Starting from the observation that in German transitive reflexive sentences of the form subject + verb + accusative reflexive pronoun are multiply ambiguous, we were searching for an analysis which derives all interpretations in a uniform way from one underlying syntactic representation. Recall that transitive reflexive sentences are ambiguous between the following interpretations:

1. Reflexive interpretation: both the subject and the reflexive pronoun are linked to the first and second semantic argument of the verb respectively. These arguments are coreferent.

2. Middle interpretation: the subject is linked to the second semantic argument of the verb. The verb’s first argument is bound by the generic quantifier (argument saturation).

3. Anticausative interpretation: the subject is linked to the second semantic argument of the verb. The verb’s first argument is deleted (argument reduction). Anticausatives are one-place predicates.

4. Inherent reflexive interpretation: equals the anticausative interpretation. The verb’s first argument is always deleted.

We called the reflexive pronoun in (i), which is linked to a semantic argument itself, argument reflexive. The reflexive pronoun in (ii)-(iv), which is not linked to a semantic argument itself, was called non-argument reflexive. Non-argument reflexives mediate the linking of the syntactic subject to the second argument of the verb. Since there is no empirical evidence for the claim that transitive reflexive sentences which differ in meaning also differ in syntax, we keep the minimal assumption that all transitive reflexive sentences are identical in syntax. As a consequence, a syntactic derivation of the different interpretations of transitive reflexive sentences is not available. Likewise, a lexical derivation cannot be empirically motivated either. In German, middle formation is not morphologically marked on the verb and it turned out to be a very productive operation which is not lexically restricted to certain classes of verbs (the only exception are individual-level predicates which are incompatible with the generic quantifier and thus excluded for independent reasons). Therefore, we argued for a third kind of analysis that has generally been neglected so far. The ambiguity of transitive reflexive sentences is derived at the interface between syntax and semantics.

Our analysis is based on the distinction between structural and oblique case and on the observation that the morphosyntactic features of reflexive pronouns are maximally underspecified. As a consequence, reflexive pronouns are not lexically specified for the morphosyntactic feature [R]. These two assumptions enable us to derive the semantic ambiguity of transitive reflexive sentences. The ambiguity illustrated in (i)-(iv) is restricted to reflexive pronouns that are assigned structural case. Unlike accusative case, dative case is oblique in German. Therefore dative reflexive pronouns are not ambiguous. Thus our analysis correctly predicts that in German only an accusative reflexive pronoun in the position of the direct object is ambiguous between an argument and a non-argument interpretation. We called the accusative reflexive pronoun in this position a morphosyntactic middle marker. Many Indo-European languages use weak reflexive pronouns as indicators of valency reduction. It seems to be a universal property that the middle marker is always the pronominal element which is morphologically less specified. However, middle markers in different languages have different morphosyntactic properties. We saw that the middle marker in German is an independent word, i.e. the accusative reflexive pronoun in the position of the direct object. In other languages middle markers are weak reflexive pronouns, verbal clitics, verbal affixes or verbal inflection. Hence, unlike most Indo-European languages, German has a morphosyntactically ‘strong’ middle marker. Besides, the German middle marker relies on structural case. Therefore, the analysis we proposed for the middle marker in German (i.e. for transitive reflexive sentences) does not necessarily hold for the middle voice in other languages. Middle formation might be lexical in some languages and syntactic in others, but it generally involves weak reflexive pronouns. Thus every analysis of the middle voice has to embed this universal property into the language-specific context, which determines the specific morphosyntactic properties of the middle voice in each language.

In the following we briefly summarize the main findings of our analysis. In chapter 2 we argued that German is a middle marking language in the sense of Kemmer (1993). The ambiguity of transitive reflexive sentences in German is not exceptional. In many Indo-European languages the weak reflexive pronoun is a middle marker, which is usually ambiguous between a reflexive, passive, middle, anticausative, and inherent reflexive interpretation among others. Note that the possible interpretations for a weak reflexive pronoun may differ from language to language, cf. also below. Although German, unlike most Indo-European languages, is a one-form language, which does not distinguish weak from strong reflexive pronouns, it has also a morphosyntactic middle marker, the accusative reflexive pronoun in the position of the direct object, which is responsible for the ambiguity of transitive reflexive sentences. Besides, we illustrated that middle formation is not restricted to certain classes of verbs. On the one hand, we saw that middle formation is possible with all kinds of one-, two, or three-place predicates. Only zero-place predicates and individual-level predicates are excluded from middle formation for independent reasons. On the other hand, we have seen that dative objects must not undergo middle formation. Verbs selecting a dative object can only occur in impersonal middle constructions. As opposed to middle constructions, anticausatives are more restricted. In addition, German has reflexive and non-reflexive anticausatives. Non-reflexive anticausatives are syntactically unaccusative. In chapter 3, we discussed several syntactic and lexical analyses of middle constructions. We were concentrating on analyses of middle constructions, because middle formation is the most productive operation and it is most controversially debated. In 3.1 and 3.2 we argued that both lexical and syntactic theories fail to offer a conclusive analysis of middle constructions. Syntactic analyses derive middle constructions, like passives, by means of A-movement. Lexical analyses, on the other hand, assume either a lexical rule of middle formation or a middle template. Neither of these analyses can account for the presence of an accusative reflexive pronoun. Besides, they need additional ad hoc stipulations and they do not provide a uniform analysis for the systematic ambiguity of transitive reflexive sentences. And finally they are neither empirically nor conceptually motivated. Therefore, we pursued a different (and, as far as we can see, a new) path. We derive the ambiguity of transitive reflexive sentences from a uniform syntactic representation at the interface between syntax and semantics. This postsyntactic approach was developed in chapters 4, 5 and 6. Chapter 4 dealt with the syntax of transitive reflexive sentences. We argued that transitive reflexive sentences differ in their semantic interpretation but not in syntax. Section 4.1 showed that the accusative reflexive pronoun is always subject to the same restrictions on word order in the middle field, regardless of whether it is an argument reflexive (i.e. (i) above) or non-argument reflexive (i.e. (ii)-(iv) above). Section 4.2 illustrated that the differences between argument and non-argument reflexives concerning coordination, focus, and fronting follow from their different semantics. Only argument reflexives are linked to a semantic argument variable themselves, which is a necessary condition on coordination, focus, and fronting of the reflexive pronoun. Hence, there are good reasons for treating all transitive reflexive sentences the same way in syntax. Chapter 5 investigated the interpretation of reflexive pronouns in German. We followed Reinhart and Reuland (1993) and Pollard and Sag (1994), who argued that binding should be defined relative to the syntactic and semantic arguments of a verb. Only NPs that are assigned nominative and accusative case are syntactic arguments (A-expressions) in German. In addition, reflexive pronouns are not lexically specified for the feature [R]. Therefore, they can be bound either in syntax (i.e. [–R]-reflexives) or in semantics (i.e. [+R]-reflexives). And finally, reflexive pronouns that cannot be bound by a co-argument of the same predicate are exempt from binding. This leads to the following threefold picture of binding.

1. syntactic binding (A-chain formation, restricted to accusative [–R]-RPs)

2. semantic binding (O-binding, restricted to [+R]-RPs & co-arguments of a predicate)

3. logophoric binding (restricted to [+R]-RPs which are exempt from (i) and (ii))

A theory that is based on A-chain formation and the distinction between [+/–R] reflexive pronouns correctly accounts for the ambiguity of the accusative reflexive pronoun in the position of the direct object. Recall that reflexive pronouns are lexically underspecified. Thus their specification of the feature [R] depends on the syntactic context. The reflexive pronoun can either be specified as [+R] or [–R]. The [+R] reflexive pronoun must head its own chain whereas the [–R] reflexive pronoun must be included in another A-chain which is headed by a [+R]-expression. This is illustrated in (1.b) and (1.c).

(1)

A-chains and [+/–R]-expressions in German

The [–R] reflexive pronoun in (1.b) must be included in a complex A-chain, which is linked to the second semantic argument of the verb via its base in the complement position of the verb. By contrast, the [+R] reflexive pronoun in (1.c) heads its own chain, which is linked again to the second argument. The second chain in (1.c), which is headed by the subject, is linked to the first semantic argument via its base position in VP,Spec. The [–R] reflexive pronoun in (1.b) is the non-argument reflexive and the [+R] reflexive pronoun in (1.c) the argument reflexive. This analysis of is based on the following two linking principles for syntactic arguments in German.

(2)

a.

VP,Spec is linked to the first argument of the verb

b.

The complement of V° is linked to the second argument of the verb

In sum, our analysis correctly accounts for the two essential features of middle markers in German. A middle marker must be assigned structural case and it must be a reflexive pronoun. Thus we can derive the observation we made in chapter 2.

(3)

Only a accusative reflexive pronoun in the position of the direct object is a middle marker in German

The difference between middle constructions and anticausatives results from two semantic operations. Implicit semantic arguments that are not linked to syntax can either be saturated (i.e. bound by a quantifier) or reduced (i.e. deleted). Both operations apply to free argument variables. Argument saturation is responsible for the middle interpretation and argument reduction for the anticausative and the inherent reflexive interpretation. We gave several empirical arguments for the claim that dative case is oblique in German. We argued that dative objects are A’-elements in syntax. This analysis correctly accounts for the differences between accusative and dative objects and explains why dative reflexive pronouns are excluded from middle formation. Dative reflexive pronouns are always linked to a semantic argument of the verb. Our analysis predicts that middle constructions are simple transitive reflexive sentences. Therefore, the generic interpretation of middle constructions and the quasi-obligatory adverbial modification should follow from the semantics (and pragmatics) of middle constructions. In chapter 7 we illustrated how a postsyntactic approach can account for the generic quantification and the adverbial modification. Middle constructions involve generic quantification over events/situations and the implicit argument. As a consequence, middle constructions without adverbial modification are restricted to very specific contexts. The adverbial modification can thus be derived from pragmatic licensing conditions. Finally, we discussed adjunct middle constructions, which are subject to non-configurational (semantic) licensing conditions.

We hope that the present study casts new light on the interaction between syntax and semantics in general and on the analysis of middle constructions and anticausatives in particular. We confined ourselves to the discussion of the syntax and semantics of transitive reflexive sentences in German. We were mainly interested in the problem of argument linking in reflexive constructions. In this connection we also discussed case theory, focus theory, restrictions on fronting and word order in the middle field, the distinction between weak and strong reflexive pronouns, and further issues of binding theory. Of course, many questions remained unanswered. Besides, additional interesting questions arise if we accept the approach proposed in this book. First of all one would like to know to what extend our analysis can be applied to other languages. Languages differ with respect to their morphosyntactic properties. We saw in chapter 2 that middle markers can be quite different crosslinguistically. Therefore, we expect morphological, syntactic, and semantics differences between the middle voice systems in different languages. Consider, for example, English. Table (4) illustrates the possible interpretations for intransitive and transitive reflexive sentences.

(4)

Intransitive and transitive reflexive sentences in English

Unlike most Indo-European languages, English does not have an overt middle marker. However, table (4) illustrates that intransitive sentences in English receive typical middle interpetations: anticausative, middle, and reflexive. In German, these interpretations are connected with the transitive reflexive sentence in (4.b). One could argue that in English weak reflexive pronouns do not have a morphosyntactic realization (or that they are morphologically empty forms, cf. Keyser and Roeper (1984)). According to this assumption the ambiguity of intransitive sentences in English is quite regular. It can be derived if we assume the following linking principle for the subject in English:

(5)

VP,Spec is linked to the first (external or internal) and/or second semantic argument

of a predicate.

Thus, English differs from German in two respects: (i) English is a two-form language with a morphologically empty weak reflexive pronoun and (ii) only in English the subject can also be linked to the second semantic argument of the predicate, cf. Steinbach (to appear) for more discussion.

Another interesting issue concerns the interpretation of the middle voice in different languages. Recall from chapter 2 that some languages permit a passive interpretation for the weak reflexive pronoun. Russian, for example, does not morphosyntactically distinguish between middle constructions and passives, and French and Italian middle constructions can receive an eventive interpetation that seems to be almost identical to the interpretation of the periphrastic passive in these languages. Consider again table (47) from chapter 2 here repeated as (6).

(6)

Possible interpretations for an overt (weak) reflexive marker

Two closely related issues are of interest in this connection: (i) Why is the passive interpretation of the (weak) reflexive marker not licensed in a language like German? (ii) Are there any dependencies between the different interpretations of (weak) reflexive pronouns. Note that 16 different constellations are possible in principle if we restrict the investigation to the four meanings illustrated in table (6). There seems to be an interesting correlation between the morphological ‘weight’ of a weak reflexive marker and the functions it encodes: reflexive markers that are verbal affixes like the Russian -sja seem to encode more functions/interpretations than reflexive markers that are independent words. Verbal clitics are somewhere in between: French se- and Italian si-constructions, for example, can get a passive (or at least a passive-like) interpretation, which is impossible for the German middle marker. As opposed to se and si, the German reflexive pronoun sich is an independent word in syntax. At first sight the passive interpretation seems to depend on the middle interpretation in (6). This issue is also closely related to the phenomenon of grammaticalization and the historical development of ‘neo-middle constructions’ in modern Indo-European languages. Besides, this book mainly dealt with German, a one-form language. Hence, further investigation will show to what extend the analysis proposed in this book also applies to two-form languages. Finally, this analysis of the middle voice in German should be embedded in psycholinguistic research on the acquisition and processing of reflexive pronouns in reflexive constructions, middle constructions, anticausatives, and inherent reflexives.