In her investigation of the semantics and the cross-linguistic realization of PM, Lekakou claims that the following properties of PM are shared across languages (Lekakou 2002; 2004; 2005a; 2008; see also A & S 2006; 2015):

Lekakou (2004; 2005; 2008) argues that we can reduce the properties in (9) to a single one, which amounts to a characterization of PM as disposition ascriptions, as in (10); cf. Cohen (this volume) for a characterization of PM as passive dispositions:

Since the notion of a disposition ascription is taken as central for the semantics and the syntax of PM, we briefly present how Lekakou (op.cit.) treats disposition ascriptions, and how the proposal applies to PM. In section 2.2. we turn to the morphosyntax of PM in German and Dutch.

Building on literature that investigates the semantics of generic sentences (see e.g. Krifka et al. 1995), Lekakou (2004; 2005a) identifies disposition ascriptions (DA) as a type of generic sentence, or an interpretation that generic sentences may receive. Crucially, this interpretation is distinct from the habitual one.4 Sentences such as the ones in (11), from Krifka et al. (1995) (see also Menéndez-Benito 2013 for more recent discussion) for instance, are ambiguous between a habitual reading, which involves quantification over events, and a dispositional interpretation, which states the potential for the subject referent to act in the manner stated by the predicate. In contrast to the habitual interpretation, on the dispositional reading, the truth of the sentence does not require that such events have (regularly) taken place. For instance, in (11a) it is not the case that the regular or habitual speed of the car is 160 mph, and (11b) can be true in the absence of any mail from Antarctica, solely on the basis of Mary’s job description.

Lekakou (op.cit.) argues that PM pattern semantically with dispositional generics. She notes that in English, non-episodic passives and PM are not semantically identical, despite their apparent similarity in terms of their non-episodic nature, as well as the similar argument structure realization. Lekakou (2005a: 99) suggests that non-episodic passives in English only receive a habitual interpretation, and PM the dispositional interpretation. She illustrates with the pair of examples in (12), which are not paraphrases of each other: (12a) states that in general people do not recycle plastic cups easily, and (12b) states that plastic cups have a property that makes recycling them difficult.

Since PM and passives do not make the same meaning contribution, conjoining the two as in (13) does not lead to a contradictory statement (Lekakou 2005a: 100):

Lekakou argues that the two constructions have a different semantic structure (and, for languages like English, a different syntactic structure, too). For habituals, a quantificational treatment is adopted that employs a covert sentence-level operator with quasi-universal force, which quantifies over possible worlds (cf. Krifka et al. 1995). In line with the Kratzerian semantics for modal operators, this generic operator is restricted to quantify over worlds in the modal base, and the worlds quantified over are ordered with respect to an ordering source (see Kratzer 1981; 1991 and Krifka et al. 1995 for details). For PM, and dispositional generics more in general, Lekakou proposes that a different generic operator is involved, whose semantics is modeled on the semantics proposed by Brennan (1993) for dynamic will. We provide here the gist of Brennan’s analysis, taken over by Lekakou. For the details, the interested reader is referred to the original works (Brennan 1993; Lekakou 2005a: Chapter 2).

Brennan (1993) distinguishes dynamic modals (namely, ability can and dispositional will) and epistemic modals semantically and syntactically. Whereas epistemic modals are sentence-level operators, dynamic modals relate properties and individuals and, as such, are VP-modifiers. Ability can and dispositional will realize VP-level operators (the former existential in force and the latter universal). An additional difference between epistemic and dynamic modals lies in the nature of their respective modal bases: whereas S-level operators are restricted by propositions (introduced by in view of, as in Kratzer’s original proposal), VP-level operators are restricted by properties of the subject introduced by ‘in virtue of’. An example of a conversational background which restricts a VP-level operator (in this case, by making reference to the subject’s physical properties) is given in (14) (from Brennan 1993: 65):

Directly related to the proposal that VP-level operators are restricted by properties of the subject is the fact that dynamic modals are subject-oriented. The main evidence presented here on this point is the distribution of in virtue of phrases (IVOs) with different kinds of modals.

IVOs in general need not be subject-oriented, as shown by example (15). The sentences in (16) show that, when combined with S-level modals (e.g. epistemics), IVOs cannot be subject-oriented. By contrast, when combined with VP-level operators, IVOs are obligatorily subject-oriented. The pairs in (17) and (18) illustrate this.

Lekakou suggests that what Brennan argues for dynamic modals carries over to PM (and to dispositional generics in general). The intuition, in other words, is that the generalization made by PM, as in e.g. (12b), relies on properties inherent in the subject referent. Concretely, Lekakou’s proposal is that the generic operator in middles is a VP-level operator with semantics identical to that of dispositional will; in this case too, the modal base comprises properties of the subject. The semantics of the VP-level operator, abbreviated here as DISP, is modeled on Brennan’s proposal for so-called “property-level” must and will, and is given in (19) below:5

As evidence for the proposed similarity between dynamic modals à la Brennan (1993) and the generic operator of middles, Lekakou notes the subject-orientedness of IVOs in PM:6

If PM are DAs and as such involve a VP-level generic operator, the three core properties identified above fall out.7 First, since dispositionality is a type of genericity, the generic nature of middles follows. Second, if DAs are subject-oriented, it follows that a DA to an internal argument will only be possible if that argument appears in subject position. Third, since there can only be one subject, the Agent is obligatorily demoted in PM. Its interpretation seems to also be linked to the genericity of the sentence. Lekakou suggests that the Agent in PM is an inherently generic indefinite, in the sense of Condoravdi (1989b); for instance, it cannot be anaphorically bound by a previously mentioned set of individuals. Consider the following context: Imagine you are a lecturer of linguistics. The MA students this year have virtually no background in semantics, and are for the most part non-native speakers of English. In this context, the implicit agent of the PM in (22) cannot be interpreted as referring to the previously mentioned set of students, but can only refer to people in general.

Summing up, on this approach to middles, the key component is their dispositionality as a particular flavor of genericity. In the following section, we look more closely at the (morpho-)syntax of PM in the languages under consideration.

Based on the syncretism in the morphological marking of anticausatives and middles, Schäfer (2008) proposes that middles are syntactically identical to marked anticausatives. We briefly illustrate what this means for German, where anticausatives come in two types: (reflexively-)marked (23b) and unmarked (23a). As (23c) shows, PM are formally identical to marked anticausatives.

Syntactically, (23a) is straightforwardly unaccusative; the sole argument is base-generated as the internal argument and surfaces as the subject of the sentence. Semantically, the marked anticausative in (23b) patterns just the same: no Agent is involved, and the structural subject realizes the Theme argument. In line with a configurational theta-theory, this entails that the subject in (23b) is base-generated as the internal argument (but see e.g. Koontz-Garboden 2009 for a different analysis). The difference in the morphological marking between (23a,b), Schäfer argues, correlates with a syntactic difference: while unmarked anticausatives lack the projection that introduces the external argument (v*P or VoiceP, see e.g. Kratzer 1996), this projection is present in marked anticausatives, and hosts the reflexive pronoun in its specifier. Thus, marked anticausatives instantiate a syntax-semantics mismatch: while they are semantically intransitive, they project two argument positions in the syntax (Spec,VoiceP and V,DP) and can therefore be called syntactically transitive. To account for the semantic absence of the Agent in marked anticausatives, despite the syntactic presence of an external argument position, Schäfer (2008) introduces a subtype of Voice head: expletive Voice. Expletive Voice (in German) projects a specifier, but, unlike agentive Voice, does not assign a theta-role to it. In other words, Voice_EXPLETIVE is semantically vacuous (i.e., it denotes the identity function; see also Wood 2015; Wood & Marantz 2017), and Schäfer argues that a reflexive pronoun, due to its referential defectiveness, is the only element that can legitimately occupy this non-thematic argument position.

Turning to PM in (23c), Schäfer proposes that just as in the case of (marked) anticausatives, the structural subject is interpreted as the Theme, and is consequently base-generated as the internal argument (pace Fagan 1992; A & S 1994; 1995; 2006; Lekakou 2005a; 2005b). Furthermore, it is a well-known fact that although an Agent is implied in PM, its status is quite different from the status of the implicit Agent of passives. As already mentioned in section 1, for example, only the latter licenses by-phrases (see (24)), or controls into purposes clauses (as in (25)).

In lexicalist frameworks, this difference has often been captured by a presyntactic middle-forming operation that blocks the Agent from projecting into the syntax (see e.g. Fagan 1992; A&S 1994; 1995; 2006; Lekakou 2005a; b; among others; also section 3.1). In Schäfer’s analysis, the lack of a syntactic Agent in PM follows from the assumption that PM, like marked anticausatives, involve expletive Voice, which does not encode agentivity, and thus blocks the projection of a (potential) thematic external argument. The identical morphological marking between marked anticausatives and middles is straightforwardly captured by this syntactic parallel, since both constructions involve expletive Voice. Finally, Schäfer argues that the presence of the reflexive in Spec,VoiceP is responsible for the fact that PM and marked anticausatives behave syntactically like transitive structures (for example by failing the unaccusativity diagnostics, see below).8

The part of the syntax that is shared by PM and marked anticausatives under Schäfer’s proposal is provided in (26). Following basic assumptions of Distributed Morphology, Schäfer assumes category-neutral roots, which are categorized by dominating functional heads – v identifies verbs, n nouns, etc. (see, e.g., Marantz 2001). As for the specific decomposition of the verbal layer in the form of a VoiceP-vP split, see Pylkkänen (2008); Harley (2013); Alexiadou, Anangnostopoulou & Schäfer (2015), among others.

On this approach, PM (and marked anticausatives) fail traditional unaccusativity diagnostics, not because they are unergative, but because they are syntactically transitive. In fact, Schäfer (2008) argues that traditional diagnostics such as auxiliary selection and the distribution of prenominal past participles are sensitive to the presence of Spec,VoiceP, rather than the base-position of the structural subject. Since the syntax of PM in (26) contains a filled Spec,VoiceP, the failure of the unaccusativity diagnostics follows. As soon as tests are employed that diagnose the base-position of the structural subject, this DP demonstrably behaves like an internal argument; see in particular Pitteroff & Schäfer (2014) for relevant discussion.

Adopting Schäfer’s proposal for the morphosyntax of German PM, we combine it (as he does) with Lekakou’s low generic operator, DISP, which, as discussed in section 2.1, creates a dispositional predicate. Unlike Schäfer, we assume that DISP is a transitive head – which means that it projects a specifier (the latter property represented via the selectional [uD]-feature in (27)), and has unvalued phi-features that need to be valued. DISP thus attracts to its specifier the closest active DP with valued phi-features, which then functions as the subject of the dispositional predicate.9 We arrive at the structure in (27) for PM in German (with higher functional projections such as TP omitted).

The tree in (27) represents what we assume is the core syntax of German PM. For the moment, we can also assume that (27) is the underlying syntax of Dutch PM, although this raises the question of what occupies Spec,VoiceP_EXPLETIVE (recall that Dutch PM disallow a reflexive pronoun). This issue is not directly relevant for an analysis of IM, so we defer discussion to section 4.

So far we have discussed the semantics and syntax of PM. We can now resume our focus on IM. If IM are similar to PM, and if PM are disposition ascriptions, a number of non-trivial questions arise. The questions, repeated from section 1, are summarized in (28):

In section 4, we provide an analysis of IM that is able to answer these questions. Before doing so, we briefly review previous approaches to IM, highlighting what we take to be their shortcomings.

Ackema & Schoorlemmer (A & S) (1994) propose a presyntactic derivation of PM in English and Dutch and extend it to IM. Adopting Jackendoff’s (1990) system, operations at a presyntactic level, namely the level of Lexical Conceptual Structure (LCS), may affect the argument-taking properties of predicates in the syntax. Within A & S’s set of assumptions, projection from LCS to syntax proper is not obligatory; however, non-projection of arguments is subject to restrictions. One option for an argument in the lexical conceptual structure (lcs) of a verb to not project to syntax proper is that it be assigned arbitrary reference. This is what A & S claim is happening in PM. For these authors, middle formation reduces to (29) (cf. Fagan 1992). (We simplify somewhat by abstracting away from the distinction between Agent and Actor that A & S assume, and illustrate with the more widely assumed notion of Agent.)

The effect of (29) is that the Agent is free to not project syntactically. A & S further assume a thematic hierarchy, which regulates in what order arguments that are eligible to project to syntax do so. After the application of (29) to the lcs of a transitive verb, the argument that follows the Agent in the thematic hierarchy projects syntactically to the external argument position. This essentially derives the argument structure properties exhibited by PM. Turning to IM, when (29) applies to intransitive unergative verbs, no argument is left to project externally. A & S claim that in this case an expletive is inserted as a dummy. In other words, for A & S, the subject pronoun in IM is a syntactic, but not a semantic argument of the middle verb.

The proposed analysis is successful in maintaining a uniform analysis of PM and IM in Dutch (and also adjunct middles, which we address briefly in section 6). Moreover, the analysis derives the syntactic behaviour of all middle constructions in Dutch, which, as has been mentioned already, fail canonical unaccusativity diagnostics (A & S 1995 argue that middles in Dutch and English are unergative, but see the discussion in section 2.2 on German). However, several aspects, especially of the semantics of the constructions, do not follow from (29).10 For one thing, it is difficult to see how the dispositional semantics, which we assume characterizes middles, can be linked to any aspect of the analysis. In connection to this, A&S claim that middles are individual-level predicates, and argue that this aspect of the purported middle semantics can be made to follow from the proposed derivation of middles. However, Lekakou (2005a: Chapter 2) argues that the characterization of middles as disposition ascriptions is superior to one in terms of individual-level predicates; see also Marelj (2004: 104ff) for critical discussion of the relevance of individual-level-hood. Furthermore, and more crucially for IM, the need for an additional modifier remains unexplained under this approach.

Focusing on different occurrences of the reflexive pronoun sich in German, Haider (1985) also discusses middles. His analysis is mainly concerned with the syntax of middles, and therefore the semantic properties of this construction (i.e. questions (28a,c)) are not addressed. Regarding the subject pronoun, Haider argues that it is a pure expletive and that its obligatory presence in IM is due to the need of the reflexive pronoun to have a (formal) antecedent (see also Abraham 1987; 1994; Steinbach 2002). Since with one-place predicates no such antecedent can be provided after the agent has been suppressed, insertion of the pronoun prevents a Principle A violation.

This answer to (part of the) question in (28b) suffers from two problems. The first one is that Dutch IM lack a reflexive pronoun, but nevertheless require the subject pronoun (30).

If the sole purpose of the pronoun is to provide a binder for an otherwise unbound anaphor, its obligatoriness in Dutch IM is unexpected. This is true even if Dutch, unlike German, has an EPP-feature on T that needs to be checked, and the pronoun in IM is a true expletive inserted to fulfill this function. As we discuss in section 4.1, the pronoun that occurs in IM does not behave like a true expletive, rendering this alternative implausible.

A second problem is that other configurations exist in German, in which there is no apparent binder for the reflexive pronoun, but no Principle A violation occurs. This is illustrated in (31a) for passives of reflexive verbs (e.g. Schäfer 2012), and impersonal let-middles in (31b) (see Pitteroff 2014; 2015 for discussion).

Given the acceptability of the examples in (31a) and (31b), it is not at all clear why an expletive must be inserted in IM (cf. (31c)), but not in the other types of constructions.

Cardinaletti (1990) develops an account of IM that differs from the analysis by Haider in treating the pronoun as the overt realization of arbitrary pro, which functions as the external argument. The generic interpretation of middles follows for her from the external argument receiving arbitrary reference (see also Fagan 1992; A&S 1994; 1995; 2006; Steinbach 2002 for different implementations of this idea). Cardinaletti’s account crucially relies on an alleged difference between PM and IM: whereas the external argument is not syntactically active in the former, it is syntactically projected – as arbitrary pro/ es ‘it’ – in the latter.

One problem for this account is empirical. Contrary to what is claimed, IM and PM do not differ in terms of the syntactic activity of the Agent, as e.g. signaled by the licensing of depictive secondary predicates, reflexive pronouns, or control into purpose clauses. For reasons of space, we only exemplify the latter. The examples in (32) are taken from Cardinaletti, including her reported grammaticality judgments:

However, neither the native German author of this paper, nor other native speakers we have consulted confirm this contrast in grammaticality: both sentences in (32) are equally unacceptable. The lack of a difference in grammaticality is also true of the other alleged contrasts; we conclude that PM and IM do not differ in the way proposed by Cardinaletti.

A further problem relates to the morphological marking. Cardinaletti takes the reflexive pronoun in German middles to be a morphological marker of subject dethematization. Since the alleged difference between IM and PM is tied to the respective (non-)application of subject dethematization, she predicts that the reflexive pronoun should only occur in PM, and not in IM – contrary to fact.

A number of researchers (especially Hoekstra & Roberts 1993; Broekhuis & Corver 2015; see also Steinbach 2002; Lekakou 2005a; Pitteroff 2014) have claimed that IM attribute a property to the instrument, location or time referred to by the (nominal inside the) adverbial PP. In other words, these analyses of IM assume that what functions as the logical subject of the middle sentence is what we have called the additional modifier, i.e. the PP, or the NP within it (the different accounts are vague in this regard). According to such analyses, in a sentence like (33), for instance, the logical subject is the PP in this disco, or the nominal this disco. The meaning of (33), following the dispositional analysis of middles, would roughly be: ‘in virtue of inherent properties of this disco, one dances well there’.11

However, if one adopts Lekakou’s approach to middles as dispositional sentences, particularly her analysis of dispositionals as subject-oriented generics, there is an immediate problem: what appears in the subject position of IM is not the purported logical subject, i.e. the PP (or the nominal within it), but an (expletive) pronoun.12 How is it that what looks like an adjunct functions as the subject of the IM? The answer to this question builds on a relation between the syntactic subject and the purported semantic subject of the sentence: the obligatory pronoun associates with the modifier, thus linking it semantically to the subject position. The claim is that the type of linking mechanism that exists in IM (indicated in (34a) via co-indexation) is the same one found in other contexts, such as extraposition (34b) (Bennis 1986), or certain copula constructions (34c) (see, e.g., Bennis & Wehrmann 1987):

This analysis of IM has two advantages. Firstly, the presence of the pronoun and the obligatory additional modifier are treated as two sides of the same coin, since both are present to license the disposition ascription. Through the association relation, the pronoun, i.e. the syntactic subject, enables the (nominal inside the) PP to function as the semantic subject, and, on Lekakou’s terms, to be ascribed the dispositional property. Secondly, the syntactic mechanism that links the pronoun and the modifier appears not to be middle-specific, being independently required by the grammar (34b,c). However, as we will see in this section, this kind of approach faces a number of problems, which suggests that an alternative should be sought for.

The first problem for this type of approach is that it predicts semantically implausible subjects for the dispositional sentence. Consider for instance (35). The predicted meaning of (35) is as follows: ‘in virtue of some inherent property of blankets, one sleeps well in their absence’. It seems, however, odd to ascribe a disposition to an entity, which only manifests itself in the absence of that particular entity. After all, dispositions are defined as precisely those properties that are attributed to entities in virtue of their inherent make-up; it is hard to imagine how this sort of disposition could exist, since it would be a statement about the very absence of the dispositional subject.13

A second, related problem is that it is not only (locative) PPs that can appear in IM; other syntactic categories, such as AdvPs and AdjPs can also occur as the “additional” modifier. This is illustrated by the examples in (36) which feature a manner adverbial, a time adverbial and a depictive secondary predicate (predicated of, presumably, the implicit Agent).14 The examples are given in two versions to further illustrate that the additional modifier is not restricted to occur preverbally.

It is unlikely that such modifiers can act as the dispositional subject, given that they do not introduce an event participant that could be construed as the target of the disposition ascription. It is therefore unclear what the semantic subject would be in cases such as (36).

A third problem arises from the claim that the mechanism linking the modifier to the subject position via the pronoun is not middle-specific. For example, even if the modifier in the examples in (36) could in some way function as the dispositional subject semantically, the linking mechanism involved must be different from the one in e.g. copula constructions. As (37) shows, modifiers of the type involved in (36) cannot function as the semantic subject in copula constructions.

The pronoun es in (37) must have an entity-referring interpretation (as indicated in the translation). This is clearly a different use of the pronoun than the one found in IM, and it would have to be explained why this difference exists, if the same linking mechanism is involved. As an anonymous reviewer points out with regard to (37a), this sentence does have an additional impersonal reading (meaning something like It is nicely quiet there). On this reading, however, a covert locative pronoun (there) is implied, which functions as the (implicit) associate of the pronoun es. The question therefore remains, why, if the same mechanism is involved, the associate in a copular construction cannot be, say, a manner adverb, while this is possible in IM. Note finally, that if one were to assume, based on (37), that the pronoun cannot associate with e.g. manner adverbials in IM either, the middles in (36) would remain without a semantic subject.

The parallel with extraposition (cf. (34b)) is also problematic. Generally, in extraposition the presence of the pronoun blocks extraction out of its associate, arguably because the latter occurs in an adjunct position and therefore constitutes an island (Bennis 1986; Haider 2010; Wood 2012; Fischer 2016). This is shown in (38): the pronoun is in principle optional (see (38a)), but when extraction occurs, it has to be absent (see (38b)):

Turning to IM, we see that extraction from a sentential complement is licit (39). This is despite the fact that the pronoun, unlike in (38b), cannot be omitted.

The contrast between (38b) and (39b) suggests that the association procedure in IM differs from canonical cases of extraposition: (39b) indicates that the complement clause is not an adjunct in IM, but occupies an A-position (i.e. the complement position of the verb). As a consequence, the linking mechanism involved in (38) and (39) must be different. What is more, the data above indicate that the linking mechanism would have to be non-uniform across IM, sometimes targeting adjuncts, and sometimes arguments. This latter point is also illustrated by other cases of IM, such as the ones in (40), which involve a lexically case-marked internal argument. As shown in (40b), the pronoun is obligatory even in these cases – and no additional modifier is required. Under the approach discussed in this section, this would mean that the pronoun in (40) links an argument to subject position, in contrast to what happens (i) in extraposition or copula constructions, and, even more importantly, (ii) in other instances of IM, such as (33) or (36).

Summing up, the approach to IM that takes the additional modifier as the semantic subject of the sentence, linked to the subject position via an association relation with the subject pronoun, is problematic both semantically and syntactically. Semantically, it predicts implausible subjects for the dispositional predicate, and syntactically it relies on a mechanism that on the one hand is middle-specific, and on the other hand fails to provide a uniform analysis for all instances of IM, once one considers examples that go beyond the ones involving a spatial PP.

In the literature on existential constructions, it has frequently been proposed that the expletive involved (typically (a variant of) the there-type expletive) is not a pure expletive, but functions as an overt realization of the event/spatio-temporal argument (e.g., Heycock 1994a; Ramchand 1996; Felser & Rupp 2001 among others).16 We propose that the it-type expletive which surfaces in IM has a comparable function in so far as it relates to the event(-uality) denoted by the main predicate; this enables it to function as the semantic and syntactic subject of the middle. Unlike there expletives in existential constructions, however, we claim that it in IM only refers to the event anaphorically, i.e. via an association relation with the verbal domain. In that regard, then, the pronoun in IM is comparable to the one found in extraposition contexts or certain copula constructions (see section 3.4), the sole difference being the (semantic) type of its associate. Our proposal can roughly be represented as in (42), where coindexation indicates the association relation (we use an embedded sentence to prevent the V2 requirement from affecting word order).

Informally, (42) thus attributes to a dancing event that takes place in a specific disco the dispositional property of being a good dancing event. Following Pitteroff & Schäfer (2017), we assume that the subject pronoun inherits the semantic type of its associate via the association relation that links the two (see also Ruys 2010, who claims that the pronoun is a variable that is operator bound by its associate). While in the case of extraposition, the pronoun stands in for something propositional, via its association relation with the vP in IM, its interpretation is that of an event (comp. the (neo-) Davidsonian treatment of verbs as predicates of events).

Our proposal makes certain predictions about the behavior and the nature of the pronoun involved in IM. First, this pronoun is correctly predicted to be obligatory in IM (43b), and therefore to behave fundamentally different from the one found in impersonal passives, which, being a true expletive, is barred from a sentence-internal position (43a).

The obligatory presence of the pronoun in IM illustrated in (43b), then, suggests that it behaves like a quasi-argument, rather than a true expletive (which in German is restricted to sentence initial position). Dutch provides further support for this view. In Dutch, unlike in German, the true expletive and the quasi-argument are distinct in form, the former being realized as er, the latter as het. As (44) shows, it is the quasi-argument that surfaces in IM (Hoekstra and Roberts 1993; A & S 1994; Lekakou 2005a; Pitteroff 2014 among others). As noted already, the same pronoun appears in the context of extraposition and the relevant copula constructions (see Bennis and Wehrmann 1987).

Following Bennis (1986); Vikner (1995); Ruys (2010), we assume that het-type pronouns differ from expletive er in being argumental and fully specified for phi-features (e.g., Cardinaletti 1990; Richards & Biberauer 2005; see Deal 2009 for an analysis in which expletives also have phi-features, albeit unvalued ones). Recall from section 2.2 that our analysis of PM in German involves movement of the dispositional subject to Spec,DispP, with subsequent phi-feature valuation of DISP. Due to its nominal nature and its phi-specification, the pronoun that occurs in IM can value the unvalued phi-features on DISP, occupy its specifier, and thus function as the dispositional subject. The question now arises whether the subject pronoun in IM is base-generated in Spec,DispP, or is moved there. We will approach this question by eliminating potential positions.

Since we assume with Schäfer (2008) that the external argument position in German middles is occupied by the reflexive pronoun, this is not a possible base position for the subject pronoun in IM, given that both elements obligatorily co-occur (45).

There is in principle the option of merging the pronoun in the object position under a pseudo-transitive analysis of IM (e.g. Haider & Rindler-Schjerve 1987; see Dobrovie-Sorin 1998; Cabredo-Hofherr 1999; 2006 for a pseudo-transitive analysis of impersonal passives). This might seem reasonable, since most of the predicates that occur in IM optionally license (cognate) objects.

Yet, as shown in (47), exceptions can be found, which renders a pseudo-transitive analysis of IM implausible (see also Pitz 1988). Particularly (47c) appears problematic for this possibility, given that an internal argument is already present and co-occurs with the pronoun. (Although NOM-ACC-DAT verbs do exist in German, helfen is not one of them. It is therefore unlikely that es in (47c) is merged as a second internal argument).

An anonymous reviewer points out that the verbs in (47a) and (47b) allow resultative constructions with spurious sich (48a, b).

As (48) shows, such a resultative construction licenses a reflexive pronoun, as well as an additional, accusative-marked argument (the latter possibly functioning as the subject of a small clause complement). One could thus suggest that IM have a similar syntax, so that the subject pronoun can be base-generated VP-internally as the subject of the small clause. Yet, such a structural similarity is implausible given the lack of a semantic motivation for it: the IMs in (47a’), (47b’) are not interpreted as resultatives (thanks to Marcel den Dikken, p.c., for pointing this out to us). There is thus no evidence for a small clause in IM that could host the it-type pronoun.

Since no position for the pronoun is available inside the narrow verbal domain (i.e. Voice and below), we propose that the pronoun in IM is directly merged in Spec,DispP, i.e. as the subject of the dispositional predicate.17

The syntactic structure we propose for German IM is given in (48). Note that in the tree we omit representation of the evaluative/manner adverb easily, which is not directly relevant for our current concerns (it can, for instance, be an outer adjunct). For the sake of simplicity, we also abstract away from V-to-C movement. Thus, the tree in (49) does not represent how the structure is linearized, which is (also) a PF-issue, and thus not entirely determined in narrow syntax.

As in the case of (marked) anticausatives and PM, the syntax of IM involves expletive Voice, which hosts the reflexive pronoun in its specifier and blocks the introduction of a potential thematic external argument. Since IM are generally based on unergative predicates (see the end of this section for some refinement), IM differ from anticausatives or PM in that no internal argument is projected. In other words, while marked anticausatives and PM are syntactically transitive, IM are syntactically intransitive, and, in fact, unergative (due to the reflexive pronoun in Spec,VoiceP). As a consequence, IM are correctly predicted to fail the unaccusativity diagnostics: for instance, they select have and cannot form prenominal past participles.

We wish to stress at this point that under the proposed analysis, IM is actually a misnomer: what we call IM are, via the pronoun in subject position, just as personal as “canonical” instances of PM. In other words, since the disposition always has to be assigned to something, subjectless middles do not exist; every instance of a middle involves a referential subject and is therefore personal. Support for this view comes from the distribution of implicit control. Pitteroff & Schäfer (2017) show that there is a systematic correlation between languages that allow implicit control with try-type matrix predicates (implicit predicative control (50b); see Landau 2015) and languages that license impersonal passives (50a). Furthermore, they show that all languages in their investigation allowed implicit control with decide-type matrix predicates (implicit logophoric control) in (50c):

Although the details need not concern us here, we would like to stress the contrast between (50a) and (50c). Pitteroff & Schäfer argue extensively that the matrix passive in (50c) cannot be an impersonal passive, as otherwise the unacceptability of (50a) and (50b) could not be explained. These authors therefore conclude that (50c) involves a personal passive, and since personal passives are acceptable in all languages under investigation, the fact that implicit logophoric control is possible does not come as a surprise. Pitteroff & Schäfer consequently argue that the subject pronoun in (50c) is not expletive, but argumental and theta-marked, so that its movement to subject position derives a personal passive, completely in parallel to “canonical” cases like This issue was decided yesterday. They further show that the pronoun starts out as the internal argument and stands in an association relation with the infinitival clause, from which it inherits its semantic type – it is thus of the same sort as the pronoun we argued is involved in IM, the sole difference being the semantic type of the associate (proposition vs. event). Therefore, IM are personal in the same way the passive in (50c) is.

Turning now to Dutch, the question is whether the structure in (49) carries over to this language too. Recall that Dutch IM (and PM alike) disallow the weak reflexive zich. This raises the following question: does expletive Voice also exist in this language, and, if so, what lexicalizes it?

With respect to this question, Schäfer (2008) points out that, due to its semantic vacuousness, an expletive VoiceP without any overt realization of either its specifier or its head creates learnability issues: a structure with such an “empty” expletive Voice would effectively be indistinguishable from a structure that lacks VoiceP altogether. He therefore proposes that Dutch PM also involve expletive Voice, but unlike in German, the internal argument moves to its specifier, thereby licensing it. This movement does not violate the Theta-Criterion because the target position is non-thematic, and allows Schäfer to maintain the same underlying structure for German and Dutch PM, which is, furthermore, in line with a configurational theta theory, as the Theme argument is base-generated VP-internally in both languages. Finally, in addition to giving content to expletive Voice, the movement to Spec, VoiceP could potentially be implicated in the claimed unergative behavior of Dutch middles (on which see Fagan 1992; A&S 1994; 1995; Broekhuis & Corver 2015).

Even if we were to assume Schäfer’s solution for Dutch PM, extending this kind of approach to Dutch IM is problematic. Since we propose to base-generate the subject pronoun in Spec,DispP, and since there is no (other) argument projected in IM, the specifier of expletive Voice is left empty (recall that the pseudo- transitive analysis of unergatives is untenable). We will briefly consider two potential ways how to reconcile IM with Schäfer’s analysis of PM. It will turn out that each one has its own problems.

One option is to base-generate the subject pronoun as an internal argument in Dutch IM. This may appear initially plausible, given the claim in Broekhuis & Corver (2015) that Dutch IM license locative, but not directional PP-modifiers as in (51):

Hoekstra & Mulder (1990) propose that structures containing directional PPs are unaccusative: the PP functions as the predicate of a small clause complement to the verb, taking the single argument of the verb as its subject. If, now, the subject pronoun in IM was base-generated as an internal argument, too, it would be in complementary distribution with the small clause that introduces directional PPs.18 As it turns out, however, the empirical claim made by Broekhuis & Corver (op.cit.) is not correct: the directional interpretation is available in other examples involving spatial PPs such as (52) and (53) (Hedde Zeijlstra, p.c., and Olaf Koeneman, p.c.):

There is thus no evidence in favour of merging the pronoun in Dutch IM as an internal argument.

A second possibility is to have Spec,VoiceP host the subject pronoun (which subsequently would move to Spec,DispP). This makes possible an otherwise uniform syntax for IM in the two languages, as both would involve expletive Voice. Note, however, that this comes at the cost of a difference between German and Dutch it-type pronouns in IM (i.e. the former are base-generated in Spec,DisP, the latter in Spec,VoiceP) for which no independent evidence is (at this point at least) available.

There is an obvious alternative to consider, namely that Dutch middles lack expletive Voice. This would be consonant with the fact that Dutch systematically does not employ zich in any “valency-reducing” structures (such as anticausatives, PM, or IM) — see Steinbach (2002); Lekakou (2005a; b); Schäfer (2008) for relevant discussion. On this alternative, the syntax of IM in Dutch is as in (54):

We consider the option in (54) more in keeping with the evidence currently available. Given that we treat IM as belonging to the same family as PM, it becomes pertinent to address the issue of whether PM in Dutch also do not involve expletive Voice (contra Schäfer 2008). The investigation of this question goes beyond the limits of this paper, and we therefore leave it open for future investigation.

Before we proceed to the discussion of the additional modifier, let us briefly consider a concern expressed by an anonymous reviewer. He points out that nothing in our account of IM so far blocks the co-occurrence of it with a transitive predicate (55).

What derives the fact that IM must be based on intransitive predicates? It seems to us that the answer must rely on the empirical generalization that a language with IM also has PM (but not the other way round) and that a PM has to be derived where possible, i.e. the possibility of a PM blocks the derivation of an IM. A reason for this could be case. It has been argued that case assignment is determined at the VoiceP-level (Sigurðsson 2000; 2003; Schäfer 2008; 2012). Adopting a dependent case approach (see Marantz 2000 for the original formulation), the assignment mechanism works as follows. The argument that has valued the phi-features on Voice receives nominative, the one that has not receives dependent accusative case (unless it is specified for lexical case). Since the reflexive pronoun has unvalued phi-features itself (cf. Burzio 1991; Kratzer 2009; Schäfer 2012 for similar proposals), it cannot value the phi-features on Voice. Therefore, any internal argument that is eligible for structural case will enter an Agree-relation with Voice and value its phi-features. As a consequence, this DP will surface with nominative case, deriving a PM. An IM such as (55), with a transitive predicate and an accusative-marked internal argument, can therefore simply not be generated.

Support for this approach comes from the fact that IM can be based on transitive predicates (where “transitive” means the presence of an external and an internal argument), but only if the internal argument cannot be promoted to subject position, e.g. when it receives lexical case (47c) (repeated here for the sake of convenience).

In (47c), the phi-features on Voice cannot be valued, either by the internal argument, which is inactive due to its lexical case, or by the reflexive pronoun in Spec,VoiceP, which lacks valued phi-features. Therefore, Voice agrees with DISP (via cyclic Agree, see Legate 2005), and the pronoun in Spec,DispP values this agreement chain, receiving nominative case.

We have so far provided answers to two of the three questions in (28): (i) the pronoun in IM is required because it functions as the syntactic and semantic subject of the disposition ascription; (ii) via the pronoun’s association relation with the vP, the disposition is assigned to an event(uality), rather than an event participant. The question that remains unanswered is why IM require the additional modifier, if it is not itself the dispositional subject. We propose that this modifier is needed for semantic/pragmatic reasons, namely to restrict the generalization made by the sentence. Our approach to the additional modifier is reminiscent of pragmatic analyses of the manner/evaluative modifier in PM (e.g. Goldberg & Ackerman 2001). Although we do not think that a pragmatic approach to “easily” is tenable for PM (see Lekakou 2005a; 2006 for a critical evaluation), we believe that it is correct for the additional modifier in IM.

Let us consider an IM without an additional modifier, as in (56).

Sentence (56), we argue, is syntactically well-formed, but semantically problematic (whence the #-notation). The sentence expresses that in virtue of their internal make-up, dancing events are such that dancing is good. The generalization made by (56), therefore, is simply too broad – and inevitably false. What is required for (56) to be felicitous is a proper restriction of the generalization made. This restriction is contributed by what we have been referring to as “the additional modifier” (which, recall, can be a PP, but also a secondary predicate, or even an oblique argument). It seems that it is difficult to construe events, in and of themselves, as having inherent properties that can affect the way they are realized. Whereas, for instance, books have several properties that may affect the way they are read (language, style, font size, etc), rendering This book reads easily acceptable, events, we contend, cannot easily determine their realization in a comparable way. Yet, once the event is restricted, properties that can affect its realization can be identified. For instance, in (56) the restriction could be to dancing events in a certain location, on a crowded dance floor, to the right kind of music, etc.19

Our claim about the semantic/pragmatic function of the additional modifier suggests a parallel with non-middle sentences that express, so to speak, a comparably uninformative generalization. The status of IM in (57) is the same as that of the impersonal man-construction in (58).

In the absence of the PP-modifier or the depictive predicate, the sentences in (58) lack informative content. Crucially, it is the addition of exactly the same type of modifiers that are required in IMs that renders the examples in (58) acceptable. This parallelism between (57) and (58) speaks in favor of our approach to the additional modifier in IM.20

We provide two types of arguments to support our view regarding the function of the additional modifier. The first argument comes from distribution. What we predict is that any type of modifier is licensed in IM, as long as it modifies the event, and is thus able to restrict the generalization made. For purposes of illustration we adopt the classification of modifiers by Maienborn and Schäfer (2011), given in Figure 1. Of the categories listed, the ones that are capable of modifying (and thus restricting) the event are (a) participant-oriented adverbs and (b) verb-related adverbs.

The data in (59) show that it is exactly these categories of adverbs that are licit in IM.

We have also seen other examples of modifiers that are acceptable in IM. This is because they can appropriately restrict the generalization made by IM. In (60a), an example of a depictive is repeated, and (60b) illustrates the acceptability of a conditional clause.

By contrast, VP-external modifiers (sentence or functional adverbials in Maienborn & Schäfer’s classification) do not restrict the event, and are correctly predicted to be insufficient to render an IM acceptable. Note that if a suitable adverbial were added to the (b) examples below, they would be well-formed.

A second argument in favour of our proposal comes from the fact that the additional modifier may be omitted if the required restriction is contributed contextually. Consider (65), where the context includes explicit mention of the “additional modifier”, namely ‘this new disco’ in (65a) and ‘being naked’ in (65b):

Although individual speaker judgments may vary, speakers generally accept examples such as the ones in (65). For instance, we included (65a) as a filler item in a questionnaire study and it patterned with the class of fully acceptable test items (raw mean rating: 5.25 on a 7-point Likert scale, with 1 = unacceptable and 7 = fully acceptable; n = 71).

Importantly, under the analysis that treats the additional modifier as the dispositional subject (see section 3.4), the data in (65) cannot be easily explained. If the additional modifier is required because it functions as the dispositional subject, the acceptability of the IM in (65) is unexpected. One would be forced to assume some covert modifier that it associates with and that functions as the semantic subject, which does not seem plausible. Nothing like that is required under our analysis, which effectively predicts the acceptability of such data.