This book investigates the syntax and semantics of middle and related constructions in German. We are mainly interested in the following four questions: (i) Why are middle constructions in German and many other languages reflexive? (ii) Can we derive the linking of arguments in middle and related constructions form general principles or do we need construction specific lexical and/or syntactic assumptions? (iii) Which universal and language-specific linking-principles determine the categorial realization of semantic arguments or, to put it the other way round, the semantic interpretation of syntactic arguments in middle and related constructions? (iv) Does the analysis of middle and related constructions provide evidence in favor of either a lexical, a syntactic or a (postsyntactic) semantic approach to argument linking?

The middle construction in German is a particularly interesting case study for the correlation between semantic and categorial selection.

In this introductory chapter we use the terms ‘semantic selection’ (s-selection) and ‘categorial selection’ (c-selection) without any theoretical implications. It will become clear in the course of this study that we do not think that the c-selectional properties of a lexical item can entirely be determined in the lexicon. Apart from that, we use the more neutral term argument linking instead of theta-role assignment, because (i) we will argue that theta-roles are irrelevant for the linking of arguments in middle constructions and (ii) we follow Dowty (1991) in his criticism of the traditional concepts of thematic roles. We will deal with proto-roles and proto-role properties in Chapter 6.

(1)

Dieses Buch liest sich leicht

This book-nom reads reflexive-pronoun-acc easily

‘This book reads easily’

(2)

Das Buch wurde gelesen

The book-nom was read

‘The book was read’

(3)

Hans liest dieses Buch

Hans-nom reads this book-acc

‘Hans is reading this book’

Besides, middle constructions show an interesting divergence in their c-selectional and s-selectional properties. On the one hand, there is an implicit semantic argument in middle constructions like (1), which is not linked to a syntactic constituent. This implicit ‘logical subject’, which is the first argument of the predicate, is realized as a NP with nominative case (i.e. as the syntactic subject) in the corresponding simple active sentence in (3). Instead of the verb’s first argument its second argument is linked to the subject position (i.e. to a NP assigned nominative case) in middle constructions. This second argument is normally linked to a NP assigned accusative case (i.e. the direct object) in the active counterpart (3).

So-called adjunct middle constructions might be one exception. We deal with adjunct middles in chapter 2 and chapter 7.

Further interesting properties of middle constructions are adverbial modification, which seems to be required by most middle constructions, and the characteristic ‘generic’ interpretation of the implicit argument, which differs from the interpretation of implicit arguments in passives, which are usually interpreted existentially. Hence, we might wonder whether (predicates in) middle constructions idiosyncratically select some kind of adverbial and a ‘generic’ logical subject, which must not be linked to syntax, or whether these properties of middle constructions can be derived from independent principles of grammar. The next chapter will be concerned with the properties of middle constructions in greater detail.

Recall that arguments which are only s-selected but not c-selected, and vice versa, can also be found in other contexts. In passives the first semantic argument of a predicate can optionally be realized as a by-phrase (cf. 4.a). Furthermore, some semantic arguments of some predicates are only optionally c-selected. They can but need not be linked to a syntactic constituent, as can be seen in (4.b).

The third argument of the verb verlesen (‘to read/call out’), i.e. the goal of the reading out, might be an example for an argument that must not project in syntax at all, cf. Höhle (1978). For an analysis of implicit arguments in German see Jacobs (1994) and Rapp (1999).

(4)

a.

Das Buch wurde (von Hans) gelesen

The book-nom was (by Hans) read

‘The book was read (by Hans)’

b.

Hans schreibt (ein Buch)

Hans-nom writes (a book-acc)

‘Hans is writing (a book)’

c.

… weil es mich friert

… because it-nom me-acc cold-is

‘… because I am cold’

Middle constructions like (1) combine both an implicit semantic argument (i.e. the ‘logical subject’ or first argument of the verb) and an obligatorily c-selected constituent that is not linked to a (s-selected) semantic argument of the verb (i.e. the reflexive pronoun). Thus middle constructions include the suppression of a semantic argument as well as the addition of a syntactic argument. The selectional properties of middle constructions are illustrated in (5) – ‘Ø’ means that this element does not correspond either to a c-selected or to a s-selected entity.

(5)

s-selection:

x

y

Ø

c-selection:

Ø

NP_NOM

Reflexive Pronoun_ACC

Hence, middle constructions have only one element that is both s-selected and c-selected but two elements that are not linked at all. We will see in chapter 2 that so-called impersonal middle constructions actually contain no linked element at all. The German middle construction has so far proved a good example for the study of non-trivial cases of argument linking. It will also turn out to be a good example for illustrating and discussing the differences between lexical, syntactic, and postsyntactic theories of argument-linking.

However, middle constructions become even more interesting as soon as we compare them with morphosyntactically identical constructions. We will see that deeper insight into this issue can be gained if the study of middle constructions is embedded in the more general context of transitive reflexive sentences. In German middle constructions are morphosyntactically common transitive sentences in the active with an accusative reflexive pronoun in the position of the direct object. But the middle interpretation is not the only possible interpretation for transitive reflexive sentences. In addition to the middle interpretation, transitive reflexive sentences can also get a reflexive, an anticausative and an inherent reflexive interpretation. Examples for these four interpretations are given in (6).

(6)

a.

Reflexive interpretation:

Herr Rossi rasiert sich

Mr. Rossi-nom shaves reflexive-pronoun-acc

‘Mr. Rossi is shaving (himself)’

b.

Middle interpretation:

Das Buch liest sich leicht

The book-nom reads reflexive-pronoun-acc easily

‘The book reads easily’

c.

Anticausative interpretation:

Die Tür öffnet sich

The door-nom opens reflexive-pronoun-acc

‘The door opens’

d.

Inherent reflexive interpretation:

Herr Rossi erkältet sich

Mr. Rossi-nom catches-a-cold reflexive-pronoun-acc

‘Mr. Rossi catches a cold’

The reflexive interpretation in (6.a) differs from the other three interpretations in one respect. Only in (6.a) the reflexive pronoun is both c- and s-selected and therefore interpreted as a semantic argument of the verb, which is bound by and coreferent with the subject of the sentence. This means that both the subject and the direct object are linked to a semantic argument variable of the verb each. In (6.b), (6.c) and (6.d), on the other hand, the reflexive pronoun is only c-selected but not s-selected and therefore it is not interpreted as a semantic argument of the verb. The reflexive pronoun indicates valency reduction of the first semantic argument of the verb, its logical subject, which is not linked to syntax. Hence, the accusative reflexive pronoun in (6) can but need not be linked to a semantic argument of the verb. We call the reflexive pronoun in (6.a) argument reflexive and the one in (6.b-d) non-argument reflexive.

Nothing depends on this terminology. We use it only to distinguish these two different interpretations of reflexive pronouns. The argument reflexive is also called ‘echt reflexiv’ or ‘anaphorical sich’ and the non-argument reflexive ‘unecht reflexiv’, ‘lexical sich’, and ‘Fügungs-sich’, cf. Haider (1982) and Reis (1981). Fagan (1992) calls the argument reflexive ‘referential’ and the non-argument reflexive ‘nonreferential’. This terminology is somewhat misleading, because reflexive pronouns are never referentially independent.

Beside this first ambiguity of the reflexive pronoun, another ambiguity is responsible for the interpretation of sentence (6.b) on the one hand and sentence (6.c) and (6.d) on the other. This second ambiguity is due to the interpretation of the suppressed semantic argument. We already saw that the first semantic argument of the verb is not linked to syntax if the non-argument interpretation is chosen. In this case it can either be saturated or reduced. The latter operation (argument reduction) completely removes the suppressed argument from the semantic representation. Argument reduction is the more restrictive operation and yields the anticausative and inherent reflexive interpretations. Anticausatives like (6.c) are one-place predicates that are systematically derived from underlying two-place predicates. Argument saturation, on the other hand, means that the implicit argument is bound by a semantic quantifier/operator. The resulting interpretation of argument saturation is the middle interpretation. Argument saturation is less restrictive than argument reduction. It can be applied to most verbs selecting at least one argument. Simplified semantic representations of all four sentences are given in (7). (‘x’ is the implicit argument in MCs, which is bound by a generic operator, cf. chapter 7, and ‘1’ stands for the deleted first argument in anticausatives and inherent reflexives).

(7)

a.

W < r₁, r₁ >

W = waschen, r = Rossi

(reflexive interpretation)

b.

S < x, b >

S = schneiden, b = Brot

(middle interpretation)

c.

O <1, f >

O = öffnen, f = Fenster

(anticausative interpretation)

d.

E <1, r >

E = erkälten, r = Rossi

(inh. refl. interpretation)

The ambiguity of reflexive constructions is a widespread phenomenon that can be observed in many Indo-European and non-Indo-European languages. All four interpretations in (6) come under the notion of middle voice and have more or less identical counterparts in various other languages. Like many other languages, German uses a reflexive marker (the accusative reflexive pronoun) to encode a variety of related interpretations. Thus, German middle constructions are one possible interpretation of what we will call the middle voice.

This book deals with the ambiguity of the reflexive pronoun and the interpretation of implicit arguments in transitive reflexive sentences. As will be discussed at length in chapter 3, nearly all recent approaches to middle constructions and anticausatives derive their s-selectional and c-selectional properties in the lexicon and/or in the syntax. They are based on the assumption that the syntax of verbal arguments is completely determined by the selectional properties of a verb in co-operation with general linking-principles. Therefore, these analyses require some lexical and/or syntactic manipulation of the argument structure of the underlying verbs in (6.b), (6.c) and (6.d). They must somehow prevent the linking of the first semantic argument to the syntactic subject, promote the second semantic argument and add a reflexive pronoun, which does not correspond to a semantic argument of the verb. By contrast, we argue for a new approach that takes the syntactic surface more seriously. We keep the minimal assumption that all transitive reflexive sentences are equal, because we do not find any empirical (or conceptual) evidence for the assumption that the transitive reflexive sentences in (6) must be assigned different syntactic structures or that they contain lexically different kinds of verbs. This change in viewpoint enables us to develop a uniform analysis that derives all four interpretations of transitive reflexive sentences illustrated in (6) from the same underlying syntactic representation. This makes the theory proposed in this book superior to lexical and syntactic approaches, which both do not offer a conclusive and uniform analysis of the ambiguity of transitive reflexive sentences in German. Besides, both lexical and syntactic approaches need additional ad hoc stipulations, which are not necessary in the postsyntactic analysis developed in this book.

We will argue in chapter 5 that (weak) reflexive pronouns are not lexically specified for the morphosyntactic feature [R], because their Φ-features are maximally underspecified. Elements that are not lexically specified for [R] can but need not be interpreted as semantic arguments. Therefore the linking of the syntactic subject can be mediated by a (weak) reflexive pronoun, if the former binds the latter. In this case, the syntactic subject is not linked to the first but to the second semantic argument of the verb. Roughly speaking, the reflexive pronouns in (6.b), (6.c), and (6.d) (i.e. the non-argument reflexives) function as a place-holder for the linking of the syntactic subject, which is thus linked to the second argument variable of the verb. Furthermore, we will argue in chapter 5 that in German, non-argument reflexives must be assigned structural case. Hence, only transitive reflexive sentences permit the syntactic subject to be linked to the second argument of the verb. The first assumption that (weak) reflexive pronouns are lexically underspecified for the feature [R] seems to be valid crosslinguistically, whereas the second assumption that non-argument reflexives must receive structural case is language-specific. We do not think that all languages draw similar distinctions between structural and oblique case. On the other hand, (weak) reflexive pronouns universally seem to be the less specified pronominal elements and various languages use (weak) reflexive pronouns to indicate valency reduction. It seems to be a universal property that the morphologically less specified elements need not be interpreted as semantic arguments. Note, however, that (weak) reflexive pronouns have quite different morphosyntactic properties crosslinguistically, cf. chapter 2. Thus our postsyntactic analysis we develop for German transitive reflexive sentences does not necessarily hold for the middle voice in other langages. We come back to this issue in the final chapter .

As a consequence of our analysis, syntactic arguments that are not specified for [R] need not be linked to a semantic argument of the verb themselves. Any other syntactic argument (i.e. personal pronouns, demonstratives, definite NPs, and indefinite NPs) corresponds to a semantic argument of the verb. We will see that every syntactic argument must be linked to a semantic argument, except for non-argument reflexives and impersonal subjects. This is in accordance with the first part of the theta-criterion (i.e. each argument bears one and only one theta-role). Note, however, that our account permits two strictly defined exceptions. Both exceptions are pronominal elements that serve a specific grammatical purpose. The second part of the theta-criterion, on the other hand, is no longer a valid linking-principle, as we will argue throughout this book. We will see that the s-selectional properties are not a wellformedness condition for syntactic representations or, more generally, for the computational system, although they are an essential part of the meaning of a lexical item determining the semantic interpretation of predicates. This shift towards the semantic interpretation of syntactic arguments allows for a uniform analysis of all four sentences in (6). Moreover, we can account for the implicit argument and (the c-selection of) the reflexive pronoun in middle constructions, anticausatives, and inherent reflexives without additional stipulations. According to our analysis, the non-argument reflexive is not lexically selected by the verb. Instead it is licensed by the grammatical function it fulfills. Hence, our approach predicts that the categorial properties of arguments are not always and completely determined in the lexicon. Finally let us briefly look at another example that confirms this prediction: V2 complement clauses in German. In German the propositional argument of some verbs can be realized as a V2-clause. Therefore, we might claim that some verbs, as for example glauben (‘believe’) in (8), can c-select either a complementizer-initial verb-final clause (8.a) or, alternatively, a V2 complement clause (8.b). Verbs like bedauern (‘regret’), on the other hand, can only c-select complementizer-initial verb-final clauses (9.a). Embedded V2-clauses are ungrammatical with these verbs, as can be seen in (9.b).

(8)

a.

Ich glaube, dass ein Sturm aufkommt

I believe that a storm up-comes

‘I believe that a storm breaks’

b.

Ich glaube, ein Sturm kommt auf

I believe a storm comes up

‘I believe that a storm breaks’

(9)

a.

Ich bedaure, dass ein Sturm aufkommt

I regret that a storm up-comes

‘I regret that a storm breaks’

b.

*Ich bedaure, ein Sturm kommt auf

I regret a storm comes up

‘I regret that a storm breaks’

This is, however, only part of the story. Consider the examples in (10) below. As soon as we add a negation or a dative object to the sentences in (8), the second example (8.b) with the V2 complement clause becomes ungrammatical. Hence, embedded V2-clauses cannot simply be c-selected in the lexicon by some verbs s-selecting a propositional argument. V2-clauses have, roughly speaking, specific semantic properties which restrict the contexts they can occur in. As opposed to complementizer-initial verb-final clauses, V2 complement clauses are always assertional (or [-presuppositional]). Therefore they cannot be interpreted immediately in the scope of the negation in (10.b) or the negative predicate bedauern (‘regret’) in (9.b). This would contradict their assertional character.

Gärtner (1998) points out that V2 adverbial clauses and V2 relative clauses are subject to the same restriction. As opposed to the V-final adverbial clause in (i.b), the V2 adverbial clause in (i.a) cannot be interpreted in the scope of negation. Therefore, only sentence (i.b) is ambiguous. Sentence (i.a), on the other hand, can only mean that she did not go to Frankfurt and the reason for this is that she is ill. The same holds for V2 relative clauses, which are also ungrammatical in the scope of negation, cf. (ii.b).

(i)

a.

Sie fuhr nicht nach Frankfurt [ weil sie ist krank ]

She went not to Frankfurt because she is ill

‘She didn't go to Frankfurt because she is ill’

b.

Sie fuhr nicht nach Frankfurt [ weil sie krank ist ]

(ii)

a.

Peter hat einen Freund, der ist krank

Peter has a friend who is ill

b.

*Peter hat keinen Freund, der ist krank

Peter has no friend who is ill