6  Suppressed Arguments and Dative Objects

In the previous chapter we argued that the first ambiguity of transitive reflexive sentences results from the underspecification of reflexive pronouns and A-chain formation. An accusative reflexive pronoun in the position of the direct object can but need not be linked to a semantic argument itself. On the one hand middle constructions, anticausatives, and inherent reflexives involve the non-argument reflexive, which is a [–R]-expression. [+R] reflexive pronouns on the other hand yield the reflexive interpretation. In the first part of this chapter we turn to the difference between middle constructions and anticausatives. We will argue that in each construction a different semantic operation applies to the unbound argument variable, i.e. the implicit first argument. The second part deals with dative objects. A-chain formation is a necessary prerequisite for the non-argument interpretation of reflexive pronouns. A-chain formation itself is limited to A-elements, i.e. NPs that are assigned structural cases. We mentioned already that only nominative and accusative case are structural in German. Therefore, oblique case forms like dative objects are excluded from middle formation. We provide empirical evidence to support this distinction. We will see that dative objects differ from accusative objects in various respects. All of the evidence indicates that structural case differs (not only syntactically) from oblique case. Since dative case in German is oblique, dative reflexive pronouns cannot be included in a complex A-chain and thus no ambiguity can arise. They are always linked to an argument variable of the verb.

Recall from chapter 5 that the semantic interpretation of an anticausative like (1.a), which contains a non-argument reflexive, is (1.b). According to the linking-principles for syntactic arguments, the head of the complex A-chain, the subject Tür (‘door’), is linked to the second argument of the verb öffnen (‘open’) via the base position of the complex A-chain. The first argument of the predicate is not linked to syntax because VP,Spec is occupied by an intermediate link of the complex A-chain. As a consequence, the semantic representation in (1.b) contains an unbound argument variable. Middle constructions equal unaccusatives in this respect.

But what can we do with an unbound semantic argument variable? The next subsection investigates this issue and subsection 6.1.2 discusses some consequences of the linking-principles we proposed in chapter 5.

We follow Chierchia (1989) and Reinhart (1996) in assuming that two operations on unbound semantic argument variables are available. They can either be bound by a semantic operator (OP) or can be completely removed from the semantic representation. The former operation is called argument saturation, the latter argument reduction, cf. Chierchia (1989). Both operations are illustrated in (2).

[page 143↓]

Argument reduction is much more restricted than argument saturation and depends on the lexical meaning of the verb. The class of verbs that permit reduction is a subclass of the verbs that permit saturation. Reduction changes a two-place predicate (a relation between two arguments) into a one-place predicate (a property of one argument). Saturation, on the other hand, introduces a semantic operator that binds the free argument variable.¹ Argument reduction is responsible for the anticausative and inherently reflexive interpretations, whereas argument saturation yields the middle interpretation.² We will argue in chapter 7 that in middle constructions the free argument variable is bound by a generic operator. In this section we are concentrating on argument reduction. Note that only some two-place verbs permit the anticausative interpretation. For example, the first argument of zeichnen (‘draw’) in (3.a) can only be saturated but not reduced, whereas öffnen (‘open’) in (3.b) forms a perfectly grammatical anticausative. Öffnen permits both saturation and reduction.

Reinhart (1996) argues that these verbs differ in one respect. The subject of both verbs is lexically specified for the basic semantic feature [+ causing-change]. But only the subject of zeichnen in (3.a) is additionally specified for [+ mental state involved]. Therefore, öffnen permits all kinds of [+ causing-change] subjects, e.g. agents, instruments, or other causing entities, because the verb does not obligatorily select the feature [+ mental state involved]. As opposed to öffnen, the very similar verb zeichnen only selects agents, which fulfill both specifications, [+ causing-change] and [+ mental state involved]. Note that Reinhart’s specification corresponds to Dowty’s (1991) contributing properties for the agent proto-role. Zeichnen has the following two entailments for its subject: a. volitional involved in the event or state and b. causing an event or change of state in another participant (cf. Dowty 1991:572). Öffnen, on the other hand, has only the second entailment for its subject.

The basic intuition is that the anticausative interpretation of a basically two-place predicate requires that the event described by the verb can be conceptualized as taking place without an explicitly mentioned cause. The opening of a door, for example, can be perceived as taking [page 144↓]place all by itself. This does not hold for the drawing of a picture. (5) illustrates this difference with additional examples.³

This restriction on the anticausative interpretation of two-place predicates is part of the lexical meaning of a verb. The anticausative interpretation of trinken, zeichnen or lesen contradicts our knowledge about drinking, drawing,or reading events. Thus mental state verbs (or verbs that entail volitional involvement in the event or state of their first argument) are excluded from the semantic operation of argument reduction that applies to two-place predicates. Reduction of the first argument is only possible if it is not specified as [+ mental state involved] (if the verb does not entail volitional involvement for this argument). This can be stated in the following way (cf. Reinhart 1996: 19).

Like anticausatives, inherent reflexive verbs can be derived from an underlying two-place representation. The reflexive pronoun indicates that the syntactic subject is linked to the second argument of the verb. They differ, however, from the former in one crucial respect. Inherent reflexives only permit the anticausative interpretation. Their first argument must obligatorily be reduced. Inherent reflexive verbs are inherent anticausative verbs. The first argument (the cause of the event) of these basically two-place verbs can neither be linked to syntax nor can it be bound by a semantic operator.

This very special property of inherent reflexive verbs may either follow from their meaning or it must be stipulated in the lexicon. Under the assumption that first-language acquisition takes place without negative evidence, this restriction must follow from the meaning of inherent reflexive verbs. Imagine that children have only the input in (7.a). How can they learn that the German verb erkälten (‘catch a cold’) must be used inherently reflexive? Children do not have access to the information that sentences like (7.b) are ungrammatical. What they can conclude from (7.a) is that the verb erkälten basically selects two arguments. This follows from every theory that treats the reflexive pronoun either as an indicator of valency reduction or as a semantic argument of the verb. Together with the meaning of the verb children must learn that it is impossible to express the first argument of the verb, which must obligatorily be reduced. Alternatively, one could, of course, argue that all inherent reflexive verbs are idiomatic expressions (or frozen forms) that must be learned as a whole. More research on inherent reflexive verbs in German and other Indo-European languages and their semantics, historical development and acquisition is necessary in order to decide this issue with a clear conscience.

The analysis of non-argument reflexives in chapter 5 is based on two linking-principles for syntactic arguments which are repeated here in (8).⁴ (9) illustrates the linking of the first and the second argument to the VP,Spec and the V-complement position, respectively.

Our analysis predicts that unaccusatives and unergatives share the same VP-structure. The linking-principles in (8) contradict a theory, which reflects the lexical distinction between unergatives and unaccusatives in syntax. Let us assume that the lexical representation of unaccusatives like ankommen (‘arrive’)differs from that of unergatives like lachen (‘laugh’). We call the sole argument of unergative predicates external and the that of the unaccusative predicates internal. The external argument has typical proto-agent properties, whereas the internal argument has typical proto-patient properties.⁵

For the unaccusative in (10.b) two different VP-structures are at hand. Either we analyze unaccusatives parallel to unergatives in syntax, as is illustrated in (11.a), or the sole argument of unaccusatives corresponds to the complement-position of V° and VP,Spec is empty or does not project, cf. (11.b).⁶

[page 146↓]

Structure (11.a) is in line with our derivation of non-argument reflexives. The sole syntactic argument contained in the VP, the nominative NP, must always be linked to the sole (and hence first) argument of the one-place predicate, regardless of whether this argument is external or internal in the lexical representation. Matters are a little more complicated if we choose structure (11.b). We might slightly modify the linking-principles in (8). According to this modification, the complement position of V° would be linked to the internal argument of the predicate whereas the specifier-position is linked to the external argument. As a consequence, unaccusative predicates are expected to have two grammatical syntactic realizations. The internal argument can be linked either to a NP in the complement position or to a complex A-chain. Hence, unaccusatives should be syntactically intransitive or transitive reflexive. This issue is illustrated in (12) and (13).

The structure in (12) contains a complex A-chain. Like subjects in middle constructions and anticausatives, the subject in (12) can in principle be interpreted in the complement position. Therefore the derivation in (12.b) should be grammatical, contrary to facts. If we accept structure (11.b) above and the modified linking principles, we expect sentence (12.a) to yield the same interpretation as the unaccusative in (13), which is derived in syntax by A-movement of the complement of the verb to AgrS,Spec, cf. (11.b).

This may be seen as evidence that German does not distinguish between unaccusatives and unergatives in syntax. All intransitive sentences have the same underlying VP-structure in (11.a) This VP contains only one NP, which is always linked to the first semantic argument of the one-place verb. In addition, our analysis predicts that non-reflexive anticausatives as, for example, rollen (‘roll’) in (14) are basically unaccusative one-place predicates. In this case the causative variant is derived from an underlying anticausative/unaccusative verb by causativization, i.e. by addition of an external argument. In this respect verbs like rollen are the mirror [page 147↓]image of reflexive anticausatives such as öffnen (‘open’) in (15). The latter are basically two-place predicates and the (one-place) anticausative variant is derived from the underlying two-place representation. Of course, further research on the semantic differences between these two kinds of predicates is necessary (cf. also Wunderlich 1993).

The binding theory we defined in chapter 6 predicts that the non-argument interpretation of reflexive pronouns is restricted to syntactic arguments that can be included in a complex A-chain, which is headed by another syntactic argument. Thus only reflexive pronouns that are assigned (or check) structural case are able undergo middle formation. Recall the definition of the notion syntactic argument in (42.1), chapter 6, here repeated as (16).

In this section, we give empirical evidence for the distinction between structural and oblique case in German. We argue that only nominative and accusative are structural in German, whereas dative case is oblique. As a consequence, dative objects cannot undergo middle formation. They are only grammatical in impersonal middle constructions like (17.c). In impersonal middle constructions the dative object preserves its case and the accusative reflexive pronoun indicates valency reduction as usual.⁷

We combine the differences between nominative and accusative case on the one hand and dative case on the other in two groups. The first group subsumes differences related to morphology. Syntactic differences are subsumed under the second group. Essential parts of this section relate to joint work with Ralf Vogel and the analysis proposed in Vogel and Steinbach (1995 and 1998). We are concentrating on accusative and dative object case and exclude case-[page 148↓]assignment by prepositions and nominals. Table (18) summarizes these differences, which will be discussed in the following two subsections.

This subsection deals with several phenomena that can be related to differences in the morphological specification of accusative/nominative case on the one hand and dative case on the other, cf. the left column in table (18) above. First we briefly discuss case morphology in German, before we turn to several examples that can be related to the morphological distinction between marked and unmarked case.

(i)Marked and unmarked case forms: German draws a clear distinction between unmarked structural case and marked oblique case. Consider the following table, which illustrates the case pattern of definite NPs (first line) and pronouns (second line). Except for the masculine singular form, accusative case is always morphologically identical to nominative case. Thus nominative and accusative are equally unmarked on the assumption that nominative case is the unmarked form in a language like German (cf. e.g. Bittner and Hale 1996 and Weerman 1996). Dative case does not pattern with the unmarked structural case forms. Only the feminine singular form of the article is identical to the genitive, which is also an oblique case. Besides, only the oblique forms contain nominal affixes, which are written in bold face. Homonymous forms are written in italics in (19).^{8 , 9}

[page 149↓]

The difference between morphologically marked and unmarked case may be reflected in the phrase structure of the respective nominal constituents. According to Bittner and Hale (1996) and Bader et.al. (1996) unmarked (i.e. structural) case forms project simply DPs, whereas marked (i.e. oblique) case forms are KPs. We return to this issue below.

(ii)‘nichts’ and ‘genug’: German has a small class of uninflectable indefinite NPs as, for example, genug (‘enough’) and nichts (‘nothing’). Gallmann (1995) points out that these indefinite expressions can only be used as accusative but not as dative objects, cf. (21) and (22).

(iii)Sentential complements: A similar contrast is reported in Fanselow and Felix (1987b: 85f). They observe that complement clauses can only be assigned nominative or accusative case. Most verbs that select propositional complements assign nominative or accusative case to them, but some verbs assign dative or genitive case to their propositional complement (we illustrate this issue only for dative case; for examples with genitive case see Fanselow and Felix 1987b and Vogel and Steinbach 1995).

Fanselow and Felix observe that only verbs assigning (nominative and) accusative case to their propositional object can realize this object as a CP. Dative objects cannot be sentential. They draw the conclusion that complement clauses can only be assigned structural case. Oblique case must not be assigned to clauses.

We already saw that structural case is morphologically unmarked. Hence, complement clauses, which do not inflect for case in German, are compatible with the zero morphology of the structural cases. Therefore accusative clauses, cf. (24.a), as well as nominative clauses, cf. (25.c), are grammatical in German. As opposed to structural case, dative and genitive are morphologically marked. This is what rules out dative and genitive complement clauses. A sentential complement that is assigned dative case usually requires an additional expletive pronominal that inflects for dative case and is coindexed with the extraposed clause as in (22.a). Alternatively, these sentential complements can also be realized as nominative subjects of the so-called kriegen-‘passive’ (or dative-passive) as in (25.c). Thus the sentential complement has two ways of avoiding dative case. The latter example is from Webelhuth (1990).

(iv)Free relatives: Relative pronouns in free relatives may come into a case conflict, because they receive case twice. Case assignment in the matrix clause may differ from case assignment in the embedded relative clause as is illustrated in (26).¹⁰ In (26.a) the matrix verb assigns nominative, whereas the embedded verb assigns accusatives. According to Vogel (2000), such case conflicts are resolved by two constraints in German.¹¹ (i) The relative pronoun receives the case that is assigned in the embedded relative clause if the matrix case is structural. Thus structural case can be overwritten. This is illustrated in (26.a), (26.b), and (26.c). In (26.c) the matrix accusative is ‘outranked’ by the embedded dative. (ii) The relative pronoun does not receive case if the matrix case is oblique, because oblique case cannot be overwritten and accusative does not outrank dative, cf. (26.d). Free relatives that are assigned [page 151↓]oblique case in the matrix clause are only grammatical if the relative pronoun is assigned the same oblique case in the relative clause, as is illustrated in (26.e) – ‘M’ does not stand for murder but for ‘matrix clause’, ‘E’ for ‘embedded clause’, and ‘RP’ for ‘relative pronoun’).

Once again, structural case differs from oblique case. Only the former can be overwritten, whereas the latter does everything to preserve its case.

(v)Nominalization: This point includes three related observations. First, verb-object-compounds are only grammatical if the object receives accusative but not if it receives dative.¹² Dative objects are excluded in principle, no matter whether they are the second or third semantic argument of the verb. Sentence (27.e), for example, contains a two-place predicate, which assigns dative case to its object Kinder (‘children’). Nevertheless, nominalization is ungrammatical with the object.

One might argue that nominalization is a lexical process, cf. e.g. Chomsky (1970) that excludes morphologically marked constituents. Recall that Bittner and Hale (1996) assume that datives project a KP. Alternatively, one might argue that dative objects, unlike accusative objects, are not licensed in a syntactic position adjacent to the verb (i.e. the complement of V°). We come back to these two proposals in section 6.2.3 below.

Second, Bader et.al. (1996) mention yet another difference between structural and oblique case. Nominative and accusative, unlike dative, alternate with a postnominal genitive in [page 152↓]nominalizations. Again, only structural case permits conversion into any other case (cf. also Maling 1999:19f.).¹³

Third,so-called ‘Rektionskomposita’ as (29.a) are less productive with objects assigned dative case than with objects assigned accusative case, cf. Rivet (1999) for discussion. Although incorporation of dative objects is not completely impossible, as is illustrated in (29.d), it is much more restricted than incorporation of accusative objects. Again this difference between accusative and dative objects might be related to morphological and/or syntactic differences between structural and oblique case forms.

(vi)Idioms: Vogel and Steinbach (1995: 114) mention that a certain kind of idioms which can be compared to take care of or take advantage of in English, can only be formed with accusative objects. Examples with dative which are expected to be grammatical are, however, ungrammatical.

The direct object in (30.a) and (30.b) seems to be (prosodically) incorporated into the verb, which is only possible if the incorporated NP is not marked for case and/or adjacent to the verb. Dative case, unlike accusative case, is morphologically marked and tends to preserve its case morphology. Besides, we argued that accusative objects are base generated in the complement position of V°, which is adjacent to the verb. Steinbach and Vogel (1998) argue that dative objects are adjuncts in syntax. Hence, dative objects never occur in a base position that is structurally adjacent to the verb, cf. also (v) above.

(vii)Reflexive pronouns: In section 2.3 we mentioned dative reflexive pronouns developed very late in German. Until the end of the 17th century German had only one pronominal form for dative case. This still holds for some German dialects. As opposed to dative reflexive pronouns, accusative reflexive pronouns are attested for earlier stages of German.

This subsection discusses further differences between structural accusative objects and oblique dative objects, which relate to syntax. These differences concern word order, binding, valency change, and sentence processing.

(i)Word order: Recall from chapter 4 that the linearization of arguments in the middle field depends on various constraints on unmarked word order.¹⁴ Dative NPs can precede nominative and accusative NPs in the unmarked word order in certain contexts because the unmarked position of dative objects is sensitive to animacy: a NP that refers to an animate entity precedes a NP that refers to an inanimate entity.¹⁵ As opposed to dative objects the unmarked position of accusative objects is ‘structurally’ fixed. The following example illustrates this difference between dative and accusative objects. Begegnen (‘meet’) in (31.a) assigns dative and treffen (‘meet’) in (31.b) accusative. Note that the meaning of these two verbs is almost identical.¹⁶

As opposed to dative objects, accusative objects usually follow the subject. However, a small class of accusative verbs also permit the unmarked order accusative > nominative. But unlike dative objects, the accusative object can also follow the subject in unmarked order.

Gärtner and Steinbach (1997) observe a similar contrast between dative and accusative objects. In sentence-initial position dative objects can refer to inanimate entities more easily than accusative objects. Note that discourse linking of personal pronouns becomes easier if they refer to animate entities, which seem to be more salient. This holds especially for sentences with marked word order. Unlike dative objects, accusative objects are usually marked in sentence-initial position, cf. section 4.2 and Cardinaletti and Starke (1994).¹⁷

(ii)Binding: In the previous chapter we argued that binding is subject to an obliqueness hierarchy. More oblique arguments cannot bind less oblique arguments. Therefore, dative objects are asymmetrically bound by accusative objects (and subjects) and they asymmetrically bind more oblique objects. We repeat the relevant examples in (34) and (35).

Besides, dative objects can serve as A’-binders. And in this case, they are able to (A’-) bind an accusative object provided that they c-command it. This is illustrated by the following examples from Vogel and Steinbach (1995:107). These examples include quantifier-pronoun relations, each...other constructions and negative polarity items.

[page 155↓]

(iii)Middle constructions: It goes without saying that dative objects are excluded from middle formation. Dative reflexive pronouns cannot indicate valency reduction. It is only the accusative reflexive pronoun in the position of the direct object that can be called a middle marker in German.

(iv)Tough-movement: The picture is the same with tough-movement constructions. Again only accusative objects must be promoted to subject, as can be seen in (37.a) and (37.b). The dative NP of a corresponding active sentence must not be substituted by a nominative NP in the tough-movement construction. (37.d) equals an impersonal middle construction.¹⁸

(v)Passive and ‘D.c.I.’: The third difference related to valency reduction concerns passive. It is a wellknown fact that dative objects, unlike accusative objects, cannot be passivized. They cannot be promoted to subject again and thus it is not surprising that they are only grammatical in the impersonal passive in (38.c) - ‘pass’ stands for the passive auxiliary.

[page 156↓]

Note, however, that some authors claim that dative objects can also be structurally passivized. As opposed to accusative objects, dative objects require the passive auxiliary kriegen or bekommen instead of werden:

Besides, free datives can be ‘passivized’ as well, although they are not selected by the verb.

Kriegen-passive is debated controversially. So far it has not become clear whether sentences like (39) and (40) should be analyzed as a passive (either in syntax or in the lexicon) or as a predicative construction, cf. for example Höhle (1978), Haider (1984), Reis (1985), Wunderlich (1985), or Kathol (1995). Apart from that Vogel and Steinbach (1998) argue that (39) and (40) may provide evidence that dative case can be absorbed in kriegen-passives. But case absorption may not be an exclusive property of structural case. In contrast, case movement (i.e. movement into a case position) seems to be an exclusive property of structural case. German has only movement into a structural nominative position (passive and raising) and into a structural accusative position (ECM or A.c.I.) but no movement into a (structural) dative position (e.g. ECM with dative case or D.c.I.). Furthermore, Bader et.al. (1996) observe that accusative is the default case assigned to objects. One-place predicates like schlafen (‘sleep’) may select so-called cognate object, which are always assigned accusative case. This case assignment seems to be ‘the result of the presence of a structurally defined [case] position’ (Bader et.al. 1996: 11):

In sum, there is a clear asymmetry between nominative and accusative on the one hand and dative on the other hand with respect to case movement/case assignment.

(vi)Free datives and multiple datives: German has so-called free datives , which can be inserted in sentences rather freely. Typically, they are interpreted as beneficiary or possessor, but very often they have more than one interpretation. The ambiguity of free datives is illustrated in (42).

[page 157↓]

In addition, the examples in (43) show that more than one dative object can be licensed within one clause. Multiple occurrences of dative objects might be rare but not ungrammatical. They seem to restricted semantically because each dative requires a different interpretation.

Free or multiple occurrences of accusative (or nominative) objects cannot be found in German. There exist only few verbs that select two accusative objects like lehren (‘teach’) or abfragen (‘test’). Multiple accusatives are limited to these exceptions, and they are not productive (anymore).

(vii)Extraction: Although intuitions are not always totally clear, accusative NPs are obviously more transparent for extraction than dative NPs. WH- and PP-extraction out of accusative objects is generally much better than extraction out of dative objects (cf. also Müller 1993 and Pafel 1996).¹⁹

(viii)Coherent infinitives: Some verbs selecting a zu-infinitive can form a morphosyntactic unit with its infinitival complement, which is called ‘coherent infinitive’, cf. Bech (1995/1957). The complex verb, which consists of the matrix verb and the zu-infinitive, seems to be one single verb in syntax and thus assigns only one nominative and one accusative (recall from (vi) that German does not have multiple nominatives and accusatives). Therefore, [page 158↓]verbs selecting a coherent infinitive are expected to be intransitive. They cannot assign accusative case to an additional complement, as is illustrated in (45.a). However, Haider (1988) shows that dative objects can occur in the matrix clause. Verbs that select a dative object are able to form a coherent infinitive, cf. (45.b).²⁰ This difference is further evidence for an analysis that distinguishes between structural and oblique case.

(ix)Sentence processing: Bader et.al. (1996) demonstrate in an off-line study and an ERP-experiment that processing difficulties arise if an object NP in clause-initial position which is not morphologically marked for case (i.e. ambiguous between accusative and dative case) must be assigned dative instead of accusative. A garden-path effect arises especially if the distance between the sentence-initial NP and the verb that assigns either accusative or dative is long enough. This is illustrated in (46).

The syntactic structures of both sentences in (46) do not crucially differ. Thus the garden-path effect in (46) is solely related to case assignment. The first sentence contains a verb that assigns accusative case and is much easier to parse than the second sentence, which contains a verb assigning dative. Bader et.al. argue that this difference can be explained on the basis of the same distinction that we draw: the parser prefers assignment of structural case over oblique case (in their terminology abstract case is preferred over lexical case). The lexicon must only be reaccessed in sentences like (46.b) which contain a NP which turns out to be assigned dative case. This results in a garden-path effect. The relevant assumptions from Bader et.al. are summarized in (47).²¹

[page 159↓]

All examples we discussed in sections 6.2.1 and 6.2.2 indicate that German makes a distinction between structural and oblique case. Moreover, we found evidence for an additional asymmetry between nominative and accusative case: nominative is ranked higher than accusative just as dative case is ranked higher than other oblique cases.²² This distinction supports the picture that we gave at the end of chapter 5.

The differences between accusative and dative objects can be explained under the assumption that (i) structural case is morphologically unmarked, whereas oblique case is morphologically marked and (ii) structural and oblique case differ in syntax. Analyses like those proposed in Fanselow (1995), Gallmann (1992), Sabel (1995), and Wegner (1991), who treat dative as structural case, predict that accusative and dative objects behave alike. The difference between structural and oblique case can be implemented in various ways. Müller (1993), for examples, argues for a derivational account. Dative objects move from their VP-internal Θ-position into the specifier of μP, a VP-shell, for case reasons. In this position they receive dative case. Unlike accusative case, dative case is not assigned in the VP-internal base position by the verb.

Müller claims that in German μP,Spec is a A’-position per definition. With this he can account for the A’-properties of dative objects and the asymmetry between structural and oblique case. One could, for example, argue that A-movement (i.e. passivization, middle formation, or tough-movement) is excluded for dative objects because this would lead to a violation of the principle of unambiguous binding, cf. Müller and Sternefeld (1993).²³ Müller’s analysis is based on Chomsky’s (1981) theory of binding. According to this theory, a dative reflexive pronoun must be c-commanded by its nominative or accusative antecedent. Recall from chapter 5 and section 6.2.2 (ii) above that nominative and accusative NPs asymmetrically bind dative NPs. In order to account for this asymmetry, Müller assumes that (dative) reflexive pronouns need not be case-assigned. A dative reflexive pronoun stays in its VP-internal base-position and can thus be bound by an accusative object, which c-commands the reflexive pronoun in this position, cf. structure (49). In sum, Müller’s analysis might account for the differences between accusative and dative objects. Nevertheless, we reject the assumption of μP for two reasons. First, the stipulation that (dative) reflexive pronouns need not be assigned case cannot be maintained. We already saw that oblique case forms are morphologically marked. Moreover, in the first and second person singular the dative form of reflexive [page 160↓]pronoun (mir, dir) clearly differs from the accuative form (mich, dich). Thus dative reflexive pronouns are morphologically case marked (at least in the first and second person). Besides, even reflexive pronouns are expected to be subject to the case filter. Second, Müller incorrectly predicts that free datives must not be bound by an accusative object, because they are inserted directly into μP,Spec. However, we already saw in chapter 6 that free datives can be bound by accusative objects. We repeat the example in (50).

Therefore we propose a different analysis of dative objects. One might argue that the phrasal category of dative NPs differs from the one of accusative and nominative NPs. Only the former project a KP. Nominative and accusative simply project NPs. Thus, oblique case forms are syntactically more complex, which might account for certain asymmetries between structural and oblique case forms. Apart from that, one could argue that only arguments which are linked to NPs can undergo argument structure alternations such as passivization, middle formation, and tough-movement. This line of argumentation can be found in Bittner and Hale (1996) and Bader et.al. (1996). Alternatively, one might argue that oblique all dative objects are adjuncts in syntax. As opposed to Müller (1993), Vogel and Steinbach (1998) propose that dative objects do not move into an A’-position but are base-generated as adjuncts. Note that these two alternative approaches do not exclude each other. Dative objects might be syntactic adjuncts because they are KPs. Vogel and Steinbach argue that dative objects are adjoined either to VP, to AgrOP or to AgrSP, cf. structure (51) below. The insertion into one of these positions depends on the constraints regulating word order in the German middle-field. The position of accusative objects (as well as nominative subjects) is structurally fixed. Accusative objects must check their (structural) case feature in the extended projection of the verb, cf. Vogel and Steinbach (1995 and 1998) for a more detailed discussion of this issue. In (51) TP and the heads of the functional projections are omitted.

(51) [_AgrSP (dat) [_AgrSPnom₁ [_AgrOP (dat) [_AgrOPacc₂ [_VP (dat) [_VP t₁[_V' t₂ V ]]]]

This analysis correctly predicts that dative objects cannot undergo ‘middle formation’. A-chain formation as defined in chapter 5 is impossible for dative objects because they are A’-elements in German. Dative reflexive pronouns cannot be bound by the subject in syntax and therefore the non-argument interpretation is excluded. Dative reflexive pronouns are always linked to a semantic argument of the verb and they are bound by a less oblique co-argument of the predicate, i.e. either the nominative subject or the accusative object. Reduction of the first argument and promotion of the second argument is generally restricted to arguments that are linked to structural case positions. Of course, a wide-ranging study of structural and oblique case forms and their morphological, syntactic, and semantic properties is necessary for a better understanding of the interaction of case and valency change/argument structure alternations.²⁴ [page 161↓]For the present study of middle voice in German the illustration that the distinction between structural and oblique case is independently motivated will be sufficient.

The first part of this chapter dealt with the difference between middle constructions and anticausatives. We argued that the implicit first argument can either be saturated or reduced. Then, we compared unaccusatives to impersonal middle constructions and showed that our analysis provides an indirect argument against a syntactic analysis of unaccusatives. In the second part we gave various empirical arguments in support of the distinction between structural and oblique case. Unlike nominative and accusative case, dative case is oblique and dative objects cannot undergo middle formation, because they are not syntactic arguments.

Footnotes and Endnotes

¹ Argument saturation does also apply to implicit objects/second arguments. In German implicit objects are not morphosyntactically marked, because they do not change the linking principles, as opposed to implicit subjects (the subject is linked to the first semantic argument of the predicate):

² Argument saturation also yields the passive interpretation in languages that have reflexive passives, cf. section 2.3 and chapter 7.

³ The German verb schneiden (‘cut’) is an interesting example.

⁴ 

In a case-based linking theory, we can state the following linking-principles for active sentences:

(i) a nominative subject is linked to the first argument of the verb

(ii) an accusative object is linked to the second argument of the verb

We refer the reader to Vogel (1998) for a more detailed discussion of this issue.

⁵ No problem arises in case both kinds of one-place predicates have the underlying representation (45.a).

⁶ For a purely lexical analysis of unaccusatives, cf. e.g. Haider (1985) or Wunderlich (1985). In contrast to these two authors, Grewendorf (1983 and 1989) argues that unaccusatives and unergatives also differ in syntax. Fagan (1992) argues against a syntactic and a lexical distinction between these two types of one-place predicates. Recall from section 2.1 that unaccusatives differ in certain respects from unergatives, although the diagnostics do not always yield a clear classification. Unaccusatives typically have proto-patient entailments for their subjects. Unergatives have more proto-agent entailments for their subjects. These differences are semantic and at least for German there is no clear evidence for an additional syntactic distinction between unergative and unaccusative one-place predicates.

⁷ Additionally, dative and accusative reflexive pronouns are assigned case by a preposition are excluded from middle formation, because prepositional objects are also oblique in German, cf. also section 2.4.

⁸ 

Many German dialects have an even more restricted pattern. They distinguish only between unmarked (structural) and marked (oblique) forms. The following paradigm of Zurich German, a German dialect spoken in Switzerland, is from Cooper (1994:15). Swabian is almost identical to Zurich German in this respect.

(i)

⁹ Some nouns additionally distinguish between weak and strong forms. In this case, the masculine singular nominative form of the noun differs from the the corresponding accusative and dative forms. The noun Beamte (‘civil servant’) is one example. The nominative singular form is der Beamte/ein Beamter, whereas the accusative and dative form is den Beamten/einen Beamten and dem Beamten/einem Beamten. Besides, the old dative form of masculine and neuter singular contains the nominal affix –e, i.e. dem Manne and dem Kinde.

¹⁰ Free relatives are always grammatical when both the matrix and the embedded case are identical.

¹¹ Such case conflicts are resolved quite differently in different languages. For detailed discussion see Vogel (2000).

¹² Note that the complex nominal is one phonological word, which only receives one primary accent.

¹³ Note that the postnominal genitive can also realize the agent role in (28.b’) and (28.c’).

¹⁴ Only the unmarked word order permits maximal focus spreading/projection, cf. e.g. Höhle (1982:126).

¹⁵ This constraint is rather weak. Therefore, all of the other intervening constraints must be neutralized. This holds especially for the constraint based on agentivity: the NP that refer to an agent precedes other constituents.

¹⁶ For a more detailed discussion cf. Reis (1987), Vogel and Steinbach (1998), Haider and Rosengren (1999), and Gärtner and Steinbach (2000).

¹⁷ Accusative pronouns in clause-initial position may refer also to inanimate entities provided the sentence is embedded in an appropriate context. Dative pronouns do not require this additional context, cf. Gärtner and Steinbach (1997) for further discussion.

¹⁸ Unlike impersonal middle constructions, ‘impersonal’ tough-movement constructions must not have an impersonal subject. In this respect they are like impersonal passives in German, cf. section 2.1.2.

¹⁹ This contrast may have to do with processing asymmetries, which are discussed below. A parser seems to prefer accusative objects over dative objects for the reconstruction of the fronted constituent.

²⁰ Scrambling the embedded object in front of the matrix subject is only possible with coherent infinitives. According to this criterion, both sentences in (45) are coherent constructions. Besides, coherent infinitives are also grammatical with two dative objects, cf. Vogel and Steinbach (1998).

²¹ The second assumption can be derived from the fact that every sentence that contains an accusative object also contains a subject. ECM-constructions may be the only exception. In ECM-constructions the subject of the embedded sentence is assigned accusative case. But even in this case, a subject assigned nominative is present in the matrix clause. Moreover, we saw in the previous section that in subject oriented languages like German implicit or reduced accusative objects do not require morphosyntactic marking whereas implicit or reduced subjects do (e.g. in passives, anticausatives, or middle constructions). Thus only the omission of the subject is morphosyntactically marked.

²² This asymmetry follows from the syntactic analysis of nominative and accusative case which is standard in GB-theory: nominative NPs always c-command accusative NPs. Note that in German finite verbs show agreement only with a nominative subject. Moreover, only nominative case seems to depend on the specification of tense. A [+finite] active clause obligatorily requires a nominative DP.

²³ The principle of unambiguous binding demands that every intermediate trace of a movement chain must be of the same type as the head of the chain, i.e. either A or A’.

²⁴ 

We are aware of the fact that this analysis of dative objects also requires a modified account to their semantics. In our framework dative objects receive a semantic interpretation like other oblique case forms as e.g. von-PPs (by-phrases) or mit-PPs (with-phrases, see e.g. Strigin 1995), which are also adjuncts in syntax. Many oblique forms can be linked to a semantic argument of the verb but they can also receive a non-argument interpretation. Structural case-positions (or VP-internal argument positions) are only one possibility of argument realization/linking. Besides, an explanation of the (verb-independent) semantics of free datives is independently needed. Possibly, this leads to a unified analysis of the semantics of dative (and other oblique) objects in general. Wegener (1985) observes that all datives seem to have some underspecified meaning in common, which can roughly be outlined as follows: the entity the dative object refers to seems to be ‘personally affected’ by the event/action. Consider the following two examples.

Similar examples are embedded V2-clauses and parentheticals in German. We argued in chapter 1 that embedded V2-clauses cannot be selected by the verb. Instead they are licensed by general semantic and pragmatic conditions. Steinbach (1999) argues that parenthetical constructions are yet another example for ‘non-standard’ argument linking, cf. also Vogel (1998) for further examles.