Verbs in regular middle constructions are related to transitive verbs. The examples in (161) show that regular middles are like regular passives in that their subject corresponds to the direct object of the corresponding transitive verb. This is clear from the form of the pronoun in the (a)-examples and the subject-verb agreement in the (b)-examples.

The regular middles in (161) differ from passives in that they do not allow the subject of the corresponding transitive verb to be expressed by an agentive door-PP, as shown in (162). Nevertheless, the notion of agent still seems to be implied in the middle constructions, in the sense that the implied experiencer of the evaluative modifier is typically interpreted as a potential agent of the activity denoted by the transitive variant of the verb; we will return to this in Subsection C.

Regular middle formation can have subtle side effects. For instance, the regular middle construction in (163b) differs from the transitive construction in (163a) in that it licenses the use of the particle wegaway, which seems to be used especially in contexts of (excessive) consumption; cf. also the discussion of example (207c) in Subsection E. To our knowledge, such side effects have not yet been investigated, and we therefore leave them to future research.

Intransitive verbs such as lachento laugh in (164a) and PO-verbs such as wachten (op)to wait (for) in (164b) cannot undergo regular middle formation, which shows that the verb must have a nominal object that can appear as the subject of the middle construction (although Section 3.2.2.4 will show that under certain strict conditions impersonal middles can still occur).

The examples in (165) show that monadic unaccusative verbs such as vertrekkento leave and undative verbs such as weten/kennento know also resist regular middle formation. This shows that the verb must have an external argument in order to allow regular middle formation, and that it is not sufficient for a verb to have an internal theme argument; it must also be possible to realize this argument as a direct object. In fact, Subsection E will argue that it is not the notion of internal argument that is relevant for middle formation, but the notion of direct object.

Since the above has shown that regular middle formation is possible for a verb that takes a direct object, we would expect that ditransitive verbs also allow regular middle formation, but example (166b) shows that this expectation is not borne out: regular middle formation is excluded when the input verb takes a nominal indirect object. The primed (b)-example is added to show that regular middle constructions in which the indirect object is promoted to subject are also excluded.

The examples in (167) show that regular middle formation is blocked not only in double object constructions, but also in constructions with a periphrastic indirect object; regular middle constructions such as (167b) are at best marginally acceptable with the aan-PP present.

The promotion of indirect objects to subject is never allowed, not even for speakers who allow them to be promoted to subject in passive constructions, as with the verb verzoekento request when it takes an infinitival direct object clause, as in (168); while the passive construction in (168b) is possible for some speakers, the middle construction in (168c) is rejected by all speakers.

The meaning expressed by the regular middle is quite complex. The construction as a whole refers to an inherent property of the subject referent; for instance, example (169a) expresses that the wall has the inherent property that it can be painted. The adverbially used adjective gemakkelijkeasily functions as an evaluative modifier of this property attributed to the subject of the clause: the implicit experiencer of the adjective functions as a universal quantifier that extends to all relevant entities in the domain of discourse. All in all, this means that the meaning of example (169a) can be paraphrased as in (169b).

Another example is given in (170a). The proper noun Vergilius refers to a body of literary work which has the inherent property that it is easy to translate (for those with sufficient knowledge of Latin). The meaning of this example can therefore be paraphrased as in (170b), where iedereen refers to a set of persons in an established domain of discourse (e.g. students of Latin).

In short, regular middle constructions are generic in nature; the verb phrase functions as an individual-level predicate in the sense that it does not refer to a specific state of affairs, but describes an inherent property of the subject of the construction. This is further supported by the following facts.

Regular middle constructions usually contain an adverbial phrase like gemakkelijkeasily or moeilijkwith difficulty, which acts as an evaluative modifier of the property expressed by the middle verb. In (179) we provide a small sample of adjectives that can be used as adverbial modifiers in regular middles.

The adjectives in (179) are all predicative and can be predicated of an embedded clause, as shown by the fact that they can all be used as complementives in a copular construction: the adjective in (180b) is predicated of the anticipatory pronoun het, which introduces the infinitival subject clause in clause-final position.

It seems that this ability to be used as a predicate is a prerequisite for entering into the middle construction; the examples in (180) show that evaluative adverbial phrases such met gemakwith ease, which cannot be used as the complementive in a copular construction, cannot be used in middles either.

The evaluation expressed by the adjective in the regular middle construction can be positive, as in (182a), or negative, as in (182b).

The default interpretation is that the evaluation given is that of the speaker, but the examples in (183) show that this can be overridden by adding a PP headed by volgensaccording to; the nominal complement of this preposition is taken to refer to the source of the evaluation.

Adjectives such as gemakkelijk belong to a set of adjectives that optionally take an experiencer voor-PP, which is taken as the norm for the evaluation expressed by the adjective; cf. Deze som is gemakkelijk voor Janthis calculation is easy for Jan. However, the examples in (184) show that experiencer voor-PPs cannot normally be expressed overtly in middle constructions.

This may be related to the fact that the experiencer of the adjective is also interpreted as a potential agent of the transitive verb that served as input for middle formation. As a result, the restriction expressed by the experiencer PP voor Jan in (184) may be incompatible with the generic interpretation of the middle construction as a whole: if a wall paints easily or if a book translates easily, this is said to hold for all potential agents, not just for Jan. This account of the unacceptability of the voor-PPs in (184) seems to be supported by the fact that the results improve considerably when we replace the complement of the voor-PPs by a generic noun phrase, as in (185).

Note that example (184a) also improves if the voor-PP is placed in a position before the adverb, as in (186a), is given a contrastive accent, as in (186b), or is preceded by the focus particle zelfseven, as in (186c). These examples no longer have a generic interpretation: it is only for Jan that the wall is said to be easy to paint. However, there is reason to think that we are not dealing with an experiencer voor-PP in these examples, since voor-PPs can also be used as restrictive adverbial modifiers; cf. Section N15.2.1 for discussion.

That we are dealing with adverbial phrases is clear from the fact, illustrated by (187), that the adverbs in the regular middle construction allow modification. When the degree modifier tetoo is used, the experiencer can optionally be expressed as a dative phrase; since the experiencer in (187c) is also taken as a potential agent of the input verb, such cases are exceptional in that the construction is not necessarily interpreted generically.

The adverb can also appear as an equative, a comparative or a superlative, as shown in the examples in (188).

There is a smaller subset of regular middles that need not contain a modifier of the type in (179). First, there is a small set of adjectives that can be used in regular middles even though they do not normally select an experiencer voor-PP, including snelquickly, traagslowly, lichtwithout difficulty and zwaarwith difficulty. Like the adjectives in (179), they express some inherent property of the subject from the perspective of the speaker or some other entity in the domain of discourse.

Second, regular middles can contain the negative adverb nietnot. In such cases, it is expressed that the subject of the sentence does not have the property denoted by the verb phrase: example (190a) expresses that the wall cannot be painted. In (190b), the negation is contained in the idiomatic expression voor geen meterhardly at all (lit. “for no meter”).

The evaluative modifier may also be absent if the evaluation is expressed in some other way. In (191a), the contrastive accent on the verb expresses that the subject has a high degree of the property denoted by the verb. In (191b), the evaluation is expressed by comparison; the thesis is claimed to be exciting and highly readable. The comparison is sometimes idiomatic in nature; an example is given in (191c), in which the phrase als een treinlike a train expresses that the thesis has the property that it can be read quickly and smoothly.

Subsection A has shown that the input verb for regular middle formation must be transitive: intransitive, unaccusative, undative, and ditransitive verbs are all excluded. This subsection shows that there are also a number of restrictions of a more semantic nature.

All subjects of the regular middle constructions discussed in the previous subsections correspond to the theme argument of the corresponding transitive verb. From this we might hypothesize that the subject of the regular middle construction must be the internal theme argument of the verb. This in turn would predict that the middle construction leads to an ungrammatical result if the object in the corresponding transitive construction is selected by some other element in the clause. This subsection will show that, despite appearances, this prediction is not correct. First, we will consider the examples in (199) and (200), which seem to support the proposed hypothesis. The English examples in (199) show that the subject of the regular middle construction cannot correspond to an accusative noun phrase functioning as the subject of an infinitival clause.

Comparable examples cannot be given for Dutch, since it does not allow the construction in (199a). However, example (200a), in which the direct object is generally taken to be the external argument of the complementive (i.e. the predicative noun phrase/AP), has the Dutch counterpart in (200b); the primed examples show that English and Dutch behave similarly in not allowing regular middle counterparts of such constructions.

The examples in (201) further show that the same holds for AcI-constructions in which the accusative object functions as the subject of the embedded infinitive; note, incidentally, that such constructions do not enter passivization either.

The examples in (199) to (201) thus support the hypothesis that the subject of a middle verb must be an internal argument of the corresponding transitive verb. However, there is also a problem for this hypothesis: some resultative constructions do have middle counterparts, as shown by the examples in (202).

To solve this problem for this hypothesis, it has been claimed that the accusative noun phrase the metal/het metaal enters into a predicative relation not only with the adjective flat/plat but also with the verb to hammer/slaan, i.e. the accusative noun phrase is not only the logical subject of the adjective but also an internal argument of the verb; cf. Levin & Rappaport Hovav (1995: §2.2.1) and the references cited there. The empirical support for this assumption is that English (202a) is also acceptable without the adjective: John hammers the metal. However, a problem is that the supposed thematic relation between verb and accusative noun phrase is not present in all resultative constructions that have regular middle counterparts; dropping the complementive platflat in the Dutch example in (202b) actually leads to a degraded result at best, and is completely impossible in examples such as (203a). This shows that the object is not assigned a thematic role by the verbs in these examples, and thus refutes the claim that the object must be an internal argument of the transitive verb for middle formation to be possible.

Similarly, the fact that particles are often obligatory in particle-verb constructions has been used to argue that accusative noun phrases in such constructions are arguments of the particles and not of the verbs. But again, middle constructions can easily arise with such particle verbs; cf. the primed examples in (204).

For completeness’ sake, the examples in (205) show that resultative unaccusative constructions are like simple unaccusative constructions in that they do not allow regular middle formation. This again shows that middle formation requires the input verb to have an external argument.

The fact that the object promoted to subject need not be assigned a thematic role by the verb but can be introduced by some other predicative element shows conclusively that regular middle formation is not a lexical but a syntactic process. This does not mean, of course, that there are no semantic constraints on middle formation: on the contrary, the contrast between the examples in (199)/(200) and those in (202) can be made to follow from the previously established restriction that the input verb must denote an activity. Likewise, the contrast between the examples in (204) and those in (206) below shows that the middle verb must refer to some inherent property of the derived subject; cf. the discussions in Subsections A and D.

Finally, it should be noted that there are more restrictions on middle formation than those mentioned above: although opetento eat up denotes an activity and we could easily imagine that gemakkelijk opeten would refer to some inherent property of rice, the middle formation in (207b) is nevertheless excluded. This may be due to the fact that middle verbs denoting some form of consumption are often used with the particle weg; cf. example (207c) and the discussion of example (163) in Subsection A.

But there is certainly more going on, since middle formation is also excluded in resultative constructions such as (208), in which the PP onder de tafel is predicated of the noun phrase die studententhose students.

Since we have no further insights to offer at present, we leave the formulation of the precise conditions under which resultative construction may or may not undergo middle formation to future research.

Subsection B has shown that regular middle constructions normally receive a generic interpretation, as is clear from the fact that punctual time adverbs such as gisterenyesterday cannot be used, unlike time adverbs referring to a longer span of time; cf. (171b&c), repeated here as (209).

However, example (209a) improves considerably if the adverbial phrase gemakkelijk is modified by a degree modifier such as ergvery, as in (210a), or is given in a form other than positive degree, as in (210b). In such examples, the verb phrase no longer denotes an individual-level but a stage-level property of the subject; the examples explicitly compare the degree of “paintability” at different times. Adverbs like nogstill or alalready can also improve the result by explicitly stating that the relevant property has changed.

While it is easy to imagine that the degree of paintability of a given object depends on additional circumstances and can thus be construed as a stage-level property, this is less likely to be so when it comes to the degree of “translatability” of a given body of literary work. The fact that marginal examples such as ??Vergilius vertaalde gisteren gemakkelijk (lit. Vergil translated yesterday easily) in (172b) cannot be improved by degree modification or comparative formation is therefore not really surprising; examples such as (211) are unacceptable if Vergilius is construed as referring to the whole body of his works (which is the default reading when these examples are uttered out of the blue). However, these examples seem to become perfectly acceptable if the context allows one to construe Vergil’s work as divided into a number of texts of varying degrees of translatability, and if Vergilius refers to only one of these texts.

The discussion in this subsection suggests that the generic interpretation of the verb phrase in regular middles can be overridden if the adverbial evaluative modifier provides sufficient evidence that a stage-level interpretation is intended.

Unaccusative constructions can easily occur in the perfect tense. Although there are also unaccusative verbs that select the perfect auxiliary hebben, the fact that the verbs in (212b) and (213b) select zijn is sufficient to show that they are unaccusative; cf. Section 2.1.2, sub III. Note that the acceptable versions of the examples in (217) can still be ambiguous; without the adverbs automatischautomatically and vanzelfspontaneously they allow not only an unaccusative but also a passive interpretation, due to the fact that verbs like sluitento close and brekento break can also be used as transitive verbs.

Regular middle constructions do not occur easily in the perfect. This is probably due to their generic reading, since we find the same in generic examples such as (218a). However, generic sentences in the perfect tense seem to improve when we add an adverbial phrase such as altijd alall along, as in (218b).

The examples in (219a&b) show that perfect regular middle constructions also improve by adding this adverbial phrase. For our present purpose, it is important to note that the auxiliary verb in these examples is hebben. Example (219c) shows that the use of hebben is not forced by the generic meaning of the middle construction, since the generic unaccusative construction in (216a) still takes zijn.

We can conclude from the above examples that the choice of the perfect auxiliary can be used as a test to distinguish between unaccusative and middle verbs.

Past participles of unaccusative verbs can often be used predicatively in copular constructions; cf. Section A31.3. However, this is not possible with past participles corresponding to verbs in regular middle constructions, as is shown by the contrast between the primeless and primed examples in (220).

Similarly, past participles of unaccusative verbs can easily be used attributively (cf. Section A31.2), whereas this is excluded in the case of middle verbs. This is shown by the contrast between the primed and primeless examples in (221).

However, the primeless examples in (222) show that, unlike in English, present participles of both unaccusative and middle verbs can be used attributively.

For completeness, the examples in (223) show that, as usual, present participles of middle verbs cannot be used predicatively.

The examples in this subsection thus show that the ability of past (but not present) participles to occur in predicative and/or attributive position can be used as a test for distinguishing unaccusative and middle verbs.

Unaccusative and regular middle constructions differ in that the former usually refer to a specific state of affairs, whereas the latter are predominantly generic in the sense that the verb phrase denotes an inherent (i.e. time-independent) property of the referent of the subject of the construction. As a consequence, unaccusative and regular middle constructions differ systematically with respect to the properties of middles discussed in Subsection IB. First, the examples in (224) show that unaccusative and regular middle constructions differ in that only the former can easily be combined with punctual adverbial phrases; the use of such time adverbs in regular middle constructions conflicts with the fact that the verb phrase must refer to some inherent property of the subject of the construction.

Second, unaccusative and regular middle constructions differ in that only the former can be used in the progressive aan het + infinitive construction; the latter are categorically excluded in this construction.

Third, unaccusative and regular middle constructions differ in that only the former can occur as the complement of a perception verb. Subsection IB accounted for this by assuming that the tense of the infinitival clause must be linked to the tense of the main verb, and that this conflicts with the generic interpretation of the middle construction.

Note that the AcI-construction with the verb vindento consider in (227a) is perfectly acceptable. This seems consistent with the fact that the verb vinden can take propositional complements headed by individual-level predicates such as intelligent; vinden thus differs from the perception verbs in that it is compatible with the generic, individual-level meaning of the middle verbs; cf. Subsection IB for discussion.

The obligatory adverbial phrase in the regular middle construction is interpreted as an evaluative modifier of the inherent property denoted by the verb phrase. This modifier thus implicitly introduces an experiencer. As noted in Subsection IC, this implied experiencer is interpreted by default as a potential agent of the transitive verb that served as the input for middle formation. This explains why regular middle verbs cannot be used with adverbial phrases like automatischautomatically and vanzelfspontaneously, since these suggest the absence of such an agent.

The previous subsections have shown that unaccusative and middle verbs differ systematically in several ways. First, whereas unaccusative verbs can take either zijn or hebben in the perfect tense (depending on their aspectual properties), middle verbs invariably take hebben. Second, whereas past participles of unaccusative verbs can easily be used as predicates or attributive modifiers, this is not possible with past participles of middle verbs. Third, whereas unaccusatives can refer to an actual event, middles have a generic interpretation; this means that the former, but not the latter, can be modified by punctual time adverbs, occur in the progressive aan het + infinitive construction, or occur as the infinitival complement of a perception verb. Finally, modification of the middle construction by adverbs like automatischautomatically or vanzelfspontaneously leads to a degraded result, because this conflicts with the fact that the implicit experiencer introduced by the evaluative adverbial is interpreted as a potential agent of the transitive verb used as input for middle formation.

	unaccusative	middle
auxiliary selection	zijn/hebben	hebben
predicative/attributive use of past participles	+	—
event denotation	+	—
vanzelf/automatisch	+	—

We conclude our discussion of the differences between unaccusative and middle constructions with a discussion of a potentially ambiguous case. Compare the two examples in (229). The examples are similar in that they both require the adverb to be present, but we will see that they behave differently with respect to the tests in Table 2.

The first test does not distinguish between the two constructions; they both take the perfect auxiliary hebben, which is compatible with both unaccusative and middle verbs. However, the fact that (230a) is also perfectly acceptable without the adverbial modifier altijd alall along is worth mentioning, because we have seen that this adverbial is usually needed to license the perfect-tense form of generic constructions, but is an issue independent of auxiliary selection.

The examples in (231) show that the two examples in (229) also differ with respect to the predicative and attributive use of the past participle. This suggests that (229a), but not (229b), exhibits properties of unaccusative verbs.

The primeless examples in (232) strongly suggest that example (229a) refers to an actual event: although the construction does not seem very felicitous in the progressive aan het + infinitive construction, it can be used with a punctual time adverb such as vanmorgenthis morning and as the infinitival complement of the perception verb zien.

Finally, the examples in (233) show that the adverb vanzelfspontaneously can easily be added to example (229a), but not to example (229b). On the basis of the discussion above, we can safely conclude that example (229b) is a genuine case of the regular middle construction. The discussion also suggests that verkopen in (229a) functions as an unaccusative verb, despite the obligatory presence of the adverb goedwell: a conclusion also arrived at on different grounds for the English verb to sell in Keyser and Roeper (1984:394).

Finally, we note that verbs like verkopento sell and verhurento rent out differ from other transaction verbs like kopento buy, hurento rent and lenento borrow: the latter cannot enter the unaccusative construction in (234a), nor the regular middle construction in (234b). It might be useful to investigate whether the contrast is related to the fact that the former involve an (implicit) recipient/goal, while the latter involve an (implicit) source.

The regular middle construction has some similarities to the easy-to-please construction in (235b), which is discussed in detail in Section A28.5, sub IVA.

There is also a certain similarity between middles and the modal-infinitive constructions discussed in Section A28.5, sub IVB. For instance, the examples in (238) show that, like middle verbs, modal infinitives must denote an activity.

However, the unacceptability of the modal infinitives in (239) shows that the set of adverbially used adjectives that can be found in modal-infinitive constructions is considerably smaller than the set of adverbially used adjectives that can enter the regular middle construction; cf. the list of adverbs in (179).

We leave it to future research to examine the similarities and differences between the two constructions in more detail.