The Interpretation of Reflexive Pronouns in German

So far, we argued that the different interpretations of transitive reflexive sentences are not related to different syntactic representation. Both the argument and the non-argument reflexive are analyzed as a direct (or accusative) object. In this chapter we will show that the ambiguity of transitive reflexive sentences can be derived at the interface between syntax and semantics. The accusative reflexive pronoun need not be linked to a semantic argument of the verb. This very specific property of the ‘middle marker‘ in German results from the interaction between a universal property of (weak) reflexive pronouns and a language specific distinction between structural and oblique case forms. We will argue below that (weak) reflexive pronouns are ‘referentially’ underspecified. As a consequence, (weak) reflexive pronouns can either be linked to a semantic argument themselves or they mediate the linking of another syntactic argument (usually the subject). We called the former argument reflexive and the latter non-argument reflexive. In German, the latter interpretation is only available for reflexive pronouns that are assigned structural case. Argument reflexives receive a reflexive interpretation whereas non-argument reflexives are indicators of valency-reuction that receive a middle, a anticausative or a inherent reflexive interpretation. The non-argument reflexive indicates that the first argument of the verb is not linked to syntax. It can either be implicitly present in the semantic representation of the sentence (argument saturation) or deleted (argument reduction). Figure (1) illustrates this ambiguity of transitive reflexive sentences in German.

(1)

Syntax:

[NP_NOM V refl_ACC]

I.

Interpretation of the reflexive pronoun:

1. argument reflexive

reflexive

2. non-argument reflexive

valency reduction (cf. II)

II.

Interpretation of the implicit argument

1. argument reduction

anticausative & inherent reflexive

2. argument saturation

middle

In the previous chapter we argued that all transitive sentences with an accusative object that is a reflexive pronoun are syntactically identical. It turned out that certain differences between argument and non-argument reflexives concerning coordination, focus, and fronting are semantic rather than syntactic. In this chapter we investigate the first ambiguity of the accusative reflexive pronoun in transitive sentences. We turn to the second ambiguity in chapter 6. We argue that the morphosyntactic features of reflexive pronouns are maximally underspecified. Thus they can either head their own A-chain or they can be included in a complex A-chain, which is headed by another syntactic argument that c-commands the accusative reflexive pronoun (i.e. the subject). Besides, we argue that A-chains are subject to the following to linking-principles: (i) VP,Spec is linked to the first argument position of the verb and (ii) the complement position of V° is linked to the second argument position of the verb. According to (i) and (ii), the complex A-chain that includes the subject and the reflexive pronoun is linked to the second argument position via its base, the complement position of V°. Hence, the non-argument reflexive mediates the linking of the subject to the second argument position. As opposed to the non-argument reflexive, the argument reflexive heads its own chain and is thus linked to a semantic argument itself. We will see that this approach offers a uniform analysis of the reflexive, middle, anticausative and inherent reflexive interpretation.

Before we turn to reflexive pronouns in transitive reflexive sentences, we must introduce the basic notions of binding theory first. Of course the crucial element for the interpretation of transitive reflexive sentences is the reflexive pronoun in the position of the accusative object, which we call a morphosyntactic ‘middle marker’.

In the following presentation, we mostly use only the third person singular form.

In most Indo-European languages (weak) reflexive pronouns are ambiguous between the argument and the non-argument interpretation. The latter interpretation of reflexive pronouns (i.e. its function as an indicator of valency reduction) has been strongly neglected in the discussion of binding, which was mainly dealing with argument reflexives. Thus a modification of the existing theories of binding will be necessary. We limit the discussion to reflexive pronouns. Otherwise we would lose sight of our topic.

We are not interested in personal pronouns, possessive pronouns and R-expressions. In languages like German or English, personal possessive pronouns are ambiguous between a bound variable interpretation and pragmatic coreference (cf. e.g. Reinhart 1983 and 1991) whereas other languages have also reflexive possessive pronouns. In the following presentation, we ignore this issue as well as the status of Chomsky’s binding principle C. Additional issues of binding theory that are irrelevant for the problem under discussion are long distance anaphors, reciprocal pronouns, distributive vs. collective interpretations of plural antecedents, or different kinds of conceptualization of binding relations - e.g. the body or body parts, personality, pictures, reflections, images or statues.

A theory of binding must consider at least morphological, syntactic, semantic and discourse-theoretical aspects, which cannot be discussed in detail here. The following points will be relevant to the discussion of transitive reflexive sentences (cf. also the brief outline of the discussion at the end of the next section).

German does not always draw a morphological distinction between personal pronouns and reflexive pronouns
Unlike most Indo-European languages, German is a one-form language that does not distinguish weak from strong reflexive pronouns
structural binding of reflexive pronouns differs from logophoric binding
the distinction between structural and oblique case forms turns out to be relevant for binding theory

We will be concentrating on the interpretation of locally bound pronoun, irrespective of whether the paradigm has always two morphologically different forms for reflexive and personal pronous. Furthermore, German has only one kind of reflexive pronoun, as opposed to many other languages. German is a so-called one-form language (Kemmer 1993), that does not morphologically distinguish weak from strong reflexive pronouns. We will put the first two issues (i) and (ii) aside for the present, but we come back to the morphology of reflexive pronouns in the course of the this chapter. In the next subsection we turn to (iii) first and in the following sections we are dealing with point (iv).

Logophoric or exempt anaphors

Reflexive pronouns in German are subject to principle A of Chomsky’s (1981) binding theory.

In Chomsky (1981:188) binding theory is defined as follows:

(A) An anaphor is bound in its governing category

(B) A pronominal is free in its governing category

The governing category, i.e. the local domain for binding, is defined on the base of a governor and a subject available for the pronominal element. Chomsky (1986) replaces this term by the term complete functional complex (see Haegeman 1992 or Gärtner 1991 among others).

In the last years Chomsky’s theory of binding based on the notion of c-command has given rise to criticism. Reinhart and Reuland (1993) as well as Pollard and Sag (1994) - henceforth R&R and P&S - discuss various shortcomings of Chomsky’s version of binding theory that lead them to develop their own proposals. Both theories share essential features. Structural binding of reflexive pronouns is restricted to coreference of arguments of the same predicate. Therefore, reflexive pronouns are not always subject to the principles of binding theory. P&S (1994) and R&R (1993) give a sample of sentences that are grammatical although they violate principle A of Chomsky’s binding theory.

For further empirical and conceptual arguments against Chomsky’s theory of binding cf. P&S (1994) and R&R (1993).

In all examples in (2) the reflexive pronoun is not bound within its governing category (cf. also Ross 1970 and Jackendoff 1972):

(2)

a.

[John and Mary]₁ knew that [the journal had rejected [each other’s]₁ paper]

b.

John suggested that [tiny gilt-framed portraits of [each other]₁ would make ideal gifts for [the twins]₁]

c.

John₁ told Mary₂ that there were some pictures of themselves₁₊₂ inside

d.

[A picture of myself] would be nice on the wall

e.

[The picture of himself₁] that John₁ saw in the post office was ugly

In each case, the reflexive pronoun and its antecedent are not (co-)arguments of the same predicate. Furthermore the reflexive pronoun cannot be bound by a co-argument of the same predicate (the noun), because it is its sole argument. In addition, Chomsky’s principles of binding predict strict complementary distribution of personal pronouns and reflexive pronouns. But this contradicts the observation that in certain contexts the reflexive pronoun can be replaced by a personal pronoun as can be seen in (3.a) and (3.b) for German.

Grewendorf (1983) suggests that non-configurational aspects may play a role in these cases. Under his analysis two different governing categories are available in certain examples and hence each pronominal element is grammatical relevant to one of these two governing categories.

Similar examples can be found in Dutch (cf. 3.c) and English (cf. 3.d).

Examples like (3.b) have been taken as empirical evidence that the subject of so-called psych-verbs must base-generated somewhere below the object to account for the binding of the reflexive pronoun in line with Chomsky’s principle A (cf. Belletti and Rizzi 1988 or Stroik 1994). However, this kind of analysis predicts that the reflexive pronoun cannot be replaced by a personal pronoun. Additionally, it cannot be applied to sentence (2.b) because make is certainly not a psych-verb and the object ideal gifts for the twins is not an experiencer argument. Under R&R’s and P&S’s theory of binding these anaphors are exempt from the binding conditions in both cases.

(3)

a.

[Bilder von sich₁/ihm₁] machen Peter₁ glücklich

Pictures-nom of himself/him make Peter-acc happy

b.

Den Hans ängstigen [nur Geschichten über sich₁/ihn₁]

The H.-acc frighten only the stories-nom about himself/him

c.

Der Chef₁ läßt die Leute₂ für sich₁/ihn₁ arbeiten

(Grewendorf 1983)

The chief let the people for rp/him work

d.

Max₁ legt het boek achter zich₁/hem₁

(R&R 1995: 243)

Max puts the book behind himself/him

e.

Max₁ saw a ghost next to him₁/himself₁

The interpretation of reflexive pronouns in ellipses provide another argument for this distinction. A reflexive pronoun that is not bound by a co-argument is ambiguous between a strict and a sloppy interpretation, whereas a bound reflexive pronoun permits only the interpretation as bound variable, i.e. strict identity. This contrast is illustrated in (4.a) and (4.b) on the one hand and (4.c) on the other.

(4)

a.

Peter hat [das Bild von sich] in der Zeitung gesehen und Maria auch

Peter has the picture of rp in the newspaper seen and Maria too

b.

Den Kanzler ärgert [die neue Karikatur von sich] und den Oppositionsführer auch

The chancellor annoys the new caricature of rp and the opposition leader too

c.

Maria wäscht sich und Peter auch

Maria shaves rp and Peter too

R&R call reflexive pronouns that are exempt from the (structural) conditions on binding logophoric anaphors, P&S call them exempt anaphors.

Crosslinguistically, different types of reflexive pronouns are used as logophoric anaphors. Consider three closely related languages (cf. Reinhart and Reuland 1995). Dutch, for example, distinguishes weak from strong reflexive pronouns which both are independent words (zich vs. zichzelf). The weak form zich is used logophorically in Dutch. As opposed to Dutch, English uses the strong form him-/her-/itself as logophoric anaphor because the weak form of the reflexive pronoun is morphologically empty (cf. section 2.3). In Frisian the weak form of the reflexive pronoun is identical to the personal pronoun (in both cases him/har) - note that the third person feminine has an additional pronominal form se that can only be used as a personal pronoun, i.e. it cannot be locally bound as opposed to him/har. In Frisian we cannot morphologically distinguish logophoric anaphors from personal pronouns. The following table summarizes this (the logophoric anaphor is underlined):
(i)

Logophoric anaphors

weak form

strong form

Dutch

zich

zichzelf

Frisian

him/har (≈ pronoun)

him-/harsels

English

Ø

him-/her-/itself

The conditions for logophoric anaphors vary from language to language as well. On the one hand. it seems to be easier to use English reflexive pronouns logophorically than e.g. German reflexive pronouns (cf. (i) with (ii)). On the other hand, in German a reflexive pronoun is obligatory in contexts that allow also the personal pronoun in Dutch.
(i)

Max₁ said that the queen invited both Lucie and himself₁ for tea

(ii)

Max₁ sagte dass die Königin Lucie und *sich₁/ihn₁ zum Tee eingeladen hat
Max said that the queen Lucie and rp/him for tea invited has

The antecedents of logophoric or exempt anaphors are not determined by binding theory but by other, non-syntactic factors as, for example, processing and discourse constraints. Intervening constituents or point of view might influence the grammaticality of certain coindexations.

Note that a NP-internal subject (agent or possessor) like in (i)-(iii) blocks the coindexation between a NP-internal reflexive pronoun and a NP-external antecedent in picture-noun-phrases:

(i) */? Lucie₁ liked [your picture of herself₁]

(ii) */? [Your picture of myself] would be nice on the wall

(iii) */?^{[Deine Bilder von sich₁] gefallen dem Hans₁ (i.e. Your pictures of rp like the Hans-dat)}

One could argue that in (i)-(iii) the head noun picture or Bilder selects two arguments, which must be coindexed, because they are subject to R&R’s condition A or P&S’s principle A (cf. next section). In this case, local coindexation is obligatory and prevents the logophorical use of the reflexive pronoun. Alternatively, one could argue that the reflexive pronoun is still exempt from the conditions on binding, but the intervening NP-internal subject blocks the coindexation with an antecedent outside the DP similar to the examples in (5) below (cf. R&R 1993: 681f.).

We briefly illustrate this point in (5). In the first two sentences (5.a) and (5.b) the reflexive pronoun in the embedded clause is obligatorily coindexed with and bound by the closest subject. The second two examples (4.c) and (4.d) show that additional binding relations are also grammatical. When the first subject refers to an inanimate entity the reflexive pronoun can also be the matrix subject. In (4.c) and (4.d) the reflexive pronoun is clearly not locally bound.

(5)

a.

John₁ found [a picture of himself₁]

b.

Bill₂ remembered that Tom₁ saw [a picture of himself_1/*2] in the post office

c^.

^??Bill₁ remembered that the Times had printed [a picture of himself₁] in the Sunday edition

d.

Bill₁ thought that nothing could make [a picture of himself₁in the Times] acceptable to Sandy

So far, we separate logophoric from structural binding relations as stated in (iii) above. Logophoric or exempt anaphors provide arguments for a reformulation of binding theory. In the following we ignore logophoric or exempt anaphors and concentrate on structural binding, because only the structural restrictions on reflexive pronouns will be relevant for the anaylsis of argument and non-argument reflexives. Logophoric anaphors are always argument reflexives, whereas the accusative reflexive pronoun in transitive sentences (i.e. the middle marker in German) is always (structurally) bound by the subject of its sentence. We discuss R&R’s binding theory first because it is based on a distinction between syntactic and semantic arguments that turns out to be very fruitful for our explanation of the ambiguity of accusative reflexive pronouns. In the next section we give an outline of R&R’s theory, which will be necessarily incomplete. In section 5.3 we modify the syntactic part of R&R’s binding theory. Besides, R&R’s (1993) analysis is based mainly on Dutch, which is a two-form language. Therefore, a modification of the semantic part of their binding theory will also be necessary in order to apply it to German. This will be done in section 5.4 and 5.5. In this connection we also refer to P&S’s binding theory. P&S do not distinguish between syntactic and semantic arguments but they assume a hierarchy of grammatical relations, which also turns out to be fundamental for the interpretation of reflexive pronouns in German. In sum, the theory we develop in this chapter is based on the assumption that only co-arguments of the same predicate are subject to principle A of Chomsky’s binding theory. Reflexive pronouns that cannot be bound by a co-argument are exempt from (structural) binding. In addition, we postulate that accusative reflexive pronouns are also ‘bound’ in syntax, because arguments that check structural case are subject to the general condition on A-chains. This syntactic ‘binding’ relation, that results from chain formation, is responsible for the ambiguity of the accusative reflexive pronoun. The following discussion can be outlined as follows:

We introduce the binding theory of Reinhart and Reuland. Their distinction between syntactic and semantic binding and the General Condition on A-chains will be important in the discussion (section 5.2).
The syntactic part of R&R’s binding theory can be reduced to General Condition on A-chains, as is argued in Fox (1993). This condition will be relevant to the analysis of argument and non-argument reflexives in German. Besides, we argue that in German the definition of syntactic argument is based on structural case (also section 5.2).
R&R distingush between two types of reflexive pronouns. This distinction is irrelevant to the analysis of reflexive pronouns in German, because German is a one-form language (section 5.3).
As a consequence of (iii), the semantic part of R&R’s binding theory is replaced by principle A of Pollard and Sag’s binding theory (section 5.4).

Section 5.6 deals with the semantic ambiguity of reflexive pronouns in German and the final section 5.6 checks whether the binding theory developed for German in this chapter can also be applied to two-form languages as, for example, Dutch.

Syntactic and semantic predicates and A-chains

Recall that R&R’s binding theory is based on Dutch, which is a two-form language. As opposed to German, Dutch has two kinds of reflexive pronouns.

In the following presentation of R&R’s theory we use the term reflexive pronoun, although R&R use the term anaphor. According to their theory, both kinds of reflexive pronouns have less specified Φ-features than personal pronouns but only the strong reflexive pronoun is a reflexivizer, i.e. only the strong form is specified as [+REFL]. Thus the weak reflexive pronoun is the pronominal element that is less specified. It is neither specified with [+REFL] nor with [+R] (= referential independence), cf. table (6) below. Therefore, it does not make sense to call the weak reflexive pronoun a reflexive pronoun in R&R’s theory. As opposed to R&R, we do not distingush between [+REFL] and [–REFL]. We will keep on using the term reflexive pronoun in order to avoid confusion. Besides, the notion ‘reflexive pronoun’ is more specific than ‘anaphor’.

The third person form of the weak reflexive pronoun is the simple word zich and the strong counterpart of zich is the complex word zichzelf. According to R&R’s theory, two properties are relevant for the specification of reflexive pronouns in Dutch. The feature [+/–R] (for referential independence) distinguishes reflexive pronouns from R-expressions and personal pronouns.

In R&R’s framework, [R] is a purely morphosyntactic feature: ‘Having this property is a necessary condition for an expression to function as an independent argument, but R itself does not have anything to do with reference’ (R&R 1993: 697). The property [R] depends on the internal specification of an NP:

(i) An NP is +R iff it carries a full specification for Φ-features and structural Case

This issue will be discussed in greater detail below.

Both weak and strong reflexive pronouns are referentially deficient and therefore specified as [–R]. A second feature, [+/–REFL], distinguishes strong from weak reflexive pronouns. [+REFL] stands for a reflexivizing function that operates on predicates and their arguments. Roughly speaking the feature [+REFL] requires two arguments of the predicate to be coindexed. Only strong reflexive pronouns are specified for [+REFL], whereas weak reflexive pronouns and personal pronouns are specified as [–REFL]. R&R call the weak reflexive pronoun zich ‘SE anaphor’ and its strong counterpart zichzelf‘SELF anaphor’.

(6)

reflexive and personal pronouns in Dutch

SE anaphors are the less specified elements of the pronominal paradigm, because they are neither reflexivizers nor referentially independent elements. The distribution of pronominal elements in Dutch is illustrated in (7).

(7)

a.

Max₁ gedraagt zich₁/ *zichzelf₁/ *hem₁

(inherent reflexive)

Max behaves SE/SELF/him

b.

Max₁ hoorde zichzelf₁/ *zich₁/ *hem₁

(‘strong’ reflexive)

Max heard SELF/SE/him

c.

Max₁ wast zich₁/zichzelf₁/*hem₁

(‘weak’ reflexive)

Max washes SE/SELF/him

d.

Henk₁ hoorde [zich₁/zichzelf₁/*hem₁ zingen]

(ECM)

Henk heard SE/SELF/him sing

The first example (7.a) contains the inherent reflexive verb gedragen (‘behave’), the object of which must be the weak form of the reflexive pronoun, i.e. the SE anaphor.

The same holds for reflexive anticausatives in Dutch. Dutch, like German, has inherent reflexive verbs and also some reflexive anticausatives, whereas middle constructions are formed without a reflexive pronoun like middle constructions in English, cf. section 2.3.

In contrast, the SE anaphor cannot be the object of verbs like horen (‘hear’) in (7.b). In this case the strong form, i.e. the SELF anaphor, must be used. Verbs like wassen (‘wash’) permit both the weak and the strong form of the reflexive pronoun, cf. (7.c). Thus SE anaphors are only grammatical with verbs that are inherent reflexive like gedragen in (7.a.) or verbs like wassen in (7.c) describing events that are very likely to be reflexive. The crucial difference between the English and Dutch pronominal system shows up in ECM-constructions like (7.d). As opposed to (7.b), SE anaphors are not excluded in this construction, even if the matrix verb is horen. This difference between (7.b) and (7.d) can be attributed to the status of the direct object. In (7.d), on the one hand, the reflexive pronoun is only a syntactic co-argument of its antecedent. Both the antecedent and the anaphor are assigned case by the same predicate, the matrix verb horen. Semantically, the anaphor is an argument of the embedded verb zingen (‘sing’). The reflexive pronoun is linked to its first semantic argument. On the other hand, in (7.b) the anaphor is both a syntactic and semantic argument of the same verb. Note finally that personal pronouns are ruled out in all four sentences. The following table summarizes the observations made so far.

This table is taken from the discussion of R&R’s theory in Vogel and Steinbach (1995).

(8)

The distribution of pronominal forms with the verb horen (‘hear’)

R&R formulate a theory of binding that accounts for the different distribution of weak and strong reflexive pronouns in Dutch. Their binding theory is based on the following two conditions. The relevant definitions of the notions reflexive, reflexive marked, syntactic and semantic predicate and syntactic argument are given in (10)- cf. R&R (1993: 678):

(9) and (10) are abbreviated versions of R&R’s binding theory. R&R relativize their binding conditions to an index i. This is necessary to exclude the licensing of a coindexation i by an SELF anaphor bearing the index j. The following example illustrates this:

(i) *Max₁ showed myself₂ to him₁

The predicate is reflexive marked by the anaphor myself and it is reflexive because two of its arguments are coindexed. Nevertheless, sentence (i) is ungrammatical. This restriction of the conditions to a certain index correctly excludes sentences like (i), because the index of the anaphor differs from the index of the coindexed arguments. In (9) and (10) the terms reflexivity and coindexation must be replaced by i-reflexivity and i-coindexation (cf. also Fox 1993). We continue referring to the simplified versions in (9) and (10) because the restriction to an index is not relevant for the problem under discussion.

(9)

Conditions

A:

A reflexive-marked syntactic predicate is reflexive.

B:

A reflexive semantic predicate is reflexive-marked.

(10)

Definitions

a.

The syntactic predicate formed of (a head) P is P, all its syntactic arguments, and an external argument of P (subject).

b.

The syntactic arguments of P are the projections assigned a Θ-role or case by P.

c.

The semantic predicate formed of P is P and all its arguments at the relevant semantic level.

d.

A predicate is reflexive iff two of its arguments are coindexed.

e.

A predicate is reflexive-marked iff either P is lexically reflexive or one of P’s ar-guments is a SELF anaphor.

Condition A and B correctly explain the distribution of the SE and SELF anaphors in Dutch. Condition B says that, whenever a semantic predicate is reflexive, it must be reflexive-marked. A predicate is reflexive-marked either if it is lexically reflexive or if one of its arguments is a SELF anaphor. Gedragen in (7.a) is a inherent reflexive verb that is lexically reflexive,

We do not think that inherent reflexive verbs are two-place predicates with both arguments obligatorily coindexed. Unlike two-place verbs like wassen, that are optionally reflexive, inherent reflexive verbs equal anticausatives. Both inherent reflexives and anticausatives do not select an external argument (cf. section 2.2). R&R’s theory cannot account for the fact that weak reflexive pronouns (i.e. [–REFL] anaphors) indicate valency reduction. We come back to this issue in section 5.6 and in chapter 6.

hence no SELF anaphor is required.

The SELF anaphor is ungrammatical for matters of economy. The SELF anaphor would reflexive-mark the predicate once again but binding condition B requires only one reflexive marking, cf. also below.

Horen in (7.b) is not lexically reflexive. Thus the predicate horen can only be reflexive-marked by a SELF anaphor. A verb like wassen in (7.c) shares features with both gedragen and horen. It permits SE and SELF anaphors (i.e. weak and strong reflexive pronouns). According to R&R, verbs like wassen can but need not be lexically reflexive.

R&R assume two different lexical entries for verbs of grooming like wassen. We already saw in section 2.3 that identical cases can be observed in many Indo-European languages. We come back to this issue in section 5.4 below.

Therefore, weak and strong forms are grammatical. And finally,both forms are also licensed in the position of the ECM-subject, cf (7.d), because it is only a syntactic argument of the matrix predicate to which condition A applies: the syntactic (matrix) predicate is reflexive-marked by the SELF anaphor in object position and it is reflexive because two of its arguments are coindexed. Condition B is also fulfilled: the embedded (semantic) predicate is not reflexive, hence no reflexive-marking is required and both the SE and the SELF anaphor are grammatical. Only if the embedded predicate is reflexive, a SELF anaphor is obligatorily required, because the reflexive (semantic) predicate must be reflexive-marked (but it does not matter which of the two arguments is the SELF anaphor).

(11)

a.

Jan hoorde [zich zichzelf/*zich critiseren]

b.

Jan hoorde [zichzelf/*zich zich critiseren]

Jan heard himself himself criticize

So far R&R correctly derive the distribution of SE and SELF anaphors. Recall that personal pronoun are also excluded in all examples in (7), although SE anaphors, which are also [–REFL], are grammatical in (7.a), (7.c) and (7.d). Personal pronouns are ruled out by an additional condition, that unifies the treatment of binding and A-movement. This condition will be central to the analysis of reflexive pronouns in German we propose below.

R&R give the following definition of chain, following Chomsky (1986a, b):

(i) Generalized Chain definition

C (α₁, …, α_n) is a chain iff C is the maximal sequence such that

a. there is an index j such that for all j , 1≤ j ≤ n, α_jcarries that index, and

b. for all j, 1≤ j ≤ n, α_jgoverns α_j+1.

For the term government see example (33) below.

(12)

General Condition on A-chains (GCC)

A maximal A-chain (α₁, ..., α_n) contains exactly one link - α₁ - that is both +R and case-marked.

The crucial difference between SE anaphors and personal pronouns is the feature [R]. Only personal pronouns (and R-expressions) are referentially independent, i.e. [+R], cf. table (6) above. As a consequence of (12), personal pronouns can only be the head of an A-chain. Hence they are excluded in all positions α_j with j > 1. The GCC correctly predicts that personal pronouns are ungrammatical in all examples in (7).

Fox (1993) goes even further than R&R and argues that condition A of R&R’s binding theory can be completely reduced to the GCC in (12). As opposed to R&R,

In R&R’s version only two-member chains are relevant: “Either we define [the general condition on A-chains, i.e. (13)] as a condition on chains with more than one link, or we define an A-chain as consisting of at least two coindexed links” (R&R 1993: 702).

Fox assumes singleton chains.

The predicate coindexed is reflexive: every element X bearing an index i is coindexed with itself (cf. also Chomsky 1995, chapter 3 on one-member chains).

He proposes that any sequence of coindexation that is headed by an A-element forms a maximal A-chain. As a consequence, every syntactic argument forms an A-chain of its own.

This condition is trivially met if every syntactic argument is forced to move to some Agr-position to check its case feature. As a consequence of obligatory A-movement, every a element heads its own chain, which is of course only possible for [+R]-expressions.

Moreover, every [–R] A-element must always be contained in a A-chain that is headed by a [+R] A-element. This modification will also be relevant for the derivation of the ambiguity of transitive reflexive sentences, as we argue below. Besides, it simplifies the binding theory as well as the analysis of ECM-constructions and SE anaphors. Recall that SE anaphors are only grammatical in ECM-constructions like (13.a) if the embedded (semantic) predicate is not reflexive (cf. condition B). Condition A is met, because the SE anaphor does not reflexive-mark the matrix predicate. Unlike SELF anaphors, SE anaphors are [–REFL]. Condition A does not applied to SE anaphors in ECM constructions and the syntactic arguments of the matrix predicate need not be coindexed. Thus we expect SE anaphors to behave more like pronouns or like logophoric anaphors, which is obviously not true.

Note that indices are assigned to all DPs in syntax and must not change at the interface to semantics. Otherwise, R&R (as well as Fox) cannot exclude personal pronouns on the basis of the general condition on A-chains either. A predicate is not reflexive-marked by a personal pronoun, hence no coindexation of the syntactic arguments is required by condition A (cf. Chomsky 1980, Fiengo and May 1994 for assignment of indices).

Chomsky (1995: 215, footnote 53) claims that “a theoretical apparatus that takes indices seriously as entities, allowing them to figure in operations (percolation, matching, etc.), is questionable on more general grounds. Indices are basically the expression of a relationship, not entities in their own right. They should be replaceable without loss by a structural account of the relation they annotate.”

(13)

a.

Henk₁ hoorde [zich₁/zichzelf₁/*hem₁ zingen]

Henk heard SE/SELF/him sing

b.

Henk₁ hoorde [zichzelf₁ Max critiseren]

Henk heard SELF Max criticize

A theory that reduces condition A to the modified version of the GCC predicts that the SE anaphor zich in (13) must be bound by the subject, because it is [–R] and cannot head its own A-chain. Therefore, it is forced to form a chain with a [+R] A-element c-commanding it, i.e. the subject of the matrix clause.

Besides, this modification of R&R’s binding theory renders movement of the embedded verb in ECM superfluous. On the one hand, R&R assume that the embedded verb must not stay in situ in (13.b) (cf. also footnote 17 above). Otherwise it would be reflexive marked without being reflexive, because the ECM-subject is a syntactic argument not only of the matrix predicate but also of the embedded predicate, which assigns a theta-role to the SELF anaphor, cf. (10.b) above. R&R claim that raising of the verb prevents it from being reflexive marked (R&R 1993:708f.). According to R&R, LF-raising and LF-lowering is always optionally permitted. They propose that in Dutch the embedded verb raises at S-structure and adjoins to the matrix predicate. An examples like (i) has the S-structure representation in (ii).

(i) ... dat Max zichzelf Lucie hoorde critiseren

(ii) ... dat [_IP Max [_IP zichzelf Lucie t_i] [_V hoorde_j critiseren_i]_j ] (i.e. ... that Max SELF Lucie heard criticize)

On the other hand the embedded predicate must stay in situ in (iii) - alternatively it could be lowered at LF -, because it is reflexive marked by its second argument.
(iii)

Peter hoorde [Maria₁ zichzelf₁ critiseren](reflexive-marked by zichzelf)
Peter heard Maria SELF criticize

The verb stays in situ, if two syntactic arguments must be coindexed, and it is forced to raise, if the syntactic arguments must not be coindexed. raising and lowering of the embedded verb in ECM-constructions becomes superfluous in the modified version of R&R’s theory, because it no longer relies on the term syntactic predicate. This revised version only requires grammatical A-chains and does not depend on additional verb-movement at LF.

Besides, after raising the direct object of the embedded clause, i.e. Lucie, is still a syntactic argument of the head critiseren, because it receives accusative case via the trace t _i of the verb (cf. R&R 1993: 708, footnote 49). Hence, the head P (i.e. critiseren) is available at S-structure for case-marking its arguments (cf. definition 10.b) and it is not clear why P is not available for predicate formation as well (cf. definition 10.a).

Finally we follow Vogel and Steinbach (1995 and 1998), who argue that in German the term syntactic argument should be defined on the basis of structural case. In German, only NPs that are marked with structural case (or have a structural case-feature to check) are syntactic arguments. In chapter 6 we give empirical evidence in support of a distinction between structural and oblique case. We will see that nominative and accusative are structural. As opposed to nominative and accusative, dative is oblique. Hence, A-chain formation is restricted to subjects and accusative objects. The relevant modification of R&R’s definition (10.b) is given in (14).

(14)

Definition of syntactic arguments in German

...

b.

The syntactic arguments of P are the chains that are assigned structural case in the extended projection of P.

...

So far we argued that R&R’s binding condition A should be reduced to a slightly modified version of the GCC. In addition, their definition of syntactic arguments was modified for German. In German only NPs that are asigned structural case are subject to ‘syntactic binding’, i.e. A-chain formation. In the following two sections we turn to the distinction between [+REFL] and [–REFL] expressions and R&R’s binding condition B that is responsible for ‘semantic binding’.

Weak and strong reflexive pronouns and the [+/–R]-distinction

As we mentioned in chapter 2, German is a so-called one-form language. Unlike Dutch and many other Indo-European languages, German does not distinguish weak from strong reflexive pronouns, or in R&R’s terminology SE from SELF anaphors. The sole reflexive pronoun in German morphologically correspond to the SE anaphor in Dutch (cf. the diagram (8) above and section 2.1.3).

We argued in chapter 4 that sich selbst in German is formed by adjunction of the adnominal focus particle selbst to the NP sich. In German, the expression sich selbst is not grammaticalized as opposed to Dutch zichzelf or English him-/her-/itself. Note that it can always be replaced by the stressed simplex anaphor sich, although some speakers seem to prefer sich selbst in some contexts. The Dutch simplex anaphor zich cannot be stressed at all. It is the intrinsically weak counterpart to the strong anaphor zichzelf.

Nevertheless, one might argue that the German pronominal system also draws a distinction between weak and strong reflexive pronouns. Under this assumption, the weak reflexive pronoun zich in Dutch would correspond to unstressed sich in German and the strong form zichzelf in Dutch would correspond to stressed (and thereby focussed) sich. However, the distribution of the weak and strong forms in Dutch does not match with the distribution of unstressed and stressed sich in German. While stressed sich always correspond to zichzelf in Dutch, unstressed sich differs from zich. Recall from section 4 that the reflexive pronoun in German cannot be focussed, fronted, or coordinated when it is interpreted as a non-argument reflexive, i.e. in middle constructions, anticausatives and inherent reflexives. In this respect the unstressed non-argument reflexive sich corresponds to the weak form in Dutch (zich), that cannot be stressed (i.e. focussed) and fronted either and that must also be used with anticausatives and inherent reflexives, that require a non-argument reflexive. The strong form zichzelf and sich are ungrammatical in these contexts in both languages. In addition, (7.b) above shows that zich must not be the object of verbs like horen (‘hear’), that are not lexically reflexive. This generalization seems to hold for all languages that distinguish weak from strong reflexive forms (cf. Kemmer 1993 and section 2.3). In contrast, German unstressed sich is grammatical with verbs like hören, the German counterpart of horen, as is illustrated in (15). The examples in (16) illustrate the same point: verbs that are unlikely to be (lexically) reflexive take unstressed sich as direct object. In all examples we put the focus on some other constituent to exclude a ‘strong’ interpretation of sich.

(15)

Hat sich der Peter nur gehört oder auch gesehen? Der Peter hat sich nur [_F GEHÖRT]

Has rp the Peter only heard or also seen? The Peter has rp only heard

(16)

a.

... weil sogar [_F HANS] sich nun nicht mehr belügt

because even Hans rp now no more tells a lie

b.

... weil sie sich nur [_F GEStern] im Fernsehen gesehen hat

because she rp only yesterday in-the TV seen has

In sum, although the distribution of the weak and strong forms in Dutch largely overlaps with the distribution of unstressed vs. stressed reflexive pronouns in German, the pronominal system of both languages are not completely identical. As opposed to two form languages like Dutch (and English) the reflexive pronoun in German is not sensitive to whether a predicate is likely to be reflexive or not. This difference is illustrated in the following picture for Dutch and German.

(17)

Dutch and German reflexive pronouns

Language/Example

gedragen/benehmen

wassen/waschen

horen/hören

Dutch:

zich

zich & zichzelf

zichzelf

German:

sich (- focus)

sich (± focus)

sich (± focus)

We conclude that German does not distinguish between weak and strong reflexive pronouns or SE and SELF anaphors.

R&R (1993 and 1995) argue that an intrinsic/nonintrinsic contrast shows up at least on the dative argument of three-place predicates in German. They give the following examples to illustrate this.
(i)

a.

Peter stellte sich/^??sich selbst die Statue vor

([+R] verb)

Peter imagined rp-dat a statue PARTICLE

b.

^??Peter vertraute sich seine Tochter an
Peter entrusted rp-dat his daughter PARTICLE

([-R] verb)

c.

Peter vertraute seine Tochter nur sich selbst an

Peter entrusted his daughter only RP himself

(i.e. .. place her only in his hands)

However, example (i.b) seems to be odd for pragmatic reasons. If we replace vertrauen by another [–R] verb, sentence (i.b) gets perfectly grammatical (cf. (ii.a)). Moreover, two-place predicates that require a dative object allow a simple reflexive as well, as can be seen in (iii). There is no evidence that German distingushes between [+/– REFL]-elements.
(ii)

a.

Peter hat sich wieder die besten Karten (selbst) gegeben
Peter has rp-dat again the best cards (himself) given

b.

Maria hat sich den Rest gegeben

Maria has rp-dat rest given

(i.e. Maria has finished herself off)

c.

Udo gestattet sich wieder einmal gar nichts
Udo allows rp-dat once again absolutely nothing

(iii)

Helmut hat immer nur sich geholfen
Helmut has always only rp-dat helped

One morphological form corresponds to both weak and the strong forms in Dutch.

Additionally, middle constructions in German are also reflexive, whereas Dutch middle constructions are formed without a (weak) reflexive pronoun.

The distribution of stressed and unstressed reflexive pronouns in German depends solely on the interpretation of the reflexive pronoun itself (i.e. whether it is an argument or a non-argument reflexive). Besides, a subdivision of reflexive pronouns into [+REFL] and [–REFL] would lead to the situation that nearly every verb in most languages must be listed in the lexicon twice, which is an unnecessary enlargement of the lexicon.

Note that the non-argument reflexive must be the weak [–REFL] form. Hence, every verb that can undergo middle formation must also be lexically specified for [–REFL] because its direct object is a SE anaphor that is specified as [–REFL]. In most Indo-European languages ‘middle verbs’ are lexically reflexive, cf. section 5.6 for more details.
(i)

Peter liest das Buch
Peter reads the book

(ii)

Das Buch liest sich gut
The book reads SE-[-REFL] well

Therefore, we can ignore the distinction between weak and strong or SE and SELF anaphors in the following discussion. The feature [REFL] is needless in German. We come back to two-form languages in section 5.6.

But how can we account for the difference between argument and non-argument reflexives in German. We think that another feature, that is also discussed in R&R, is relevant to the ambiguity of reflexive pronouns in German (and any other Indo-European languages). According to R&R, the feature [R] is a morphosyntactic feature, that depends on the inherent specification of a lexical item, cf. footnote 10 above. All nominal expressions are intrinsically specified for Φ-features. Moreoever, only elements that are not least specified for these features (i.e. not maximally underspecified) are [+R]. (18) and (19) illustrate the specification of the Φ-features number, person, gender, and case of personal and reflexive pronouns (homonymous forms within one paradigm are written in italics and homonymous forms in both paradigms are underlined).

(18)

Personal pronouns in German

(19)

Reflexive pronouns in German

Reflexive pronouns in Modern German differ from personal pronouns at least in two dimensions: they are not specified for gender and they lack nominative case.

In German the lack of nominative reflexive pronouns can be derived from the obliqueness hierarchy we propose below. Nominative NPs are always the least oblique elements. Therefore they cannot be bound by any other co-argument of the same predicate. Things might different for a language like Islandic. Everaert (1990) argues that in Islandic a nominative reflexive pronoun in VP-internal position is expected to be bound by a dative subject. The relevant example is given in (i). The dative personal pronoun honum locally binds the nominative NP. Instead of a potential (but nonexisting) reflexive pronoun REFL, the personal pronoun hann must be used.
(i)

Honum₁ finnst *REFL₁/hann (sjalfur)(vera) skrytinn
Him-dat finds refl/he-nom (self) (be) strange

(i)Anderson (1986:69) shows that possessive pronouns have reflexive forms for both dative and nominative. they are licensed in the subject position of embedded sentences. Nominative reflexive pronouns are excluded in these cases although they seem to be structurally licensed as well. One might argue that languages generally lack nominative forms for reflexive pronouns (cf. also Pollard and Sag 1993 and R&R 1995 on this issue).

Furthermore, a morphological distinction between accusative and dative case can be found only in first and second person singular whereas personal pronouns also distinguish third person accusative forms from third person dative forms. The personal pronoun ihr, for example, is specified as < [+ sing, - plur], [- 1.p, - 2.p, + 3.p], [- masc, + fem, - neuter], [- nom, - acc, + dat] >. The corresponding third person singular reflexive pronoun sich is not specified for number, gender, and a specific object case: < [Ø], [- 1.p, - 2.p, + 3.p], [Ø], [+ acc, + dat] >. A discussion of the correct morphological specification of individual forms within this paradigm would lead us beyond our topic.

There are at least three issues that need further discussion.

(i) homonymous forms within one paradigm (e.g. the personal pronoun sie).

(ii) homonymous forms in two paradigms (e.g. the 2. person reflexive and personal pronouns dich and dir), cf. below.

(iii) two-form languages that have two kinds of reflexive pronouns, cf. section 5.6.

The two tables in (18) and (19) are only meant to illustrate that reflexive pronouns are the pronominal items that are least specified. Therefore, they are not lexically specified for the feature [R]. The specification of this feature depends on the lexical properties of an element and the pronominal paradigm of a language. Pronominal elements with similar feature specification are compared to each other and only the element, that is maximally underspecified, is not inherently specified for [R]. We will argue in section 5.5 that the feature [R] is relevant for A-chain formation (i.e. ‘syntactic binding’) and thereby for the interpretation of non-argument reflexives. Consider, for example, the subject of impersonal middle constructions and of weather verbs, the personal pronoun es (‘it’), carries a full specification for Φ-features. It is specified as < [+ sing, - plur], [- 1.p, - 2.p, + 3.p], [- masc, - fem, + neuter], [+ nom, + acc, - dat] > and hence [+R]. Therefore, es can be the head of an A-chain and bind another pronominal element that is not specified for [R] (we come back to impersonal subjects in section 5.6).

(20)

a.

... weil es₁ sich₁ einregnet

... because it rppart-rains (i.e. The rain is settling in)

b.

Es₁ schläft sich₁ gut in diesem Bett

It sleeps rp well in this bett

As can be seen in (18) and (19) Modern German, like many other languages, does not morphologically distinguish reflexive from personal pronouns in the first and second person. Although Modern German does not always have two morphologically distinct forms, it discriminates between reflexive pronouns and personal pronouns in principle. Besides, we saw in section 2.3 that Middle High German has no dative reflexive pronouns, using personal pronouns for local binding relations, too. In addition, some languages distinguish reflexive possessive pronouns from personal possessive pronouns whereas others have only personal possessive pronouns that are used in every context - i.e. locally bound (bound variable) and locally free (pragmatic coreference).

Cf. Reinhart (1991) on possessive pronouns that are interpreted either as bound variables or via pragmatic coreference.

Burzio (1989) connects this observation to his principle of morphological economy: a NP that is locally bound must be maximally underspecified. Reflexive pronouns are less specified than personal pronouns (and personal pronouns are less specified than R-expressions). Whenever a language discriminates between reflexive and personal pronouns they must be used in different (syntactic) contexts. On the other hand, if there is only one pronominal form, it can be used in every context a pronominal form can occur in, cf. e.g. the first and second person object pronouns in Modern German. The logic of this argument can be illustrated as follows.

(21)

X is locally bound

→

X is maximally underspecified (i.e. there is no Y that is less specified than X than X)

(21) does not imply that every maximally underspecified element must be locally bound. In Modern German the first and second person pronominal forms are maximally underspecified (there is no comparable pronominal form that is even less specified), but they need not be locally bound (cf. also dative pronouns in Middle High German, and logophoric or exempt anaphors). The following definition accounts for this.

(22)

a.

A pronominal element X is not lexically specified for [R] iff

(i)

it is equally specified for the features < F₁, ..., F_n > like another pronominal element Y and

(ii)

only Y is additionally specified for a feature < F_n+1 >.

b.

A pronominal element Y is lexically specified as [+R] iff

(i)

it is equally specified for the features < F₁, ..., F_n > like another pronominal element X and

(ii)

only Y is additionally specified for a feature < F_n+1 >.

Hence, a maximally underspecified pronominal element is not lexically specified for [R]. A pronominal element that is not maximally underspecified is [+R]. If we compare, for example, the third person singular personal pronoun ihr (‘her’) in (23.a) with a third person singular reflexive pronoun sich in (23.b) it becomes clear that the reflexive pronoun in (23.b) is less specified than the personal pronoun in (23.a). Thus only the personal pronoun is [+R].

(23)

a.

ihr

< [+ s, - p], [- 1.p, - 2.p, + 3.p], [- m, + f, - n.], [- nom, - acc, + dat] >

b.

sich

< [Ø], [- 1.p, - 2.p, + 3.p], [Ø], [+ acc, + dat] >

We conclude that the semantic ambiguity of (weak) reflexive pronouns – or more general of the elements that are least specified for Φ-features – results from the intrinsic lexical specification of these elements and the pronominal systems of a language. Expressions that are not specified for [R] can be either [+R] or [–R]. This ambiguity seems to be a universal property of (weak) reflexive pronouns, which can also be observed in German. All reflexive pronouns in German are inherently underspecified for the feature [R]. As a consequence, they are ambiguous and can be used as [+R] and as [–R]-expressions. The GCC discussed in the previous section requires that reflexive pronouns that are [+R] must head a chain of their own, which is linked to a semantic argument. Thus [+R] reflexive pronouns yield the argument reflexive interpretation. In contrast, [–R] reflexive pronouns cannot head their own chain. Instead they must be included in another A-chain, which is headed by a [+R]-expression. The [–R] reflexive pronoun yields the non-argument reflexive interpretation. The derivation of this ambiguity will be discussed in section 5.5 in greater detail. Recall that our approach differs from R&R’s binding theory on two points. First, (weak) reflexive pronouns (or SE anaphors in R&R’s terms) can but need not be [–R] expressions. Second, German does not discriminate between [+REFL] and [–REFL] expressions. Because of this second difference, R&R’s condition B cannot applied to German, cf. (9) above. The next section deals with the final modification of the semantic part of R&R’s binding theory.

O-command and o-binding in German

This final modification follows P&S (1994), who argue that the definition of binding should be based on the relative obliqueness of arguments. In the previous section we saw that in German the feature [REFL] is needless. Therefore, we cannot apply R&R’s binding condition B, repeated here as (24), to German.

(24) B: A reflexive semantic predicate is reflexive-marked

We replace binding condition B by P&S’s condition on reflexive pronouns, which will be needed anyway to account for certain binding asymmetries in German. Let us first introduce P&S’s binding principle A and B:

Note that P&S use the term anaphor instead of reflexive pronouns We can replace P&S’s original principle A in (21) by (21’):

(21’) Principle A: A locally o-commanded reflexive pronoun must be locally o-bound

(25)

Principle A:

A locally o-commanded anaphor must be locally o-bound

Principle B:

A personal pronoun must be locally o-free

Local o-command and local o-binding are defined in the following way.

(26)

Definitions

a.

Let Y and Z be synsem objects with distinct LOCAL values, Y referential. Then Y locally o-commands Z just in case Y is less oblique than Z.

b.

Y locally o-binds Z just in case Y and Z are coindexed and Y locally o-commands Z. If Z is not locally o-bound, then it is said to be locally o-free.

In P&S the relative obliqueness of grammatical functions is reflected by the order of the syntactic arguments on the SUBCAT list: a synsem object Y is less oblique than a synsem object Z iff Y precedes Z on the SUBCAT list of the same lexical head. Consider, for example, the following SUBCAT-list in (27). The first synsem value can bind the second and the third one, because NP [NOM] is less oblique than NP [ACC] and NP [DAT], and the second synsem value can bind the third one, but not vice versa. Hence, a dative NP (or indirect object) can be bound by an accusative NP (or direct object) and a nominative NP (or subject) whereas an accusative NP can be bound only by a nominative NP. These asymmetries will be illustrated immediately.

(27) [SUBCAT < NP [NOM], NP [ACC], NP [DAT] >]

As opposed to P&S, we do not think that the obliqueness of arguments has to be fixed in each lexical entry separately. Instead we assume the general obliqueness hierarchy for German.

This might be connected to processing asymmetries between different kinds of syntactic constituents. We will see in section 6.1.6 that the parser prefers structural case to oblique case and nominative to accusative.

This asymmetry between nominative, accusative, and dative is manifested in various phenomena, which will be discussed in the next chapter.

(28)

Obliqueness hierarchy

nominative (subject) > accusative object > dative object > other oblique objects

Besides, P&S lexical definition of obliqueness is incompatible with our analysis of non-argument reflexives. Their theory requires a lexical derivation of non-argument reflexives, because the relative obliqueness is defined on the basis of subcat lists. However, we argued at length in chapter 3 that a lexical approach fails to account for non-argument reflexives. Especially impersonal middle constructions as (29), can be hardly derived in the lexicon. Recall that middle constructions in German always select two syntactic arguments (a nominative subject and an accusative object) even if the basic verb subcategorizes only for one syntactic and semantic argument.

(29)

a.

In diesem Bett schläft es sich gut

In this bed sleeps it rp well

b.

schlafen <1>

A similar issue might be free dative objects in German. They are unlikely to be subcategorized for by the verb. They can be inserted rather freely and they are very often ambiguous between various interpretations as, for example, possessor or beneficiary. Besides, German has multiple occurences of dative objects, cf. chapter 6.

(30)

a.

Ich backe meiner Mutter einen Kuchen

I bake my mother-dat a cake-acc

b.

Du legst deinem Vater das Buch auf den Tisch

You put your father-dat the book-acc on the table

Free datives show the same syntactic behaviour and the same binding properties as ‘subcategorized’ dative objects. They can be asymmetrically bound by the nominative subject or accusative object of the same sentence, cf. (31.a) and (31.b) and they asymmetrically bind more oblique objects, cf. (31.c) and (31.d). Again, a lexical binding theory based on the selectional properties of predicates fails to account for free dative objects.

A lexical theory could introduce a lexical rule of ‘free dative insertion’ that adds a free dative object to the SUBCAT-list of a basic verb. However, a lexical rule of dative insertion does not provide an argument against the general obliqueness hierarchy we proposed for German, because all datives (no matter whether they are subcategorized or free, i.e. introduced by a lexical rule) share the same syntactic properties.

Note that free dative objects also pose a problem for configurational binding theories that define the binding conditions on the basis of c-command and VP-internal A-positions. They are forced to base-generate free datives along with subcategorized dative objects in a position c-commanded by nominative subjects and accusative objects. However, this base position is motivated mainly by the selectional properties of a verb (the verb assigns a theta-role to this position). Below we give further arguments that dative objects cannot be analyzed parallel to accusative objects, cf. Müller (1993) and Vogel and Steinbach (1995 and 1998) and chapter 6.

(31)

a.

Hans erzählte dass Peter₁-nom sich₁-dat einen Kaffee machte

a'.

*Hans erzählte dass dem Peter₁-dat sich₁-nom einen Kaffee machte

Hans told that the Peter rp a cup of coffee-acc made

b.

Maria setzte die Kinder₁-acc einander₁-dat auf den Schoß

b'.

*Maria setzte den Kindern₁-dat einander₁-acc auf den Schoß

Maria sat the children each other on the lap

c.

Ich öffne ihr₁-dat [über sich₁] die Augen

c'.

*Ich öffne [über sie₁] sich₁-dat die Augen

I open on her herself the eyes

‘I open her eyes to the truth’

d.

Ich habe ihr₁-dat gestern die Wohnzimmerwand [mit sich₁ selbst] bemalt

d'.

*Ich habe [mit ihr₁] sich₁-dat gestern die Wohnzimmerwand bemalt

I have with her herself yesterday the living room-wall painted

‘Yesterday, I painted her (picture) on the wall of the living room for herself’

To summarize, the binding conditions for German should not be defined relative to lexical synsem objects but to semantic arguments of a predicates. In addition, the obliqueness hierarchy as stated in (28) does not depend on individual lexical entries but is a general property of German. Before we formulate the final version of our binding theory we briefly discuss the relevant German examples, that illustrate the asymmetry in binding relations in German. Consider first simple two-place predicates. The examples show that a nominative subject can bind less oblique objects as, for example, the accusative object in (32.a) or the dative objects in (32.b). Both examples include an argument reflexive. Besides, accusative reflexive pronouns need not be linked to a semantic argument of the verb, cf. the anticausative in (32.c). In addition, the accusative object can also be bound in resultative constructions by the subject, cf. (32.d). In resultatives, the accusative object is linked to a semantic argument of the complex resultative predicate.

(32)

a.

Hannelore₁ wäscht sich₁ nie

Hannelore-nom washes rp-acc never

b.

Helmut₁ widerspricht sich₁

Helmut-nom contradicts rp-dat

c.

Das Fenster öffnet sich

The window opens rp-acc

d.

Peter trinkt sich unter den Tisch

Peter drinks rp-acc under the table

Besides, the nominative subject can also bind a dative object in ditransitive clauses like (33). In this case binding across a less oblique accusative object is possible because the relation ‘x is less oblique than y’ is transitive.

(33)

Der Manager₁ überwies sich₁ das ganze Geld auf sein Konto

The manager transferred rp-dat the whole money to his account

It is a well known fact that in German accusative objects can bind dative objects but not vice versa as can be seen in (34), cf. Grewendorf (1988), Müller (1993) and Vogel and Steinbach (1998). This is correctly predicted by the obliqueness hierarchy in (28).

(34)

a.

Gestern hat man die Gäste₁ einander₁ vorgestellt

Yesterday has one the guests-acc each other-dat introduced

b.

*Gestern hat man den Gästen₁ einander₁ vorgestellt

Yesterday has one the guests-dat each other-acc introduced

The examples in (35) illustrate that accusative and dative objects can asymmetrically bind other more oblique objects. We already saw in (31) that free datives equal subcategorized datives in this respect.

(35)

a.

Ich bringe den Jungen₁ zu sich₁

I take the boy-acc to rp

b.

*Ich bringe zu dem Jungen₁ sich₁

I take to the boy rp-acc

c.

... weil (es) den beiden Henkern₁ vor einander₁ graute

... because (it) the two executioners-dat of each other dreaded

d.

*... weil (es) vor den beiden Henkern₁ einander₁ graute

... because (it) of the two executioners each other-dat dreaded

Note that an explanation that relies on R&R’s (1993) binding condition B (a reflexive semantic predicate is reflexive-marked) would incorrectly predict that both sentences should be grammatical, because condition B does not refer to an order of obliqueness. Thus even if we would relate our explanation to R&R’s original version of the binding theory, that we discussed in section 5.2, some additional condition on the obliqueness of arguments would be necessary. The next kind of example can be already found in P&S (1994: 275f.). In German, like in English, a reflexive pronoun included in a by-phrase can be bound by the subject of the sentence – ‘pass’ is the passive axiliary.

(36)

a.

The only barber who was shaved by himself was Figaro

b.

Kater Karlo und Zwerg Zwetschge waren die einzigen Ganoven, die₁ jemals von

Kater Karlo and Zwerg Zwetschge were the only crooks that ever by

sich₁ (selbst)/*ihnen₁ ausgeraubt wurden

rp (self)/them robbed pass

c.

Helmut₁ wurde wieder nur von sich₁ (selbst)/*ihm₁ gewählt

Helmut pass again only by rp (self)/him voted-for

d.

Die BRD₁ sieht sich_1/*2 von der DDR₂ betrogen (cf. Grewendorf 1983)

The BRD sees rp/her by the DDR cheated

These examples clearly show that o-command must be defined relative to case (or grammatical relations) instead of thematic roles. A binding theory based on a thematic hierarchy cannot account for the reflexive pronoun in (34).

Note that under this theory, reflexive pronouns should be ungrammatical in passives at all. This theory incorrectly predicts that the reflexive pronouns cannot bear the agent-role because this would violate the thematic hierarchy. On the other hand, if the reflexive pronouns would correspond to the theme or patient of the verb, they are ungrammatical either because English and German lack a nominative form for reflexive pronouns..

A similar situation can be found in embedded passives in ECM-constructions as (34.d). Again, the agent of the embedded clause (von der DDR ‘by the DDR’) cannot bind the theme/patient although the reflexive pronoun receives accusative instead of nominative case and is thus not excluded for case reasons. Finally we come to ECM- (or A.c.I.-) constructions. German has only a few ECM (or A.c.I.) verbs as, for example, some verbs of perception like sehen (‘see’) or hören (‘hear’) and the verb lassen (‘let’/‘have s.o. do s.th.’), which assign accusative case to the embedded subject. The following brief discussion will be limited to verbs of perception.

Lassen differs from verbs of perception in certain respects (cf. e.g. Grewendorf 1983). Firstly, lassen, unlike sehen and hören, does not select a corresponding finite complementizer-initial embedded clause.
(i)

a.

*Peter ließ dass der Mechaniker das Auto reparierte
Peter lets that the mechanic the car repaired

b.

Peter sah dass der Mechaniker das Auto reparierte

Secondly, as opposed to sehen and hören, lassen selects embedded passives without passive morphology.
(i)

a.

Peter ließ das Auto (von einem Mechaniker) reparieren
Peter lets the car (by the mechanic) repair

b.

Peter sah das Auto (von einem Mechaniker) repariert werden

Thirdly, only lassen selects bare infinitives (iii), resultative constructions (iv), and so-called lassen-middles (v).
(iii)

a.

Der Chef läßt arbeiten
The boss lets work

b.

Der Chef sieht arbeiten

(iv)

a.

Der Pfarrer läßt die Kinder in Ruhe
The minister lets the children quiet
‘The minister leaves the children alone’

b.

*Der Pfarrer sieht die Kinder in Ruhe

(v)

a.

Das Buch läßt sich leicht lesen
The book lets RP easily read
‘The book is easy to read’

b.

*Das Buch sieht sich gut lesen

Wunderlich (1985) argues that the ECM-subject is also a semantic argument of lassen. We will argue below that the same is true for perception verbs. Therefore, the ECM-subject can be bound by a less oblique co-argument, i.e. the subject of the matrix clause.
(vi)

Peter läßt sich waschen
Peter lets rp wash

ECM constructions are a particular interesting case for two reasons. Firstly, the ECM-subject can be bound by the matrix subject. Secondly, the embedded object can only be bound by the ECM-subject, although the latter is not less oblique. Both the binder and the reflexive pronoun are assigned accusative case.

This issue is rather more complex. Some data are very subtle and require more discussion (cf. for example Reis 1976 and Grewendorf 1983 for detailed analyses). In the following we present only the core cases..

(37)

a.

Im Traum sah die Frau₁ sich₁ den Geliebten waschen

In-the dream saw the woman-nomrp-acc the sweetheart-acc wash

b.

Im Traum sah die Frau₁ den Geliebten₂ sich_*1/2 waschen

In-the dream saw the woman-nom the sweetheart-accrp-acc wash

c.

Im Traum sah der Mann₁ die Politiker₂ wieder mal nur sich_*1/2 helfen

In-the dream saw the man-nom the politicians-acc again only rp-dat help

We turn to the the matrix clause first. Recall that R&R argue that the ECM-subject (the reflexive pronoun in (37.a) is a syntactic argument of the matrix verb, because it assigns accusative case to the ECM-subject. However, the ECM-subject is not only a syntactic but also a semantic argument of the matrix verb.

See also Pollard and Sag (1994:132), who argue that equi verbs select an embedded VP, the unexpressed subject of which is also a semantic argument of equi verb itself.

Consider he following contrast. The ECM-construction in (38.a) implies that Peter sees Maria. In contrast, the complementizer-initial embedded clause in (38.b) does not trigger this implication. (37.b) can also mean that Peter realizes, that Maria took the baby for a walk, because Maria and the baby carriage are gone. The same contrast can also be found in (38.c) and (38.d) for the verb hören (‘hear’).

(38)

a.

Peter sah Maria mit dem Kind spazieren gehen

―→

Peter sieht Maria

P. saw M.-acc with the child stroll went

‘Maria saw Maria taking the baby for a walk’

b.

Peter sah, dass Maria mit dem Kind spazieren ging

– /→

Peter sieht Maria

P. saw that M.-nom with the child stroll went

‘P. saw that M. took the baby for a walk’

c.

Peter hörte den Präsidenten singen

―→

Peter hört den Präsidenten

Peter hears the president-acc sing

d.

Peter hörte, dass der Präsident singt

– /→

Peter hört den Präsidenten

Peter hears that the president-nom sings

Likewise negation in the ECM-construction implies that subject of the matrix clause does not see or hear the ECM-subject at all. As opposed to (39.a), (39.b) does not imply that Hans did not see Hans. Only the latter sentence can be uttered in a situation, where Peter is sitting next to Hans in the living room.

In addition, both sentences permit a narrow scope reading of the negation. In this case only the constituent im Garten (‘in the garden’) is in the scope of the negation and the sentences mean that it is not in the garden but somewhere else where Peter saw Hans playing.

(39)

a.

Peter sah Hans nicht im Garten spielen

―→

Peter sieht Hans nicht

Peter saw Hans-acc not in-the garden play

‘Peter did not see Hans playing in the garden’

b.

Peter sah, dass Hans nicht im Garten spielte

– /→

Peter sieht Hans nicht

Peter saw that Hans-acc not in-the garden play

‘Peter saw that Hans did not play in the garden’

We conclude that in ECM-constructions the matrix object is not only a syntactic but also a semantic argument of the matrix verb. The ECM-subject can be bound by the matrix subject, because they are co-arguments of the matrix predicate (sehen or hören) and the matrix subject is less oblique than the ECM-subject. Hence, the binding relations in the matrix clause follow from the obliqueness hierarchy proposed in (28) above. We complete the discussion with the embedded clause of ECM-constructions. (37.b) above and the similar example in (40) illustrate that the embedded object can only be bound by the ECM-subject. Recall that the ECM-subject is also a semantic argument of the embedded predicate. Hence, binding should be possible. However, the antecedent is not less oblique than the reflexive pronoun, because both arguments are assigned accusative case. These examples demonstrate that o-command must not be defined relative to less oblique constituents. Instead a constituent X o-commands a constituent Y iff X is not more oblique than Y. Thus the antecedent must not be more oblique than the reflexive pronoun. This modification would also be necessary for the original version of P&S’s binding theory.

Alternatively, one could exempt the reflexive pronoun in (40) from binding, because it is not locally o-bound by a less oblique co-argument. This proposal would predict that sich in (40) is a logophoric reflexive pronoun that should be subject to processing and discourse constraints (cf. P&S 1994:266f.). However, binding of the embedded reflexive pronoun by the matrix subject seems to be only marginally possible even if the intervening ECM-subject is excluded from binding (‘pp’stands for personal pronoun).
(i)

a.

Die Königin läßt nur mich ??sich/sie waschen
The queen lets only me RP/PP wash

b.

Die Katzen lassen mich immer wieder ??sich/sie waschen
The cats let me again and-again RP/PP wash

Note that the binding relation between the ECM-subject and the embedded object cannot be reversed, because a reflexive ECM-subject must also be bound by the matrix subject.

(40)

Peter sah den Barbier sich rasieren

Peter saw the barber rp-acc shave

‘Peter saw the barber (himself) shaving’

We are now in the position to define the final version of our binding theory for German. The following table illustrates the different kinds of arguments and the relevant binding relations we defined for German. The structural (or configurational) binding in (i) and (42.1) below will be relevant for the analysis of non-argument reflexives in the following section.

(41)

Syntactic, semantic and logophoric binding in German

Reflexive pronoun

Binding condition

Interpretation

(i)

syntactic argument

GCC, cf. (42.1)

non-argument

(ii)

semantic argument

Principle A, cf. (42.2) and (42.2)

argument

(iii)

non-argument

exempt from Principle A, cf. (42.2)

argument

Note that (ii) deals with to the standard examples of binding, argument reflexives that are coreferent to another argument of the same predicate, whereas (iii) accounts for the examples we discussed in section 5.1. The only pure syntactic condition in (i) explains the distribution of non-argument reflexives in German. The distinction between (i) and (ii) is based on the observation that (weak) reflexive pronouns are lexically underspecified for the feature [R]. Syntactic arguments are defined on the basis of structural case: only NPs that are structurally case-marked (or that check a structural case-feature) count as A-elements. Hence, not all semantic arguments of a verb are also syntactic arguments: dative objects, mit-phrases, für-phrases, or von-phrases are A’-elements in syntax, that can be arguments of the verb in semantics. [–R] reflexive pronouns that check structural case must be included in an A-chain headed by a [+R]-expression, because they are subject to the GCC. The distribution of [+R] reflexive pronouns is generally controlled by principle A. Either they are locally o-commanded or they are logophorical. Our binding theory is summarized in (42).

(42)

1.

General Condition on A-Chains (GCC)

A maximal A-chain contains exactly one link - α₁ - that is both +R and case-marked.

A maximal A-chain is any sequence of coindexation of syntactic arguments that satisfies antecedent government.

The syntactic arguments of P are the chains that are assigned structural case in the extended projections of P.

2.

Binding principles

Principle A: A locally o-commanded reflexive pronoun must be locally o-bound

Principle B: A personal pronoun must be locally o-free

Let Y and Z be different semantic arguments of the same predicate, Y referential. Then Y locally o-commands Z just in case the syntactic constituent Y is linked to is not more oblique than the syntactic constituent Z is linked to.

Y locally o-binds Z just in case Y and Z are coindexed and Y locally o-commands Z. If Z is not locally o-bound, then it is said to be locally o-free.

3.

Obliqueness hierarchy in German

nominative (subject) > accusative (direct) object > dative (indirect) object > other oblique objects

The interpretation of accusative reflexive pronouns

Having defined the conditions on o-binding and A-chain formation, we can now turn to the interpretation of accusative reflexive pronouns in the position of the direct object. In the final part of this chapter we turn to the first ambiguity of accusative reflexive pronouns illustrated in figure (1) above. The accusative reflexive pronoun in the position of the direct object can but need not be linked to a semantic argument of the verb. The examples are repeated in (43) on the one hand and (44) on the other. We will argue that this ambiguity can be derived from the binding theory defined in (42) above. The ambiguity between (44.a) and (44.b), which results from the interpretation of the implicit argument, will be discussed in the next chapter. This second ambiguity is not directly related to the interpretation of the accusative reflexive pronoun.

(43)

a.

Peter wäscht sich

(argument-reflexive)

Peter-nom washes rp-acc

(44)

a.

Das Buch liest sich gut

(non-argument reflexive, argument-saturation)

The book-nom reads rp-acc well

b.

Die Tür öffnet sich

(non-argument reflexive, argument-reduction)

The door-nom opens rp-acc

Following Chomsky (1995:chapter 3) a DP must check its (structural) case feature in the extended projection of V (cf. also chapter 4 above). In German, only nominative and accusative case is structural.

In chapter 4 we argued that our approach to reflexivity does not depend on a special analysis of structural case. All we need is a (syntactic) distinction between structural and oblique case. We follow Chomsky (1995: chapter 3) because this approach enables us to account for the differences between structural and oblique case straightforwardly (we turn to the syntactic analysis of oblique case forms in the next section).

An accusative object moves to AgrOP,Spec and a nominative subject moves to AgrSP,Spec.

A-movement can take place either before or after Spell-Out depending on the case feature: strong features must be checked before Spell-Out, weak features need not be checked before Spell-Out (and according to Procrastinate must not be checked before Spell-Out), cf. Chomsky 1995.

Hence, A-movement creates two A-chains in a transitive sentence, as is illustrated in corresponding structure (45).

(45)

[_AgrSP NP₁ (Subject) [_AgrOP NP₂ (Object) [_VP t₁[_V' t₂ V ]]]]

chain 1 = NP₁ - t₁ ; chain 2 = NP₂ - t₂

In (45) both A-chains check their case-feature in the head-position and they are interpreted in the tail- (or base-) position. Both chains are headed by a [+R]-expression, which is linked via the base position to a semantic argument of the verb. Chain number 1 is interpreted in its VP-internal position t₁, i.e. VP,Spec. This position is always linked to the first argument of the verb.

We assume that the arguments of a two-place predicate are ordered pairs.

Chain number 2 is interpreted in t₂, the complement-position of V°. This position is linked to the second argument of the verb. Consider the following simple transitive sentence without a reflexive pronoun (in the following presentation all NPs are interpreted as individual type variables for the sake of simplicity):

(46)

a.

Peter read ‘War and Peace’

b.

[_AgrSP Peter₁[_AgrOP ‘War and Peace’₂[_VP t₁[_V' t₂ read ]]]]

chain 1 = Peter₁ - t₁ ; chain 2 = ‘War and Peace’₂ - t₂

c.

R < x, y >

d.

1y (R < x, y >) (w&p)

1

R < x, w&p >

e.

1x (R < x, s >) (p)

1

R < p, w&p >

Note that in German the linking-principles for syntactic argument are very simple: (i) Spec of VP is linked to the first argument variable x and (ii) the complement position of V° is linked to the second argument variable y. In the next chapter we come back to these two principles. Let us now turn to reflexive pronouns in AgrOP,Spec. Reflexive pronouns are not lexically specified for the feature [R]. Therefore, they can be either [+R] or [–R]. According to the General Condition on A-Chains (GCC), only the latter must be a proper part of a maximal A-chain that is headed by another A-element that is specified as [+R]. The [–R] reflexive pronoun is forced to form a chain with the nominative subject, otherwise the resulting syntactic structure would be ungrammatical. Thus whenever the second A-chain in (46) is headed by a [–R]-expression, it must be coindexed with the first chain, which results in the complex A-chain 3. The complex chain meets the GCC, because it is headed by the [+R]-expression in AgrSP,Spec. On the other hand, chain 2 in (46) must not be coindexed with chain 1 if the former is headed by a [+R] reflexive pronoun. Otherwise this would lead to a violation of the GCC, which excludes A-chains that contain two [+R]-expressions.

Condition (42.1) states that a maximal A-chain is any sequence of coindexation of syntactic arguments that satisfies antecedent government. The syntactic arguments are chains themselves. In (38) they consist of two members, that are coindexed. In case two syntactic arguments (i.e. chains) are coindexed, all four members bear the same index. Alternatively, one can define maximal A-chains in the following way: A maximal A-chain is any sequence of coindexation of A-positions that satisfies antecedent government. An A-position is any position/link of a chain to which structural case is assigned. This slightly revised definition yields the same result: a [–R]-expression in an A-position must not be the head of the A-chain it is part of.

(47)

[_AgrSP NP–[+R]₁ [_AgrOP RP–[+/–R]₂ [_VP t₁[_V' t₂ V ]]]]

chain 1 = NP–[+R]₁ - t₁ ; chain 2 = NP–[+R]₂ - t₂

^{chain 3 = NP}–[+R]₁ - NP–[–R]₂ - t₁ - t₂

As a consequence, the [–R] reflexive pronoun must be included in a complex A-chain, the head of which is linked via the chain’s VP-internal base position t₁ (i.e. the complement position of V°). Thus a complex A-chain is always linked to the second argument of the verb, whereas the verb’s first argument cannot be linked to syntax. Unlike [–R] reflexive pronouns, [+R] reflexive pronouns head an A-chain of their own and both A-chains are linked to the first and second argument position respectively. To summarize, a accusative reflexive pronoun in object position can either head its own chain or it is included in another (complex) A-chain, which is headed by a [+R]-expression. The former option is only available for [+R] reflexive pronouns and the latter one for [–R] reflexive pronouns. Thus the reflexive pronoun can be linked to a semantic argument only if it heads its own chain. According to the GCC, this is only possible for reflexive pronouns that are specified as [+R]. The ambiguity of accusative reflexive pronouns is illustrated in the second and third line of table (48).

(48)

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

Let us take a closer look at [+R] reflexive pronouns first. The lexically underspecified reflexive pronoun can only be the head of an A-chain if it is specified as [+R]. The resulting syntactic representation in (49) equals the one in (46) above. Both structures contain two A-chains that are linked to a semantic argument each. The reflexive pronoun is linked to the second semantic argument of the two-place verb and the subject to the first one.

(49)

a.

Peter wäscht sich

Peter washes rp-acc

b.

[_AgrSP Peter₁[_AgrOP RP–[+R]₂[_VP t₁[_V' t₂ wäscht ]]]]

chain 1 = Peter₁ - t₁ ; chain 2 = RP–[+R]₂ - t₂

c.

W < x, y >

d.

1y (W < x, y >) (RP)

1

W < x, RP >

e.

1x (W < x, s >) (p)

1

W < p, RP >

Hence, the [+R] reflexive pronoun is linked to a semantic argument itself and it must be bound by another argument of the same predicate. Recall the binding principles from the previous section, which are defined relative to semantic arguments. Principle A states that a locally o-commanded reflexive pronoun must be locally o-bound bound, i.e. it must be bound by a co-argument that is not more oblique than the reflexive pronoun itself. In (49) the subject is less oblique than the reflexive pronoun. Thus it locally o-commands and locally o-binds the reflexive pronoun. Thus both arguments are coindexed.

(50) W < p_i, rp_i >

Reinhart (1983) defines a translation mechanism for bound reflexive pronouns.

Reinhart’s (1983: 160) original version is limited to S’-expressions: [_S’ _{Φ ] ⇒ [_S’β(λx ( Φ^β/x ))]. This definition follows the binding theory in Chomsky’s (1981) which is defined on the basis of (syntactic) governing categories. Our theory of binding is, however, defined on semantic predicates. Therefore, we omit this restriction.}

Unlike Reinhart (1983), our approach relates binding to semantic co-arguments. Therefore, we assume that Φ in (51) is a semantic predicate. Thus rule (51) operates on a semantic predicate Φ, λ-abstracts on the antecedent (a referential expression) and converts all arguments that are coindexed with the antecedent into variables bound by the λ-operator. The final semantic representation of sentence (49) is given in (52).

(51)

Interpretation of bound variables (following Reinhart 1983)

[_{Φ ] ⇒ [ β(λx ( Φ^β/x ))]}

(52)

(λx W < x, x >) (p)

So far our theory predict the correct semantic representation for argument reflexives. In the next step we derive the interpretation of non-argument reflexives. We already saw that [–R] reflexive pronouns, unlike their [+R] counterparts, must be included in another A-chain that is headed by a [+R]-expression. This complex A-chain is again interpreted in its base-position, the complement of V°, which corresponds to the second argument position of the verb. Note that in (53) VP,Spec cannot be linked to a semantic argument, because it is not the base position of the chain. Therefore the first argument position of the verb is not linked to syntax. The only [+R]-expression is linked to the second argument. We will argue in the next chapter that the implicit argument can either be reduced or saturated. Argument reduction removes the free argument variable from the semantic representation, whereas argument saturation binds the argument variable (‘OP’ stands for operator and ‘Ø’ for the deleted first argument).

(53)

a.

Die Tür öffnet sich

The door opens rp-acc

b.

[_AgrSP Die Tür₁[_AgrOP RP–[–R]₂[_VP t₁[_V' t₂ öffnet ]]]]

chain = Peter₁ - RP–[–R]₂ - t₁ - t₂

c.

O < x, y >

d.

1y (O < x, y >) (t)

1

O < x, t >

e.

Op x (O < x, t >) or O < Ø, t >

Impersonal middles can be derived in the same way. In (54) the reflexive pronoun is bound by a [+R]-element, the impersonal subject es (‘it’) in Spec of AgrSP. We mentioned in chapter 4 that this [+R]-element might base-generated in the VP-internal subject-position, because a VP must contain at least one [+R]-expression. Structure (54) contains one complex A-chain, which must be linked again to the second argument of the verb. However, the impersonal subject es cannot be linked to a semantic argument. Although it is not maximally underspecified, it is only a quasi-argument that is not interpreted in semantics. Note that the impersonal subject, which bears the default values third person singular neuter is the nominative counterpart of the non-argument reflexive. Within the nominative paradigm, the third person singular personal pronoun es is the ‘weakest’ element. Thus it can be used as a pleonastic element or dummy that only fulfills a syntactic function.

Es (‘it’) can fulfill rather different functions in German. It is used as impersonal subject in impersonal middle constructions, with weather verbs (i.a), and in active sentences that do not select a subject (i.b), cf. also Bierwisch (1996).
(i)

a.

… weil es regnet
… because it rains

b.

… weil es in dieser Stadt viele Autos gibt
… because it in this town many cars gave
‘… because there are many cars in this town’

Besides, it is an expletive element, that can optionally occur in the middle field if the corresponding clausal complement is extraposed.
(ii)

a.

.... weil es₁ mir sehr gut gefallen hat, [dass ihr gekommen seid]₁
... because it to-me very much pleased has that you come have

b.

.… weil der Hans es allen erzählt hat, dass die Ökosteuer nicht ausgesetzt wird
… it seems that Hans contented with himself is

And thirdly, es can also be used as so-called ‘Vorfeld-es’ in sentence-initial position.
(iii)

Es steht ein Männlein im Walde
It stands a little-man in-the forest

Note finally, that es, unlike other personal pronouns, cannot be stressed and coordinated. It seems to be intrinsically ‘weak’. For further discussion see Hall (1998) and Cardinaletti and Starke (1994).

As a consequence, the A-chain is not linked to the second argument of the verb. Note, however, that the A-chain is not superfluous in impersonal middle constructions. Its purpose is to prevent the first and sole argument of the one-place predicate from linking to syntax, which must be bound by a semantic operator (argument saturation). Argument reduction is impossible because the sole argument of a one-place predicate cannot be deleted, cf. chapter 6 for further discussion.

(54)

a.

Es schläft sich (gut in diesem Bett)

It sleeps rp-acc

b.

[_AgrSP Es₁[_AgrOP RP–[–R]₂[_VP t₁[_V' t₂ schläft ]]]]

chain = Es₁ - RP–[–R]₂ - t₁ - t₂

c.

S < x >

d.

Op x (S < x >)

We conclude this section with a final remark about ECM-constructions. Our analysis predicts that the non-argument reflexive can also occur in ECM-constructions. This seems to be correct. Reflexive pronouns in ECM-constructions show the same ambiguity between an argument and a non-argument interpretation. An embedded [–R] reflexive pronoun can form an A-chain with the ECM-subject and a [–R] reflexive pronoun in the position of the ECM-subject (i.e. accusative object of the matrix clause) is governed by the matrix subject. Thus the reflexive pronoun in both positions can be included in a maximal A-chain that is headed by a [+R]-expression. The examples in (55) illustrate the ambiguity between argument- and non-argument reflexives for accusative reflexive pronouns in the embedded clause.

(55)

[_CP NP-nom V [ [+R]-NP-acc₁ [–R]-rp-acc₁ V]]

a.

Der Engel sah Maria₁ sich₁ kämmen

(argument reflexive)

The angel sees Maria-accrp-acc comb

b

Peter hörte die Tür₁ sich₁ öffnen

(non-argument reflexive)

Peter hears the door-acc rp-acc open

c

Hans sah den Fjord₁ sich₁ verengen

(non-argument reflexive)

Hans sees the fiord-acc rp-acc narrow

In the position of the ECM-subject non-argument reflexives are harder to find. The reason for this is that ECM- (or A.c.I.-) verbs in general do not form perfect in middle constructions. Besides, they are ungrammatical in anticausatives (cf. next chapter for a brief discussion of the semantics of anticausatives). The ECM-middle construction in (56.b) seems to be as acceptable as the corresponding simple middle construction (56.b’). One exception are so-called let-middle like (56.c).

The embedded predicate lesen (‘read’) is interpreted as passive in this case, although it has no passive morphology. Recall from footnote 33 above that embedded passives of lassen A.c.I.-constructions must not be marked for passive.

(56)

[_CP DP-nom₁ V [ rp-acc₁ DP-acc V]]

a.

Im Spiegel sieht der Engel sich Maria waschen

(argument reflexive)

In-the mirror sees the angel rp Maria wash

b.

^??Von hier aus sieht sich der gegnerische Stürmer viel besser den Elfmeter schießen

(non-argument reflexive)

From here sees rp the opposing forward much better the penalty take

b’.

^??Von hier aus sieht sich das gegnerische Tor viel besser

From here sees rp the opponent’s team much better

c.

Das Buch läßt sich einfach lesen

(non-argument reflexive)

The book lets rp easily read

‘The book is easy to read’

In this section we argued that the non-argument interpretation of the accusative reflexive pronoun in the position of the accusative object depends solely on A-chain formation. According to the GCC, an A-chain must be headed by an [+R]-expression. Therefore, [–R]-expression must not be the head of an A-chain. A non-argument reflexive, which are [–R]-expressions, in the position of the accusative object is forced to form an A-chain with the nominative subject. Argument reflexives on the other hand are [+R]-expressions, which must be the head of their own chain. In addition, we saw that A-chains are linked to semantic arguments of the verb. Therefore, both the subject and the direct object can only be linked a semantic argument if transitive reflexive sentences contain an argument reflexive. (57) summarizes the possible interpretations of one- and two-place predicates in transitive, intransitive and transitive reflexive sentences in German (‘OP’ stands again for semantic operator).

(57) a.

Transitive active sentences in German

b.

Intransitive active sentences in German

Weak and strong pronouns

Recall from section 2.3 and 5.3 that, as opposed to German, most Indo-European languages distinguish not only personal pronouns from reflexive pronouns but also weak from strong reflexive pronouns. Weak reflexive forms are generally less specified than their strong counterparts. They are the morphologically ‘weaker’ expressions. Consider table (58) for this distinction.

(58)

Weak and strong reflexive pronouns

language

weak form

strong form

English:

zero

(∅)

complex word

(him-/her-/itself)

Russian:

verbal affix

(-sja)

word

(sebja)

French:

verbal clitic

(se)

complex word

(lui-/elle-/soi-même)

Dutch:

simple word

(zich)

complex word

(zichzelf)

Only the weak reflexive form is ambiguous between an argument and a non-argument reading (cf. the examples in section 2.3 above). Besides, the weak form permits a reflexive interpretation only with certain verbs that describe actions of grooming or body care or change in body posture, cf. Kemmer 1993, chapter 3. The relevant examples from Dutch are repeated in (59).

(59)

a.

Max₁ gedraagt zich₁/*zichzelf₁

Max behaves SE/SELF

b.

Max₁ wast zich₁/zichzelf₁

Max washes SE/SELF

c.

Max₁ hoorde *zich₁/zichzelf₁

Max heard SE/SELF

The binding theory we defined in section 5.4 for German, a one-form language, does not refer the feature [+/- REFL], that is responsible for the distinction between weak and strong reflexive pronouns in R&R’s theory. Therefore we are finally interested in whether we can build a bridge between one-form languages and two-form languages that accounts for the different distribution of weak and strong reflexive pronouns within our framework? In a one-form language like German a reflexive pronoun (in an A-position) can but need not be linked to a semantic argument of the verb. The restrictions on focus, fronting, and coordinate of the reflexive pronoun can be derived from the its semantic ambiguity. Things are slightly different in two-form languages. Consider first that strong reflexive pronouns are always linked to an argument of the verb. Strong reflexive pronouns are lexically specified as [+R]. As opposed to this, weak reflexive pronouns can but need not be linked. The Φ-features of weak forms are maximally underspecified. Like the reflexive pronoun in German, weak forms can be [+R] and [–R]. Therefore they must be used in contexts that require a non-argument reflexive, i.e. inherent reflexive verbs, anticausatives, middle constructions, and passives (cf. section 2.3). Hence, the strong form is excluded in example (60.a). We turn now to example (59.b), (59.c), and (59.d). R&R assume that verbs like wassen (‘wash’) have two distinct lexical entries: wassen1 is specified as [–REFL] and wassen2 is specified as [+REFL] whereas horen (‘hear’) is only specified as [–REFL]. We want to argue that the difference between verbs like wassen, on the one hand, and horen, on the other hand, need not be stipulated in the lexicon. It can be derived from our conceptual knowledge about events of washing and events of hearing. The weak form of the reflexive pronoun can only be used if the verb describes an action or event that is very likely to be reflexive, i.e. if both the hearer and the speaker expect only one participant. Whenever two (different) participants are expected, the strong form must be used. Unlike the weak form, the strong form seems to be intrinsically contrastive (and because of it must always be linked to an argument variable).

Note that the strong forms in Dutch and English consist of a pronominal part (zich or him/her/it/...) and the adnominal intensifier self. Historically, the adnominal intensifier has been a focus particle. This can be still seen in German, which has not incorporated the intensifier in the pronominal element (cf. chapter 4). Hence, it is very likely that the strong form of the reflexive pronoun intrinsically forms a contrast to other type-identical salient alternatives. The minimal alternative to the reflexive interpretation would be the non-reflexive interpretation, which is only expected with verbs like horen.

The strong reflexive pronoun must be used in (59.c), because horen is not expected to be reflexive.

Recall that the weak reflexive pronoun can also be used in the position of the ECM-subject in ECM-constructions. As opposed to (59.c) ECM-constructions describe complex events in which someone sees/hears someone doing something. One could argue that ECM-constructions are not in a strict sense reflexive, because the the ECM-subject is a semantic argument of both the matrix and the embedded verb as we argued above. However, this issue requires further research.

We conclude this discussion with a final remark on the feature [REFL], although much more should be said about the morphological properties of weak and strong reflexive pronouns and their distribution in different languages. Let us assume that we made a distinction between [+REFL] and [–REFL] reflexive pronuns/anaphors in German as well, contrary to what we said in section 5.3. In German the weak and the strong form would be homophones. In addition, it is the [–REFL] form that must be used in inherent reflexives, anticausatives, middle constructions, and with verbs like waschen, that are likely do be reflexive (cf. section 5.2 above). Hence, all verb that occur in middle constructions and anticausatives must be lexically reflexive in order to license the weak [–REFL] form. As a consequence, nearly every verb in German would have to be lexically specified for both features [+REFL] and [–REFL].

The only exception might be inherent reflexive verbs, which are always specified for [+REFL].

The same is true for two-from languages like Russian.

Note that in Russian passives, anticausatives, and middle constructions are formed with the weak form -sja (which must also be specified as [–REFL]).

Nearly every verb would have to be listed in the lexicon twice as [–REFL] and [+REFL] in order to meet condition A of R&R’s binding theory. But this kind of specification is redundant. Recall that we argued against a lexical derivation of middle constructions in German. Hence, there is no lexical rule that turns non-reflexive verbs into lexically reflexive ‘middle verbs’. Note finally that this version of the binding theory does not account for the fact, that weak reflexive pronouns or SE anaphors in R&R’s terminology are ambiguous between an argument and a non-argument interpretation. According to condition B in (9) above and definition (10.d), two arguments of a reflexive semantic predicate are coindexed. But we saw in this chapter that non-argument reflexives are not linked to a semantic predicate of the verb. This excludes coindexation in anticausatives, inherent reflexives and middle constructions.

Conclusion

In the first part of this chapter we argued that binding theory should be defined relative to syntactic and semantic arguments of a verb. Only NPs that are assigned nominative and accusative case are syntactic arguments (A-expressions) in German. In addition, reflexive pronouns are not lexically specified for the feature [R]. Therefore, they can be either bound in syntax (i.e. [–R]-reflexives) or in semantics (i.e. [+R]-reflexives). And finally, reflexive pronouns that cannot be bound by a co-argument of the same predicate are exempt from binding. This leads to the following threefold picture of binding.

syntactic binding (A-chain formation, restricted to accusative [–R]-RP)
semantic binding (o-binding, restricted to [+R]-RPs and co-arguments of a predicate)
logophoric binding (restricted to reflexive pronouns that are exempt from (i) and (ii))

In the seond part of this chapter we argued that a theory that is based on A-chain formation and the distinction between [+/–R] reflexive pronouns correctly predicts the ambiguity of accusative reflexive pronouns in the position of the direct object. Furthermore, our analysis correctly accounts for the two essential features of middle markers in German. A middle marker must be assigned structural case and it must be a reflexive pronoun. Thus we are now in the position to derive the observation we made in chapter 2:

(60)

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

Recall that it is only the accusative reflexive pronouns in the position of the direct object that can be both a syntactic argument and a [–R]-expression which must be bound by another syntactic argument in syntax. Besides, we postulated the following two simple linking-principles for syntactic arguments in German.

(61)

a.

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

b.

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

In the next chapter we turn first to the difference between middle constructions and anticausatives. In addition, we briefly discuss some consequences of the linking-principles in (65). In the second part we give empirical evidence for a distinction into structural and oblique case in German. We argue that dative case is not structural but oblique. As a consequence, dative objects cannot undergo middle formation in German. Dative reflexive pronouns are always interpreted as a semantic argument of the verb and they must be bound by a another co-argument which must not be more oblique.