Probably the best known and most influential classification of main verbs is the one developed in Vendler (1957), which distinguishes the four aspectual classes in (54).

Vendler argues that activities and accomplishments can be grouped together as processes and that states and achievements can be grouped together as non-processes, as shown in Figure 2.

The distinctions shown in Figure 2 are based on a number of semantic properties, which will be discussed in the following subsections.

We have labeled the top node in Figure 2, which is repeated below for convenience, not verbs but states of affairs. The reason for this is that, although Vendler seems to have set out to develop a classification of verbs, he actually came up with a classification of states of affairs; cf. e.g. Verkuyl (1972) and Dowty (1979).

For example, it seems impossible to classify the verb schrijvento write without additional information about its syntactic environment. The judgments on the use of the adverbial phrases of time in example (63) show that schrijven functions as an activity when it is used as an intransitive verb, but as an accomplishment when it is used as a transitive verb.

However, it is not simply a matter of the adicity of the verb. First, the examples in (64) show that properties (here: specificity) of the direct object can also play a role; cf. Verkuyl (1972/1993), Dowty (1979) and Dik (1997). The (a)-examples illustrate this with the contrast between the bare plural noun phrase boekenbooks and the plural noun phrase preceded by a cardinal numeral, twee boekentwo books. The (b)-examples illustrate the same with the contrast between noun phrases headed by the non-count noun spaghetti and the singular count noun bordplate, respectively.

A similar effect can occur with verbs such as ontploffento explode. If the subject is a singular noun phrase, we are dealing with a momentaneous ( extremely brief) event, i.e. an achievement. However, if the subject is a definite plural, the adverbial test suggests that we may also be dealing with an activity, and if the subject is an indefinite plural noun phrase, the adverbial test suggests that we must be dealing with an activity.

Second, the addition of elements other than objects also has an effect on interpretation; for instance, the examples in (66) show that the addition of a complementive such as naar huisto home or a verbal particle such as terugback turns an activity into an accomplishment.

Third, the examples in (67) illustrate that the categorial status of the complement of the verb can also affect the aspectual nature of the event: whereas the nominal complement in (67b) triggers an accomplishment reading, the PP-complement triggers an activity reading.

The examples in (68) show a somewhat similar alternation between states and activities. The (a)-examples show that if the verb denkento think takes a propositional complement like a clause, it cannot occur in the progressive aan het + infinitive + zijn construction, and we may therefore conclude that we are dealing with a state. The (b)-examples show that if the verb denken selects a PP-complement, it can occur in the progressive construction, and that we are thus dealing with an activity. The (c)-examples show that we get a similar meaning shift if we supplement the verb with the verbal particle na.

This subsection has shown that Vendler’s classification does not involve verbs, but larger constituents including at least the internal arguments of the verb; it provides a classification of the state of affairs denoted by these phrases.

The previous subsections have briefly discussed some distinctive semantic properties of verbs and events that Vendler (1957) used to motivate his classification in Figure 2. This discussion leads to the following characterizations of the four subclasses.

One problem with this classification is that the features used are in fact more broadly applicable than just for making the distinctions given in (69). The feature [±bounded], for example, may be as relevant to states and achievements as it is to activities and accomplishments: [±bounded] can group states and activities as unbounded, and accomplishments and achievements as bounded states of affairs. This can be demonstrated by the examples in (70), which show that states behave like activities in that they can be used in perfective questions of the form Hoe lang auxfinite ... V? For how long did ... V ...?, whereas accomplishments and achievements cannot.

If, on the other hand, an interrogative phrase refers to a specific time, the acceptability judgments are reversed. This is shown in (71) with the adverbial phrase hoe laatat what time.

Distribution patterns like these suggest that the four verb classes can be defined by a binary feature system of the form given in Table 4, where the features [±bounded] and [±continuous tense] can be construed as given in Figure 2; cf. Verkuyl (1993).

	–bounded	+bounded
–continuous tense	states	achievements
+continuous tense	activities	accomplishments

Note that the feature [±bounded] seems to correlate with other semantic properties of the events. Accomplishments such as opetento eat up and achievements such as ontploffento explode in (72) both indicate that some participant in the event (here, the object and the subject, respectively) undergoes a change of state and that the attainment of the new state marks the end of the event; the only difference is that the transformation takes some time in the former case, but is perceived as instantaneous in the latter case.

Activities and states, on the other hand, typically do not involve a change of state and refer to more or less homogeneous states of affairs with the result that the end of these states of affairs is more or less arbitrarily determined. This shows that it is not clear a priori whether the feature [±bounded] would be the correct feature; it could just as well be [±change of state], as shown in Table 5. It is therefore not surprising that there are a variety of binary feature systems available; cf. Rosen (2003: §1.3) for a brief discussion of some other proposals.

	–change of state	+change of state
–continuous tense	states	achievements
+continuous tense	activities	accomplishments

Other alternatives to Vendler’s classification come easily to mind. In Figure 3, based on Smith (1991) and Dik (1997), the basic division is between states and events: states lack internal dynamism in that they require no input of energy, since nothing changes while they exist (Lehmann 1999:44), while events have some form of internal dynamism. Events can be further subdivided on the basis of their boundedness: activities are not inherently bounded, whereas accomplishments and achievements are. The latter two differ in that only the former evolve over time. This leads to the hierarchical, or at least more layered, classification in Figure 3.

As this is not the place to discuss the pros and cons of the available feature systems, we simply summarize some of the conspicuous properties of the verb classes as distinguished in Vendler (1957) in Table (73); we refer the reader to Miller (1999) and Rosen (2003) for more discussion.

	state	activity	accomplishment	achievement
dynamic	—	+	+	+
bounded/change-of-state	—	—	+	+
punctual	—	—	—	+
continuous tense	—	+	+	—

Subsection II has shown that Vendler’s classification can also be expressed by appealing to the roles that nominal arguments play in the event and it considered a number of advantages of this shift in perspective. Another potential advantage is that activities, achievements and accomplishments no longer need to be defined in terms of the temporal feature [±time extension]. This allows us to solve the problem for Vendler’s original proposal that there is a class of achievements that have a temporal extension: verbs like afkoelento cool, smeltento melt, and zinkento sink are not momentaneous, but involve a gradual change of state; cf. Dowty (1979: §2.3.5). Moreover, we can now also define the class of so-called semelfactive verbs like kloppento knock, krabbento scratch and kuchento cough as instantaneous activities. Finally, we can also understand that accomplishments like een boek schrijvento write a book and een raam brekento break a window differ in their temporal extension. In short, the aspectual feature [±time extension] can be used to divide all three main event types into two subclasses.

	[-time extension]	[+time extension]
activities	kloppen ‘to knock’ kuchen ‘to cough’ knipogen ‘to wink’ rukken ‘to jerk’	dragen ‘to carry’ lachen ‘to laugh’ luisteren ‘to listen’ wachten (op) ‘to wait (for)’
achievements	arriveren ‘to arrive’ herkennen ‘to recognize’ ontploffen ‘to explode’ overlijden ‘to die’	afkoelen ‘to cool’ smelten ‘to melt’ verdorren ‘to wither’ zinken ‘to sink’
accomplishments	doorslikken ‘to swallow’ omstoten ‘to knock over’ verraden ‘to betray’ wegslaan ‘to hit away’	bouwen ‘to build’ opeten ‘to eat up’ oversteken ‘to cross’ verbergen ‘to hide’

Note that our discussion above abstracted from the fact that properties of the nominal arguments of the verb can affect the temporal interpretation: for example, crossing a square has a temporal extension, while crossing a line is instantaneous. The three classes of non-momentaneous verbs in Table (84) are easy to recognize, since they can always be the complement of the inchoative verb beginnento begin.

The momentaneous verbs, on the other hand, usually do not allow this, except when they can be repeated and thus receive an iterative reading when combined with a durative adverbial phrase; cf. the examples in (86).

Since momentaneous activities differ from momentaneous achievements and accomplishments in that they can typically be repeated, it is the former type, not the latter, that is typically used as the complement of beginnen.

Another way to extend Vendler’s classification is to add Dik’s (1997) feature [±control]. This feature denotes a property of the subject of the clause and expresses whether the referent of the subject can initiate or terminate the event. The examples in (88) show that this feature can be superimposed on all four subclasses; the states of affairs in the primeless examples are all controlled, whereas those in the primed examples are not.

Dik provides a number of tests that can be used to determine whether the subject is able to control the event. The first involves the use of the imperative: while controlled events allow the imperative, non-controlled events do not. This is shown in (89).

This finding is interesting because Vendler (1957) and Dowty (1979) have claimed that states cannot occur in the imperative form in their prototypical use: an example such as Ken uw rechten!Know your rights! was explained by saying that this example did not involve an order/advice to know something, but to do something that would lead to the state of knowing something. Similarly, a command such as Zit!Sit! would be interpreted as an instruction to perform an activity that would lead to assuming the desired posture. However, if gelovento believe indeed denotes a state, this cannot be maintained. Other typical states that can occur in imperatives are copular constructions with zijnto be, provided that the predicative element is a stage-level predicate (i.e. denotes a transitory property); individual-level predicates, which denote more permanent properties, usually give infelicitous results in the imperative construction. The restrictions illustrated in (89) and (90) follow if the imperative is only possible with states of affairs that are [+control].

Another context in which the difference between controlled and non-controlled events comes out clearly is in infinitival constructions such as (91), in which the implied subject PRO of the infinitival clause is interpreted as coreferential with the subject of the main verb belovento promise.

Note that this runs counter to earlier claims (e.g. Dowty 1979) that states cannot occur in this environment. The examples in (92) show that the difference between stage-level and individual-level predicates which we observed in the copular constructions in (90) is also relevant in this context.

Although some verbs require a [+control] or [-control] subject, other verbs may be more permissive in this respect; a verb such as rollento roll in (93), for example, is compatible with both a [+control] and a [-control] subject. That the referent of Jan in (93a) is a controller, but the referent of de steenthe stone in (93b) is not, is clear from the fact that the adverbial phrases opzettelijk/vrijwilligon purpose/voluntarily can only be used with the former. The examples also show that [+control] subjects are typically animate (with the possible exception of certain machines).

For completeness’ sake, note that controllability and volitionality are often seen as defining properties of the thematic role of agent; cf. Levin & Rappaport Hovav (2005: §2.3.1). The fact that the subjects of states and achievements are typically not agents, but can still have these properties if they are animate, casts serious doubt on such proposals.