Middle Constructions and Middle Voice — What does a Middle look like?

Transitive reflexive sentences in German can be related to major topics like diathesis, genus verbi or grammatical voice, valency reduction, or argument structure alternations. Argument structure alternations systematically change the selectional properties of verbs. That is, they can change the categorial properties of arguments, they can add or delete syntactic and/or semantic arguments and they can add secondary predicates. Finally, they can also change the meaning of the underlying verb and the morhosyntactic form of verbs e.g. by adding an affix or a seperable verbal particle to the verbal stem. Middle constructions do not significantly change the underlying basic meaning of verb. In the middle construction like (1) the verb schneiden (‘cut’) still denotes a two-place relation between a cutting person (the implicit argument) and something (‘the bread’) that is cut (‘RP’ stands for reflexive pronoun).

In the following presentation, most examples have glosses only. These examples are always interpreted like middle constructions in English and other languages. The morphosyntactic realization of the middle construction differs from language to language (cf. 2.3. below), but their semantics is homogeneous across the Indo-European languages. All middle constructions have the thematic interpretation outlined in the brief introduction of this chapter. Throughout this study we use the term reflexive pronoun (which is more specific) instead of anaphor, but nothing hinges on this.

(1)

Das Brot schneidet sich leicht

The bread-nom cuts rp-acc easily

‘The bread cuts easily’

However, the middle construction changes the selectional properties of the underlying verb. The syntactic subject in (1.a), the nominative NP das Brot, is not linked to the first (or external) but to the second (or internal) semantic argument of the verb schneiden – i.e. das Brot is the thing that is cut. The first (or external argument) of the verb, the cutter, is not linked to a syntactic argument. This semantic argument is only implicitly present in the semantic representation of the sentence. In German middle constructions the suppression of the first semantic argument is indicated by the accusative reflexive pronoun. Hence, the accusative reflexive pronoun in (1) can be analyzed as a morphosyntactic middle marker, i.e. an indicator of valency reduction. Note, however, that it is not a verbal affix or clitic but an independent word. In this respect, the German middle marker differs from middle markers in most Indo-European languages. We will discuss this issue in greater detail in section 2.3 below.

German has a second construction beside the middle that also involves suppression of the first semantic argument, the passive. Passives and middle constructions have one thing in common: in both the middle construction in (1) and the passive in (2) the first or external semantic argument of the verb is not linked to the subject.

(2)

Das Brot wird geschnitten

The bread-nom is-pas cut

‘The bread is being cut’

The passive in (2) and the middle construction in (1) correspond to the active counterpart in (3), in which both semantic arguments of the two-place predicate are linked to syntactic arguments. The active voice is the unmarked case of argument linking.

In most languages the active voice is the morphologically or syntactically unmarked form (cf. Benveniste 1972, Kemmer 1993, Klaiman 1991, or Beekes 1995).

In (3) the nominative subject of the middle construction and passive is linked to the accusative object and the suppressed implicit semantic argument of the middle construction and the passive is linked to the nominative subject.

Adjunct middles are the only exception to this correspondence between the subject of middle construction and the object of the active counterpart. Here the middle-subject corresponds to a DP included in a PP. We come back to these examples immediately. They are discussed in more detail in chapter 7.

The non-argument reflexive is not present in the active counterpart in (3).

Note that middle constructions usually induce an additional modality effect, which will be discussed below.

(3)

...dass jemand das Brot (leicht) schneidet

…that someone-nom the bread-acc easily cut

‘that someone (easily) cuts the bread’

By ‘active’ we mean the grammatical voice and not the morphological form of the verb. Both middle constructions and the corresponding sentence in active voice contain morphologically unmarked ‘active’ forms of the verb. Moreover, middle constructions, like the corresponding active sentence in (3), are syntactically transitive as opposed to passives. Passives must not have an accusative reflexive pronoun or some other accusative NP in German.

The only exception might be passivized reflexive sentences like (i). In this case the reflexive pronoun cannot be promoted to subject because German does not have nominative reflexive pronouns, cf. chapter 5.

(i)

Jetzt wird sich gewaschen

Now is rp-acc washed

‘Everybody is going to wash him- and herself now’

In addition to the middle interpretation in (1), transitive reflexive sentences have yet another interpretation. Both the subject and the reflexive pronoun can be linked to the first and second semantic argument respectively, as can be seen in example (4). In this case the reflexive pronoun does not indicate valency reduction. We call this ‘active’ interpretation of transitive reflexive sentences reflexive interpretation.

The attentive reader may have noticed that example (4) is ambiguous between the (active) reflexive interpretation ‘Peter is washing himself quickly’ and a middle interpretation ‘Peter washes quickly’. Because both the middle interpretation and the reflexive interpretation are two possible readings of transitive sentences with an accusative reflexive pronoun in object position, such sentences are ambiguous if both interpretations make sense.

(4)

Peter wäscht sich schnell

Peter-nom washes rp-acc (i.e. himself) quickly

‘Peter is washing (himself) quickly’

A third interpretation for transitive reflexive sentences like (1) and (4) has also been mentioned. The anticausative variant of verbs like öffnen (‘open’) is also reflexive. Hence, in German the accusative reflexive pronoun does not only indicate valency reduction in middle constructions but also in anticausatives like (5). But unlike middle constructions, anticausatives do not include an implicit semantic argument. Sentence (5) does not imply that someone or something is opening the door.

(5)

Die Tür öffnet sich

The door-nom opens rp-acc

‘The door opens’

Many Indo-European and non-Indo-European languages use reflexive elements for more or less identical purposes. In this chapter we present a descriptive survey of reflexive constructions in some Indo-European languages. We are mainly concentrating on reflexive constructions indicating valency reduction like middle constructions in (1) or anticausatives in (5). Moreover, we show that reflexive constructions are always ambiguous between an argument and a non-argument interpretation of the reflexive pronoun. Thus in all languages under discussion they also receive the reflexive interpretation illustrated in (4). That is, reflexive constructions are systematically ambiguous in many languages. We start off small and enlarge the picture of the middle step by step. In section 2.1 we briefly illustrate the syntactic and semantic properties of middle constructions in German. In section 2.2 we discuss anticausatives and inherent reflexives. In section 2.3 we turn to further Indo-European languages and show that the correlation between valency reduction and reflexivity we observe in German is not accidental. We argue that there is good crosslinguistic evidence to subsume the analysis of transitive reflexive sentences in German under the major phenomenon of middle voice.

In this chapter we do not want to make any theoretical claims whether we prefer a lexical, syntactic, or (postsyntactic) semantic analysis of middle and related constructions. Nevertheless, every description of facts partly depends on underlying theoretical concepts and their specific terminology.

The terminology is mainly influenced by lexical and syntactic theories of middle formation. Lexical theories postulate some lexical manipulation of the argument structure (i.e. a lexical rule of argument suppression or a middle template). Syntactic theories derive middle constructions like passives by case movement: a deep structure object is moved to subject position at surface structure. We discuss these theories and their shortcomings at length in chapter 3. For the time being we want to describe the properties of middle and related constructions as neutral as possible.

Therefore, we have to be careful with the terminology we are going to use. In the following presentation we distinguish between syntactic and semantic properties of a sentence. Hence, we must also distinguish between syntactic and semantic terminology. (In-) transitivity, (nominative) subject, accusative object or dative object are used as descriptive syntactic terms.

We do not talk about direct and indirect objects because German distingushes objects on the basis of case. For example, the second argument of two-place verbs can be assigned either accusative, dative or genitive case (and it can also be linked to a prepositional phrase). There are good arguments that grammatical functions should not be an essential part of the grammar of German (cf. e.g. Reis 1986 and Sternefeld 1985).

On the semantic side, we use one-place predicate, two-place predicate, semantic argument (variable) and external and internal semantic argument as descriptive semantic terms. The valency (or argument structure) of a predicate is simply represented as an ordered set of one, two, or three arguments. The first semantic argument of two-place predicates is often called the ‘logical subject’ and the second semantic argument the ‘logical object’ of the verb. However, we do not use the terms ‘logical subject’ and ‘logical object’, because subject and object belong to syntactic terminology. One-place predicates can be subdivided in unergatives and unaccusatives. The only semantic argument of unergatives is external whereas the only semantic argument of unaccusatives is internal. Besides, we avoid the term argument in syntax. Instead of arguments, we talk about subjects and accusative, dative, or prepositional objects.

Finally, a comment on grammaticality judgements seems to be necessary. We already saw that transitive sentences with a reflexive pronoun in the position of the accusative (or direct) object are multiply ambiguous between a middle, anticausative, inherent reflexive or reflexive interpretation. Note that sometimes the middle reading is hard to get. This might be due to the following two obervations which seem to be relevant for the interpretation of semantically ambiguous sentences.

Semantic representations which do not require valency change seem to be the unmarked case. Hence, the preferred interpretation for a transitive reflexive sentence seems to be as follows: the verb is interpreted as a two-place predicate and both the syntactic subject and the syntactic object are linked to one of the verb’s semantic arguments. Especially with subjects that refer to animate entities, the reflexive interpretation is preferred, because they are very likely to be interpreted as proto-agents.
Semantic representations without implicit arguments are the unmarked case. According to (ii), the anticausative interpretation of transitive reflexive sentences is preferred over the middle interpretation.

Besides, reflexives and anticausatives, like passives, usually refer to specific events whereas the middle interpretation involves generic quantification. This might additionally hinder the discourse linking of middle construction. Hence, the middle interpretation is often the least preferred option for a transitive reflexive sentence. Sometimes it takes some time to grasp the middle reading. Furthermore, middle constructions usually require additional adverbial modification (but cf. section 2.1.4 and chapter 7). Therefore, especially middle constructions without any adverbial modification need an appropriate context to be licensed. Last but not least, middle constructions in German are semantically closely related to three further constructions. Lassen-middles like (6.a) and tough-movement construction llike (6.b) are semantically nearly identical to common middle constructions, cf. e.g. Fagan (1992: 210f.) for lassen middles. The ‘active’ sentence with the indefinite ‘generic’ subject man (‘one’) and the modal können (‘can’) in (6.c) is also a close paraphrase of the middle construction in (1). Note that (6.c) does not involve valency reduction.

(6)

a.

Das Brot läßt sich gut schneiden

(lassen-middle)

The bread lets rp well cut

‘The bread cuts easily’

b.

Das Brot ist gut zu schneiden

(tough-movement)

The bread is good to cut

c.

Man kann das Brot gut schneiden

(indefinite pronoun)

One can the bread easily cut

‘One can cut the bread easily’

The middle construction in German

Before we turn to the plot of the middle-story we must introduce the protagonists. Although a lot has been written about middle constructions in German and related languages (cf above all the detailed overview in Fragan 1992 and Abraham 1995b)

See also Wagner (1977). Some data of the following presentation are from Fagan’s book on middle constructions. Most of the other examples are taken from German newspapers or books. It will become clear in chapter 3 that we disagree in several respects with Fagan’s description of the relevant properties of middle constructions in German. Furthermore, Fagan does not mention adjunct middles in German. Our disagreement with Fagan and other authors concerns among others the constraints on Aktionsarten, aspectuality, adverbial modification, the implicit subject, and the so-called ‘static’ interpretation middle constructions are supposed to have. We will discuss these shortcomings in detail in chapter 3 and 7.

the following survey is necessary for four reasons: firstly, many of the restrictions suggested for middle constructions in German turn out to be more complex at second sight; secondly, constructions like adjunct middles have not been noticed for German yet; thirdly, recent analysis do not put middle constructions into the context of transitive reflexive sentences; and fourthly, the argumentation in the following chapters will make use of the examples introduced in this section. In 2.1.1 we focus on verbs in middle constructions. In 2.1.2 we turn to the syntactic subject of middle constructions. Section 2.1.3 deals with the reflexive pronoun, and in 2.1.4 we discuss adverbials and further (semantic) issues.

Verbs in the middle construction

(7) are further examples of typical ‘personal’ or ‘transitive’ middle constructions that correspond to transitive sentences in the active voice. All examples contain verbs that select two semantic arguments. In the following we mention the (in-) transitivity of the corresponding active sentences in parenthesis in each case (example (7.a) is taken from Bernhard Schlink, Der Vorleser, and (7.c) is from Harry Rowohlt, Pooh’s Corner).

The valency of many verbs varies. Extrem examples are polyvalent verbs like e.g. rollen (‘roll’) or schlagen (‘hit’), cf. Vogel (1998). By two-place predicate, for instance, we mean that the respective verb is interpreted as a two-place predicate in this context. The same holds for many terms used in this chapter like ‘achievement’, ‘accomplishment’ or ‘activity’, cf. also footnote 18 below.

(7)

Two-place predicates (transitive):

a.

Aber richtig war, dass [der Bericht] sich anders las

Präteritum

But correct was, that the report rp differently read

‘However, it was correct that the report read differently’

b.

Das Buch wird sich wie ein Kriminalroman lesen

Future Tense

The book will rp like a crime story read

c.

… frierend schreibt sich irgendwie besser hin

Present Tense

… being-cold writes rp somehow better verbal-particle

d.

Das Klavier hat sich schlecht gespielt

Perfekt

The piano has rp badly played

We already mentioned that middle constructions in German are syntactically transitive themselves, which will be discussed in great detail in chapter 4. Note that the middle construction does not change the morphological form of the verb. Middle constructions are gramamtical in present, past (Präteritum) or future tense and with perfect aspect (Perfekt and Plusquamperfekt). In addition to personal middle constructions, German has also so-called impersonal or ‘intransitive’ middle constructions (cf. Fagan 1992: 44).

The term ‘intransitive’ middle construction is somewhat misleading, because middle constructions are always transitive. Impersonal or ‘intransitive’ middle constructions only correspond to intransitive active sentences.

They correspond to intransitive sentences and have a pleaonastic nonreferential element (the third person neuter pronoun es) in the position of the grammatical subject.

Abraham (1995b) mentions another kind of impersonal middle construction. In this case the verb agrees with impersonal subject es (third person singular), although the middle construction is derived from a two-place predicate. The second argument (the plural NP solche Beamte) is not linked to the syntactic subject of the sentence but receives accusative case. This construction equals the impersonal si-construction in Italian (cf. 3.1.2).
(i)

a.

???…weil es sich solche Autos gut fährt

… because it (sg.) rp such cars-acc (pl.) well drives (sg.)

b.

???… weil es sich diese Bücher gut liest

… because it rp these books-acc well reads

c.

???…weil es sich einen solchen Beamten leicht besticht

… because it rp such an official-acceasily bribes

Most native speakers we asked find these impersonal constructions hardly acceptable. Nevertheless, we share Abraham’s judgements and think that they are not ungrammatical. Note, however, that all three sentences in (i) sound old-fashioned and stilted, and they are very uncommon in Modern German. This kind of impersonal middle construction lies beyond the scope of all the analyses of middle constructions we discuss in chapter 3. Additional principles seem to be necessary in any case to explain this specific impersonal construction. We neglect it in the following discussion.

The sole argument of the one-place verb is suppressed and there is no argument left that can be linked to the subject position. (8) are a few examples with typical intransitiv unergative verbs like wohnen (‘live/reside’), schlafen (‘sleep’) or jodeln (‘yodel’). The first example is taken from Franz Hessel, Ein Flaneur in Berlin, (new edition of Spazieren in Berlin, Berlin 1927), the second example is from Waltraut Lewin, Louise, Hinterhof Nord, the third one is from Fagan (1992: 243), and the last example is from the Frankfurter Allgemeine Zeitung, 11.9.2000.

(8)

One-place predicates (intransitive - unergative):

a.

Hier wohnt sich's altertümlicher und heimlicher als in den belebten Straßen am südlichen Tiergartenrand

Here lives rp itmore ancient and homey than in the busy streets at-the southern edge of the Tiergarten

b.

Nun schläft es sich doch ein bischen besser

Now sleeps it rp well particle a littlet better

c.

Mit der Heimat im Herzen jodelt es sich überall gut

With home in the heart yodels it rp everywhere well

‘With home in your heart, you can yodel well everywhere’

d.

Mit blauen Augen flirtet es sich leichter

With blue eyes flirts it rp more-easily

Besides unergative one-place predicates, unaccusative/ergative predicates are also grammatical in middle constructions. German has two classes of one-place verbs that differ in many respects: unergatives and unaccusatives/ergatives. Among other things, unaccusative/ergative verbs select the auxiliary sein (‘be’)

German also has few two-place verbs that select sein (‘be’) instead of haben (‘have’).
(i)

Ich bin die ganze Stadt abgelaufen

I be the whole city down-walked

Van Riemsdijk (1978) argues that examples like (i) involve postposition-incorporation. The head of the postpositional phrase die ganze Stadt ab incorporates into the unaccusative/ergative verb laufen and behaves like a separable particle (cf. also Fagan 1992 and Ackema and Schoorlemmer 1995).

and their subject or first semantic argument can be attributively modified by the past participle and the present participle.

Further tests are –er-nominalization, VP-topicalization, impersonal passives and stress assignment. Note that these tests do not always give a clear classification and that they do not provide compelling evidence for a syntactic derivation. Besides, further aspects has to be taken into account as, for example, aspectuality in the context of auxiliary selection. We return to unaccusatives in chapter 6.

Unergatives, on the other hand, select haben (‘have’), and their subject cannot be modified by the past participle but only by the present participle, cf. Grewendorf (1989b) and Fagan (1992) for further discussion. According to these two tests, the verbs in (9.a) and (9.b) are clear examples for ergative or unaccusative verbs: they select sein and their past participle can modify the subject. One-place verbs of movement like reisen (‘travel’) in (9.c) or fahren (‘drive’) in (9.d) have less proto-patient properties than sterben or einschlafen, cf. Dowty (1991). Nevertheless, they also select sein and modification of the subject by the past participle is also possible if we add an adverbial or a directional PP, e.g. ein weit gereister Künstler (‘an artist who travelled far’) or der nach Hamburg gefahren Zug (‘the train that goes to Hamburg’).Example (9.a) is from the Berliner Zeitung, 22/23.11.97 and example (9.b) is from Fagan (1992: 243).

The examples in (i) illustrate the be/have selection with perfect tense in German. Unaccusative verbs like arrive (i.a) select be while unergative verbs like sleep (i.b) select have.
(i)

a.

Peter ist auf der Autobahn gefahren

Peter be on the highway driven

b.

Peter hat im Bett geschlafen

Peter have in the bed slept

Example (ii) illustrates that past participles can be used only to modify the accusative object of a transitive sentence (i.e. the second or internal argument of the verb) (ii.b), whereas present participles modify the subject of the sentence (i.e. the first or external argument) (ii.c).
(ii)

a.

Der Mann trinkt ein Bier

The man-NOM drinks a beer-ACC

b.

das getrunkene/*trinkende Bier

The PAST-PART./PRESENT-PART. beer

c.

Der *getrunkene/trinkende Mann

The PAST-PART./PRESENT-PART. man

The crucial difference between unaccusatives and unergatives is illustrated in (iii): the subject of unaccusatives can be modified by both the past participle and the present participle. The subject of an unergative, on the other hand, can only be modified by the present participle like the subject of the transitive sentence in (ii).
(iii)

a.

Der gestorbene/sterbende Mann

The PAST-PART./PRESENT-PART. man

b.

Der *geschlafene/schlafende Mann

The PAST-PART./PRESENT-PART. man

(9)

One-place predicates (intransitive - unaccusative):

a.

Gesundheitsstudie: In welchem Bezirk stirbt es sich am frühesten

Study on health: In which district dies it rp at the earliest

b.

Bei hellem Licht schläft sich’s nicht so gut ein

With bright light fall-asleep rp it not that well

c.

Dann...reist es sich besser

Then...travel it rp better

d.

Es fährt sich gut auf der Autobahn

It drives rp well on the highway

The analysis of ergativity is an interesting issue of its own, which lies beyond the scope of this study. However, we will argue in chapter 6 that our analysis of middle constructions provides an argument against a syntactic analysis of unaccusative/ergative verbs. From a lexical point of view, middle formation can be described as a function that takes as input a predicate with at least one argument and ‘demotes’ the first semantic argument of the verb. This argument need not be external.

Perlmutter (1978), Burzio (1986), or Grimshaw (1990) among others argue that unaccusative verbs select an internal argument. The syntactic subject of a sentence with an unaccusative verb is linked to the internal argument of this verb. This linking-configuration can also be found in the causative-alternation. It is the the syntactic subject of the unaccusative verb break/zerbrechen in (i.a) and (ii.a) that is realized as object of the corresponding causative variant in (i.b) and (ii.b). In both examples, the NP the vase/die Vase is linked to the same semantic argument (cf. also Levin and Rappaport 1995).
(i)

a.

The vase broke

b.

His mother-in-law broke the vase

(ii)

a.

Die Vase zerbricht

b.

Seine Schwiegermutter zerbricht die Vase

Consequently, verbs like regnen (‘rain’) or tauen (‘thaw’), which do not select a semantic argument, are ungrammatical in middle constructions.

Most of these zero-place verbs can be also used with a semantic argument.
(i)

a.

Es regnet

and

b.

Blätter regnen auf das Dach

It rains (it is raining)

Leafs are raining on the roof

(ii)

a.

Es taut

and

b.

Der Schnee taut

It thaws (it is thawing)

the snow is thawing

Under our perspective these verbs can be input to middle formation if they yield the interpretation in (i.b) and (ii.b). In fact, sentence (10.b) can receive a middle interpretation. In this case the verb must be interpreted as an one-place predicate with an implicit argument. This interpretation resembles the interpretation of unaccusatives in middle constructions and is of course absurd. But imagine a fairytale with two snowflakes talking to each other. The topic of the conversation is their experience of thawing in different cities. In this context one of the snowflakes can actually utter sentence (10.b). But for this we must interpret the verb as one-place predicate.

(10)

a.

Im Frühjahr taut es in Berlin sehr schnell

In spring thaws it in Berlin very quickly

b.

*Im Frühjahr taut es sich in Berlin sehr schnell

In spring thaws it rp in Berlin very quickly

The middle construction itself is unergative. Personal and impersonal middle constructions select haben (‘have’) as their auxiliary and the past participle cannot attributively modify the subject. Instead, we have to use the present participle, cf. also Ackema and Schoorlemmer (1994:61f.) for Dutch.

(11)

a.

Unter den Linden hat es sich schon immer gut flaniert

‘Unter den Linden’ has it rp always well strolled

b.

Das sich gut lesende Buch vs.*Das sich gut gelesene Buch

The rp well reading book The rp well read book

(12) and (13) are further examples for personal middle constructions. (13.a) – (13.c) are middle constructions derived from three-place predicates. It is again the accusative object of the active counterpart that corresponds to the subject of the middle construction.

(12)

three-place predicates (ditransitive)

a.

Fahrräder laden sich jetzt leichter in unsere Wagen

Bikes load rp now more easily in our carriages

b.

... weil sich süßer Hustensaft kleinen Kindern besser einflößt

... because rp sweet cough syrup small children-dative better fills-in-their-mouth

c.

Dieses Buch verkauft sich (den Nonnen) hervorragend (an Nonnen)

This book sells rp (the nons-dative) excellently (to nuns)

In addition to three-place predicates, we also find resultatives in middle constructions in German. Both, adverbial (13.a) and prepositional (13.b) and (13.c) secondary predicates are grammatical.

(13)

Resultatives:

a.

Warmes Metall hämmert sich einfacher flach

Warm metal hammers rp more easy flat

b.

Diese Füllspachtel quetscht sich sehr gut in die Fugen

This filling compound squeezes rp very well into the joints

c.

Kleine Menschen trinken sich schnell unter den Tisch

Small people drink rp quicker under the table

We conclude that all kinds of predicates that select at least one semantic argument (i.e one-place unergatives and unaccusatives, two-place and three-place predicates and resultatives) are generally perfectly grammatical in middle constructions and that middle constructions are not restricted to present tense. Apart from that, there is a second restriction on middle formation: individual-level predicates like wissen (‘know’), können (‘be able’, ‘know’), heißen (‘be called’) or abstammen (‘be descended’) cannot undergo middle formation at all. We will argue in chapter 7 that middle constructions involve generic quantification over the first semantic argument and the event (or situation) variable. Hence individual-level predicates that do not select an event/situation variable are excluded from middle formation.

(14)

a.

*Diese Antwort weiß sich leicht

This answer knows rp easily

b.

*Spanisch kann sich einfach

Spanish knows rp easily

c.

*So wie mein Vater heißt es sich nicht so leicht

Like my father names it rp not that easily

d.

*Vom Gorilla stammt es sich nicht so leicht ab

From the Gorilla be-descendedit rp not that easily part

Individual-level predicates can be subsumed under Vendler’s (1967) class of states (see Dowty 1979 and Fagan 1992:89f.). Note, however, that only individual-level predicates are excluded from middle formation. (15) would be an example of a middle construction with a stative verb provided that we classify a verb like sitzen (‘sit’) as a state (instead of an activity).

(15)

Auf diesem Stuhl sitzt es sich weitaus bequemer

On this chair sits it rp far more comfortable

In sum, middle formation is only possible with stage-level predicates. We find all kind of stage-level predicates in middle constructions.

The so-called aspectual properties of verbs (Aktionsarten) are not (necessarily) inherent lexical properties of single verbs but very often result from the interpretation of more complex structures including especially the verb and the direct or accusative object, cf. Dowty (1991) or Tenny (1994), among many others for further discussion of this issue.

A clear example of an achievement is etwas ausschalten (‘switch something off’) in (16.a). (16.b) is an example for a middle construction with an activity verb and (16.c) with an accomplishment.

(16)

a.

Der Fernseher schaltet sich schnell aus

The TV switches rp quickly off

b.

In dieser Ecke des Sees schwimmt es sich am besten

In this corner of the lake swims it self best

c.

Große Wände bemalen sich nicht so leicht

Big walls paint rp not that easily

Let us conclude that so far there are only two fundamental restrictions on the predicates that are grammatical in middle constructions: they must provide at least one semantic argument and they must not belong to the class of individual-level predicates.

The subject of middle constructions

Usually the syntactic subject of the middle construction corresponds to the accusative object of the active counterpart. Impersonal middle constructions are derived from one-place verbs and correspond to intransitive active sentences without an accusative object. They have an pleonastic or impersonal subject that is not linked to a semantic argument of the verb. In this resprect impersonal middle constructions differ from impersonal passives, which do not have a subject at all. In impersonal passives the third person neuter pronoun es is only grammatical in sentence-initial position of matrix-clauses. The pronoun in (17.a) is called Vorfeld-es (cf. Grewendorf 1988).

The third person neuter pronoun es fulfills quite different functions: referential personal pronoun, place-holder for complement-clauses, impersonal subject, and Vorfeld-es. An example for the Vorfeld-es in a simple active sentence is given in (i):
(i)

a.

Es ging ein Mann durch denn Wald und …

It went a man throuh the woods and …

b.

Durch den Wald ging (*es) ein Mann und …

c.

… weil (*es) ein Mann durch denn Wald ging und …

Like the reflexive pronoun in transitive reflexive sentences es can but need not be interpreted as an argument of the verb. It need not be referential. Therefore, the third person neuter pronoun, unlike other personal pronouns, can also be used for certain grammatical functions. For further differences between es and all the other personal pronouns see Cardinaletti and Starke (1994) and Gärtner and Steinbach (1996 and 2000).

As opposed to the pronoun in impersonal passives, the pronoun in impersonal middle constructions is a genuine (impersonal or pleonastic) subject that also occurs in the middle-field.

(17)

a.

Es wird hier getanzt

(impersonal passive)

It is here danced

‘People are dancing here’

b.

Gestern wurde (*es) getanzt

Yesterday was it danced

c.

... weil (*es) hier getanzt wird

...because it here danced is

(18)

a.

Es tanzt sich gut hier

(impersonal middle construction)

It dances rp well here

b.

Hier tanzt *(es) sich gut

Here dances it rp well

c.

...weil *(es) sich gut tanzt hier

...because it rp well dances here

The contrast between impersonal passives and impersonal middle constructions is yet another argument for an analysis that treats middle constructions as simple transitive sentences that are morphosyntactically ‘active’. We will come back to this issue in chapter 3 and 4.

So far, we saw that the syntactic subject in personal middle constructions corresponds to the accusative object of the active counterpart. There is, however, one exception to this correlation between the middle-subject and the accusative object of the active voice. Hoekstra and Roberts (1993) and Ackema and Schoorlemmer (1994) discuss another interesting kind of middle construction in Dutch which they call adjunct middles. The subject of the adjunct middle in (19.a), (20.a), and (21.a) corresponds to the PP-adjunct in the impersonal middle constructions in (19.b), (20.b), and (21.b) or in the active counterparts in (19.c), (20.c), and (21.c). Adjunct middles seem to be quite productive in Dutch. In German they seem to be more restricted.

Miller (1993: 183) notes that middle formation from an underlying or corresponding PP is possible in English as well. In this case the preposition seems to incorporate into the verb:
(i)

a.

?That tree climbs up quickly

b.

*That tree climbs quickly up

(ii)

That stove melts over quickly

At first glance native speakers of German sometimes judge adjunct middles not to be perfectly grammatical.

(19)

a.

^?Diese Schuhe laufen sich aber nicht sehr bequem

These shoes walk rp yet not very comfortably

b.

In diesen Schuhen läuft es sich nicht sehr bequem

In these shoes walks it rp not very comfortably

c.

In diesen Schuhen läuft man nicht sehr bequem

In these shoes walks one not very comfortably

(20)

a.

^?Mein neuer Füller schreibt sich gut

My new pen writes rp well

b.

Mit meinem neuen Füller schreibt es sich gut

With my new pen writes it rp well

c.

Mit meinem neuen Füller schreibt man gut

With my new pencil writes one well

(21)

a.

^?Diese Wolle strickt sich sehr angenehm

This wool knits rp very comfortably

b.

Mit dieser Wolle strickt es sich angenehm

With this wool knits it rp very comfortably

c.

Mit dieser Wolle strickt man angenehm

With this wool knits one comfortably

Not every adjunct feeds middle formation. Adjunct middles formation is subject to additional constraints, that restrict the kind of adjuncts that can undergo middle formation. (22.a) and (22.b) are two examples for adjunct middles that are not acceptable in German.

(22)

a.

*Diese Lampe liest sich besser

This lamp reads rp better

a’.

Mit/unter dieser Lampe liest es sich besser

With/under this lamp reads it rp better

b.

*Verstand schreibt sich leichter

Intellect writes rp more easily

b’.

Mit Verstand schreibt es sich leichter

With intellect writes it rp more easily

Hence, for some adjuncts German has two alternative middle constructions, the adjunct middle or the impersonal middle construction. The subject NP of the adjunct middle is still contained in the adjunct-PP in the corresponding impersonal middle construction, the subject position of which is again occupied by the impersonal subject es. Note, however, that the meaning of the adjunct middles in (19.a), (20.a) and, (21.a) is not totally identical to the meaning of the corresponding impersonal middle construction in (19.b), (20.b) and, (21.b). We will discuss adjunct middles in greater detail in chapter 7.

German middle constructions are subject to another important restriction on their subject. The subject of the middle construction cannot correspond to dative objects in the active counterpart. Hence, dative objects must not be ‘promoted’ to subject in middle constructions. Middle formation with arguments that are linked to a dative object in the active voice is ungrammatical in general. On the one hand, sentence (23.b) is ungrammatical if the reflexive pronoun is assigned accusative case. On the other hand, (23.b) does not yield a middle interpretation if the reflexive pronoun bears dative case, because dative reflexive pronouns cannot indicate valency reduction in German. (23.b) can only receives a reflexive interpretation. A verb selecting a dative object can, however, occur in impersonal middle constructions, cf. (23.c). In impersonal middle constructions, the dative objects preserves its case and the accusative reflexive pronoun indicates again valency reduction. Hence, verbs selecting dative objects are not excluded from middle formation in principle. We conclude that dative reflexive pronouns cannot indicate valency reduction and dative objects cannot be promoted to subject in middle constructions.

The same holds for the rare cases of genitive objects ((ii) is an impersonal middle construction):
(i)

a.

*Diese Gewohnheit enträt sich leicht

This habit-GEN do-without RP easily

b.

Dieser Gewohnheit enträt es sich leicht

This habit-NOM do-without it RP easily

(23)

a.

Wir helfen einem Obdachlosen

We-nom help a homeless-person-dat

b.

Ein Obdachloser hilft sich leicht

A homeless-person-nom helps rp-*acc/dat easily

‘A homeless person is helping himself easily’

(reflexive interpretation)

*‘A homeless person helps easily’

(middle interpretation)

c.

Einem Obdachlosen hilft es sich leicht

(impersonal middle constr.)

A homeless-person-acc helps it-nomrp-acc easily

Van Oosten (1977), Fagan (1992: 76f.), and Greenspon (1996) observe a further restriction on the subject of middle constructions, which can be outlined as follows: some property of the subject must be ‘responsible’ for the event described by the verb. This property of the subject is called (primary) responsibility. Fagan’s observation that there exists a contrast between buy and sell as well as between the corresponding German verbs kaufen and verkaufen illustrates this condition nicely. While sell and verkaufen form acceptable middle constructions, buy and kaufen sound odd.

(24)

a.

*These books buy well

/ *Diese Bücher kaufen sich gut

b.

These books sell well

/ Diese Bücher verkaufen sich gut

The properties of the object for sale may influence the act of selling. A best-seller can be sold more easily than a shelfwarmer. A parallel situation is hard to imagine for buy. However, a person’s selling abilities or the availability of an entity can have a positive or negative effect on the act of buying. In this context sentence (24.a) gets much better.

Sentence (25.b) is due to Manfred Bierwisch.

(25)

a.

Bei fachlich geschultem Personal kauft sich die richtige Software letztlich doch schneller als im Discounter

With qualified personnel buys rp the right software in the end particle faster than in a discount store

‘In the end the right software buys faster with qualified personnel than in the d.s.’

b.

Standardgrößen kaufen sich leichter als Sondergrößen

Standard-size buys rp more easily then extra-size

The significance of the promoted second semantic argument (i.e. the subject of the middle construction) for the event described by the verb affects the acceptability of middle constructions. Again, (primary) responsibility seems to be a property of the ‘subjects’ of active sentences. As opposed to subjects in middle constructions, subjects in passives are not subject to this restriction (cf. Lakoff 1977 and Greenspon 1996). Impersonal middle constructions usually contain another constituent (e.g. a prepositional phrase) that is ‘responsible’ for the event. This can be seen in example (26): the quality of a bed can be very important for the way we sleep. In section 7.1 we come back to this issue.

(26)

In diesem Bett schläft es sich hervorragend

In this bed sleeps it rp excellently

It has often been claimed that only verbs with ‘affected’ internal arguments may undergo middle formation (cf. Roberts 1987, Hale and Keyser 1987, Hoekstra and Roberts 1993, or Rapoport 1993). This restriction is, however, much too strong. Verbs without ‘affected’ internal arguments are also grammatical in middle constructions. This is illustrated by the following sentences (cf. also section 3.1.1 and 3.2.1). The first two examples are taken from Fagan (1992: 65). The corresponding German examples are also grammatical. Example (27.e) is from the Schwäbische Tagblatt, 27.11.1999.

(27)

a.

This book reads easily / Dieses Buch liest sich leicht

b.

She photographs well / Sie photographiert sich gut

c.

Diese Geschichte vergißt sich nicht so leicht

This story forgets rp not that easily

d.

Von hier aus sieht sich das gegnerische Tor viel besser

From here sees rp the opponent’s goal much better

e.

Im Schaufenster sieht sich Weihnachten noch schöner an als in der Wirklichkeit

In the display-window look-at rp christmas even nicer verb.part. than in the

reality

The reflexive pronoun

Unlike middle constructions in English and Dutch, which must not have a reflexive pronoun in object position, personal and impersonal middle constructions in German are ungrammatical without the accusative reflexive pronoun. The reflexive pronoun is always bound in syntax by the subject of the sentence, cf. also Abraham (1995b:14f.) for examples similar to (28). In impersonal middle constructions it is the expletive subject es that binds the reflexive pronoun.

(28)

a.

Ich₁ schreibe mich₁ mit ‘st’

I-1.sg write rp-1.sg with ‘st’ (i.e. ‘my name has to be written with ‘st’’)

b.

Du₁ verkaufst dich₁ gut - ich meine, dein Buch₂ verkauft sich₂ gut (Reis 1981)

You-2.sg sell rp-2.sg well - I mean, your book-3.sg sells rp-3.sg well

c.

Auf dieser Party tanzt es₁ sich₁ prima

At this party dances it-3.sg rp-3.sg fantastically

Table (29) illustrates that in the first and second person, reflexive pronouns cannot be distinguished from personal pronouns.

(29)

Accusative and dative personal pronouns and reflexive pronouns in German

Furthermore, accusative and dative forms can only be distinguished in first and second person singular. Hence, we can tell accusative from dative pronouns in the first and second singular slots and reflexive pronouns from personal pronouns in the third person singular and plural slots. One might either assume that personal pronouns and reflexive pronouns in German are homonymous in the first and second person or that German does not distinguish reflexive pronouns from personal pronouns in the first and second person. The formal identity of first and second person might be due to the fact that the antecedent is always clearly identifiable in the discourse. Hence, it makes no difference whether the pronoun is locally bound (principle A) or locally free (principle B). Burzio (1989) argues that a personal pronoun can be locally bound if the paradigm has no reflexive pronoun (or anaphor) or, to put it the other way round, that it is always the least specified element in that paradigm that can be locally bound. According to Burzio’s morphological economy a NP that is bound must be maximally underspecified. Reflexive pronouns (anaphora) have less specified Φ-features than personal pronouns. Therefore, locally bound personal pronouns are grammatical only if a language does not distinguish reflexive pronouns from personal pronouns, cf. also the discussion in Gärtner (1991). In section 2.4 we present further evidence for Burzio’s theory from Middle High German. His theory will also be relevant for the definition of the middle marker which we develop in chapter 5.

Note that the politeness form Sie of the second person is morphologically identical with the third person female singular and the third person plural personal pronoun sie. The politness form is totally regular. The corresponding reflexive pronoun is the third person reflexive pronoun sich:
(i)

Können Sie *Sie/sich bitte hierhin setzen?

Could you RP please here sit down

Adverbials, ‘genericity’, and the implicit argument

It has often been claimed that middle constructions in German and English require some additional adverbial modification, cf. Fagan (1992), Haider (1982), Hoekstra and Roberts (1983), or Bierwisch (1997). However, this seems to be an overgeneralization because we also find middle constructions without any adverbial modification, as can be seen in (30), cf. also Ackema and Schoorlemmer (1994) and Iwata (1999). Example (30.d) is from Fagan (1992: 43), example (30.f) from Theodor Fontane, Irrungen, Wirrungen, and example (30.g) from Der Spiegel 28/1999.

(30)

a.

Welche Tür öffnet sich?

Which door opens rp

‘Which door can be opened?’

b.

Nimm diese Tür da, die öffnet sich!
See Fagan (1992:157) for a similar example in English.

Take that one over there, it opens rp

‘Take that one. It can be opened’

c.

Nur keine Angst. Dein Ohring wird sich finden

‘Anything but fear’. Your earring will rp find

d.

Dieses Kleid hat keinen Reißverschluß. Es knöpft sich zu

This dress has no zip. It buttons rp particle

e.

Jetzt ist es schwer, aber es vergißt sich alles.

Now is it hard, but it forgets rp everything

f.

Die entscheidende Frage nach den Kriterien der Amerikaner [i.e amerikanischen

Lektoren] beim Ankauf eines Manuskripts fand regelmäßig die verblüffend ein

fache Antwort:

When asking what Americans [i.e. American editors] consider to be crucial crite

ria for accepting a manuskript you always get the same answer:

“Dass es uns gefällt und dass es sich verkauft.”

That it us pleases and that it rp sells

‘That we like it and that it sells’

We will argue in chapter 7 that the observation that middle constructions usually need some kind of adverbial modification follows from certain conditions on assertions to be pragmatically licensed. Take, for example the middle construction in (30.g), which is not very informative in many situations because a book normally can be sold. Therefore, we are usually interested in whether or not a book sells well. But in some situations (e.g. if we are interested in whether or not a book is a shelf-warmer) the middle construction in (30.g) makes sense. We will argue the adverbial modification is not crucial for the grammaticality of middle constructions, cf. e.g. Ackema and Schoorlemmer (1994) for the same observation with respect to English and Dutch middle constructions. Other examples without adverbial modification are middle constructions with negation and modal auxiliaries in the subjunctive. Example (31.a) is attested by a friend.

(31)

a.

... und Tabellen, die sich nicht drucken

... and tables, that rp not print (i.e. that does not print)

b.

Diese Tür könnte sich öffnen

This door might rp open

Typical adverbials in middle constructions are manner adverbials like gut (‘well’) or leicht (‘easily’)

Sometimes we also find strange adverbial modifications. In example (i) the adverbials describe a property not of the reading event but of the subject itself (Die Zeit, 22.1.98).
(i)

Seine Geschichten aus dem Bürgertum (Süd-) Europas lesen sich gebildet, lebens- und redegewandt, kosmopolitisch

His stories situated in the middle class of (Southern) Europe reads RP educated, streetwise and articulated, cosmopolitan

whereas subject-oriented adverbials are ungrammatical, cf. (32.a). In this respect middle constructions differ from passives, cf. (32.b).

(32)

a.

*Das Brot schneidet sich absichtlich

The bread cuts rp on purpose

b.

Das Brot wurde absichtlich geschnitten

The bread was-pas on purpose cut

Furthermore, adverbial phrases like wie Butter (‘like Butter’) or comparative adverbials are also possible in middle constructions. Reference to the suppressed ‘logical’ subject is again ungrammatical.

(33)

a.

Warmes Brot schneidet sich wie Butter

Warm bread cuts rp like butter

b.

Das Buch liest sich besser als du denkst

The book reads rp better than you think

c.

*Sein Auto fährt sich wie Niki Lauda (cf. Peter fährt sein Auto wie Niki Lauda)

His car drives rp like Niki Lauda (cf. Peter drives his car like N. L.)

A further common assumption is either that middle constructions are ‘generic sentences’ or that ‘middle-verbs’ are individual-level predicates. According to the first assumption middle constructions are characterizing sentences in the sense of Carlson and Pelletier (1995), which involve ‘generic’ quantification over the implicit argument and the event/situation variable. A related issue is modality. Middle constructions normally involve some modal interpretation including ability, possibility, and sometimes necessity, cf. Fagan (1992). The middle construction in (34) is true if a situation in which someone reads this lovestory is very likely to be a situation in which this lovestory is good to read for this person.

(34)

Diese Liebesgeschichte liest sich gut

This lovestory rp reads well

We will argue in chapter 3 and 7 that middle constructions, unlike passives, are characterizing sentences. The generic quantifier binds the implicit argument and the situation variable.

Fagan (1992: 159) points out that middle constructions in French can be eventive, i.e. sometimes they can receive a passive interpretation. In this sense they resemble our German example in (35). French middle constructions (as well as Italian middle constructions, cf. section 2.3.2 and 3.1.2 below) can, however, refer more freely to particular events. Most of the French (and Italian) examples would be ungrammatical in German. This might be due to the morphosyntactic difference between the German and the French/Italian reflexive element: whereas the reflexive pronoun in German is an independent word in syntax, it is a verbal clitic in French/Italian. There seems to be a correlation between the passive interpretation of reflexive constructions and the degree of ‘grammaticalization’ of the reflexive element.
(i)

La question s’est discutée hier dans la salle du conseil

‘The issue was discussed yesterday at the council hall’

(ii)

Les vivres se distribueront tout à l’heure au premier étage

‘The food will be distributed in a while on the first floor’

Furthermore, middle constructions in English can sometimes be used in progressive form to refer to a specific event:
(iii)

Yesterday afternoon your new book was selling like hell

Besides, Fagan (p. 59) observes that French middle constructions - in contrast to English and German middle constructions - are not necessarily associated with the notion of modality.

As a consequence, ‘middle verbs’ are not individual-level predicates. Note that generic quantification in middle constructions can be restricted to specific periods of time, as can be seen in (35). The generic and modal interpretation of middle constructions will be discussed in detail in section 7.1. Thanks to Hans-Martin Gärtner for example (35).

(35)

Der Bach hat sich gestern Abend ausnahmsweise mal ganz gut gespielt

The Bach has rp yesterday evening exceptionally once quite well played

‘Exceptionally, this piece by Bach played well yesterday evening’

A third general assumption concerns the suppressed external argument. It is widely accepted that unlike in passives the suppressed argument cannot be realized in overt syntax in middle constructions. In passives the external argument can be syntactically expressed by a von- (‘by-’) phrase (36.a). By-phrases are ungrammatical in middle constructions, cf. (36.b).

(36)

a.

Castorp wurde von der russischen Patientin verführt

Castorp was-pas by the russian patient seduced

b

*Dieses Buch verkauft sich von Hans ohne Probleme

This book sells rp by Hans without problems

The preposition von (‘by’) cannot be linked to the suppressed argument in middle constructions. But sometimes this job can be done by the preposition für (‘for’), cf. also Stroik (1992) for English and Condoravdi (1989) for Greek. In (37) the small children are the ones who can read these books well. Note, however, that linking of the implicit argument is highly restricted in middle constructions. This might be due to the generic quantification over the implicit argument. As opposed to middle constructions, passives do not involve generic quantification and their implicit argument can much more easily be linked to syntax (cf. the contrast in example (38) and chapter 7 for further discussion of this issue).

(37)

Ich finde, dass sich diese Bücher auch für kleine Kinder ganz gut lesen

I think that rp these books also for small children quite well read

(38)

a.

Diese Bücher werden von meinem Kollegen verkauft

(passive)

These books are by my colleague sold

b.

??Diese Bücher verkaufen sich für meinen Kollegen nicht so gut

(middle c.)

These books sell rp for my colleague not so well

Our final remark concerns the thematic interpretation of the suppressed argument. We already saw that it need not be the external argument. Recall that unaccusative verbs form perfectly grammatical middle constructions. The subject of unaccusative verbs have typical patient properties, cf. Dowty (1991). Hence, the suppressed argument need not be the actor or agent of the verb/event. Other examples that confirm this observation are two-place verbs like verlieren (‘lose’), vergessen (‘forget’), or finden (‘find’). They do not assign the thematic role agent or actor to their external or first argument. Nevertheless, they form perfectly grammatical middle constructions (the second example (39.b) is from Stefan Zweig, Joseph Fouché, and the third example (39.c) from the Frankfurter Rundschau, 27.9.97).

(39)

a.

Diese neuen kleinen Münzen verlieren sich aber sehr einfach

These new small coins lose rp really very easily

b.

Solche Erinnerungen vergessen sich nicht

Such reminiscences forget rp not

c.

... ein Telefonbuch fand sich nicht

... a phonebook found rp not

The following figure summarizes the presentation of middle constructions in German. The syntactic elements that belong to a middle construction are given in the first column. The corresponding conditions discussed in this section are given in the second and third column.

(40)Middle constructions in German

NP _NOMINATIVE

A. The subject is either (i) es
(imp.m.) or (ii) referential NP
(pers.m.).
B. If (ii), then the corresponding
constituent can be: (i) accusative object
(standard middle construction) (ii)
complement of a preposition (adjunct
middle constr.). *Dative objects and
*genitive objects are ungrammatical.

C. ‘Responsibility’ (in case of impersonal
middle constructions there
must be another element e.g. a
prepositional phrase).
D. The suppressed argument can
sometimes be realized in a für-PP and
it does not have to be an external
argument, agent, or actor.

Verb

- at least one argument
- generic interpretation
- *individual-level predicates

RP _ACCUSATIVE

- bound by the subject
- not linked to a semantic argument

(Adverbial)

- not obligatory

- manner adverbial
- comparative adv.
- *subject oriented

Anticausatives and inherent reflexives

Besides middle constructions, anticausatives also systematically use the accusative reflexive pronoun to indicate valency reduction. The anticausative variant (41.b) of verbs like biegen (‘bend’) that can undergo the causative-anticausative alternation must be formed with a reflexive pronoun similar to the middle construction. In (41.c) we list more verbs that behave the same way.

(41)

a.

Hans-Georg biegt den Stock

Hans-Georg bends the stick

b.

Der Stock biegt *(sich)

The stick bends rp-acc

c.

further examples of class I verbs: öffnen (‘open’), schließen (‘close’), füllen (‘fill‘), leeren (‘empty’), aufwärmen (‘warm up’), aufklären (‘solve’), falten (‘fold’), glätten (‘smooth’), erhellen (‘light up’), verdunkeln (‘darken’), vergrössern (‘enlarge’), verkleinern (‘reduce’), stabilisieren (‘stabilize’), beruhigen (‘calm down’), drehen (‘turn’), …

In addition to this class of ‘reflexive-anticausative’ verbs, German has yet another class of verbs the anticausative variant of which is formed without an accusative reflexive pronoun. They equal anticausatives in English. As opposed to the reflexive-anticausatives in (41) the non-reflexive-anticausatives in (42) are unaccusative (cf. section 2.1.1 above).

(42)

a.

Hans-Georg bricht den Stock

Hans-Georg breaks the stick

b.

Der Stock bricht (*sich)

The stick breaks rp-acc

c.

further examples of class II verbs: rollen (‘roll’), fliegen (‘fly’), trocknen (‘dry’), zerbrechen (‘smash’), zerknittern (‘crumple’), abbrechen (‘break off’), einfrieren (‘freeze’), auftauen (‘thaw’), …

In (41) the reflexive pronoun indicates valency reduction again. The verbs of the first class are two-place predicates, which undergo valency reduction, i.e. the anticausative in (41.b) is derived from the underlying two-place verb in (41.a). As for the second class the situation appears to be the opposite. In this case one could argue that the one-place predicate (42.b) is the underlying form and the causative variant (42.a) is derived from this unaccusative one-place verb by adding a first or external argument, cf. Wunderlich (1993). Both middle formation and anticausative formation with class I verbs involve valency reduction indicated by an accusative reflexive pronoun.

There is, however, one crucial difference between anticausatives and middle constructions: the former, unlike the latter, have no implicit semantic argument at all. The first semantic argument is not only suppressed but also completely removed from the semantic representation. It is part of our knowledge about the world that there must be some cause for events (a human being, a physical force or natural force, ...), but in contrast to middle constructions, the causing entity (which corresponds to the first or external argument) is not implied in anticausatives, possibly because it cannot be observed in the event described by the verb. In sentence (43.a), for example, it is not implied that someone is rolling the ball. It simply describes a situation where a ball is rolling down the hill. The same holds for (43.b). Peter might be the person who opens the door, but this is again not implied. It might also be possible that the door opens automatically or that it is opened by a servant. Therefore, only anticausatives can be modified by von selbst (‘all by itself’), cf. example (44) and Fagan (1992: 20).

Sometimes middle constructions can be modified by von selbst. But these sentences do not literally mean that the described event takes place all by itself. Instead we observe some additional pragmatic effect. Sentence (i) means that it is obvious which representative team must be put together. (ii) means that the answer to this question is so obvious, that everybody must know it or that something happens that answers the question, and in (iii) we are talking about a best seller. A seller need not do very much for the sale of this book, cf. also Greenspon (1996) for similar examples in English.
(i)

Die Auswahl, der Ribbeck seine Premiere als Teamchef anvertraut,

The representative-team, to-which Ribbeck his premiere as choach entrust,

hat sich fast von selbst aufgestellt

(Berliner Zeitung)

has RP nearly all by itself nominated

(ii)

Die Frage beantwortet sich von selbst

(Alan Isler, Der Prinz der West End Avenue)

This question answers RP all by itself

(iii)

Dieses Buch verkauft sich ganz von selbst

This book sells RP all by itself

(43)

a.

Der Ball rollt den Berg hinunter

The ball rolls the hill down (i.e. ‘rolls down the hill’)

b.

Die Tür öffnete sich und herein kam Peter

The door opens rp-acc and in came Peter

(44)

a.

Die Tür öffnete sich von selbst

(anticausative)

The door opens rp-acc all by itself

b.

#Das Brot schneidet sich gut von selbst

(middle construction)

The bread cuts rp-acc well all by itself

(45.a-c) are further examples that illustrate the deletion of the first or external argument of the causative counterpart in anticausatives. Of course, there are reasons for the global warming, the widening of the fjord, or the filling of the glass. But all three examples only describe a change of state without implying an implicit cause that is responsible for these events.

Because no implicit first argument is present at all, only the syntactic subject, i.e. the only semantic argument of the anticausative, can control the pro subject of an infinitive clause. However, this would result in nonsense, as can be seen in (i). In contrast to this, the suppressed argument of the passive in (iii) and middle construction in (iv) can control the pro subject of the infinitive clause (cf. chapter 3 for further discussion of this issue):
(i)

#Das Glas füllt sich [um Maria zu ärgern]

(anticausative)

The glass fills RP in order Maria to annoy

The glass annoys Maria

(ii)

Peter füllt das Glas [um Maria zu ärgern]

(active/causative)

Peter fills the glass in order Maria to annoy

Peter annoys Maria

(iii)

Das Glas wurde (von Peter) gefüllt [um Maria zu ärgern]

(passive)

The glass was (by Peter) filled in order Maria to annoy

‘Impl.arg.’/Peter annoys Maria

(iv)

Das Buch liest sich nur gut [um einzuschlafen]

(middle construction)

The book reads RP only well in order to fall asleep

‘Impl.arg.’ falls asleep

(45)

a.

Die Atmosphäre hat sich in den letzten Jahren etwas aufgewärmt

The atmosphere has rp-acc in the last years a little warmed-up

b.

Der Fjord verbreitert sich wieder

The fjord widens rp-acc again

c.

Das Glas füllt sich mit Wasser

The glass fills rp-acc with water

Finally, we want to mention a second class of verbs that is related to the issue under discussion. German has so-called inherent reflexive verbs (absolut or echt reflexive Verben) that are ungrammatical without an accusative reflexive pronoun in object position, although they do not correspond to a causative counterpart. The verb schämen (‘be ashamed’) is the prototypical example.

German has a subclass of inherent reflexive verbs that can yield a reciprocal interpretation with plural subjects. They differ from the verbs under (46) in being relational, i.e. they establish a relation between two entities whereas verbs like schämen (‘be ashamed’) express a property of only one entity. One example is the verb verkrachen (‘fall out’) which is inherently reflexive:
(i)

Peter hat sich mit Maria verkracht

Peter has RP with Maria fall-out

The corresponding sentence with a plural subject can either mean that we fall out with someone else (ii) or that we fall out with each other (iii):
(ii)

Wir haben uns gestern mit Maria verkracht

We have RP yesterday with Maria fall-out

(iii)

Wir beide haben uns gestern verkracht

We two have RP yesterday fall-out

Further examples are: anfreunden (‘become friends’), einigen (‘agree’), überwerfen (‘fall out’), verbrüdern (‘fraternise’).

(46)

a.

Carlo schämt sich

Carlo is-ashamed rp-acc

b.

Further examples: irren (‘be wrong’), verirren (‘lose one’s way’), sehnen (‘long for‘), gedulden (‘be patient’), beeilen (‘hurry’), erholen (‘recover’), erkälten (‘catch a cold’), auskennen (‘know all about’), …

Like middle constructions or anticausatives, inherent reflexives are transitive in syntax. Again the reflexive pronoun is not linked to a semantic argument of the verb. Inherent reflexive verbs seem to be class I anticausatives that have lost their underlying causative form, i.e. the corresponding two-place predicate. Note finally that inherent reflexives can be inserted into impersonal middle constructions.

(47)

a.

Im dunklen Wald verirrt sich’s schnell

In the wood loses-the-way rp-acc it quickly

b.

Nackt im Schnee erkältet es sich leicht

Naked in the snow catches-a-cold it rp-acc easily

c.

Großer Geldmengen bemächtigt sich’s nicht so einfach

Big amounts of money take-possession-of rp-acc it not that easily

‘One cannot take possession of big amounts of money that easily’

So far we have seen that the accusative reflexive pronoun in the position of the direct object has various functions in German. It can either be linked to the internal or second argument of the verb or it is a morphosyntactic ‘marker’ for valency reduction. The second interpretation, which we called non-argument reflexive, is the one we discussed in this section. Non-argument reflexives can be found in middle constructions, anticausatives, and inherent reflexives. The correlation between reflexivity and valency reduction we observed in German is not unique and can be found crosslinguistically in many languages. In the next section we present examples from further Indo-European languages to illustrate this point. And in the final section of this chapter we argue that middle constructions as well as anticausatives in German are part of a major phenomenon: the diathesis of the middle voice.

The interpretation of weak reflexive pronouns in Indo-European languages

In this section we confine ourselves to morpholosyntactically and semantically related constructions in other modern Indo-European languages. We will see that the ambiguity of the reflexive pronoun observed in German is a widespread phenomenon that can be found in many Indo-European (as well as non-Indo-European) languages. Crosslinguistically, the ambiguity of transitive reflexive sentences will turn out to be quite regular. We limit the following discussion to Indo-European languages, the family of languages German belongs to.

For the middle voice in further (non-Indo-European) languages see Geniusiene (1987), Kemmer (1993), Miller (1993) and Klaiman (1991).

Furthermore, we are exclusively interested in the correlation between reflexivity and valency reduction we found in German. Therefore, we only look at middle constructions, anticausatives, reflexives and passives. Moreover, we ignore further specific properties these constructions have in different languages. Therefore, we only note in passing whether a language has impersonal middle constructions or adjunct middles or whether a middle construction in a certain language can syntactically realize the suppressed argument, must have an adverbial or a ‘generic’ interpretation. Both reflexivity and reduction of the first argument (or logical subject) have been subsumed under the notion middle voice. A discussion of all phenomena that are related to this complex notion is far beyond the scope of the present study, cf. Geniusienne (1987). Our main concern is the systematic ambiguity of (weak) reflexive pronouns in many languages. In the first part we give a brief description of the term middle voice, cf. Abraham (1995b) for further discussion. Data from five modern Indo-European languages are presented in the second part. And in the final section 2.4 we argue that the accusative reflexive pronoun in the position of the direct object is a morphosyntactic ‘middle marker’ in German.

Benveniste (1972) argues that the threefold distinction between active, passive and middle voice can be attributed to the historically basic dichotomy of active voice and middle voice. The passive voice is a variety of and has developed diachronically from the middle voice. Bosch (1983: 52) states, that “reflexive pronouns, just like reciprocals, are relatively new features in Indo-European languages. Both reflexivity and reciprocity used to be expressed by the medium inflection in the finite verb. (reflexive pronoun forms only arise in classical Greek).” In some Indo-European languages middle voice and passive voice are morphosyntactically still indistinguishable.

These forms are often called mediopassive in the literature. Russian, which we briefly discuss below, is one example. In Old Greek passive and middle voice are also almost identical. They differ only in future tense and aorist (Ars Graeca 1981, cf. also Klaiman 1991: 82f., or Beekes 1995). But a split in passive and middle voice was never carried out in Greek. Modern Greek equals the Russian ‘mediopassive’: middle and passive are formally indistinguishable. In Latin-Romance the weak reflexive pronoun takes over some of the major functions of the ‘old’ Latin middle morphology (a verbal affix). This process leads to a morphosyntactic differentiation of middle and passive voice (for the last two points cf. Miller 1993: 224f.).

The familiar distinction between active and passive is a result of modern linguistic theory. We already saw in the previous sections that this distinction is not sufficient to describe all the phenomena we illustrated for German.

According to Benveniste, Indo-European languages are generally subject oriented. They do not have object agreement. The diathesis of the verb (or the genus verbi) indicates the attitude of the subject to the event described by the verb. He roughly outlines the opposition active-middle from the perspective of the syntactic subject as follows (see also Lyons 1968, Kemmer 1993, or Klaiman 1991):

Klaiman distinguishes three kinds of applications for the term grammatical voice:

(i) alternations in the verb’s argument structure

(ii) alternations in the subject’s participant status

(iii) alternations in clause-level pragmatic salience

(i) is the most general use of the term grammatical voice for all kinds of argument structure/linking alternations. (ii) corresponds to the way we use this term in the following presentation. Pragmatic voice in (iii) is a distinct type of voice which is relevant to languages where verbal morphemes signal a special pragmatic salience of some constituent. This type of voice will be irrelevant for the ongoing discussion of German and further Indo-European data.

the active voice describes an action that proceeds from the (syntactic) subject and does not include it, whereas the middle voice describes an action that takes the (syntactic) subject as its centre, i.e. the subject is included in the action. This specification is of course very general.

The middle voice in Classical Greek is semantically very complex and yields a variety of different interppretations. The following examples illustrate only a few interpretations. The examples are from Benveniste (1972) (m = middle voice, a = active voice).
causative:

ορχεομαι (m)

-ορχεω (a)

'I dance'

I make someone else dance

reflexive:

νομουσ τιθεναι (a)

-νομουσ τιθεσθαι(m)

‘lay down laws’

‘lay down laws for oneself’

'indirect reflexive’/possessive:

λυειτονιππον (a)

-λυεταιτονιππον (m)

‘untie the horse’

‘untie the horse of one’s own‘

'exchange':

μισθουν (a)

-μισθουσθαι (m)

‘let’

‘rent’

We are mainly interested in the interpretations introduced in the preceding discussion of transitive reflexive sentences in German. In the following we compare five modern Indo-European languages (Modern Greek, Russian, Italian, French, and English), which equal German at least in one respect: middle constructions, anticausatives and reflexives can be expressed by the same morphosyntactic form. Before we turn to the discussion of the relevant examples we would like to mention two general points:

Many languages distinguish weak from strong reflexive forms. Kemmer (1993) calls them two-form languages. In Russian, for example, the weak form is a verbal affix (i.e. -sja) and the strong form a pronominal NP (i.e. sebja). In Dutch both forms are pronominal NPs but the weak reflexive marker (i.e. zich) is a simple and the strong reflexive marker (i.e. zichzelf) a complex word.
Another example is Modern Greek that has two additional strong forms of reflexive markers: apart from the weak reflexive form, i.e. the middle marker. Modern Greek has (a) a reflexive prefix afto- which can occur in addition to the middle marker (cf. Tsimpli 1989) and (b) a reflexive DP of the form ton eafto + possessive pronoun which can be translated as: ‘the self + possessive pronoun’ (‘the self mine’, ‘the self his/her/its’, ...). In Modern Greek, like in Classical Greek, the strong reflexive pronoun can occur together with the weak mediopassive affix (cf. Papakyriacou 1997):
(i)

Peripiithike mono ton eafto tis

care-PA-3s only the self her

‘she cares only for herself’

The strong form is usually used in coordination, with focus or modification (see chapter 4 for semantic restrictions on coordination of and focus on reflexive pronouns). Furthermore, in two-form languages the weak reflexive pronoun (i.e. the middle marker) yields a reflexive interpretation only in ‘body part reflexives’, i.e. verbs of personal grooming like comb, wash, anoint, dress,... These verbs describe actions that are very likely to be reflexive. In this case one could argue that the reflexive interpretation is salient. With other verbs (e.g. hear oneself, love oneself, kill oneself, ...) the strong form is obligatory to express reflexivity (cf. Kemmer 1993, Klaiman 1991:82f., and Miller 1993). One exception seems to be the reciprocal interpretation with an plural subject (like ‘they kiss/hear/embrace each other’). In this case the reciprocal interpretation can also be indicated by the weak form.

A similar difference can be found in English. Some verbs can express reflexivity without a reflexive pronoun. Other verbs must have an overt reflexive pronoun to yield a reflexive interpretation. The respective verbs are very similar to the corresponding verbs with either weak and strong reflexive forms in Greek and Russian. Therefore, one could argue that English is also a two form language. The strong form is the complex word him-/her-/itself and the weak form is morphologically empty. We come back to the English and German examples below.
The crucial observation is that the middle marker of a language must always be the weak reflexive form. We will see in chapter 4 and especially in chapter 5 that German is a one-form language, which does not distinguish between weak and strong reflexive pronouns. German has only one reflexive marker that is a simple word (recall section 2.1.3). In chapter 5 we discuss this distinction between weak and strong reflexive markers in greater detail.
Indo-European languages use quite different morphosyntactic strategies to indicate the diathesis of middle voice. Some languages have a special verbal inflection (Classical and Modern Greek) or a verbal affix (Russian), others use clitics as middle markers (Italian or French). German does not have a special verbal morphology for the middle voice but an independent word in a special syntactic position, i.e. the reflexive pronoun in the position of the accusative object, and English has no morphosyntactic middle marker at all, cf. Geniušiene (1987) and Abraham (1995b). Therefore, we use the term morphosyntactic middle marker in a very liberal way in the following presentation (cf. also Kemmer 1993). A middle marker can be a verbal affix, a verbal clitic, or an indepedent word. (48) summarizes the different kinds of middle markers and their possible interpretations to which the following discussion is limited.

(48)Middle markers and their interpretation

Possible middle markers

Possible interpretations

- verbal inflection

- passive

- verbal affix

- middle

- verbal clitic

- anticausative

- weak reflexive pronoun

- inherent reflexive

- accusative reflexive pronoun

- reflexive

In Modern Greek the middle marker is a verbal affix that indicates the reflexive, passive, middle and anticausative interpretation among others, as can be seen in (49).

Causativity is another possible interpretation of the mediopassive affix:
(i)

O Markos xirizete sto kurio tis gitonias tu

The Markos shave-PA-3S at the hairdresser’s shop the residential quarter his

‘Markos has his hair cut at the hairdresser’s shop in his neighbourhood’

Modern Greek has also a ‘strong’ form to indicate the reflexive interpretation (cf. footnote 33 above). The middle marker is the ‘weak’ reflexive element. The examples (49.a, b and d) are taken from Tsimpli (1989) and example (49.c) from Sioupi (1997).

Thanks to Artemis Alexiadou and Androulla Papakyriacou.

‘PA’ means passive affix, cf. Tsimpli (1989). Under the perspective outlined above we can also call it mediopassive affix or middle (voice) marker.

(49)

a.

Afto to vivlio dhiavas-ti-ke xtes

(passive)

this book-nom read-pa-3s yesterday

‘This book was read yesterday’

b.

Afto to vivlio dhiavas-ti-ke efharista

(middle construction)

this book-nom read-pa-3s with pleasure

‘This book reads with pleasure’

c.

To plio vithiz-et-e

(anticausative)

The ship-nom sink-pa-3s

‘The ship sinks’

d.

I Maria xteniz-et-e kathe mera

(reflexive)

Maria-nom comb-pa-3s every day

‘Maria combs herself every day’

Modern Greek has yet another way of middle and anticausative formation. Beside the mediopassive affix there exists an alternative option: some verbs form middle constructions and anticausatives in the active voice, i.e. without the middle marker (Tsimpli’s passive affix ‘PA’), cf. Condoravdi (1989). The verb anigo (‘open’) in (50) is one example for verbs that are morphologically marked active in middle constructions and anticausatives. The affiliation of a verb to one of these two classes seems to depend on the lexical meaning of the respective verbs.

Concerning middle formation verbs like anigo (‘open’), girizo (‘turn’), ligizo (‘bend’) are class 1 verbs (active) and verbs like kovo (‘cut’), diashizo (‘cross’), gializo (‘polish, shine’) belong to class 2 (mediopassive).

(50)

a.

Afti i porta anigi

this door-nom opens

b.

Afti i porta den anigi kala

this door-nom does not openwell

Russian equals Modern Greek in the middle voice. Russian, like Modern Greek, uses a morphological middle marker, the verbal affix ‑sja

In the literature the morpheme -SJA has been analysed either as a pronominal clitic or as an affix. Both analysis involve some problems, which do not concern us here. For further discussions see Junghanns (1996) and Schoorlemmer (1996).

The morpheme -SJA seems to be developed from the former accusative singular reflexive pronoun. Klaiman (1991) notes that some modern Indo-European languages have developed a ‘neo-middle construction’, which is derived from an originally reflexive marker (a pronoun or affix). In Russian the middle marker -SJA seems to result from the grammaticalization of a reflexive pronominal clitic that has been added to the verb in active voice. For similar processes in Old Norse see Miller (1993: 205f.). A similar process might take place in Romance languages like Italian and French, cf. below.

(-s’ after a vowel), for the passive and the middle (51.a and b), the reflexive (51.c), reciprocal and anticausative interpretation (cf. Junghanns 1996).

The examples are from Junghanns (1996). Special thanks to Assinja Demjjanow for her help with the Russian data. Further interpretations are ‘antipassive’ (i.e. sentences with an implicit internal argument) and causative (like in Modern Greek).
(i)

Otec rugaetsja

or

sobaka kusajetsja

father scold-SJA (the father is scolding)

dogs bite-SJA

(ii)

Ja strigus’ v parikmayerskoj

I hair-cut-SJA at hairdresser

Sentence (51.b) is ambiguous between the anticausative and the middle reading. It is quite similar to its German counterpart with the accusative reflexive pronoun, which is in the same way ambiguous as the Russian example.

With the perfective aspect Russian has a second way of passive formation. This periphrastic form consists of the auxiliary ‘be’ and the passive participle (dver’ byla zakryta nami, ‘the door was closed by us’, cf. Miller 1993: 238).

(51)

a.

Dom stroitsja (plotnikami)

(passive)

house build-SJA (by the carpenters)

‘The house is being built (by the carpenters)’

b.

Dver’ (legko) otkrylas’

(middle constr./anticausative)

Door-nom (easily) opened-SJA

‘the door opened (easily)’

c.

Ivan moetsja

(reflexive)

Ivan-nom washes-SJA

‘Ivan is washing himself’

We already noted in footnote 33 that Russian also has a second way of indicating reflexivity. In addition to the middle marker (i.e. the weak reflexive marker) there is a historically related strong reflexive marker, the reflexive pronoun sebja. Again, these two forms are not always interchangeable.

Italian and French share some properties with both Russian and German. On the one hand, the weak reflexive pronoun in Italian and French is a verbal clitic in syntax and equals the Russian -sja that has been analyzed either as verbal affix or as pronominal clitic.

The French and Italian se-/si-construction seems to lie between the Russian -sja and the German sich-construction with respect to the degree of grammaticalization. There is diachronic evidence that a weak reflexive element that has become a verbal clitic can be further reduced to a verbal affix. Furthermore, there might be a principle ‘that it is simpler to have a unified derivation of a given formative’ (Miller 1993: 220), i.e. that a formative comes to encode a variety of parallel functions - in this case middle and passive (cf. Miller: 205f. on the process of reflexive incorporation in Scandinavian languages and in Russian). In this respect the French and Italian middle construction differs from the German middle construction. We already mentioned that German, unlike French and Italian, is a one-form language. Furthermore, there is no evidence that the accusative reflexive pronoun in German is a syntactic clitic. In syntax the reflexive pronoun is an independent pronominal object (cf. Fagan 1992, Gärtner and Steinbach 1997 and 2000, and chapter 4).

As opposed to Russian, the pronominal clitic in Romance maintains the person-bound form and need not be adjacent to the main verb.

Cinque notes, however, that middle constructions in Italian are possible only with the third person singular clitic si. This does neither hold for unaccusatives/ergatives or inherent reflexives nor for French middle constructions that are possible in the first and second person as well (cf. Grimshaw 1982 and Fagan 1992).
(i)

Io mi avvicino

(ergative)

I RP am going near

(ii)

Io mi ammalo

(inherent reflexive)

I RP get ill

(iii)

*Io mi trasporto facilmente

(middle construction)

I RP transport easily

The Russian middle marker –sja has only one form for first, second, and third person singular and plural. On the other hand, a weak reflexive pronoun can only be used in middle constructions, anticausatives and reflexives to indicate valency reduction but not in passives. Both French and Italian have a periphrastic passive form.

Cinque (1988) notes that middle constructions in Italian and French need not receive a ‘generic’ interpretation (cf. also footnote 25 above). Middle constructions can describe particular events and they can be modified by e.g. agentive adverbs. With this ‘eventive’ reading the interpretation of middle constructions in Italian and French is very similar to the interpretation of (periphrastic) passives. Fagan (1992: 58) points out that a by-phrase is possible in earlier stages of French with middle constructions describing a particular event and Cinque (1988, footnote 11) notes that ‘in more rhetorical styles of Italian, by-phrases are found to cooccur with si’. In some Italian dialects middle constructions can yield a generic reading only.

Additionally, we find both reflexive and non-reflexive anticausatives. In (52) we give some examples from French.

The examples are from Miller (1993). For a more detailed survey of French middle constructions see Grimshaw (1982), Fagan (1992), and Dobrovi-Sorin (1998). The corresponding strong forms are ‘elle même (herself), lui même (himself), ...’.

(52)

a.

un veston de laine se lave facilement

(middle construction)

a jacket of wool rp washes easily

b.

la branche s’est casseé

(anticausative)

the branch rp has broken

c.

Pierre se rase

(reflexive)

Pierre rp shaves

At first glance, a language like English has a quite different way of middle, passive and anticausative formation (note, however, that some verbs in Modern Greek make use of the ‘English’ way of middle and anticausative formation). English does not mark middle voice morphosyntactically. In contrast to their counterparts in other Indo-European languages, the sentences in (53.a and b) are morphologically active without an overt middle marker. Middle constructions and anticausatives in English do not contain a reflexive marker. Nevertheless we find the same semantic effects of the diathesis of middle voice outlined above. Furthermore, the reflexive interpretation can be expressed with and without a reflexive pronoun (53.c and d).

German has only few corresponding examples: duschen (‘take a shower’) and baden (‘take a bath’) can be used with or without a reflexive pronoun. The unmarked interpretation of (i) without the reflexive pronoun is: I am taking a shower. The transitive sentence in (ii) means that I give someone (i.e. the dog) a shower. These verbs might be two-place predicates with an implicit secon argument.
(i)

Ich dusche (mich)

I take-a-shower (RP-ACC)

(ii)

Ich dusche den Hund

I give-a-shower the dog-ACC

In this respect, English equals two-from languages like Modern Greek, Russian, French or Italian. We find the same distinction between weak and strong reflexive forms.

We mentioned above that in two-form languages the weak reflexive marker (i.e. the middle marker) yields the reflexive interpretation only with a limited class of verbs usually describing actions that are mainly reflexive (or reciprocal). These verbs are to some extent identical to the English verbs that can express reflexivity without a reflexive pronoun. Hence, English seems to be also a two-form language that distinguishes weak from strong reflexive forms.

Things are completely different in German. While the preferred interpretation for the intransitive English sentence in (iii) is the reflexive interpretation (i.e. coreference of the subject and the implicit object), the corresponding German sentence in (iv) cannot receive a reflexive interpretation. You might utter sentence (iv) without the reflexive pronoun if shaving is your profession.
(iii)

I am shaving

(I am shaving myself)

(iv)

Ich rasiere

(I am shaving someone)

The same contrast between English and German can be found with many verbs, e.g. hide or meet and their German equivalents verstecken or treffen. This might be due to the fact that German, as opposed to English, has no weak reflexive pronouns. In German we cannot tell the weak from the strong form of the reflexive pronoun.

However, unexpected differences in the interpretation of intransitive sentences can be found even within one language. Semantically related verbs like dress and cloth can differ with respect to reflexivity (cf. Jackendoff 1987):
(v)

I dressed

(vi)

*I clothed

Furthermore, Miller (1993: 193) notes that the unmarked interpretation of (vii) ‘would involve laundry’, whereas sentence (viii) is ambiguous between the interpretation we get for sentence (vii) and the reflexive interpretation:
(vii)

John washed all day long

(viii)

John is washing

(He is washing himself/his laundry)

The interpretation of reflexive constructions seems to depend at least on the respective pronominal paradigms of languages and possibly on further semantic and contextual/pragmatic factors.

Hence, in English simple intransitive sentences can yield various interpretations, some of which can be subsumed under the notion of middle voice. The middle marker or weak reflexive marker is morphologically empty in English. Besides, English, like German, French, and Italian, uses periphrastic passives.

(53)

a.

This book reads easily

(middle construction)

b.

The door opens

(anticausative)

c.

Peter is washing himself

(reflexive – ‘strong’ form)

d.

Peter is shaving

(reflexive – ‘weak’ form)

Dutch resembles English as well as German. With respect to anticausative formation Dutch equals German.

We mentioned above that German has two kinds of anticausatives. Most verbs that undergo in the causative-anticausative alternation belong to one and the same class. As opposed to this, Dutch has some verbs that can be found in both classes simultaneously (cf. Everaert 1986):
(i)

De suiker lost op

The sugar dissolves up

(ii)

De suiker lost zich op in het water

The sugar dissolves rp up in the water

Dutch middle constructions, on the other hand, equal English middle constructions. Unlike their German counterparts, middle constructions in Standard Dutch are not reflexive (cf. Ackema and Schoorlemmer 1994 and 1995 and chapter 3).

Cornips (1996) notes that some southern dialects of Dutch have reflexive middle constructions. Not surprisingly these dialects are spoken in an area close to the German border.

The following table summarizes the observations made so far. It gives a brief survey of the correlation between the middle marker and its (potential) interpretations and integrates the observations on German we made above into the larger context of further Indo-European languages. We confine ourselves to passive, middle, anticausative, and reflexive interpretation. Besides, this table is anything but complete. So far we considered only five Indo-European languages and we already mentioned that some of these languages permit exceptions or alternative forms for some of these interpretations, which sometimes depend on the lexical meaning of the verb and sometimes on additional semantic conditions. And last but not least, although the arrangement in table (54) looks quite systematic, we do not want to make any hypothesis about the synchronic and diachronic correlation between reflexivity and middle voice. We do not claim that languages like English and Modern Greek are the extremes on a scale of middle marking-languages (we refer the reader to Geniusiene 1987 for a detailed crosslinguistic study of this issue).

(54)Possible interpretations for an overt (weak) reflexive marker

Interpretation

English
English has no morphosyntactic middle marker at all. The reflexive interpretation is expressed either by a (strong) reflexive pronoun or without a pronoun at all (cf. last but one footnote). But we already mentioned that English can also be analyzed as a two-form language, which distinguishes weak from strong reflexive pronouns. The weak form is morphologically empty and the corresponding sentences are simply intransitive in syntax (cf. figure (65) below). According to this assumption, English would equal French and Italian.

Dutch

German

French

Modern Greek

Russian

Passive

-

-

-

-

+

+

Middle

-

-

+

+

+

+

Anticausative

-

+

+

+

+

+

Reflexive

-

+

+

+

+

+

Let us summarize the results of this section. First, we saw that the ambiguity of transitive reflexive sentences in German is not accidential and that it can be subsumed under the notion middle voice. Indo-Europpean languages use the middle voice for quite different (semantic) functions. Reflexivity and valency reduction are main functions, see Geniusiene (1987) for a detailed overview. Second, many Indo-European languages have different kinds of morphosyntactic middle markers, but there is always a strong correlation between weak reflexive markers and middle markers. Recall that some Indo-European languages like Russian, and Old Norse have developed ‘neo-middle construction’ from weak reflexive pronouns that have been reduced to verbal affixes. Third, German, unlike Modern Greek or Russian, has no verbal middle inflection. Verbs in middle constructions, reflexives, anticausatives and inherent reflexive verbs are always morphologically active. Nevertheless, German has also a morphosyntactic middle marker, the accusative reflexive pronoun in the position of the direct object. Fourth, German is a one-form language that does not distinguish weak from strong reflexive pronouns/forms. In this respect it differs from most Indo-European languages. We also saw that the passive voice must be distinguished from middle voice in many modern languages like e.g. Italian, French, English, Dutch, and German. Hence, German has a threefold distinction between active, passive and middle voice. And finally, we must also distinguish between reflexivity and middle voice. One possible interpretation for the middle marker is the reflexive interpretation. However, reflexivity cannot be reduced to middle voice as well as middle voice cannot be reduced to reflexivity or, to put it the other way round, not every reflexive marker is also a middle marker (at least in the languages listed in table (54)).

The middle voice marker in German

As opposed to the middle markers in Modern Greek, Russian, Italian and French the middle marker in German has some specific properties. It is not a verbal affix or a verbal clitic but an independent word, a reflexive pronoun in object position, i.e. a free lexical morpheme, cf. Abraham (1995a:5). Moreover, the reflexive pronoun in German can only be called a middle marker when it occupies the position of the accusative (or direct) object. This is summarized in (55)

(55)

In German only a reflexive pronoun in the position of the accusative (or direct) object is ambiguous between an argument and a non-argument interpretation

Hence, only transitive reflexive sentences are systematically ambiguous in German. The crucial examples are repeated in (56), cf. chapter 1, example (6).

(56)

a.

Herr Rossi rasiert sich

(Reflexive interpretation)

‘Mr. Rossi is shaving’

b.

Das Buch liest sich leicht

(Middle interpretation)

‘The book reads easily’

c.

Die Tür öffnet sich

(Anticausative interpretation)

‘The door opens’

d.

Herr Rossi erkältet sich

(Inherent reflexive interpretation)

‘Mr. Rossi is catching a cold’

All four interpretations are indistinguishable in syntax as we will argue in chapter 4. A sentence of the form subject + verb + accusative reflexive pronoun is potentially four times ambiguous in German. The first ambiguity is due to the reflexive pronoun, which can either be interpreted as argument or as non-argument reflexive. The second ambiguity results from two different semantic operations on the implicit argument. The implicit argument can either be bound by a generic operator or deleted. The ambiguity of transitive reflexive sentences is systematically illustrated in figure (57).

(57)

The interpretation of the reflexive pronoun and the implicit argument in transitive reflexive sentences

As opposed to the reflexive pronoun in transitive reflexive sentences, the dative reflexive pronoun in (58.a) and the reflexive pronouns contained in PPs in (58.b) and (58.c) cannot indicate valency reduction. They only yield the reflexive interpretation and cannot be called a middle marker. Thus not every reflexive marker is automatically also a middle marker.

(58)

a.

Peter widersprach sich

Peter contradicts himself-dat

b.

Peter ist außer sich gewesen

Peter is beside himself-dat been

‘Peter was beside himself’

c.

Er achtet sehr auf sich

He takes-care of himself-acc

Middle High German provides further evidence for a distinction between reflexive markers and middle markers and for the very special status of the accusative reflexive pronoun. As opposed to the accusative reflexive pronoun, the dative reflexive pronoun has developed very late. It is not until the beginning of the 18th century that the modern usage of the dative reflexive pronoun seems to be established. At earlier stages German does not morphologically distinguish dative personal pronouns from dative reflexive pronouns. This still holds for some dialects in Geman. The personal pronoun was used for local binding (principle A of Chomsky’s (1981) binding theory) and for non-local binding (principle B). This is illustrated in (59.a-c) for Middle High German (MHG). In all three examples it is the dative personal pronoun which is locally bound (principle A), cf. Paul 1988, Moser, Stopp and Besch 1988, Grimm 1905, and Behagel 1923.

Many thanks to Marie-Christine Erb for making me aware of this point (cf. also Andersen 1993 and Hermodsson 1952).

(59)

a.

(er) machit im selben ein hus

(MHG, Grimm 1905)

he-nom makes him-dat self a house

b.

so groz manheit her an im hat

(MHG, Grimm 1905)

so big bravery he-nom at him-dat has

c.

... weil sie ihr einbildete

(Early New High German,

... because she-nom her-dat imagined

Moser, et al. 1988)

Unlike the dative reflexive pronoun the use of an accusative reflexive pronoun is stated for very early stages of German. (60) and (62.a and b) are three example from Middle High German. Besides, sentence (60) illustrates that inherent reflexive verbs can already be found in Middle High German. Hence, non-argument reflexives are also stated for early stages.

(60)

do vaffende sich aspiran

(MHG, Grimm 1905)

The weaponed-nomrp-acc came-close

Table (61) is taken from Paul (1988) to exemplify the difference between accusative and dative personal and reflexive pronouns in Middle High German.

Old English, on the other hand, does not have reflexive pronouns at all. The personal pronouns (him, her, it, ...), like the dative pronouns in German, are used for principle A and B. The reflexive pronouns (himself, herself, itself, ...) in Modern English have developed from the personal pronouns and the adnominal focus particle -self. Therefore, English never had ‘pure’ reflexive pronouns that unambigiously mark the middle voice. See section 4.2.2 for more details and references.

(61)

accusative and dative personal and reflexive pronouns in Middle High German

Furthermore, early stages of German allow embedded accusative reflexive pronouns to be bound only by the syntactic subject of the clause, whereas in Modern German (MG) the reflexive pronoun can also be bound by the syntactic object. Embedded infinitives illustrate this difference in (62) and (63).

(62)

a.

ir gast₂si₁sich_1/*2kuessen bat

(MHG, Parzival)

her guest-acc she-nom rp-acc kiss ask

‘She asked her guest to kiss her’

b.

bat er₁sich_1/*2ketrencan daz wip₂

(MHG, Behagel 1923)

ask he-nom rp-acc offer-a-drink the women-acc

‘He asked the woman to offer him some drink’

(63)

a.

Der Herr₁befahl dem Diener₂, sich_*1/2 anzukleiden

(MG)

The master-nom ordered the servant-acc to dress rp-acc

(i.e. the servant)

b.

Der Herr₁befahl dem Diener-2, ihn_1/*2 anzukleiden

(MG)

The master-nom ordered the servant-acc to dress him-acc

(i.e. the master)

These observations (no dative reflexive pronouns and strict subject orientation in MHG) are in line with an analysis that treats only the accusative reflexive pronoun in the position of the direct object as a middle marker in German. Note finally that, as opposed to Italian and French, the reflexive pronoun in Modern German is an independent word/constituent in syntax. Gärtner and Steinbach (1996 and 2000) argue at length that German has no special or syntactic clitics.

Besides, in most German dialect the reflexive pronoun cannot even be prosodically reduced, whereas personal pronouns can phonologically cliticize to an adjacent foot or syllable in phonology (e.g. ich habe ihn gesehen vs. ich hab’n gesehen, i.e. ‘I have him seen’), cf. Hall (1998:107) and also section 4.1.

So far we argued that the accusative reflexive pronoun in Modern German is not a pure middle marker and reflexivity cannot be reduced to middle voice. Nevertheless, we observe a strong correlation between reflexive markers (pronouns or verbal affixes) and middle markers or more general between reflexivity and middle voice. We call the accusative reflexive pronoun a middle marker if it is the accusative (or direct) object of the sentence. In this sense we follow Kemmer (1993) and conclude that German is a middle marking language. In contrast to Modern Greek or Russian, German does not use a verbal affix as middle marker but a syntactically independent word. Furthermore, the accusative reflexive pronoun can be used in further syntactic contexts as reflexive marker only. Dative reflexive pronouns can only be used as reflexive markers. These observations for German are summarized in the following figure.

(64)

Middle marker and reflexive marker in German

This present investigation is limited to middle marker in German and its possible interpretations, i.e. the left column in figure (64). Nevertheless, we think that our results might also be relevant for the analysis of corresponding phenomena in other languages. English, for example, differs from German in having no morphosyntactic middle marker. The weak reflexive form is morphologically empty and the middle voice is morphosyntactically unmarked in English. We already noted that from a semantic point of view English also distinguishes between active and middle voice. However, because of these morphosyntactic differences, transitive and intransitive sentences in English and German are not equally ambiguous. This is illustrated in table (65). In German transitive reflexive sentences and intransitive sentences are equally ambiguous, whereas in English the intransitive sentence in (65.a) is the most ambiguous construction, cf. also Abraham (1995a) for the differences between English and German. We will come back to intransitive sentences in English in chapter 8. The next five chapters will deal mainly with transitive reflexive sentences in German.

(65)Possible interpretations for intransitive and transitive-reflexive sentences in English and German

Syntax

Semantics

I. Transitive

English

German

Interpretations

subject + verb + rp_ACC

1. V < x < x >>

1. V < Ø < y >>
2. V < (x) < y >>
3. V < x < x >>

anticaus./inh.refl.
middle constr.
reflexive

II. Intransitive

subject + verb

1. V < x >
2. V << y >>
3. V < x < (y) >>
3. V < Ø < y >>
4. V < (x) < y >>
5. V < x < x >>

1. V < x >
2. V << y >>
3. V < x < (y) > >

unergative
unaccusative
impl.intern.arg.
This construction is sometimes called ‘antipassive’. In some languages like, for instance, Eskimo the suppression of the internal argument is morphologically marked by an extra anticausative verbal affix (cf. Miller 1993: 150f. for more details). The middle marker in Russian can also receive an antipassive interpretation.

anticausative
In section 2.2 we illustrated that German has two kinds of anticausative constructions, one of which is transitive and reflexive. The second one is simply intransitive. In chapter 5 we will argue that the second kind is not anticausative but basically unaccusative. In this case, the causative variante is derived from the ‘anticausative’. Hence, intransitive sentences in German do not yield an anticausative reading.

middle constr.
reflexive

Conclusion

In this chapter we argued that German is a middle marking language. First, we gave a descriptive survey of transitive reflexive sentences in German that involve valency reduction. We started of with the middle construction and the restrictions for each of its individual parts: the verb, the syntactic subject and the reflexive pronoun, the second semantic argument, the implicit first argument, and the adverbial modification. In addition, we discussed further semantic restrictions. Second, we turned to related constructions in German: anticausatives and inherent reflexives. Finally, we argued on the basis of other Indo-European languages that transitive reflexive sentences in German belong to the diathesis of middle voice. We are aware of the fact that broader diachronic and synchronic studies are necessary. These studies lie, however, beyond the scope of this book. We refer the reader again to the detailed studies of Geniušiene (1987) Kemmer (1993), and Miller (1993).

In the following we are mainly concentrating on the middle voice in German. The next chapter discusses several lexical and syntactic analyses of middle constructions and criticizes their shortcomings. We essentially restrict the discussion to middle constructions because they are discussed most controversially. Moreover, middle constructions are more productive than the other two transitive constructions that involve valency reduction, i.e. anticausatives and inherent reflexives. The middle construction turns out to be a good example to illustrate the shortcomings of the lexical and syntactic analysis. In chapter 4, 5 and 6 we develop an alternative proposal for the syntax and semantics of transitive reflexive sentences in German that offers an unified treatment of all four interpretations of these sentences. Hence, we propose an explanation of the middle voice in German - i.e. the left column in figure (64).