Transitive verbs like kopento buy or lezento read in (22) select two nominal arguments, one external and one internal; the external argument is realized as the subject and usually refers to an agent (or cause) of the event; the internal argument is realized as the direct object and usually refers to the theme of the event.

In general, the two arguments must be expressed overtly, as can be seen from the fact that example (23a) is severely degraded. However, there are many exceptions to this rule; for instance, example (23b) is perfectly acceptable even though there is no direct object. Note, however, that the theme argument is semantically implied in such cases, and is interpreted as a canonical object of the verb lezento read: Marie is reading some kind of text. That the theme argument is semantically implied is also clear from the fact that the pronoun het in the clause given in brackets can refer to the thing that Marie is reading. We refer the reader to Levin (1993: §1.2) and Van Hout (1993: §2.5) for further discussion.

Omitting the direct object is also possible in cases like (22c&d) and leads to a habitual or occupational reading; example (24a) expresses that Jan is a habitual smoker, and (24b) expresses that Marie is a house painter or paints pictures as a hobby. We will call the verbs in (23b) and (24) pseudo-intransitive verbs.

The properties of transitive verbs are illustrated by a very small sample of verbs. Example (25) therefore provides a slightly larger sample of verbs that behave in the same way. Of course, this sample cannot be exhaustive: the set of transitive verbs is an open class consisting of many lexical items, which can easily be extended by borrowings or new coinages.

The defining property of intransitive verbs like huilento cry and slapento sleep is that they select only an external nominal argument. This argument is usually an agent or cause, and is realized as the subject of the clause. Intransitive verbs are not normally accompanied by a direct object, as can be seen from the fact that (26a') is degraded. Occasionally, however, intransitive verbs may be accompanied by a so-called cognate object. Consider the verb slapento sleep in (26b), which implies that Marie is having a sleep. This information can be made at least marginally explicit by adding a direct object, as in (26b'), provided that the object expresses some information that is not already implied by the verb; a modifier such as verkwikkend is obligatory. Something similar is illustrated by the (c)-examples; the use of the cognate object onzinnonsense is acceptable, since it has a negative connotation that is not part of the meaning of the verb.

Example (27) gives a small sample of typical intransitive verbs. In the following discussion, we illustrate the properties of intransitive verbs using only a small subset of these examples. Note in passing that many of these verbs involve voluntary or involuntary bodily functions, which shows that the notion of agent does not imply that the activity can be controlled by the external argument.

Contrary to what traditional grammar assumes, the set of monadic verbs is not a uniform category; Subsection II will show that the intransitive verbs in (27) should be distinguished from the so-called unaccusative verbs in (28).

Example (29) provides a small sample of such verbs. Unaccusative verbs usually denote a process and the subject is usually presented not as an agent but as a theme, i.e. an entity undergoing the process.

The previous subsections have shown that certain transitive verbs can be used as pseudo-intransitive verbs, i.e. as intransitive verbs with an implied canonical object, and that certain intransitive verbs can be used transitively, i.e. with a cognate object. This shows that the distinction between transitive and intransitive verbs is not absolute, but gradual. Not surprisingly, some linguists (e.g. Hale and Keyser 1993) have argued that the two verb classes should actually be considered a single class. If so, the question as to whether a direct object is overtly expressed may depend on whether a canonical object is semantically implied by the verb; a direct object can be used only if it adds something to the meaning inherently expressed by the verb.

This can be clarified by the verb dansento dance, which can easily be used as both an intransitive and a transitive verb, as shown in (30). The reason why (30a) is marked with the direct object present is that the object is redundant: the verb dansen already semantically implies that some kind of dance was performed. Example (30b), on the other hand, is acceptable with a direct object because the object conveys information that is not implied by the verb: it provides more information about the type of dance involved.

Perhaps something similar happens with intransitive motion verbs like schaatsento skate and lopento walk. The primed examples in (31) show that cognate objects are particularly common with these verbs. They trigger a reading according to which the subject participates in a sporting activity; the cognate object then refers to a conventional distance to be covered or to a specific sporting event. For example, sentence (31a') expresses that Jan is taking part in a 5-kilometer skating race/participates in the famous Frisian skating marathon that goes through 11 Frisian towns. Example (31b') gives similar examples with the verb lopento walk.

The discussion of the examples above suggests that it may not be necessary to distinguish between intransitive and transitive verbs at all: the crucial factor is not whether the verb takes a direct object, but whether that object can express non-redundant information. Although we do not want to take a stand on the idea that intransitive and transitive verbs form a single class of verbs (and will continue to use the two terms for practical reasons), we think that the fact that the question can be raised supports the claim that the classification of verbs should not focus primarily on the adicity of the verb; the basic question is not how many arguments a given verb takes, but what kind of arguments.

In the prototypical case, transitive and intransitive verbs denote activities; subjects of such verbs are agents performing these activities: therefore, they typically refer to a [+animate] participant (or an instrument designed to perform a particular task).

Unaccusative verbs, on the other hand, generally denote processes; subjects of such verbs are themes, i.e. participants undergoing these processes. The fact that the subject of an unaccusative verb is not an agent explains why it behaves like the direct object of a transitive verb, in that it can easily refer to a [-animate] participant in the event, as shown in (33).

If we assume that agents are typically external arguments and themes are typically internal arguments, this contrast between the examples in (32) and (33) follows from the claim that subjects of intransitive verbs are external, whereas subjects of unaccusative verbs are internal arguments. This generalization will play an important role in Subsection III, where it will be argued that putatively intransitive verbs like brandento burn and smeulento smolder, which can take inanimate subjects, are in fact unaccusative.

Subsection A has shown that intransitive and transitive verbs normally denote activities; the external arguments of such verbs refer to agents, i.e. entities that perform these activities. It is therefore not surprising that many of these verbs can be input of er-nominalization, i.e. the morphological process that derives agentive person nouns by suffixing the verb stem with the affix -er (or one of its allomorphs); cf. Sections N14.3.1.5 and N15.2.3.1. In (34a&b) we give some examples with transitive verbs. Note, however, that there are also many transitive verbs, such as groetento greet in (34c), which for unclear reasons do not easily allow er-nominalization (although groeter is occasionally used in jest and has recently been introduced as a translation of English greeter, which denotes volunteers who give free tours to tourists; cf. en.wikipedia.org/wiki/Greeter).

Note that the direct object of the verb can be expressed by a postnominal van-PP in the er-nominalization in (34). Occasionally, the postnominal van-PP is omitted, and a habitual or occupational reading, also found with the pseudo-intransitive verbs in (24), is likely to emerge.

The vast majority of regular intransitive verbs also allow er-nominalization; some examples are given in (36).

Unaccusative verbs, on the other hand, usually do not allow er-nominalization, as shown in (37). Apparently, having an external (i.e. agentive) argument is a necessary condition for er-nominalization, and unaccusative verbs do not satisfy this condition.

The conclusion to be drawn from the above discussion is presented in (38). Recall that the term intransitive verb is used as a substitute for the term unergative verb, which is often preferred in comparative linguistic studies, for the reasons given in the introduction to Section 2.1.

The examples in (39) are exceptions to the generalization in (38). The unaccusative verbs stijgento ascend and dalento descend in (39a) seem to allow for er-nominalization, but the resulting er-nouns have a lexicalized meaning in that they are only used in the context of a listing or competition (as in sports, charts, or financial indexes); they can refer to a stock that has gone up/down in value, but not to the subject in an example such as Het vliegtuig/De piloot stijgtthe airplane/pilot is going up. Similarly, the noun groeier in (39b) is used with adjectives like langzaamslow to refer to e.g. ornamental plants or businesses, not just anything that grows, and the noun blijvertje in (39c) refers to something of a more permanent nature, not just any entity that stays in a certain place.

We seem to be dealing here with jargon or more or less idiomatic expressions, so we will ignore such cases and simply assume that generalization I in (38) holds in full force in the core grammar; cf. Section N14.3.1.5 for further discussion of agentive er-nouns.

Dutch perfect tenses are formed with either hebbento have or zijnto be. However, transitive verbs seem to take only hebben.

The monadic verbs differ in the auxiliary they take: intransitive verbs always take hebben, while unaccusative verbs are often assumed to typically take zijn instead.

The conclusion we can draw from the examples in (40) and (41) is that (in)transitive verbs must take the auxiliary hebben in the perfect tense. The data in (42) suggests that unaccusative verbs (i.e. verbs that do not select an external argument) must take the auxiliary zijn in the perfect tense, but we will see in Subsection III, that this latter claim is not a hard-and-fast rule; the correct generalization therefore seems to be as given in (43).

Past/passive and present participles are often used in prenominal attributive position as modifiers of a noun. This subsection will show that past/passive participles can only modify a head noun corresponding to an internal (theme) argument of the verb. This restriction does not apply to present participles.

Passivization is typically associated with (di)transitive verbs. While it is certainly not true that all transitive verbs can be passivized (cf. Section 3.2.1.1, sub III), many do allow this option; some examples with transitive verbs are given in (51).

However, it is by no means true that passivization is restricted to (di)transitive verbs; the examples in (52) show that intransitive verbs can also be passivized. Since the passive constructions in the primed examples have no subject (nominative argument), they are usually called impersonal passives. Note that the regular subject position in these impersonal passives is occupied by the expletive element erthere.

Unaccusative verbs differ from intransitive verbs in that they do not allow impersonal passivization. Some examples illustrating this are provided in (53). Note that we have selected examples with human subjects because it is often claimed that there is an animacy restriction on passivization, in the sense that clauses with a [-animate] subject cannot be passivized.

The data in this subsection suggests that the presence of an external argument is a necessary condition for the passivization of a verb. If no external argument is present, as in the case of unaccusative verbs, passivization is blocked. This leads to generalization IV.

For a more detailed discussion of the restrictions on passivization, see Section 3.2.1.

The so-called wat-voor split has played a prominent role in the literature on Dutch unaccusative verbs. A wat-voor phrase is an interrogative noun phrase consisting of the sequence wat voor (een)what for a followed by a noun. Like all interrogative phrases, the complete noun phrase can be placed in the clause-initial position, as shown in (55a). The term wat-voor split refers to the fact that it is also possible to split the wat-voor phrase and place the interrogative element wat in clause-initial position, while stranding the remainder of the phrase, as in (55b). We refer the reader to Section N17.2.2 for a more detailed discussion of wat-voor phrases.

What is relevant here is that it has been claimed that the wat-voor split is only possible when the split noun phrase is an internal argument (direct object), as in (55b). Indeed, when the split is applied to an external argument, the result seems to be severely degraded; cf. Den Besten (1985). This is shown in (56b).

If the generalization that the wat-voor split is only possible with internal arguments is correct, it is predicted that the subject of an unaccusative verb can undergo it, whereas it is blocked in the case of an intransitive verb. Things are not so simple, however, since it has been suggested that the degraded status of (56b) is not due to the fact that the wat-voor phrase is an external argument, but to the fact that it is an indefinite noun phrase; in many cases, indefinite subjects require the presence of the expletive element erthere; cf. Broekhuis (1992). Although the judgments of native speakers seem to vary, example (56b) seems to improve considerably if the expletive er is added, as in (57).

Although this observation makes it rather doubtful whether appealing to the wat-voor split can help us to distinguish between intransitive and unaccusative verbs, we will nevertheless show how these verbs behave in this respect. As expected, the primed examples in (58) show that unaccusative verbs do indeed allow the wat-voor split; when the expletive er is omitted, all the examples in (58) become unacceptable.

Applying the wat-voor split to intransitive verbs yields a result that is perhaps a bit more marked, but it seems an exaggeration to declare them ungrammatical. The examples in (59) also become degraded when the expletive er is omitted, but we have not indicated this for the sake of clarity of presentation.

The hypothesis that intransitive and unaccusative verbs differ in that the former take an external and the latter an internal argument is supported by the data in this subsection only in as far as example (57) and the primed examples in (59) are marked for some speakers. The wat-voor split is therefore not a very useful test for distinguishing intransitive from unaccusative verbs.

Table 2 summarizes the discussion in the previous subsections. Row 1 indicates whether the verb takes an external and/or an internal argument, and relates this to the thematic role of the referent of the argument(s) in the event denoted by the verb. Row 2 shows that verbs can only be input of agentive er-nominalization if they take an external argument; the derived noun refers to the entity performing the activity denoted by the verb stem. Row 3 shows whether the verb takes the auxiliary hebben or zijn in the perfect tense. Row 4 shows that the past/passive participle can only be used attributively if the modified noun corresponds to an internal argument of the base verb. Row 5 shows that (impersonal) passivization is only possible if the verb takes an external argument. Finally, Row 6 shows whether the argument(s) of the verb allow a wat-voor split.

This table nicely illustrates the relationship between the type(s) of argument that the verb takes and the properties discussed. At least the material implications in (60) seem to hold. Note that we do not include the wat-voor split in this list, because it is not obvious that it really determines whether we are dealing with an internal argument; the data is simply not clear enough to assert this.

The material implications in (60) are given in their present form on purpose; they express that the consequence (= the part after the arrow) is a sufficient but not a necessary condition for the antecedent (= the part before the arrow) to hold: for instance, (60b), expresses that a verb selecting zijn in the perfect tense cannot have an external argument, but does not exclude the possibility that additional conditions must be met in order to license zijn (i.e. that not all unaccusative verbs takes zijn). The material implications in (60) therefore correspond to the generalizations I-IV formulated in the previous subsections, repeated here as (61).

Subsection IIA has shown that intransitive and transitive verbs typically denote activities, and that the subjects of these verbs are therefore typically agentive in nature. However, this is not the case with the verbs in (63); instead, they seem to refer to a process and their subjects function as themes, i.e. refer to a participant that is undergoing the process; cf. Perlmutter (1978). This conclusion is supported by the fact that examples such as (64) show that the subject need not refer to a [+animate] participant in the event. Both the process reading and the fact that the subject can be inanimate support the hypothesis that the verbs in (63) are unaccusative in nature and thus do not take an external (agentive) argument.

Another fact supporting this hypothesis is that the verbs in (63) do not normally occur in imperatives; this is illustrated in (65) for success imperatives. Section 1.4.2 has shown that while (pseudo-)intransitive verbs can easily occur in this imperative construction, unaccusative verbs cannot; the verbs in (63) correspond to the unaccusative verbs in this respect.

Since agentive er-nominalization requires as input a verb that selects an agentive (hence external) argument, we expect that the verbs in (63) cannot undergo this process. The examples in (66) show that this expectation is indeed borne out; the intended interpretations of the er-nouns are given in square brackets.

The number signs in (66a-c) indicate that the marked er-nouns occur, but not with the intended meaning. Bloeder, for instance, is a somewhat outdated noun referring to a person suffering from haemophilia. Brander is also possible, but it denotes an instrument used to remove paint by heating it (and may in fact be derived from the causative counterpart of the verb we are discussing here). Drijver is possible on more or less the intended reading (e.g. it can be used for a quill used in fishing), but it is not the case that anything that floats can be denoted by it. This means that the verbs in (63) cannot be input for the otherwise quite productive morphological rule that derives agentive er-nouns from intransitive and transitive verbs. This is another argument for assuming that these verbs are unaccusative.

At first glance, auxiliary selection seems to provide evidence against the hypothesis that we are dealing with unaccusative verbs in (63); the examples in (67) show that these verbs select hebben just like intransitive verbs.

There is reason to believe that this difference in auxiliary selection between unaccusative verbs such as vallento fall (class I) and unaccusative verbs like bloedento bleed and drijvento float (class II) is aspectual in nature. The processes denoted by the first class of verbs are usually construed as inherently bounded in time; they are telic (from Greek telosgoal) in the sense that they are construed as involving an inherently given endpoint at which a particular resulting state is achieved. The processes denoted by the second class, on the other hand, are construed as unbounded; they are atelic in the sense that no endpoint is implied.

The contrast between the two classes of unaccusative verbs will therefore follow if we assume that the selection of zijn is a special property of telic unaccusative verbs; all other verbs select hebben. The hypothesis that telicity is involved in auxiliary selection is supported by the fact, illustrated in (68), that making the events denoted by bloeden and drijven telic, e.g. by adding a resultative predicate such as dooddead or a particle such as wegaway, forces the use of zijn; cf. also Kern (1912:220).

The fact that the examples in (68) are grammatical at all is actually a second argument for assuming that verbs like bloeden and drijven are unaccusative. With intransitive verbs, the addition of a resultative predicate goes hand in hand with the addition of a second participant in the event structure; example (69a), which involves the intransitive verb huilento cry, is ungrammatical without the noun phrase zijn ogenhis eyes. With unaccusative verbs, on the other hand, the addition of a second noun phrase is excluded, as shown in (69b); cf. Levin and Rappaport Hovav (1995: §2) for further discussion.

If verbs like branden and drijven are indeed unaccusative, we correctly predict that the introduction of a second participant in (70) will also lead to an ungrammatical result. We will return to such examples in Section 2.2.

This subsection has shown that the selection of the perfect auxiliary zijn is not a necessary but a sufficient condition for assuming unaccusative status for a verb; atelic unaccusative verbs select hebben, just like the intransitive verbs. Section 2.1.3 will further support this conclusion by showing that the so-called nom-dat verbs, which are generally considered to be dyadic unaccusative verbs, also take hebben in the perfect tense if they are atelic. The claim that the selection of zijn is not necessary for assuming unaccusative status was first made in Mulder & Wehrmann (1989) on the basis of independent evidence concerning location verbs, which will be reviewed in Section 2.2.3, sub IIC1.

Subsection IID has shown that intransitive and unaccusative verbs differ as to whether the past/passive participle of the verb can be used attributively; past/passive participles of unaccusatives can be used in this way, but those of intransitives cannot. In terms of this test, the verbs in (63) match the intransitive verbs, not the unaccusative ones.

However, the ungrammaticality of the examples in (71) is again related to a difference in telicity. An example such as de gearriveerde gasten suggests that the guests have reached the endpoint implied by the verb arriverento arrive. Since verbs like bloeden and drijven do not have such an implied endpoint, the examples in (71) are semantically anomalous. As expected under this proposal, the telic examples in (68) do allow the attributive use of the participles (provided that the secondary predicate or particle is also present):

The claim that the attributive use of past participles of unaccusative verbs is sensitive to the telicity of the verb is supported by the discussion in Section 2.1.3, where it will be shown that nom-dat verbs allow the attributive use of their past participles if they are telic, but not if they are atelic.

Subsection IIE concluded that the presence of an external argument is a necessary condition for passivization. If the verbs in (63) are indeed unaccusatives, they do not have an external argument, and therefore we expect that passivization is excluded. The examples in (73) show that this expectation is indeed borne out. Note that we have selected examples with human subjects, because it is often claimed that there is an animacy restriction on passivization, i.e. clauses with a [-animate] subject cannot be passivized.

Although we have seen that the wat-voor split is not a very reliable test for distinguishing between intransitive and unaccusative verbs, we will give the relevant data here for the sake of completeness. The questions in (74) shows that the wat-voor split is possible with the subject of unaccusative verbs like bloeden and drijven, provided that the expletive er is present.

The data in this subsection has shown that the verbs in (63) are a separate class of unaccusative verbs, which differ in their aspectual properties from the unaccusative verbs discussed in Subsection II: whereas the latter are telic, the verbs in (63) are all atelic. The fact that the verbs in (63) do not select zijn in the perfect tense is probably related to their atelicity; the same may be true for the fact that the past participle of these verbs cannot be used attributively. Further support for the claim that the verbs in (63) are unaccusative can be found in Section 2.2.3, sub IIB2.