). The examples in (39) illustrate the differences between the two types of monadic verbs identified for Dutch in Pollmann (1975: §5.4) and Hoekstra (1984a: §3.3).

The first distinguishing property is auxiliary selection in the perfect tense: the (a)-examples show that intransitive verbs such as lachen take the perfect auxiliary hebbento have, whereas unaccusative verbs such as arriveren take the auxiliary zijnto be. The second property concerns the attributive use of past/passive participles: the (b)-examples show that past/passive participles of unaccusative verbs can be used attributively to modify a head noun corresponding to the subject of the verbal construction, whereas past/passive participles of intransitive verbs lack this ability. The third property concerns impersonal passivization: the (c)-examples show that this is possible with intransitive verbs, but not with unaccusative verbs.

Like monadic verbs, dyadic verbs can be divided into two distinct subclasses. Besides the usual transitive verbs such as kussento kiss with an accusative object, we find so-called nom-dat verbs such as bevallento please with a dative object; since Dutch has no morphological case, we illustrate the case property of nom-dat verbs with the German verb gefallento please in (40a'). Lenerz (1977) and Den Besten (1985) have shown that these nom-dat verbs are special in that the subject follows the object in the unmarked case, as in the (b)-examples.

This word-order property easily distinguishes nom-dat verbs from transitive verbs, since these do not allow the subject to follow the object; transitive constructions usually have a strict nom-acc order in the middle field of the clause.

The (b)-examples in (42) show that the same word-order variation as with the nom-dat verbs in (40) is found with passivized ditransitive verbs, in which case the dat-nom order is again the unmarked one.

Den Besten (1985) analyzes the word-order variation in (40) and the (b)-examples in (42) by assuming that the theme-subject originates in the direct object position and optionally moves into the canonical subject position of the clause; cf. the representations in (43), where the em-dash indicates the empty subject position of the clause and the trace ti the original position of the nominative phrase. Broekhuis (1992/2008) has shown that the movement is not truly optional, but subject to conditions related to the information structure of the clause; the subject remains in its original position if it is part of the focus (discourse-new information) of the clause, but moves into the regular subject position if it is part of the presupposition (discourse-familiar information).

The word-order similarities between (40) and the (b)-examples in (42) show that nom-dat verbs also take “derived” theme-subjects; the dat-nom orders are all the base-generated ones and the nom-dat orders are all derived by movement of the theme-subject into the regular subject position of the clause.

There are good reasons to assume that monadic unaccusative verbs such as arriverento arrive in the primed examples in (39) are like nom-dat verbs in that they also take a theme-subject. This can be illustrated by the examples in (44). The (b)-examples show that the past/passive participle of a transitive verb such as kopento buy can be used as an attributive modifier of a noun that corresponds to the internal theme argument (here: direct object) of the verb, but not to the external argument (subject) of the verb.

The fact that the past participle of arriveren in (39b') can be used as an attributive modifier of a noun corresponding to the subject of the verb supports the claim that the subject of an unaccusative verb is also an internal theme argument of the verb. That subjects of unaccusative verbs are not assigned the prototypical thematic role of external arguments (i.e. agent) can be further supported by the fact that unaccusative verbs never allow agentive er-nominalization, i.e. cannot be used as input to the derivational process that derives person nouns with the suffix –er. The primed examples in (45) show that while many subjects of intransitive and (di)transitive verbs can usually undergo this process, unaccusative and nom-dat verbs cannot. For a detailed discussion of agentive er-nominalization, see Sections N14.3.1.5 and N15.2.3.1.

We conclude with a final argument for the claim that subjects of unaccusative verbs are internal arguments. This is provided by the causative-inchoative pair in (46), which shows that the subject of the unaccusative construction in (46b) has a similar semantic relation to the (inchoative) verb breken as the direct object of the corresponding transitive construction with the (causative) verb breken in (46a). This suggests that we are again dealing with a theme-subject in (46b); cf. Section 3.2.3 for a more detailed discussion.

The discussion in Subsection B has shown that the traditional classification of main verbs based on the number of nominal arguments they take is seriously flawed. The set of monadic verbs lumps together two sets of verbs with very different properties, and the same goes for the set of dyadic verbs. If we also take into account impersonal verbs such as sneeuwento snow which are often assumed to take no argument at all and occur with the non-referential subject hetit, we can replace the traditional classification by the finer one in Table 2, which appeals to the type of argument the verb takes, i.e. the distinction between internal and external arguments. Ignore the question marks for the time being; they will be discussed in Subsection D.

	name	external argument	internal argument(s)
no internal argument	intransitive: snurken ‘to snore’	nominative (agent)	—
	impersonal: sneeuwen ‘to snow’	—	—
one internal argument	transitive: kopen ‘to buy’	nominative (agent)	accusative (theme)
	unaccusative: arriveren ‘to arrive’	—	nominative (theme)
two internal arguments	ditransitive: aanbieden ‘to offer’	nominative (agent)	dative (goal) accusative (theme)
	nom-dat verb: bevallen ‘to please’	—	dative (experiencer) nominative (theme)
	????	—	nominative (goal) accusative (theme)

Figure 1 gives the same classification in the form of a graph, which shows more clearly that the unaccusative verbs are simply the counterparts of “regular” verbs with an external argument. This graph nicely expresses our claim that the distinction based on the presence or absence of an external argument is more fundamental than the distinction based on the adicity of the verb.

We indicated in Table 2 the prototypical thematic roles assigned to the arguments in question without intending to exclude the availability of other thematic roles; external arguments, for example, need not be agents, but can also function as external causes, as is clearly the case when the human subject in (46a) is replaced by a non-human one such as de storm : De storm brak het raamThe storm broke the window.

The classification in Table 2 includes one logical possibility which we have not yet explored, in which an internal goal argument (i.e. an argument with a thematic role similar to that assigned to the dative argument of a ditransitive verb) functions as the subject of the clause, and which we can therefore call undative verbs. Since the current linguistic literature usually does not seem to recognize the existence of this verb type, we have marked this option with question marks in Table 2. However, this subsection will argue that it does exist, and that the prototypical instantiations of this type are the verbs hebbento have, krijgento get, and houdento keep. This subsection will only show this for the verb krijgen; the verbs hebben and houden, as well as a number of other possible cases, will be discussed in Section 2.1.4.

Consider the examples in (47). It seems that the indirect object in (47a) and the subject in (47b) have a similar thematic role: they both function as recipients of the theme argument het boekthe book. The fact that the subject in (47b) is not assigned the prototypical subject role of agent/cause also suggests that the verb krijgento get does not have an external argument (although the agent/cause can be expressed in a van-PP). Taken together, these two facts suggest that the noun phrase Marie in (47b) is not an underlying but a “derived” recipient-subject. This subsection will show that there are a number of empirical facts supporting this claim. Note that the derived subjects of undative verbs can also be a possessor, experiencer, or benefactive, all of which can be assigned to indirect objects; for brevity and simplicity, we will use the term Recipient-subject to include these other thematic roles.

Example (45) in Subsection B has shown that er-nominalization is possible only if an external (agentive) argument is present; snurker (snore + -er) versus *arriveerder (arrive + -er). If the subject in (47b) is indeed an internal goal argument, we expect er-nominalization of krijgen to be also impossible. Example (48a) shows that this prediction is indeed borne out, since the noun krijger only occurs with the meaning “warrior”; this noun is derived from medieval crigento wage war, which was also input for the medieval verb gecrigento obtain with difficulty, which eventually developed into modern krijgen (Landsbergen 2009: §4). The discussion in Subsection B has also shown that unaccusative verbs cannot be passivized, which strongly suggests that the presence of an external argument is a necessary condition for passivization. If this is the case, we correctly predict that passivization of (47b) is also impossible, as shown in (48b).

Although the facts in (48) are certainly suggestive, they are of course not conclusive for the claim that krijgen is an undative verb, since we know that not all verbs with an external argument allow er-nominalization, and that there are several additional restrictions on passivization. However, there is other evidence that supports the idea that the subject of krijgen is a derived subject. For example, the claim that krijgen has a recipient-subject may explain the fact that the standard Dutch example in (49a), with the idiomatic double object construction iemand de koude rillingen bezorgento give someone the creeps, has a counterpart with krijgen, as shown in (49b).

The final and perhaps most convincing argument for the assumption that the subject of krijgen in (49b) has a similar argument status as the indirect object of bezorgen in (49a) can be based on the possessive constructions in (50). In constructions with a complementive locative PP, the possessor of the complement of the preposition can be realized either as a possessive pronoun or as a dative phrase; the two examples in (50a&b) are synonymous in the sense that the object Marie/haar in (50b) must be interpreted as the (inalienable) possessor of the noun phrase de vingers. In general, only a dative object can fulfill this function of inalienable possessor of the possessum embedded in the locative PP; cf. Section 3.3.1.4. However, the subject of the verb krijgen is an exception to this general rule; the subject Marie/zij in (50c) is also interpreted as the inalienable possessor of the noun phrase de vingers. This would follow if the subject Marie/zij in (50c) is not an external argument, but has the same syntactic function as the dative object Marie/haar in (50b); we are dealing with a possessor-subject.