The core property of the passive construction is the demotion of the external argument of the active verb to adjunct status. Since unaccusative verbs have no external argument, this immediately explains why these verbs cannot be passivized. This is illustrated in (21) for the monadic unaccusative verbs stervento die and drijvento float, which select the perfect auxiliaries zijn and hebben, respectively; cf. Section 2.1.2, sub III.

Example (22) shows the same for the nom-dat (dyadic unaccusative) verbs opvallento stand out/catch the eye and tegenstaanto pall on, which select the perfect auxiliaries zijn and hebben, respectively; cf. Section 2.1.3, sub E. The singly-primed examples show that (impersonal) worden-passivization is excluded, and the doubly-primed examples show that krijgen-passivization is also excluded.

The examples in (23) show the same for the undative verbs krijgen and hebben, which are also characterized by the lack of an external argument; cf. Section 2.1.4.

The fact that the (in)transitive/unaccusative status of the verb determines whether passivization is allowed or not, makes it impossible to give an exhaustive list of verbs that do or do not allow passivization. This can be illustrated by the verb brekento break, which can be used as both a transitive and an unaccusative verb. The primed examples in (24) show that it does not make sense to say that breken does or does not allow passivization; all that can be said is that breken allows passivization if it is used transitively, but not if it is used unaccusatively.

It is generally assumed that the pragmatic function of passivization is to background the subject of the active clause; cf. e.g. Kirsner (1976a/1976b). Of course, this is especially clear when the agent is left unexpressed, but the same effect is achieved when the agent is overtly realized as an agentive door-PP. That passivization has this effect is related to the fact that the subject position of a clause is a typical topic position; by removing the agentive argument from this position, it is less likely that its referent will be construed as the entity the discourse is about. This can be illustrated by the examples in (25); the question introduces Jan as a new discourse topic, which is presented as such in the primeless but not in the primed (b)-example.

The demoted subject of the active construction may remain implicit in the passive, but can usually also be made explicit by an optional agentive door-PP; cf. the examples in (26), repeated from (17) to (19). Note in passing that door-PPs in krijgen-passives such as (26c') often alternate with van-PPs, which differ from door-PPs in that they do not refer to an agent but to a source.

An exception to the general rule that an agentive door-PP can be added is the generic pronoun men in (27a); the reason is that men can only be used as the subject of a finite clause; cf. Section N18.2.1.1, sub I.

It is not entirely clear whether the same applies to the generic pronoun jeone. An example such as (28a) can be passivized while maintaining the generic reading, but it is less clear whether the implied agent of (28b) is understood as identical to the inalienable possessor of the teeth. The number sign in (28b) further expresses that the addition of a door-PP with the pronoun je is possible, but incompatible with the intended generic reading; it seems that the pronoun can only be interpreted as referring to the addressee.

That the agent is implicitly present, even if the door-PP is not used, is supported by the distribution of agent-oriented adverbs such as expres/opzettelijkdeliberately. Consider first the primeless examples in (29). These examples show that these adverbs require the subject of the active clause to be an agent, as in (29a); if the subject of the clause is a theme, as in (29b), the use of these adverbs leads to an unacceptable result. The fact that expres/opzettelijk can be used in passive constructions such as (29a') thus suggests that the agent of the active sentence is still implicitly present.

Something similar is shown by the interpretation of the implied subject PRO in infinitival clauses. The primeless examples in (30) show that PRO must be controlled by an appropriate constituent in the main clause; the infinitival verb pestento pester requires an agentive subject, and this condition is met in (30a), where PRO is controlled by the [+human] argument Jan, but not in (30b), where PRO is controlled by the [-animate] argument het bordthe plate. The fact that the passive construction is perfectly acceptable in (30a') again suggests that PRO is controlled by an implicit agent argument.

Slightly more controversial examples are given in (31): in (31a) the reciprocal elkaar seems to be bound by of the implicit agent, and in (31b) the supplementive naaktnude seems to be predicated of the implicit agent. The percentage signs indicate that not all speakers accept such examples.

Examples such as (31) are generally considered best in generic contexts. They also require that there be no other nominal argument present that could be the antecedent of elkaar or to which the property denoted by the supplementive could be attributed; this is clear from the fact that while the primeless examples in (32) are ambiguous, the primed examples are not. Note that we have indicated both the binding and the predication relation by using indices.

The marked status of the examples in (31), as well as the fact that it is impossible to establish a binding/predication relation with the (implicit) agent in the primed examples in (32), suggests that we are actually dealing with ungrammatical structures that are nevertheless accepted by some speakers because they can easily be assigned a feasible interpretation thanks to the presence of the implicit agent. This shows that, regardless of their precise grammaticality status, the examples in (31) provide evidence in favor of an implicit agent in passive constructions.

The implicit agent in impersonal passive constructions is preferably interpreted as [+human]. This is clear from the fact that (33b) cannot easily be interpreted as the passive counterpart of (33a); (33b) instead implies that the agent is [+human]. The only way to override this reading is to express the [-human] agent overtly by an agentive door-PP, as in (33b'). We have added the % sign to this example because similar examples are considered unacceptable in Pollmann (1970/1975) and Kirsner (1976a), but all our informants accept this example.

The claim that the implicit agent is preferably construed as [+human] also explains the fact reported in Haeseryn et al. (1997:1417) that speakers tend to object to the primed examples in (34): since the activities denoted by the verbs grazento graze and kwakento quack are not normally performed by humans, a [+human] interpretation of the implicit agent leads to a semantically incoherent result. When the [-human] agent is expressed overtly by a door-PP, these passive constructions again become perfectly acceptable to our informants.

Note that the preference for an implicit [+human] agent does not hold in constructions such as (35) in which the passive verb is transitive; these examples are perfectly acceptable to all speakers, despite the fact that the default interpretation is that the agent is non-human.

According to the more or less standard account of passivization in generative grammar (Jaeggli 1986 and Roberts 1987), the agent (external argument of the verb) is never left implicit: it is syntactically realized as passive morphology on the passive participle; cf. Subsection V for further discussion. If this is correct, the semantic effects in (33) and (34) can be accounted for by assuming that the default interpretation of the passive morphology is [+human]. However, this would raise the question as to why we do not find a similar effect in (35). This may be related to the fact that providing the right contextual information is often sufficient to override the default [+human] interpretation of the implicit argument, as can be seen in the following example taken from a story about sparrows in a bird journal by Adri de Groot (vogeldagboek.nl/html/Vogeldagboek/2002/Jun02_Lot2.html), in which the impersonal passives are italicized; the translation is given in the active form.

Note in passing that the claim that the agent is syntactically expressed by the passive morphology implies that the optional door-PP cannot be seen as an alternative realization of the agent, but simply functions as an adjunct providing additional descriptive information about the external argument expressed by the passive morphology on the participle. So, there is no need to have a syntactic rule of subject demotion that places the subject of the active clause in an agentive door-PP in the passive construction (as was assumed in early generative grammar).

Although the hypothesis seems well-founded that the presence of an external argument is a necessary condition for passivization, it is not clear whether the presence of an external argument is a sufficient condition for passivization. It may be that passivization requires the external argument to satisfy a number of additional constraints. The following subsections discuss three such constraints that have been proposed in the literature, but the conclusion will be that there is little evidence to support them.

Since passivization results in the promotion to subject of one of the objects of the active verb (if there is one), it is sometimes proposed that one of the functions of passivization is the “externalization” of internal arguments of the active verb. This hypothesis seems to account correctly for what happens in (45): if the internal theme argument of the verb onderzoeken were externalized, it would be predicted to be realized as the subject of the clause.

This hypothesis could also account for the examples in (46), but only on the assumption that the subject of the small clause is assigned a (possibly secondary) thematic role by the verb slaan’ to hit.

However, appealing to a secondary thematic role assigned by the verb seems completely beside the point in describing the change in the (b)-examples; Section 2.2 has argued that the accusative DP de kruimelsthe crumbs in (45b) is not an internal argument of the verb vegen, because it requires the presence of the complementive PP van de tafel affrom the table, i.e. it is semantically licensed as the logical subject (external argument) of the predicate of the small clause, not as the complement of the verb vegento wipe.

We conclude that subjects of passive constructions are not external arguments of the passivized verb: the only thing that the examples in (45) to (47) show is that, unlike the active verb, the passive participles are not able to assign accusative case to the noun phrases acting as direct objects in the active construction, which must therefore be promoted to subject in order to recieve nominative case.

The examples in (48) show that in passive constructions the main verb usually takes the form of a passive participle. This has led to the hypothesis that it is the morphology of the participle that is responsible for the demotion of the external argument, as well as the concomitant promotion of one of the objects in (48b&c).

As already mentioned in Subsection II, the standard approach to passivization in generative grammar (Jaeggli 1986 and Baker et al. 1989) is that the passive morphology on the participle is actually the external argument of the verb. The fact that the passive morphology reduces the case-assigning property of the main verb is then accounted for by assuming that the “missing” case is assigned to the external argument, i.e. to the passive morphology itself. This is sometimes called case absorption.

Although this hypothesis seems to account for the majority of cases, it has been contested on the basis of AcI-constructions such as (49) with an infinitival transitive verb; cf. De Geest (1972), Vanden Wyngaerd (1994) and Bennis (2000). The crucial thing is that example (49a), in which all arguments of the infinitival transitive verb zingento sing are expressed, alternates with example (49b) in which the external argument de kinderenthe children seems to be demoted: it can be omitted or expressed by an agentive door-PP.

If demotion of the external argument is indeed the defining property of passivization, we should conclude that the infinitival clause in (49b) is the passive counterpart of the infinitival clause in (49a). This conclusion is supported by the fact that the alternation is also possible with intransitive but not with unaccusative verbs; this is illustrated in (50) for the intransitive PO-verb kijken naarto look at and the unaccusative verb verdwijnento disappear.

If we are really justified in considering the infinitival clauses in the primed examples of (50) as the passive counterparts of the infinitival clauses in the corresponding primeless examples, which remains to be more firmly established, we may conclude that passive morphology is not a defining property of passivization. The question as to what determines the morphological form of the verb must then be considered an unsolved problem; we refer the reader to Section 5.2.3.3 for further discussion of examples like (49) and (50). For completeness’ sake, note that for as yet unclear reasons the alternation with intransitive verbs without a PP-complement(ive) are often marked.

Some sentences are ambiguous between a regular and an adjectival passive reading. The ambiguity is due to the fact that past/passive participles can be interpreted as either verbal or adjectival elements. The verbal/adjectival nature of the participle can be determined by its position relative to the clause-final verbs: verbal participles can either precede or follow them, while adjectival participles must precede them. See Section 6.2.2 for a more detailed discussion of the word order in the clause-final verb cluster of passive constructions.

The regular and adjectival passive also differ semantically in that the verbal passive has a dynamic reading (verbal participles denote events), whereas the adjectival passive has a stative reading (adjectival participles denote properties of the subject of the clause). This can be made clear by adding adverbial phrases that favor one of the readings. Adverbial phrases such as al jarenfor years, for example, favor the stative reading and therefore cannot be added to (52a), which is necessarily interpreted as a verbal passive. This is shown in (53).

Adverbial phrases such as gisterenyesterday, on the other hand, favor the dynamic reading and thus block the adjectival reading of (52b); example (54b) can only be interpreted as a verbal passive construction.

The examples in (55) show that the adjectival reading can also be blocked by the presence of an agentive door-PP.

The fact that the adverbial phrase al jarenfor years favors an adjectival reading of the participle, while the agentive door-PP favors a verbal reading explains why they cannot co-occur in a single clause, as shown by the unacceptability of (56a). Since the adverbial phrase gisterenyesterday and the door-PP both favor a verbal reading, these two can easily co-occur, as shown by (56b).

The adjectival passive construction is usually analyzed as a copular construction. The main difference between the passive and copular constructions lies in the status of the complement of the verbs worden and zijn: passive constructions contain a verbal participial phrase, while copular constructions contain an adjectival participial phrase. See Section A31 for a more detailed discussion of the difference between the verbal and adjectival readings of past/passive participles.