In the previous chapter we criticized lexical and syntactic analysis of middle formation. We saw that they offer neither a conclusive explanation of the specific properties of middle constructions in German nor a unified account of the ambiguity of the reflexive pronoun in transitive reflexive sentences. Therefore we argue for a postsyntactic analysis of transitive reflexive sentences in the remainder of this book. Middle constructions and anticausatives are syntactically analysed as simple transitive clauses with an accusative reflexive pronoun in the position of the accusative (or direct) object. Hence, they do not differ from sentences like (1) that receive a reflexive interpretation. Of course, the crucial element for every syntactic analysis of transitive reflexive sentences is the reflexive pronoun in object position, which can either be linked to a semantic argument of the verb, as can be seen in (1), or it can be an indicator of valency reduction, cf. (2). Therefore, large part of this chapter deals with the syntactic properties of argument and non-argument reflexives.

(1)

Der Kanzler liebt sich mehr als alles andere in der Welt

the chancellorlovesrp more than everything else in the world

(2)

Dieser Käse schneidet sich sehr gut

(middle construction)

this cheesecuts rp very good

Die Tür öffnete sich ein bißchen

(anticausative)

the door opened rp a bit

Hans schämt sich fürchterlich

(inherently reflexive verb)

Hans is-ashamed rp awfully

The syntactic parallelism between the argument and non-argument reflexive is, however, not immediately obvious. On the one hand, both kinds of reflexive pronouns - the argument reflexive and the non-argument reflexive – turn out to be subject to the same constraints on word order in the middle-field. This is discussed in section 4.1. Furthermore, they are both bound by the subject. These are good arguments to treat them the same in syntax. On the other hand, there are some crucial differences between argument and non-argument reflexives which we discuss in section 4.2. These data seem to provide empirical evidence for the claim that non-argument reflexives differ in syntax from argument reflexives, thus supporting a syntactic analysis of middle formation (cf. chapter 3.1.3. and Haider 1982, Grewendorf 1984, and Pitz 1988 and Everaert 1986 for reflexive anticausatives in Dutch). However, all differences will turn out to be due to the semantic interpretation of non-argument reflexives. They should be explained by semantic theories of focus, coordination and fronting rather than by an (additional) ad hoc distinction between two different types of reflexive pronouns in syntax. This line of argumentation can already be found in Fagan (1992) and Erb and Steinbach (1997). In this chapter we show in detail how the distribution of argument and non-argument reflexives can be derived from recent theories on focus, coordination and fronting.

Similarities between argument and non-argument reflexives

First we discuss word order phenomena that give direct evidence for an analysis that does not draw a distinction between argument and the non-argument reflexives in syntax. Both the argument and the non-argument reflexive are subject to the same restrictions on word order in the middle-field (for the sentence initial position see 4.2.3). It is a well-known fact that in German reduced and full pronouns tend to be located in the so-called Wackernagel position, the second position of a clause. This is shown in (3) and (4) for main clauses. The first sentence of each pair is always a middle construction, which containes a non-argument reflexive. The second sentence always contains a argument reflexive and receives a reflexive interpretation. Anticausatives and inherent reflexives equal middle constructions in this respect.

(3)

a.

Dieser Käse schneidet sich gut

this cheese-nom cuts rp well

b.

Der Kanzler liebt sich sehr

the chancellor-nom loves rp much

(4)

a.

*Dieser Käse schneidet gut sich

b.

* Der Kanzler liebt sehr sich

The examples in (5) show the same for embedded clauses. In (5.a-f) the subject is a definite DP, while it is a pronoun in (5.g-j).

(5)

a.

... weil dieser Käse sich gut schneidet

... because this cheese-nom rp good cuts

b.

... weil der Kanzler sich sehr liebt

... because the chancellor-nom rp much loves

c.

... weil sich dieser Käse gut schneidet

d.

... weil sich der Kanzler sehr liebt

e.

^??... weil dieser Käse gut sich schneidet

(stilted)

f.

^??... weil der Kanzler sehr sich liebt

(stilted)

g.

... weil er sich gut schneidet (er = dieser Käse)

... because he (= cheese) rp good cuts

h.

... weil er sich sehr liebt

... because he rp much loves

i.

*... weil sich er gut schneidet

j.

*... weil sich er sehr liebt

The unmarked word order in the German middle-field is restricted by various ‘weak’ constraints. The relevant constraints on the positioning of reflexive pronouns are listed in (i) – (iv), for further discussion see Lenerz (1977), Uszkoreit (1987), Cooper (1994), Vogel and Steinbach (1997), Haider and Rosengren (1998), Müller (1998), and Gärtner and Steinbach (2000).

Thematic (or backgrounded) elements precede rhematic (or focused) ones.
There are various ways to state this constraint. Another possibility would rely on Jäger’s (1995) definition of topics, i.e. the topic precedes the comment in the unmarked case.
pronominal elements precede full NPs.
This constraint might be subsumed under the following more general one: “heavy” elements follow “light” ones.
Linearization of arguments directly mirrors the GF/case-hierarchy: nominative precedes accusative in the unmarked case.

The examples (5.a-d) can be described by constraint (ii) and (iii), which are equally ‘strong’. Constraint (ii) claims that sich is the first element of the middle-field, whereas constraint (iii) claims the opposite: The nominative NP dieser Käse/der Kanzler precedes the accusative NP sich. Therefore, both sequences are equally possible (and unmarked). Both sentence (5.e) and (5.f) sound stilted, because the reflexive pronoun does not only follow the subject but also the adverbial (cf. footnote 2 above).

That these constraints are ‘weak’ can be demonstrated by the following example from an anecdote by Eckhart Henscheid: ”[...] Derjenige sollte Sieger und der beste Kritische Theoretiker sein, der das Reflexivum ‘sich’ am weitesten postponieren (nachstellen) konnte [...] Sieger wurde und sein Meisterstück machte nämlich Adorno mit dem seither geflügelten Satz: ‘Das unpersönliche Reflexivum erweist in der Tat noch zu Zeiten der Ohnmacht wie der Barbarei als Kulmination und integrales Kriterium Kritischer Theorie sich’” (Henscheid 1993: 56-57). Note that sich erweisen (‘prove to be’) is an inherently reflexive verb.

The sentences in (5.g-j) are subject to all three constraints. Accusative objects can only precede (nominative) subjects if the former are thematic (or backgrounded) and the latter rhematic (or focused). In (5.g.-j) the subject and the object are pronominal and most likely thematic. Therefore, (5.i) and (5.j) violate the third constraint that states that nominative precedes accusative in the unmarked word order. Both sentences are much better if we put contrastive focus on the personal pronoun. We slightly modify example (5.i) by replacing the verb schneiden (‘cut’) by küssen (‘kiss’), because the second argument of küssen usually refers to human entities. Reference to human entities simplifies discourse linking of pronouns.

(6)

a.

^??... weil sich sogar ER gut küßt

because rp even he well kisses

b.

^??... weil sich sogar ER sehr liebt

because rp even he much loves

Two additional constraints are relevant for the linearization of dative objects in the middle-field.

The same constraints are also relevant for the linearization of the arguments of verbs that do not select an agent like e.g. interessieren (‘to be interested in’).

The NP that refers to an agent precedes other constituents.
NPs that refer to animate entities precede NPs that refer to inanimate entities.

Constraint (v) is rather weak. Usually the subject of the middle construction (the nominative NP) is not an agent. Hence, we expect that dative NPs referring to animate entities can precede the subject in middle constructions. This can be seen in the following examples.

(7)

a.

... weil sich einer Nonne ein Gesangsbuch schnell verkauft

because rp a nun-dat a hymnbook-nom quickly sells

b.

... weil einer Nonne sich ein Gesangsbuch schnell verkauft

c.

... weil sich ein Gesangsbuch einer Nonne schnell verkauft

d.

... weil ein Gesangsbuch sich einer Nonne schnell verkauft

All these examples show that the non-argument reflexives do not differ from non-arguemtn reflexives with respect to unmarked word order in the middle-field. The non-argument reflexive is neither a clitic nor in some sense syntactically incorporated into the verb. In section 2.1.2 we already mentioned that German has no syntactic (or special) clitics. Even (phonologically) reduced pronouns are not syntactic but only phonological clitics, that must be adjoined to an adjacent foot, syllable or (under certain circumstances) prosodic word in phonology (for further discussion see Gärtner and Steinbach 1997 and 2000). Note that phonologic reduction is impossible for the third person reflexive pronoun in Standard German and in most German dialects.

Hessian and Saxonian seems to be exceptions that allows phonological reduction of the third person reflexive pronoun. Hall (1998:107) argues that a vowel preceding the [ç] cannot be reduced to schwa in German.

This difference between most personal pronouns and the third person reflexive pronoun is illustrated in (8).

(8)

a.

Personal pronoun:

Sie hat ihn/’n gestern erst gewaschen

She has him yesterday just washed

b.

Reflexive pronoun:

Er hat sich/*’s/*’ch/*’si/... gestern erst gewaschen

He has rp yesterday just washed

Reflexive pronouns also differ from elements that must be adjacent to the main verb in embedded V-final clauses. The examples in (5) above illustrate that reflexive pronouns need not be adjacent to the main verb, as opposed to verb-object-combinations like Ball spielen (‘play with a ball’) or the separable prefix in davonschleichen (‘sneak off/away’):

(9)

a.

... weil Peter gestern Ball gespielt hat

… because Peter yesterday ball played has

‘… because Peter played ball yesterday’

b.

... *weil Peter Ball gestern gespielt hat

c.

... weil Heidi sich gestern morgen davongeschlichen hat

because Heidi rp yesterday morning sneaked off has

‘… because Heidi sneaked off yesterday morning’

d.

... *weil Heidi sich davon gestern morgen geschlichen hat

In chapter 2 we saw that both the argument reflexive and the non-argument reflexive are always bound by the subject NP in syntax. This is illustrated in (10) for middle constructions.

We mentioned in section 3.1.1 that not all reflexive pronouns are bound in syntax. Reflexive pronouns that are used as logophors are not subject to the binding conditions. In chapter 5 we take a closer look at binding.

We do not find any evidence that an implicit actor or agent binds the anaphor at some level of derivation as e.g. proposed by Pitz (1988).

This example is from Reis (1981) and also mentioned in Haider (1987).

(10)

a.

Ich₁ wasche mich₁ schneller als alle anderen

I wash rp-1.s. faster than all the others

b.

Du₁ hörst dich₁ heute nicht gut an

You hear rp-2.s. today not good particle (You don’t sound good today)

c.

Du₁ verkaufst dich₁ gut - ich meine, dein Buch verkauft sich gut

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

So far, we did not find empirical evidence for two syntactically different types of reflexive pronouns in German. The following analysis of transitive reflexive sentences in German is based on Chomsky’s analysis of (structural) case checking (cf. Chomsky 1993 and 1995, chapter 3). The syntactic structure for transitive reflexive sentences with personal and impersonal subjects is given in (11.1) and (11.2) below.

There is a long discussion in the literature how to analyze the relatively free constituent order of languages like German. Essentially, three approaches can be distinguished. (i) Word order is derived via movement (scrambling, extraposition and fronting) from an underlying structure. (ii) Word order is base generated (possibly in compliance with linearization principles). (iii) Word order is derived by an extra modul for linearization. For movement theories cf. Müller and Sternefeld (1993), Müller (1993) or Grewendorf and Sabel (1994) among others. Supporters of a base generation theory are Haider (1993), Cooper (1994), Fanselow (1995 & 1997) and Uszkoreit (1987) within the framework of GPSG-theory. Vogel and Steinbach (1997) argue for a mixed approach which assumes that accusative and nominative DPs are subject to A- and A’-movement, whereas dative DPs can be inserted directly. Advocates of a linearization grammar are Kathol (1995), Reape (1994) and (1995) or Richter (1997).

Another long standing problem is verb-placement in V2-languages like Dutch and German. Discussions can be found in Vikner and Schwartz (1991), Zwart (1993) and (1998) and Gärtner and Steinbach (1994). Further theories on V2 by Bobaljik (1995) and Rohrbacher (1994) are discussed in Stanek (1995) - cf. also the references in the next footnote.

In (11) we base generated the external argument in VP and split IP into AgrSP, TP, and AgrOP. Furthermore, we omit TP. For the discussion pro and contra the presence of IP in German see Grewendorf (1989a), Haider (1993) and Sabel (1995). A summary of the arguments is given in Erb (1995). See als Haider and Rosengren (1998).

(11.1)

The syntactic structure of transitive reflexive sentences with a ‘personal’ subject

a.

… weil die Tür sich öffnet

… because the door rp opens

b.

[_CP [_C’ weil [_AgrSP die Tür_S [_AgrS’ [_AgrOP sich_O [_AgrO’ [_VP t_S [_V’ t_Ot_V ]] t_V ]] öffnet_V ]]]]

Note that the reflexive pronoun overtly moves to AgrOP,Spec if we assume that the adverbial is adjoined to VP in middle constructions (cf. 11.2 below). In a second step it possibly adjoins to AgrSP. Scrambling of the reflexive pronoun is supported by the observation we discussed above that pronouns tend to be adjacent to C°, the Wackernagel position. Impersonal middle constructions can be analysed in the same way. The corresponding structure is given in (11.2) below. The examples in (12.a-c) show that the impersonal subject es has the same syntactic distribution like referential subjects. It can occur in sentence-initial position, after the finite verb in main clauses (cf. 12.a-b’), or after the complementizer in embedded clauses (cf. 12.c and c’). Besides, the reduced form of impersonal subject can cliticize to another constituent in phonology (cf. 12.d). Note finally that the impersonal subject, like non-argument reflexives, cannot be focused and coordinated. As opposed to non-argument reflexives, impersonal subjects can be fronted because subjects can occur in sentence-initial position in unmarked word order. The next section discusses focus, coordination and fronting.

(12)

a.

Es schläft sich gut in diesem Bett

a’.

Er schläft gut in diesem Bett

It sleeps rp well in this bed

He sleeps well in this bed

b.

In diesem Bett schläft es sich gut

b’.

In diesem Bett schläft er gut

c.

...weil es sich in diesem Bett gut schläft

c’.

...weil er in diesem Bett gut schläft

d.

In diesem Bett schläft sich’s (=es) gut

e.

Peter sieht *(es) regnen

Peter sees it-acc rain

Impersonal subjects need not check nominative case, as can be seen in (12). The embedded impersonal subject cannot check nominative case in so-called A.c.I.-constructions. Nevertheless, sentence (12.e) is ungrammatical without an impersonal subject. Note that German is no pro-drop language. Impersonal subjects may be licensed by either of the following three conditions: (i) German has strong a [EPP]-feature that must be checked in Spec IP. In this case, the impersonal subject in (12.e) is inserted in IP-Spec of the embedded infinitive to check the [EPP]-feature. After that it moves to AgrOP-Spec of the matrix clause to check accusative case. (ii) A VP obligatorily requires an [+R]-expression. We will argue in chapter 5 that personal pronouns are specified as [+R] as opposed to non-argument reflexives that are specified as [-R]. Hence, the impersonal subject (i.e. the personal pronoun es (‘it’)) is the only expression that fulfills condition (ii) in impersonal middle constructions and sentences containing weather-verbs. (iii) The impersonal subject in middle construction is necessary to bind the non-argument reflexive. According to this assumption, the impersonal subject in middle constructions is subject to different licensing conditions than the impersonal subject of weather-verbs. Our analyses of impersonal middle constructions is compatible with all three conditions. In the following analysis we refer to condition (ii) but further research on impersonal subjects and impersonal passives

Recall from section 2.1.2 and 3.1.1 that impersonal passives do not have an impersonal subject at all.

will be necessary to decide this issue. Impersonal subjects, like non-argument reflexives, are only syntactic arguments. Both elements fulfill a grammatical function. The non-argument reflexive indicates valency reduction whereas the impersonal subject fulfills some subject-releated function in active sentences. We argue in chapter 5 that these different functions follow from the morphological specification of the impersonal subject on the one hand and the reflexive pronoun on the other. In the following discussion we assume that the impersonal subject es is inserted directly into Spec-VP. The resulting structure for an impersonal middle construction is given below.

(11.2)

The syntactic structure of transitive refl. sentences with an ‘impersonal’ subject

a.

… weil es sich hier gut schläft

… because it rp here well sleeps

b.

[_CP [_C’ weil [_AgrSP es_S [_AgrS’ [_AgrOP sich_O [_AgrO’ [_VP gut

[_VP t_S [_V’ t_Ot_V ]] t_V ]] schläft_V ]]]]]

We conclude this section with three general remarks on the syntax of transitive reflexive sentennces. Firstly, Gärtner and Steinbach (1994, 1997 and 2000) argue that there is no empirical and conceptual evidence for a so-called asymmetry analysis that assumes different sentence-initial positions for subjects on the one hand and fronted objects, fronted VPs, or fronted adverbials on the other. We uniformly analyze all main clauses as CPs, no matter which constituent occupies the sentence-initial position (this can either be the subject, an object, the VP, or an adverbial, cf. also 4.2.3 for more discussion).

Wilder (1993), like Zwart (1993), wants to avoid vacuous movement in case of subject-initial clauses. He assumes a ‘mixed projection’ instead. Hence, the sentence-initial position is both CP and AgrSP. As opposed to Zwart’s analysis, Wilder’s analysis is no asymmetry-analysis in the strict sense because subject-initial sentences are hybrid CP/AgrSP-structures. Therefore, subject-initial sentences as well as object- or adverbial-initial sentences are always CPs.

Besides, Wilder’s and Zwart’s analyses are motivated by the assumption that sentence-initial objects but not sentence-initial subjects are ‘topics’. We refer the reader again to Gärtner and Steinbach (1994 and 1997) who show in detail that this assumption cannot be maintained. So far no definition of the term (syntactic) topic and the corresponding topic-feature has been given that includes sentence-initial adverbials and accusative objects and excludes nominative subjects. See also section 4.2.3 for a discussion of the restrictions on the sentence-initial position in German.

Secondly. note that in our framework accusative assignment is not a specific lexical property of a verb. In addition to middle constructions and anticausatives, resultatives or ECM-constructions are further examples where accusative object is not only licensed by the verb itself but by the whole construction. Besides, most verbs that are typically one-place predicates can also assign accusative case (these objects are, however, semantically restricted):

(13)

a.

Es regnet dicke Tropfen/Konfetti

It rains big drops/confetti

b.

Er schlief den furchtbarsten Schlaf seines Lebens

He slept the most terrible sleep of his life

We assume that structural (accusative and nominative) case is ‘assigned’ by morphological feature checking in syntax, cf. Chomsky (1995).

Dative case might be an exception. In section 3.2.2 we saw that dative case poses a serious problem for a lexical analysis of middle constructions. Moreover, dative objects differ in syntax from accusative objects in many respects (cf. chapter 6 below). Besides, two-place verbs like helfen (‘help’) or folgen (‘follow’) can appear only with dative objects - cf. the minimal pair begegnen and treffen (both: ‘meet’) in (i). Hence, not all instances of dative case can be analysed as a structural case that is assigned to the third argument of the verb.
(i)

a.

Maria traf den Bundeskanzler

Maria met the-acc chancellor

b.

Maria begegnete dem Bundeskanzler

Maria met the-dat chancellor

Dative assignment might either be a lexical property - especially in the case of two-place verbs with dative objects - or it can be reduced to semantic differences between dative and accusative case. It is well known that dative case is semantically more specific than accusative (cf. e.g. Wegener 1985). But as it stands this issue is still an unsolved problem that requires further research. We come back to dative case in chapter 6.

We argue in chapter 5 and 6 that accusative and nominative are structural case forms. Accusative and nominative NPs moves to AgrO,Spec and AgrS,Spec respectively (i.e. the extended projections of the verb) to check their structural case-feature. In their VP-internal base positions both NPs bear a specific relation to the arguments of the verb which restricts their semantic interpretation. The nominative NP is always linked to the first (external or internal) argument of the verb and the accusative NP to the second one. We come back to this issue in chapter 5.

Thirdly, we do not assume that all semantic arguments are obligatorily linked to syntax as is, for example, claimed in the theta-criterion. Implicit arguments need not project to syntax. On the other hand, all syntactic arguments are linked to a semantic argument except for non-argument reflexives and impersonal subjects. We will argue below that non-argument reflexives and impersonal subjects are two well defined exceptions. Their morphosyntactic specification enables them to fulfill specific grammatical functions. Hence, both elements are licensed by the grammatical functions they fulfill. Note that some lexical approaches also assume that non-argument reflexives and impersonal subjects must not be linked to (i.e. do not bind) a semantic argument variable (cf. e.g. Bierwisch 1997). Otherwise the occurence of these elements must simply be stipulated in the lexical entries of zero-place verbs, and personal and impersonal ‘middle verbs’.

Focus, coordination, and fronting: explaining the difference

So far we only told half of the story since we did not discuss the differences between argument and non-argument reflexives. Non-argument reflexives are in fact not completely identical to argument reflexives. Haider (1982) cites Reis (1981), who observes that only argument reflexives can be coordinated. In addition, only argument reflexives can be focused and occur in the scope of a focus operator and the contrastive negation. Moreover, only argument reflexives can be replaced by another (non-reflexive) DP and they can be questioned and moved into the sentence initial position. These differences are illustrated in (14), (15) and (16). (14) illustrates that argument reflexives are grammatical in coordination and sentence-initial position and that they can be focused and replaced, cf. Hermodsson (1957) and Duden (1973:75f.).

(14)

a.

Otto wäscht sich und seine Freunde

(coordination)

Otto washes rp and his friends

b.

Otto wäscht [SICH/sich selbst]

(narrow focus)

c.

Otto wäscht nur/sogar sich

(focus particles)

Otto washes only/even rp

d.

Otto wäscht freitags nicht SICH (sondern HANS)

(contrastive negation)

Otto washes on Friday not rp (but Maria)

e.

Otto wäscht sich/Maria

(substitution)

Otto washes rp/Maria

f.

Wen wäscht Otto? sich!

(questioning)

Who washes Otto? rp!(Who is Otto washing? Himself)

g.

Sich hat Otto gestern nachmittag gewaschen

(fronting)

rp has Otto yesterday afternoon washed

This is all impossible for non-argument reflexives as can be seen in (15) for middle constructions and in (16) for anticausatives. The same holds for inherent reflexive verbs. The ‘*’ indicates that all sentences in (15) and (16) cannot receive a middle interpretation and an anticausative interpretation respectively. The only reading that is available for (15) and (16) is the reflexive interpretation which would involve linking of both the subject and the reflexive pronoun. But in this case the reflexive pronoun would be an argument reflexive and the meaning of both sentences would be nonsens (books usually do not sell themselves and doors do not open themselves or windows). In the following we argue that the sentences in (15) and (16) are in fact syntactically wellformed. However, the semantics of focus, coordination and fronting forces the reflexive pronoun to linked to a semantic argument, i.e. to be interpreted as an argument reflexive.

(15)

Middle construction

a.

*Das Buch verkauft sich und seinen Autor gut

(coordination)

The book sells rp and his author well

b.

*Das Buch verkauft [SICH/sich selbst] gut

(narrow focus)

c.

*Das Buch verkauft nur/sogar sich gut

(focus particles)

The book sells only/even rp well

d.

*Das Buch verkauft nicht SICH gut (sondern...)

(contrastive negation)

The book sells not rp well (but...)

e.

Das Buch verkauft sich/*seinen Autor gut

(substitution)

f.

*Wen hat das Buch gut verkauft? sich!

(questioning)

g.

*Sich hat das Buch gut verkauft

(fronting)

(16)

Anticausative

a.

*Die Tür öffnet sich und das Fenster

(coordination)

The door opens rp and the window

b.

*Die Tür öffnet [SICH/sich selbst]

(narrow focus)

c.

*Die Tür öffnet nur/sogar sich

(focus particles)

The door opens only/even rp

d.

*Die Tür öffnet nicht SICH (sondern...)

(contrastive negation)

The door opens not rp (but...)

e.

Die Tür öffnet sich/*das Fenster

(substitution)

f.

*Wen öffnet die Tür? sich!

(questioning)

g.

*Sich öffnet die Tür

(fronting)

According to Haider, these data in (14), (15), and (16) can be explained by a syntactic analysis that distingushes argument reflexives from non-argument reflexives (cf. chapter 3). He argues that the non-argument reflexive is an A’-element only bound in syntax by the subject of the sentence. It is adjoined to VP and receives the theta-role of the implicit first argument of the verb. As opposed to the non-argument reflexive, the argument reflexive is syntactically and semantically bound by the same antecedent, the subject of the sentence. Since non-argument reflexives are A’-elements, they cannot be coordinated with A-elements or, to put it the other way round, if they are coordinated with A-elements, they are interpreted as A-elements themselves. However, we expect that coordination of the non-argument reflexive and another A’-elements should be grammatical. But non-argument reflexives cannot be coordinated at all. Hence, additional constraints on coordination of A’-elements are necessary to exclude non-argument reflexives. In 4.2.1 we give a semantic explanation that does not rely on further constraints on coordination of A’-constituents. Besides, Haider claims that focusing and fronting of non-argument reflexives is impossible because they are not referential (as opposed to argument reflexives that are referential because they are bound by a ‘referential’ antecedent). However, non-referential pronouns can appear in sentence-initial position, as can be seen in (17). Regnen (‘rain’) in (17.a) does not select a referential argument. Nevertheless, the impersonal subject is grammatical in sentence-initial position because a nominative pronoun is usually the first element in unmarked word order. We come back to this issue in section 4.2.3. below.

(17)

a.

Es regnet

It rains

b.

Es tanzt sich sehr gut hier (vs. Hier tanzt es sehr sich gut)

It dances rp well here (i.e. You can dance very well here)

As a consequence, fronting is not a matter of referentiality. Moreover, it is not clear what Haider means by ‘referential’. Verbs and modifiers might not be referential either, but they can be focused and they can occupy the sentence-initial position. Hence, focus is not a matter of referentiality either.

(18)

a.

Peter hat das Buch [sogar gelesen]

Peter has the book even read

b.

[Selten] sind so viele Leute gekommen

Seldom are so many people come

c.

Der Atomphysiker hat das Buch wieder mal [nicht aufmerksam] gelesen

The nuclear physicist has the book once again not attentively read

Note finally that referentiality is a semantic and not a syntactic notion. Therefore, the central part of Haider’s argumentation concerns semantics. It is the semantic status of the reflexive that is responsible for the differences mentioned above. So far we saw that a syntactic distinction between argument and adjunct reflexives is not sufficient to explain the differences between (14) on the one hand and (15) and (16) on the other. In the remainder of this chapter we argue that it is also not necessary. We follow Fagan’s (1992) idea that the difference between these two types of the accusative reflexive pronoun must be explained in semantics rather than in syntax. This issue will be investigated in the following subsections in more detail. We illustrated how the differences between the argument and non-argument reflexive can be derived from independently motivated theories of coordination, focus, and fronting. The following table, which is partly taken from Haider (1982), summarizes the empirical facts and gives an overview of the structure of the following discussion.

Coordination

We repeat the relevant examples from (14), (15) and (16) that illustrate the contrast between the argument and the non-argument reflexive at the beginning of each subsection. (14.a) and (15.a), repeated as (19.a and b), show that only the argument reflexive can be coordinated.

(19)

a.

Otto wäscht sich und seine Freunde

(argument-reflexive)

Otto washes rp and his friends

b.

*Das Buch verkauft sich und seinen Autor gut

(non-argument reflexive)

The book sells rp and his author well

c.

*Die Tür öffnet sich und das Fenster

(non-argument reflexive)

The door opens rp and the window

The intuition behind Haider’s explanation of the ungrammaticality of (19.b) and (19.c) seems to be the following: only syntactically and semantically identical constituents can be coordinated. At first sight one could argue that a reflexive pronoun that is conjoined with another (‘referential’) NP must also receive a ‘referential’ interpretation, because both parts of the conjunction are interpreted alike. In (19.b) and (19.c) the second part of the coordination is a referential NP that must be linked to the second argument of the verb (i.e. must be assigned a theta-role). Therefore the first part of the conjunction (i.e. the reflexive pronoun) must also be linked to the second argument variable. But this is only possible for argument reflexives. Therefore the second and the third sentence in (19) are only grammatical if they contain an argument reflexive. Note, however, that in this case the interpretation is nonsense. This analysis is in principle correct but it does not explain why sentence (20.a) below is also ungrammatical. As opposed to the corresponding sentence in (19.c), sentence (20.a) does not involve coordination of two NPs but of two sentences one of which contains a gap in the position of the verb. (20.b) is a similar example and the ungrammaticality of both examples might be related to a zeugma-effect, which is illustrated in example (20.c) (cf. Bierwisch 1983: 92f.). Hence, we are first have to answer the question what semantically ‘identical’ exactly mean?

Sentences like (i) might be an example for syntactically asymmetric coordination. In the following discussion we are mainly dealing with symmetric coordination and the problem of semantic ‘identity’.
(i)

Gestern ging der Jäger in den Wald und schoß den Hasen

Yesterday went a hunter in the forest and shot a hare

(20)

a.

*Hans öffnet das Fenster und die Tür sich

Hans opens the window and the door rp

b.

*Maria trinkt ein großes Bier und Peter den Hans unter den Tisch

Maria drinks a big beer and Peter the Hans under the table

c.

^???Die Schule bekam einen neuen Lehrer und ein Flachdach

The school got a new teacher and a flat roof

Consider example (20.b) first. Trinken (‘drink’) is interpreted differently in each conjunct. In the second conjunct trinken does not mean that Peter is drinking something, i.e. Hans, (this might only be possible if Hans is a liquid). The resultative construction in the second conjunct means that Peter and Hans are drinking alcohol and as a result Hans ends up under the table. In the first conjunct trinken is interpreted as a simple two-place verb, i.e. it means that Maria is drinking a big beer. Similarily, (20.c) involves two different specifications of the concept Schule (‘school’). In the first conjunct the word Schule means an ‘institution’, whereas in the second conjunct it means a ‘building’ (cf. Nunberg 1979, Bierwisch 1983 and Dölling 1992a/b on the concept of polysemy and lexical underspecification).

In the following discussion we refer to Wilder (1994 and 1995), who offers a unified analysis of coordination and ellipsis. He argues that many restrictions on coordination are syntactic. We choose this analysis for two reasons: first, it allows a unified analysis of (19.c) and (20.a); second, we want to illustrate that our main proposal can also be integrated in syntactic oriented approaches to coordination. Wilder analyses coordination as an application of forward and/or backward deletion. This is illustrated in (21) for forward (he) and backward (the newspaper) deletion (the examples in (21) and (22) are from Wilder 1995).

(21)

a.

He bought and read the newspaper

b.

[He bought the newspaper] and [he read the newspaper]

According to Wilder, forward and backward deletion are subject to different restrictions in English. Whereas backward deletion (BWD) affects only right-peripheral material and is licensed at PF, forward deletion (FWD) dependencies are licenced at LF and the deleted material must occur left-peripheral in the conjuncts. Therefore, only backward deleted material must satisfy a condition on form-identity at PF:

For details see Wilder (1995: 287f.).

(22)

a.

I am drinking beer and John ___ wine

(FWD: am>is)

b.

*John said that I ______________ and

Mary said that she is the best swimmer

(*BWD: am>is)

c.

John said that I ______________ and

Mary said that she was the best swimmer

(BWD: was>was)

These constraints postulate a certain asymmetry between BWD and FWD. In FWD we expect strict LF identity, i.e. both the deleted element and its antecedent receive exactly the same interpretation. As opposed to FWD, BWD should allow the gap and its antecedent to receive different interpretations at LF. The following (weak) contrast seems to confirm this for English:

Thanks to Chris Wilder for drawing my attention to this point. Example (23.a) is from Pullum and Zwicky (1986), the Laster-example in (24) is due to Hans-Martin Gärtner. This subtle contrast might be related to processing: in FWD, the interpretation of the deleted element is perhaps fixed as soon as the parser processes the antecedent. But this is just speculation, so we leave this point open.

(23)

a.

^??At the present the project managers, but in the past the executive directors, set the research priorities

a.’

* (as for our research properties...) project managers set them in the past and executive directors at the present

The situation is less clear in German. For most native speakers, the BWD in (24.a) is as ungrammatical as the corresponding FWD in (24.b). There is, however, a clear contrast between (24.a) and (24.b) on the one hand and (24.c) on the other. Although the first two examples might not be totally ungrammatical, they are both much worse than the third one, which is perfectly grammatical. Only in (24.c) the antecedent and the gap receive exactly the same interpretation.

(24)

a.

*?Für diese Spedition fahren viele ___ und Maria plagen einige Laster

For this forwarding agency drive many trucks and Maria is-troubled-by some vices

b.

*?Viele Laster fahren für die Spedition und ____ plagen Maria

Many trucks drive for this forwarding agency and many vices worry Maria

c.

Peter plagen viele ___ und Maria wenige Laster

Peter is-troubled-by many vices and Mary is-troubled-by few vices

Speakers have a clear preference to give Laster (either ‘truck’ or ‘vice’) the same interpretation in both conjuncts. The situation is similar in (20.a) and (20.b). We try to interpret the verb drink in the second conjunct also as an action of putting some liquid in one’s mouth and swallowing this liquid because this is the interpretation the verb receives in the first conjunct. But this interpretation does not make sense in the second conjunct.

Reinterpretation seems possible to a certain degree which depends on various additional conditions. Various factors influence this.

a) syntactic parallelism facilitates reinterpretation:
(i)

^??Peter setzt sich auf ____ und Hans geht in die Bank

Peter sits down on the bench and the Hans goes into the bank

(ii)

*Peter setzt sich auf ____ und Hans beauftragt mit dieser Angelegenheit eine Bank

Peter sits down on the bench and Hans instruct with this affair a bank

b) reinterpretation seems to be more difficult, if the homonymous elements cannot be reduced to the same underspecified lexical entry.

c) reinterpretation is easier if both elements (nouns or verbs) belong to the same semantic class. Coordination of a two-place predicates with e.g. an homonymous one-place predicate is much worse than coordination of two homonymous two-place predicates:
(iii)

^??Ich weiß nicht, ob ich zu_____ oder aufhören soll (Chris Wilder, p.c.)

I don’t know, if I should listen-to (hören-zu) or stop (hören-auf)

(iv)

*Er hat das Buch ___ und die ganze Nacht gelesen

He read the book and he read the whole night

d) Likewise the copy-antecedent relation can be more easily established when the two homonymous nouns belong to the same semantic sort. Compare (24) above with (i). Bank refers to bench and bank (building) respectively in the first and second conjunct. In both interpretations it refers to a physical object. On the other hand, Laster in (24) refers to a physical object in the first conjunct (truck) and to a moral idea in the second one.

The problem of reinterpretation requires, of course, further research.

In German LF-identity is at least the highly preferred option for the interpretation of the deleted material and its antecedent in both FWD and BWD.

The following example might be evidence for LF-identity of the copy and its antecedent. Although (i) does not involve PF-deletion, the same kind of zeugma-effect can be still observed. Therefore, this effect might be a matter of further conceptual inference.
(i)

Der Krug und der Jüngling, die brechen nach dem Trunke

^??The mug and the youth, they break/vomit after the drink

With this in mind we come back to the problem under discussion. In (25) we give further examples parallel to (20.a). In all examples the interpretation of the gap and its antecedent differ:

Sentence (25.b) gets grammatical if we use und zwar (namely) instead of und. In this case, the second conjunct does not describe an independent event but restricts the meaning of the first conjunct. We ignore und zwar in the following discussion.

(25)

a.

*Peter loaded sand on the wagon and ____ the truck (with hay)

b.

*Der Wagen brummt und ____ um die Ecke

The car growl and the car growl around the corner

In (25.a) the second part of the coordination can only mean that Peter loaded sand on the truck, too. It cannot mean that Peter loaded the truck. It is impossible to take a different variant of the locative alternation verb load for each conjunct. The same effect can be observed in (25.b). The verb brummen (‘growl’) is either a one-place predicate with the meaning ‘to produce a sound’ or a verb of motion selecting a directional PP.

Cf. Jackendoff (1990), Levin (1991) and Levin and Rappaport Hovav (1991) on this issue. We do not want to discuss whether this kind of verbal polysemy must be represented in the lexicon (e.g. by so-called ‘lexical extension’ or ‘lexical subordination’) or whether it is another case of underspecification.

The one-place predicate is unergative and the two-place predicate is unaccusative. Hence, not only the verb but also the subject receives a different interpretation (or theta-role) in each conjunct. (25.b) is in conflict with the condition that the deleted copy and its antecedent must receive exactly the same interpretation in both conjuncts. The interpretations of the verb are illustrated in (26.I) for the first and (26.II) for the second conjunct.

(26)

a.

[_CP Der Wagen [_C° brummt] [_VP ]] und

[_CPDer Wagen [_C°brummt] [_VP die Straße runter]]

I.

B₁< w >

(unergative)

II.

B₂<< w >> & around-the-corner < w >

(unaccusative)

Coordination of a non-argument reflexive with some other DP leads to the same conflict. In (27) we repeat the corresponding examples from (15) and (16). (27.a’) and (27.b’) is a simplified semantic representation (‘b’, ‘a’, and ‘t’ stand for Buch, Autor, and Tür; ‘Op’ stands for the semantic operator that binds the implicit argument in middle constructions, cf. chapter 7.)

(27)

a.

*Das Buch verkauft sich und das Buch verkauft seinen Autor gut

I.

V <(Op x)< b >>

II.

V < b_{< a >>}

b.

*Die Tür öffnet sich und die Tür öffnet das Fenster

I.

O << t >>

II.

O < t_{< f >>}

We argue in the next chapter that the non-argument reflexive is not linked to an argument variable of the verb. But a verb that does not link its first argument to syntax cannot be the antecedent of a verb that links both the first and the second argmuent. Moreover, the subjects in (27.a) and (27.b) are linked to different arguments of the verb in each conjunct. In the first conjunct the subject in linked to the second argument of the verb, whereas it is linked to first argument in the second conjunct. Hence, the coordination in (27) does not only involve two different interpretations for the verb but also for the subject. LF-identity between the deleted material and its antecedent is impossible in both sentences in (27). We could either choose the anticausative or middle interpretation (first conjunct) or the reflexive interpretation (second conjunct). If we chose the former, the accusative object in the second adjunct could not be linked to a semantic argument.

We argue below that all syntactic arguments that are specified as [+R] must be linked to a semantic argument. The only element that is not inherently specified as [+R] is the reflexive pronoun. The impersonal subject must also not be linked to a semanntic argument, although it is specified as [+R], cf. chapter 5 for the principles of argument linking in German.

But if we chose the latter, both the subject and the reflexive pronoun must also be linked to the first and second argument position in the first conjunct. In this case the first conjunct would mean that the the book is selling itself and that the door is opening itself. These interpretations are, of course, nonsense. Even if we permit two different interpretations of the verb in both conjuncts (i.e. the middle or anticausative interpretation in the first and the reflexive interpretation in the second conjunct) the meaning of the second conjunct would still be nonsense since the subjects das Buch (‘the book’) and die Tür (‘the door’) are linked to the first argument of the verb. Note that coordination of an active with a passive verb also leads to ungrammaticality.

Examples like (28) become slightly better with narrow focus on the auxiliary.

(i) ^??Das Buch WURDE und Hans HAT gelesen

(28)

a.

*Das Buch wurde ____ und Hans hat gelesen

The book was read and Hans has read

b.

*Die Nachrichten wurden ____ und Hans hat die Zeitung gelesen

The news were read and Hans has the book read

Coordination of the argument reflexive with another argument DP is perfectly grammatical (cf. 14.a repeated below as 29). In (29) the verb waschen (‘wash’) receives identical interpretations in both conjuncts, as can be seen in (29.I) and (29.II).

(29)

a.

Otto wäscht sich und _____ seine Freunde

Otto washesrp (i.e. himself) and Otto washes his friends

I.

V < o < o >>

II.

V < o_{< f >>}

According to Wilder’s analysis, the traditional NP-coordination in (27) involves deletion of the verb and the subject. We argued that the interpretation of both the verb and the subject is different in each conjunct. Now we can come back to sentence (20.a), repeated as (30). Again the deleted material (the verb öffnen (‘open’)) and its antecedent are not LF-identical. Only in the first conjunct both semantic arguments are linked to syntax. Öffnen in the second conjunct is again anticausative and the subject is linked to its internal argument position. The non-argument reflexive is not linked to an argument of the verb.

(30)

*Hans öffnet das Fenster und die Tür _____ sich

Hans opens the window the door and opensrp

The same difference can be found in sentence (31), which corresponds to (27.a). The second conjunct is a middle construction, the subject of which is again linked to the second argument of the verb verkaufen (‘sell’).

The examples in (30) and (31) seem to be slightly better than the corresponding examples in (27). As opposed to (27), the sentences in (30) and (31) contain two different subjects, one for each conjunct. Therefore reinterpretation is only possible in (30) and (31). Neither conjunct in (27.a) and (27.b) can reiceive a meaningful interpretation under reinterpretation. A similar effect can be observed if we coordinate two reflexive verbs, one with an argument reflexive (rasieren – ‘shave’) and one with a non-argument reflexive (schämen – ‘be ashamed’).
(i)

^???Peter schämt und rasiert sich

Peter is-ashamed and shaves rp

(31)

*Der Autor hat seinen Namen _____ und das Buch sich gut verkauft

His name has the author well sold and the book rpwell sold

We conclude that the ungrammaticality of example (19.b) and (19.c) can be derived from the semantics of the non-argument reflexive (to which we turn in chapter 5).

Focus

We subsume the examples in (14.-16.b) to (14.-16.f) under the term focus because they are all more or less connected to this phenomenon, as will be illustrated in this subsection. Recent theories of focus divide the semantic representation of a sentence into two parts. The first part corresponds to the focus of the sentence, the second part to the background. The focus-background structure of a sentence can be represented as an ordered pair in semantics (for the structured meaning approach see von Stechow 1991, Jacobs 1991, or Krifka 1992; for alternative semantics see Rooth 1985 and 1992 or Büring 1995; cf. also Schwarzschild 1999). We introduce the main concepts of this theory of focus in a nutshell before we apply it to the interpretation of argument and non-argument reflexives.

Consider first the following example in (32), which is taken from Büring (1995). The DP the baseball is the focus of the sentence. It is dominated by the syntactic feature [F] for focus. The head-noun baseball receives a pitch accent at PF and the whole DP is translated as focus at LF.

On the assignment of focus accents see Jacobs 1992 and 1993 or Féry 1993, on focus projection see Büring 1995 or Jacobs 1993.

Büring translates the NP the baseball as an iota expression (i.e. an individual type variable) instead of a generalized quantifier to simplify the illustration.

(32)

a.

John throw [the BASEball]_F

b.

[λx. threw (John, x)] (ιz. baseball (z))

c.

Background: λx. threw (John, x)

d.

ιz. baseball (z)

The background (32.c) results from lambda abstraction. The focus of a clause is replaced by a variable bound by a lambda-operator. The semantic background (or according to Rooth (1992) the focus semantic value of a sentence or constituent) is a set of alternatives to the ordinary semantic value of the sentence. (32.c) (‘λx. threw (John, x)’) is the set of worlds where John threw some x and x is an alternative to the focus (32.d).

(33) {John threw the baseball, John threw the football, John threw his pencil, ...}

Applying the focus to the background will give us the ordinary meaning of the sentence: the set of worlds where John threw the baseball. The focus in (32) is called free focus, i.e. it is not bound by a focus sensitive operator. Jacobs (1984 and 1988) assumes that free focus is bound by an operator as well, namely the illocutionary operator of the clause, which is the assertion operator ASSERT in (32.b). This is illustrated in (34).

(34) ASSERT (<λx. threw (John, x), (ιz. baseball (z)>)

What is the meaning of the ASSERT-operator? Assertion can be seen as a modification of the ‘shared knowledge’ of the participants in the conversation, i.e. the common ground CG (cf. Stalnaker 1978). ASSERT(<α(β)>) maps a common ground CG to a common ground CG’. In a simplistic version the CG is a set of possible worlds.

In fact, the situation is more complex. Among other things we need a representation of the hearer’s assumptions of the speaker’s knowledge and vice versa. But these refinements are irrelevant for the following discussion. On the term common ground see e.g Stalnaker (1978) and especially Zeevat (1997) or Kruijff-Korbayová and Hajicová (1997) for the similar term stock of shared knowledge.

Adding a new proposition (which is also as a set of possible worlds) to CG changes CG into CG’. CG’ is the intersection of CG and our actual proposition (32) (i.e. the possible worlds that make this proposition true).

(35) CG’ = CG ∩λx. threw (John, x), (ιz. baseball (z))

We can state the felicity conditions for the ASSERT-operator now (cf. Krifka 1992: 20 and Büring 1995: 23f.) <α(β)> is the focus-background structure with α the background and β the focus.

If we assume that the semantic background (or semantic focus value) corresponds to the actual common ground CG, then condition (36.c) can be derived from (36.a). There must be at least one alternative X to the focus that fulfills (36.c), otherwise CG’ = CG (cf. Büring 1995: 32).

(36)

a.

CG’ ≠ CG (informativity)

b.

CG’ Ø (compatibility)

c.

There are X , with X ≈β and X ≠β, such that α(X) could have been asserted with respect to CG: i.e. this assertion would be informative (36.a.) and compatible (36.b) and would have yielded a different output context CG’’ with CG’ ≠ CG’’.

According to (36.a), the assertion of a new proposition with respect to CG must provide new information. (36.b) means that the truth of α(β) must not be excluded by CG, i.e. that there is at least one possible world that makes α(β) true and is part of CG or, to put it the other way round, CG’, the intersection of CG and the proposition asserted by the speaker must not be empty. The last condition (36.c) states that there are pragmatically plausible and contextually salient alternatives to the interpretation of the focus. Moreover, these alternatives must be of the same logical type and sort as the focus. This brief outline of focus theory has prepared the ground for the analysis of the interaction of argument and non-argument reflexives and focus. In the following we show that the ungrammaticality of focus on non-argument reflexives can be derived from this theory of focus. We start of with narrow focus.

Narrow focus

Consider first the sentences in (14.b), (15.b) and (16.b), repeated below as (37.a-c). In all examples sich is the focus of the sentence (indicated by the labelled brackets []_F):

(37)

a.

Otto wäscht [SICH/sich SELBST]_F

b.

*Das Buch verkauft [SICH/sich SELBST]_F gut

c.

*Die Tür öffnet [SICH/sich SELBST]_F

We put sich selbst aside for the moment and only consider the stressed reflexive pronoun SICH.

It is not important for the ongoing presentation whether this focus is presentational or contrastive. Note that weak pronouns like sich can be stressed in German. An alternative to focus on the reflexive pronoun itself (with nuclear stress on sich) is the complex form sich SELBST with stress on the adnominal particle selbst (we discuss sich SELBST right away).

Why can the non-argument reflexive not be the focus of a sentence. Recall that the non-argument reflexive does not introduce an argument variable into the semantic representation of the sentence (as opposed to argument reflexives and other nominal expressions, cf. chapter 5). Therefore, replacement of and lambda-abstraction over an argument variable are impossible and no focus-background-structure can be generated for the sentences in (37.b) and (37.c). Both sentences fail to meet condition (36.c). The argument reflexive in (37.a), on the other hand, is linked to the second argument variable. Hence, this expression can be replaced by a variable and lambda-abstraction over this variable is possible. Plausible alternatives are also at hand as can be seen in (38).

The semantic representation in (38.c) is an oversimplification. Actually, possible alternatives to sich (Otto) are not only elements of type <e> but also generalized quantifiers of type << e,t > t >. If the alternatives are required to be type-equivalent, pronouns and argument reflexives must be translated as generalized quantifiers, too.

(38)

a.

Wen hat Otto gewaschen?

b.

Otto hat [SICH]_F gewaschen (... und nicht Maria)

Otto has rp (i.e. himself) washed (... and not Maria)

c.

ASSERT (<λx. wash (o, x), o>)

Syntactic expletives, which are also not linked to a semantic argument, equal non-argument reflexives, cf. (39.a) Narrow focus is again impossible. Furthermore, narrow focus on constituents that do not have plausible type-equivalent alternatives (condition 36.c) is excluded as well (cf. 39.b). Note that every element that is (i) represented in the semantic form and (ii) has at least one plausible alternative can be the focus of the clause, no matter whether it is an A- or A’-element in syntax, cf. e.g. (39.c). It must only fulfills condition 36.c.

(39)

a.

*Bei dieser Sache handelt [ES]_F sich um eine ernste Angelegenheit

In this case concerns it (i.e. it concerns) rp a serious matter

b.

^??Ich habe den Brief [AN]_F den Vermieter geschickt

I have the letter at (i.e. to) the landlord sent

c.

Ich lege das Buch [UNTER]_F den Tisch und nicht [AUF, NEBEN, ...]_F den Tisch

I put the book under the table and not (on, next to, ...) the table

Unlike (39.c), sentence (39.b) contains no alternatives to the focused preposition. The only plausible alternative is zu (‘to’). This preposition, though, selects a dative DP (dem Vermieter), so that it is ungrammatical in this syntactic context.

Focus particles

The analysis of narrow focus can also be applied to the examples with focus particles (14.c, 15.c and 16.c). The relevant examples are repeated in (40.a-c).

(40)

a.

Otto wäscht nur/sogar sich

b.

*Das Buch kauft nur/sogar sich gut

c.

*Die Tür öffnet nur/sogar sich

Jackendoff (1972), Jacobs (1983), or Krifka (1992) among many others analyze focus particles like nur (‘only’) or sogar (‘even’) as focus sensitive operators. Focus sensitive operators, like the ASSERT operator mentioned above, bind the focus of a sentence/constituent. Example (41) illustrates this for nur (‘only’).

(41)

a.

Maria hat nur [HANS]_F geliebt

b.

ONLY (<λx. love (m, x), h>)

The meaning of ONLY can be outlined as follows:

There is a wide range of investigations in the semantics of particles like only (cf. Horn 1996 for only and the references cited there).

the background is applied to the focus Hans and to no other X , with X ≈Hans and X ≠Hans. The scalar focus operator sogar (‘even’) means that the focus β is ranked lower than every alternative X , with X ≈β and X ≠β on a scale of probability determined by the background. Furthermore, sogar implies that the respective focus alternatives make the preposition also true. We do not want to go into detail here. The crucial point has already been mentioned. Focus particles are focus sensitive operators that bind the focus. But as we have seen above binding of a focused non-argument reflexive is impossible because no focus-background structure can be generated in this case.

Contrastive negation and substitution

The relevant examples for contrastive negation (14.d, 15.d and 16.d) and substitution (14.e, 15.e and 16.e) are repeated in (42.a-c) and (43.a-c) respectively. Both require a semantic representation of the negated or replaced element, too.

(42)

a.

Otto wäscht freitags nicht SICH (sondern HANS)

b.

*Das Buch verkauft nicht SICH gut (sondern...)

c.

*Die Tür öffnet nicht SICH (sondern...)

(43)

a.

Otto wäscht sich/Maria

b.

Das Buch verkauft sich/*seinen Autor gut

c.

Die Tür öffnet sich/*das Fenster

If we replace the non-argument reflexive by some other accusative DP, nothing happens in syntax. We have yet another transitive sentence:

(44) Die Tür öffnet sich ⇒ Die Tür öffnet Peter

But in semantics things change. We can substitute one element for another element of the same logical type only if this element is present in the semantic representation. In the semantic representation of (44) we cannot replace sich by the DP Peter because sich is not present there at all whereas the DP Peter is translated as a generalized quantifier or an individual constant. This explanation is parallel to the impossiblity of non-argument reflexives to be focus discussed above. The background results from replacement of the actual focus by a variable bound by a lambda-operator. Substitution of a variable (or of a constant) for another type-equivalent element in the semantic representation is only possible if the element to be replaced is present.

According to Jacobs (1982/1991), contrastive negation (‘fokussierende Negation’) in German can be analysed parallel to focus particles. The negation attracts the focus β. Furthermore, there is an implication that an alternative X exists, with X subject to the condition in (36.c) that requires: (i) X ≈β, (ii) X ≠β, and (iii) X makes the proposition α(X) true. We need again a partition of the semantic representation into focus and background which is again impossible with focussed non-argument reflexives.

Questioning

So far two examples related to focus are still unexplained: questions and selbst. We will turn to questions first. Consider the examples (14.f, 15.f and 16.f), which we repeat in (45.a-c):

(45)

a.

Wen wäscht Otto? sich!

b.

*Wen hat das Buch gut verkauft? sich!

c.

*Wen öffnet die Tür? sich!

Semantically questions can be analysed as a set of possible answers (cf. e.g. Karttunen 1977).

(46)

a.

Wen hat Hans gewaschen?

b.

λp. ∃x [person (x) ∧ p = wash (h, x)]

c.

{Hans hat Peter gewaschen, Hans hat Maria gewaschen, Hans hat ihn gewaschen, Hans hat sich gewaschen, ...}

Hence, the meaning of a question corresponds to the background of the respective answer and the wh-word corresponds to the focus of this answer.

We omit presuppositions and further semantic issues related to questions.

Both, the meaning of a question and the background of the answer can be analyzed as a set of propositions (or possible worlds, compare (33) with (46.c)). A question characterizes the actual common ground (λp. ∃x [person (x) ∧ p = wash (h, x)] ∩ CG = CG). Every possible answer p to a question Q must be informative (p ∩ CG ≠ CG) and compatible (p ∩ CG ≠Ø), for details see Büring (1995: 32f.). According to Büring (p. 35), a sentence S can be uttered as an answer to a question Q given a common ground CG if the focus semantic value (or background) of S is identical to the meaning of the question Q. Hence, the ungrammaticality of (45.b) and (45.c) follows. We have already seen that non-argument reflexives cannot be focused because no corresponding focus-background structure (or focus semantic value) can be generated. But the focus semantic value involves a second argument. This is, however, only possible if the reflexive pronuon is linked to the second argument (47.b), which is only possible for argument reflexives. The argument reflexive interpretation in (47) is, of course, nonsense, because doors usually do not open something else.

(47)

a.

Wen/was hat die Tür geöffnet?

b.

λp. ∃x [person/thing (x) ∧ p = open (t, x)]

c.

{Die Tür hat X geöffnet, die Tür hat Y geöffnet, ...}

d.

Background: λy (open (t, y))

Selbst

Finally we turn to the analysis of selbst. We already saw that the non-argument reflexive cannot be focused and/or modified by selbst.

(48)

a.

Otto wäscht [SICH/sich SELBST]_F

b.

*Das Buch verkauft [SICH/sich SELBST]_F gut

c.

*Die Tür öffnet [SICH/sich SELBST]_F

German does not distingush between weak and strong reflexive pronouns. Sich selbst is not the strong counterpart of a weak form sich. The simple form sich with a pitch accent and the complex form sich selbst are almost identical in German and can be used in the same contexts. In this respect German differs from so-called two-form languages (cf. section 2.3 and chapter 5). The reflexive pronouns zich and zichzelf Dutch, for example, have different syntactic distribution. Only the strong form zichzelf can be focused (i.e. assigned a pitch accent) whereas the weak form zich can only be used with inherent reflexives, anticausatives and verbs that are likely to be reflexive as, for example, wash. König and Siemund (1997: 4) argue that the complex form sich selbst in German consists of two independent parts: the reflexive pronoun sich and the adnominal focus particle selbst. They distinguish between four different types of the particle selbst:

(49)

a.

Selbst RIEsen haben einmal klein angefangen

(scalar focus particle)

rp (i.e. even) giants have once small begun

b.

Der Minister SELBST war in den Skandal verwickelt

(adnominal, centering)

The minister rp (i.e. himself) was in the scandal mixed up

c.

Der Minister war SELBST in den Skandal verwickelt

(adverbial, inclusive)

The minister was rp (i.e. also) in the scandal mixed up

d.

Der Minister löste den Skandal SELBST aus

(adverbial, exclusive)

The minister caused the scandal rp (personally) verbal-particle

In (49.a) selbst is a scalar focus particle, whose meaning is very similar to that of the focus particle sogar (‘even’). Like all scalar focus particles, selbst bears no accent, precedes its focus, and is preferably adjacent to it. As opposed to the focus particle selbst in (49.b), adnominal and adverbial selbst in (49.b), (49.c), and (49.d) must be stressed.

Focs particles seem to receive the pitch accent when they follow their focus. This correslation between the syntactic position and stress assignment also applies to the focus particle allein (cf. Primus 1992:70 and König and Siemund 1997:8).

(ii) Allein in HAMburg gibt es riesige ProBLEme

Adnominal selbstalways adjoins to the right of the NP it modifies, cf. (49.b). This NP is the focus bound by the adnominal modifier. It is interpreted as the center with respect to the focus alternatives. In (49.c) and (49.d) selbst is an adverbial. These two uses of selbst are closely related but differ in their syntactic distribution and their semantics. The so-called inclusive selbst tends to precede the object in the middle field, whereas exclusive selbst preferably follows the object.

In the following transitive clause, there is a strong preference that the first selbst that precedes the indefinite object is exclusive and the second one is inclusive.
(i)

Ich habe (selbst) einen Aufsatz (selbst) gelesen

I have (selbst) a paper (selbst) gelesen

Both types of adverbial selbst are usually subject-oriented. However, inclusive selbst can also modify dative or an accusative objects that precede the subject in unmarked word order (or on the thematic hierarchy) and exclusive selbst can modifiy a by-phrase in the passive.

Experiencer verbs are typical examples for dative and accusative objects that can precede the subject. In this case inclusive selbst does not modify the subject but the object, example (i) is from König and Siemund (1996:11). For unmarked word order see also section 4.1. above and the references cited there.
(i)

Mir ist dieser Kerl selbst nicht geheuer

Me-dat is this guy selbst not sympathetic

‘Even to me this guy is eerie’

(ii)

Mich interessiert diese Frage selbst

Me-acc interests this question selbst

‘I am interested in this question myself’

Semantically the inclusive adverbial can be compared to the scalar focus particle and the exclusive adverbial to the adnominal selbst (although the exclusive selbst does not presuppose that the focus is the center with respect to the focus alternatives).

The interpretations of inclusive and exclusive selbst are highly context dependent. The inclusive interpretation requires events that are repeatable (like e.g. read a book vs. write the book) or states that are not exclusive (like e.g. speak a language). The exclusive reading only makes sense if something cannot only be done on one’s own but also with other people’s help. Furthermore, minimal pairs like (i) show that there is a certain interaction between topic/focus and the inclusive/exclusive reading: the exclusive reading is preferred if the NP is topic (cf. Jäger 1995), and the inclusive one is favored if the NP is focused.
(i)

Peter hat selbst ein Buch gelesen (inclusive)

Peter has himself a book read

(ii)

Peter hat ein Buch selbst gelesen (exclusive)

Peter has a book himself read

Syntactically, both the inclusive and the exclusive selbst can be analysed as VP-adverbials. The semantic analysis seems to be much more complicated. It is not clear yet whether both readings can possibly be reduced to one basic (possibly underspecified) lexical entry (cf. Primus 1992 and especially König and Siemund 1997:18f. for more details).

The following table summarizes the relevant properties of the four different types of selbst (‘among others’ stands for the presupposition that is triggered by selbst).

(50)

(Ad)nominal and adverbial selbst

scalar f.p.

adnominal

inclusive

exclusive

accent

no

yes

yes

yes

post XP

yes

no

no

no

adverbial

yes
Alternatively, focus particles can be analysed as cross-categorial operators (which might be subject to language-specific constraints), cf. König (1993) for an overview.

no

yes

yes

‘among others’

yes

no

yes

no

It is an interesting issue whether all occurrences of selbst can be reduced to one underlying lexical entry and whether they all belong to the same syntactic category (as e.g. proposed in Primus 1992). However, our main interest is the interaction of selbst with non-argument reflexives. Both the scalar focus particle and the adnominal selbst cannot modify the non-argument reflexive. Although inclusive and excluse adverbial selbst is grammatical in sentences that contain a non-argument reflexive it does not interact with the non-argument reflexive but with the subject of the sentence. We underline the constituent, inclusive and exclusive selbst modifies in these examples. In (59.f) the exclusive selbst applies to the implicit argument.

(51)

a.

*Die Tür hat [selbst SICH] geöffnet

(scalar f.p.)

b.

*Die Tür hat [sich SELBST] geöffnet

(adnominal)

The door opens even itself (a)/ITSELF (b)

c.

Peter schämt sich doch SELBST

(inclusive)

Peter is-ashamed particle himself

d.

(I do not need your modern off-road vehicles for this journey to the North Cape...)

Mein Auto fährt sich auf solchen Straßen SELBST sehr gut

(inclusive)

My car drives rp on that kind of street also very well

‘My car drives on that kind of street very well, too’

e.

?Die Tür hat sich gerade eben SELBST geöffnet

(exclusive)

The door has rp just now by itself opened

f.

Schwierige Aufgaben lösen sich doch meistens SELBST am besten

(exclusive)

Difficult exercises solve rp particle usually by oneself best

The explanation for (51.a) is straightforward. The semantic representation of the scalar focus particle is basically identical to that of other focus sensitive operators like nur (‘only’) or even (‘sogar’). The meaning of the focus operator selbst is almost equivalent to even (for differences in syntax and semantics see Primus 1992). Therefore, the explanation for the incompatibility between focus sensitive operators and non-argument reflexives can also be applied to this type of selbst. The explanation of (51.b) is also based on the theory of focus outlined above. Primus (1992) and König and Siemund (1997) argue that the adnominal selbst is a focus sensitive operator, too. According to Primus, adnominal selbst has also a scalar implication: the adjacent NP is ranked lower on a scale of likelihood determined by the rest of the sentence. Unlike the scalar focus particle, the adnominal selbst does not presuppose that (all/some/no) focus alternatives make the proposition also true. König and Siemund’s analysis of adnominal selbst slightly diverge from that proposed by Primus. The focus is the centre and the alternatives make up the periphery or entourage:

Note that the following examples are sometimes ambiguous between the adnominal and the adverbial exclusive reading. We indicate the adnominal interpretation with brackets.

Adnominal selbst is restricted to persons that are in some sense central or important. The following examples show that modifying non-central persons with adnominal selbst does not make sense. It is difficult to establish a meaningful centre in (i) and (ii.a):
(i)

^??Die Putzfrau SELBST hat unsere Wohnung geputzt

The cleaning lady herself has our flat cleaned

(ii)

The bus had a bad accident …

a.

^??Die Fahrgäste SELBST ist ums Leben gekommen

The passengers himself died

b.

Der Fahrer SELBST ist ums Leben gekommen

The driver himself died

(52)

Peter hat [den Chef SELBST] gesprochen

Peter has the boss rp (i.e. himself) talked

In (52) the likelihood for Peter to speak to the boss is ranked lower than the likelihood to speak to other contextually salient persons (e.g. the assistant, the secretary, …). (52) implies that Peter was not expected to talk to the boss directly. This line of argumentation can be applied to reflexive pronouns, as can be seen in the following two examples.

(53)

a.

Peter hat den [Präsidenten SELBST] im Fernsehen gesehen

Peter has the president himself in the TV seen

b.

Peter hat [sich SELBST] im Fernsehen gesehen

Peter has rp himself in the TV seen

The adnominal focus sensitive operator indicates in (53) that it is remarkable for Peter to see the president or his own self in the TV. This implication is also nicely illustrated by the following minimal pair that is taken from König and Siemund (1997: 28).

(54)

a.

[Der Anführer SELBST] hat sich verraten

The leader himself has rp (i.e. himself) betrayed

b.

Der Anführer hat [sich SELBST] verraten

The leader has rp himself betrayed

By means of his function the leader is the centre of his gang. (54.a) implies that the most important or most clever and cautious person of the gang betrayed him- or herself. (54.b), on the other hand, sets the actual victim (the leader him- or herself) in contrast to other potential victims (the rest of the gang). (54.b) implies that it is noteworthy that the leader has betrayed him- or herself and not the rest of the gang, i.e. that it was a self-betrayal. The scalar implication of adnominal selbst is not always easy to make out, especially if it modifies a reflexive pronoun. In some of these examples selbst does not necessarily imply a ranking between the centre on the one hand and the alternatives on the other. In these cases reflexive + selbst is equivalent to a reflexive with narrow focus (example (55) is from Primus: 75):

(55)

Maria kauft für die ganze Familie etwas Schönes, für Mutter und Brüderchen

Maria buys for the whole family something beautiful, for mother and brother

Pralinen, für Vater und [sich SELBST] Schnaps

chocolate, for father and herself liquor

Buying liquor for herself need not be more remarkable for Maria than buying liquor for her father, mother, or brother. This might be due to the fact that the reflexive pronoun and selbst, unlike some NP and selbst, seem to form an idiomatic unit. The fusion of (reflexive) pronouns and adnominal focus sensitive operators like selbst can be observed in the history of many languages.

Many languages have reflexive pronouns that incorporate a selbst-like element (e.g. English him-/her-/itself - pronoun and scalar expression -, Dutch zichzelf - anaphor and scalar expression - or Hungarian maga). These reflexive pronouns seem to have developed from a pronoun or anaphor and an adnominal scalar expression. The basis for this might have been predicates that favor a disjoint reference interpretation of the arguments over a coreference reading (i.e. the reflexive use is more remarkable than the non-reflexive use). In these contexts, scalar expressions seem to occur regularly together with pronouns or anaphora. In addition to this, this fusion leads to the disambiguation of the binding domains in languages like English. The pronoun him is subject only to Principle B while the reflexive pronoun him-/her-/itself takes on the work of Principle A. For more details see Primus (1992) and König and Siemund (1997) and the references cited there.

The use of German sich SELBST as the focused counterpart of bare sich might be supported by the fact that reflexive pronouns tend to avoid heavy stress. Note finally that indefinites cannot be modified by adnominal selbst. It is impossible for indefinites to establish a centre and a periphery because they do not refer to an unique and specific individual.

(56)

a.

*[Wer SELBST] entging dem Gefängnis

Who -self escaped the prison

b.

*[Jemand SELBST] zeigte der Polizei ein Bild von Anna (Primus: 72)

Someone -self showed the police a picture of Anna

c.

*Mir hat gestern [einer SELBST] das Geld geklaut

Me has yesterday one -self the money stolen

d.

*Maria hat [einen Präsidenten SELBST] auf der Wahlveranstaltung gesehen

Maria has a president himself on the election rally seen

We conclude that both the adnominal and the prenominal focus particle selbst establish a set of semantic alternatives to the denotation of the focus, i.e. the NP they are adjoined to. This is again only possible for argument reflexives. Non-argument reflexives are correctly excluded in the middle construction (48.b) and the anticausative (48.c), (51.a), and (51.b) above. Many issues have been touched only in passing, but we hope that we convincingly argued that a (focus) semantic analysis of the examples discussed in this subsection is on the right track.

Fronting

One issue is still unexplained. Besides their inability to be coordinated and focused, non-argument reflexives cannot be fronted either. The sentence-initial position is not simply a mirror image of the initial position of the middle field. In German various elements can occupy the sentence-initial position (CP,Spec in GB terminology). It is neither a pure topic- nor a pure focus position and neither topic nor focus must occupy this position. Elements that move to CP,Spec are subject to various conditions, that are necessary but not sufficient. These conditions mainly relate to information structuring and discourse semantics (cf. Gärtner and Steinbach 1997 and 2000). (57) lists the the most relevant restrictions for the sentence-initial position.

We correlate unmarked word order with focus projection. Only unmarked word order allows for maximal focus spreading (cf. Höhle 1982, Jacobs 1992 and 1993 and Vogel and Steinbach 1998).

(57)

a)

the first argument of the unmarked word order (in the middle field) can appear sentence-initially in the specifier of CP

b)

the focus can appear in CP,Spec

c)

the ‘topic’ can appear in CP,Spec

d)

further conditions may depend on the structuring of the text or discourse

We will not be concerned with condition (57.d). Reflexive pronouns cannot refer to discourse referents directly because they are usually bound within the sentence they are included (condition A of binding theory). We will see in the next section that non-argument reflexives must be bound by the subject of the same sentence. Therefore, we only discuss condition (57.a), (57.b) and (57.c).

ad a) To be the first element of the unmarked word order is always a good reason for a constituent to appear in sentence-initial position. We already saw that reflexive pronouns can precede the subject (i.e. their antecedent) in the unmarked order in the middle-field only if the subject is not a pronoun itself (the relevant examples are repeated in (58)).

(58)

a.

... weil der Kanzler sich sehr liebt

... because the chancellor-nom rp much loves

b.

... weil sich der Kanzler sehr liebt

c.

... weil er sich sehr liebt

... because he rp much loves

d.

*... weil sich er sehr liebt

Pronominal elements tend to be right-adjacent to C°, the Wackernagel position (cf. Anderson 1993 for a crosslinguistic study of the so-called Wackernagel-Effekt). Therefore pronominal elements precede full NPs. This constraint overrides the constraint which states that nominative precedes accusative. However, the Wackernagel-effect can only be observed in the middle-field, because C° is the left border (Linke Satzklammer) of the middle-field. Hence this constraint does not help us on the sentence initial position.

Besides these rare cases, German has some idiomatic expressions that permit accusative-nominative order (e.g. ihn hat der Schlag getroffen - ‘he was floored’).

We already saw in section 4.1 nominative NPs precede accusative and dative NPs. Thus nominative NPs are always good candidates for the sentence-initial position, as can be seen in (59.a). With some verbs dative NPs can precede the nominative NP. These objects also appear unmarked in sentence-initial position, cf. (59.b). In addition, some verbs permit the accusative to precede the nominative. However, such verbs are rare and the reverse word order (nom precedes acc) is always also unmarked. Verbs in middle constructions as well as anticausatives, and inherent reflexives do not belong to this very small class.

Note that the experiencer verb interessieren (‘be intertested in’) in (59.c) becomes inherent reflexive when we substitute a reflexive pronoun for the accusative NP ein Mädchen (‘a girl’).
(i)

Sie interessierte sich für ein Buch

She is-interested-in rp a book

(59)

a.

Ein Junge hat einem Mädchen ein Buch gegeben

(unmarked)

A boy-nom has a girl-dat a book-acc given

b.

Einem Kind ist ein Stein aufgefallen

(unmarked)

A child-dat is a stone-nom attracted attention

i.e. ‘A stone attracted a child’s attention’

c.

Ein Mädchen hat ein Buch interessiert

(unmarked)

A girl-acc has a book-nom interested

‘A girl was interested in a book’

Besides, ‘stage-setting’ adverbials like gestern (‘yesterday’) can also appear unmarked sentence initially (i.e. they need not be focus or topic).

(60)

Gestern hat der Gärtner die Gitarre aus dem Schrank geholt

Yesterday has the gardener the guitar out of the closet taken

ad b) Focused constituents can also occur in sentence-initial position.

(61)

a.

Who did Hans wash?

b.

[SICH/sich SELBST]_F hat Hans gewaschen

rp (i.e. himself) has Hans washed

c

*[SICH]_F hat die Tür geöffnet

rp has the door opened

(61.a) illustrates that a reflexive pronoun can be moved to CP-Spec, if it is the focus of the sentence. We saw in section 4.2.2. above that non-argument reflexives cannot be focused at all because they are not linked to a semantic argument variable of the verb. Hence this difference between argument and non-argument reflexives is again due to the (semantic) inability of non-argument reflexives to be focused.

Similarly, all fronted constituents that have to be focus because they cannot appear unmarked in sentence initial position (i.a) need some semantic content. We explained in 4.2.2, example (39.b), why focus on the preposition is ungrammatical in (ib), which is only grammatical with narrow focus on the pronoun sie (i.c).
(i)

a.

*[An sie] habe ich einen Brief geschrieben

b.

*[AN sie] habe ich einen Brief geschrieben

c.

[An SIE] habe ich einen Brief geschrieben
to her have I a letter written

ad c) Besides focus, there is an additional condition that can be subsumed under the notion of topic. Recall that the semantic representation of a sentence is divided into two parts, the focus and the background. Vallduví (1992) argues that the background (ground in his terminology) is further subdivided into topic and comment (link and the tail). The link (or topic) is a designated element, i.e. ‘an address pointer in the sense that it directs the hearer to a given address [...] in the hearer’s knowledge store, under which the information carried by the sentence is entered’ (p. 47).

This partition might be too static. At a certain stage of a discourse different expressions can be equally salient or ‘designated’, so that it is sometimes not clear which element of the background is the actual address pointer for the ongoing discourse. See Kruijff-Korbayová and Hajicová (1997) for a dynamic approach.

Based on this assumption, Büring (1995) investigates the syntax, phonology and semantics of topics in German in detail. He gives a well defined notion of sentence-internal topics (S-topics). S-topics share some semantic properties with focus. Both have a similar accent (topics have a rising pitch L*H, focus has a falling one H*L) and the semantics of S-topics is very similar to the semantics of focus. We briefly illustrate this semantic similarity with an example taken from Büring (1995: 49) - ‘/’ and ‘\’ indicate the rising and falling accents respectively:

(62)

a.

What did the popstars wear?

b.

Die [/weiblichen]_T Popstars trugen [Kaftane\]_F

The female pop stars wore caftans

c.

And what about the male pop stars?

S-topics can be used to pick up entities mentioned in the preceding discourse, to narrow down a given dicourse topic (partial topic), to indicate that there are some alternatives to talk about (implicational topic) or ‘to move the conversation away from an entity given in the previous discourse’ (p. 49) (contrastive topic). Sentence (62.b) is an example for a partial topic and sentence (63.b) for an implicational topic:

(63)

a.

Did your wife kiss other men?

b.

[/Meine]_T Frau hat [keine\]_F fremden Männer geküßt

My wife has no other men kissed

c.

But what about YOUR wife?

We do not want to go into detail here. The reader is referred to the detailed presentation in Büring (1995). Two points are of interest here. Firstly, S-topic can appear in sentence initial position.

S-topics have to precede the focus at surface structure. Therefore, CP-Spec is one (maybe the preferred) option for S-topics. They can also appear on the left periphery of the middle-field.

Secondly, the semantics of S-topics is a ‘typed-up’ focus semantics. The second point is illustrated in the following. S-topics induce alternatives similar to the focus. We already know that the focus value (or background) of e.g. example (62.b) is a set like (62.b’):

(62)

b’.

{the female pop stars wore caftans, the female pop stars wore dresses, the female pop stars wore overalls, ....}

According to Büring (p. 56f.), the topic value is computed by replacing the S-topic in (62.b’) with type-equivalent salient alternatives. Thus the topic value of a sentence is a set of such sets:

(62)

b’’.

{{the female pop stars wore caftans, the female pop stars wore dresses, the female pop stars wore overalls, ....}

{the male pop stars wore caftans, the male pop stars wore dresses, the male pop stars wore overalls, ....}

{the female or male pop stars wore caftans, the female or male pop stars wore dresses, the female or male pop stars wore overalls, ....}

{the italian pop stars wore caftans, the italian pop stars wore dresses, the italian pop stars wore overalls, ....}

...}

Furthermore, S-topics have an implication that can be informally outlined as follows: The use of a S-topic in a sentence A implies that there is an element Q from the topic value (62.b’’), such that Q is still under consideration after uttering A (i.e. some elements in one of these sets in (62.b’’) must be non-absurd and informative with respect to CG). After uttering (62.b), we know that there must be some element from (62.b’’) that is still under consideration. The most salient element in (62.b’’) seems to be the following set: {the male pop stars wore caftans, the male pop stars wore dresses, the male pop stars wore overalls, ....}. This means that this set of propositions (i.e. the question: What did the male pop stars wear?) serves as the residual topic in (62). The residual topic of (63.b) (YOUR wife as opposed to MINE) can be derived in the same way. Before we look at an example with argument reflexives we want to capture the fact that non-argument reflexives cannot be S-topics. S-topics induce focus alternatives similar to the focus. But we already saw in section 4.2.2 that non-argument reflexives cannot be the focus and they cannot be S-topic either. A topic value cannot be generated for non-argument reflexives. As opposed to non-argument reflexives argument-reflexives are expected to be S-topics. In addition, they are also expected to occur in sentence-initial position. This is confirmed by example (64).

(64)

a.

Where did Peter take a picture of himself (and his friends)?

b.

[/SICH]_T hat er [vor dem Matterhorn]_F photographiert

Himself has he in front of the Matterhorn photographed

The focus alternatives are salient type-identical elements as we saw in section 4.2.2 above. The topic value (of the reflexive pronoun) results from a second substitution: in all focus alternatives the S-topic (the argument reflexive) is replaced by type-equivalent salient alternatives (cf. 64.d).

(64)

c.

{Er hat sich vor dem Matterhorn photographiert, er hat sich vor dem Montblanc photographiert, er hat sich vor der Eiger Nordwand photographiert, ...}

d.

{{Er hat sich vor dem Matterhorn photographiert, er hat sich vor dem Montblanc photographiert, er hat sich vor der Eiger Nordwand photographiert, ...}

{Er hat seine Freunde vor dem Matterhorn photographiert, er hat seine Freunde vor dem Montblanc photographiert, er hat seine Freunde vor der Eiger Nordwand photographiert, ...}

{Er hat sich und seine Freunde vor dem Matterhorn photographiert, er hat sich und seine Freunde vor dem Montblanc photographiert, er hat sich und seine Freunde vor der Eiger Nordwand photographiert, ...}

...}

According to the implication of topics, the residual topic must be informative. The second element of the topic value is a good candidate for the residual topic: ‘and where did he photograph his friends?’

Sentence (64.b) gets worse if we ask for the subject, i.e. the antecedent of the reflexive pronoun:
(i)

a. Who photographed himself in front of the Matterhorn?

b. ?[/Sich] hat [der Peter\] vor dem Matterhorn fotografiert

The residual topic is something like: and who photographed not himself but something else (X photographed Y with X ≠ Y). In this case we contrast a reflexive action with a non-reflexive action. The only thing that is under discussion here is the intrinsic meaning of the reflexive pronoun, its reflexive function. But for that we need a very special context.

Note that we can replace sich in sentence initial position by sich selbst without change of meaning (cf. 4.2.2).

We conclude that the accusative reflexive pronoun cannot occur in the sentence-initial position in unmarked word order. Therefore, it must either be focus or S-topic to occupy CP,Spec. Neither of these options is available for non-argument reflexives, because they are not linked to an argument variable of the verb. And finally, non-argument reflexives must be bound by the subject within their sentence. Therefore they cannot be subject to further conditions which may depend on the structuring of the text or discourse. In sum, non-argument reflexives do not meet the conditions on fronting in German.

1. unmarked word order:

nominative precedes accusative (the ‘Wackernagel-effect’ is limited to the middle-field)

2. focus:

* non-argument reflexives (cf. 4.2.2.)

3. S-topic:

* non-argument reflexives (parallel to 2.)

4. further conditions on the structuring of texts/discourses:

reflexive pronouns cannot pick up discourse referents from outside their sentence