On the Relevance of Unambiguous Argument Identification1 Judith M. Kainhofer Abstract This article is a contribution to the still controversial issue of the appropriate characterization of the syntactic structure of the Germanic OV languages. In Kainhofer (to-appear) the essentials of an approach to clause structure which regards the headedness of verbal projections, a parametrized directionality of licensing, and the availability of rich vs. poor case morphology as crucial for argument identification (licensing) as well as for the structural organization of verbal projections itself were discussed. There I proposed Unambiguity of argument identification as the relevant condition. This article now discusses some consequences of an approach to structure building respecting Unambiguity of argument identification. 1. Introduction There is an ongoing controversy about the syntactic structure of the Germanic OV languages, especially German. Basically there are two main positions to be distinguished: On the one hand there are those assuming a uniform syntactic organization of all languages (to a large extent relying on Kayne's 1994 Linear Corresponding Axiom) irrespective of their headedness (VO, OV) on the surface (cf. Chomsky's influential work and research based on it). Here it is argued that the basic syntactic organization is uniformly CP-TP-VP(or vP) across languages. This is in fact the predominant position in the recent syntactic research literature, especially in minimalist approaches. On the other hand, however, there are approaches which take the surface difference in headedness as symptomatic of a principled difference in the syntactic organization of clause structure. Haider (1993, 1997, 2004, 2005) is one of them. The main argument for this latter position comes from the fact that the syntactic behaviour of the Germanic OV languages does not coincide with that of the Germanic VO languages in important respects (cf. Haider's work for discussion). In this paper I rely on a minimalist approach to syntactic structure, I assume a minimalist framework assuming phases (cf. Chomsky 2001a). In more concrete terms, I take as a basis the common assumption that the clausal architecture consists of (at least) a CP layer, a TP layer, and a VP layer (contra Haider who takes this as a fundamental property of VO languages only). 2 Furthermore I assume that syntactic feature matching takes place in agreement relations (Agree) between functional heads (probes) and nominal elements (goals). However, I do not follow the assumption that Kayne's (1994) Linear Correspondence Axiom should be an integral part of the minimalist approach. Rather, I assume, following Haider (1993, 1997, 2004, 2005), that the Germanic OV languages are base-generated, i.e. underlyingly, OV. 1 This research has been partly supported by the DOC program (grant #21397) of the Austrian Academy of Sciences. 2 For sake of simplicity I assume a simple tripartition of the clausal architecture into CP-TP-vP/VP here. It might, of course, be the case that the CP and TP layer are actually organized in a more fine-grained fashion (cf. splitCP, split-IP approaches). In Kainhofer (to-appear) I have outlined the essentials of an approach to clause structure which allows to capture the different ordering patterns of weak object pronouns (in the sense of Cardinaletti & Starke 1994, 1996, etc.) in relation to non-pronominal subjects in Dutch and German in a principled way. I argued that the differences are not due to different landing positions of weak object pronouns, but they are symptomatic of a more principled difference in the organization of the clause structure of these languages: German and Dutch differ with respect to the layering of their verbal projection which results in different distributional patterns of the weak object pronouns in the two languages. Although the approach was developed with the example of the distribution of weak object pronouns in Dutch and German it is not meant to be a theory of pronoun distribution, however. Rather it is a formalization of the well-known observation that there seems to be some correlation between the existence (or non-existence) of case morphology and the variability (or non-variability) of word order (cf. Haider & Rosengren 2003: 247: the "lack of overt case in Dutch necessitates a positional system of identification, arguments must keep their relative order", see Neeleman & Weerman 1999: 78 for Dutch).3 In the following section I will give a short outline of the essentials of this proposal. 2. The Proposal of Kainhofer (to-appear): Unambiguity of Argument Identification and its Relevance for the Internal Organization of Verb Phrases Let us start with come background assumptions. As already mentioned, I assume, following Haider, that the Germanic OV languages are underlyingly, i.e. base generated, OV (contra e.g. Kayne 1994, 1998).4 Haider has argued in a number of papers that right-headed (OV) languages are less complex in their organization than left-headed (VO) languages for principled reasons. As reasons he detects headedness of the verb phrase in combination with a parametrized directionality of licensing (left-to-right, i.e. , in VO languages vs. right-to-left, i.e. , in OV languages). As a consequence, in OV languages all arguments are within the directional licensing domain of the verb and can, therefore, all be directionally licensed from the base position of the verb. Consequently, only one verbal head position needs to be instantiated in OV languages in order to guarantee directional licensing of the arguments. Head-initial projections (VO), on the other hand, require a functional identifier for the pre-head argument in the verb phrase (due to directionality of licensing). The verb phrase structures as proposed in Haider (2004, 2005) are illustrated in (1a) for head-initial verb phrases, and in (1b) for head final ones.5 Note that Haider uses the GB framework, not a minimalist approach to syntactic theory. 3 It has generally been observed in the literature that the presence of morphological case allows nominal constituents to occur in a relatively free order within the clause. One source for this observation is the fact that the loss of morphological case systems and the loss of free word order generally seems to be closely linked in diachronic developments (cf. for example, Meillet 1921; Sapir 1921; Jespersen 1922 for such observations in traditional work). A non-accidental correlation between case morphology and word order is also assumed in more recent work, as for example Hawkins (1986); Baker (1996); Haider & Rosengren (1998, 2003); Haeberli (2002), among others. 4 See, however, Vicente (2004) for argumentation that Basque constructs its OV order in a different way. 5 According to Haider, the different structures for VO and OV languages are due to argument identification (licensing) for which he regards the following factors as crucial: (i) endocentricity (V° is merged in the deepest position of VP), (ii) directionality of licensing (the canonical licensing direction is parametrized, symbolized by an arrow: VO: vs. OV: ; cf. also Baker 2001 and literature cited there for the claim that phrasal heads are (1) (a) structure of head initial verb phrases according to Haider (2004): VP 2 V° VP < > 2 XP V' <x> 2 vi° YP <x,y> (b) structure of head final verb phrases according to Haider (2004): VP 2 XP V' <x> 2 YP V° <x,y> The structures I am proposing for transitive verb phrases in the Germanic OV languages do not fully coincide with (1b). Rather, I argue that (1b) is the appropriate verb phrase structure for German only, however not for Dutch. This is put down to a condition on argument identification (Unambiguity of argument identification) which is sensitive to morphosyntactic case features. Furthermore, I assume that argument identification is a mechanism distinct from case checking (matching) (cf. Marantz 1991; Harley 1995; Sigurðsson 2003; McFadden 2004, among others, for argumentations that case and licensing should be separated). With respect to the syntax of weak pronouns I assume that weak object pronouns are basegenerated (first merged) in the canonical base positions of objects (i.e. vP/VP-internally) and undergo a movement operation (weak pronoun movement). This movement operation applies freely in the sense that it is not driven by syntactic feature-checking (cf. Chomsky 2001b: internal/external Merge comes "free"), and it is evaluated at the higher phase level. The motivation (or trigger) for weak pronoun movement has to be sought among the properties of information structure (possibly background, old information, topicality, or definiteness; cf. e.g. Lenerz 1977; Haider & Rosengren 1998 for proposals along these lines). Structurally, weak pronoun movement is either adjunction to the lexical projection of the verb (cf. scrambling, if considered in GB terms; cf. Haider & Rosengren 1998), or, in minimalist terms, movement to the edge (outer specifier) of the verb phrase. Importantly, however, it is not movement to a functional specifier position. Evidence for this conclusion comes from non-opacity for extraction as well as from topicalization (cf. Kainhofer 2004, in-prep.). As illustrated in (3a), weak object pronouns cannot precede the subject of a transitive clause in Dutch, while this ordering is unproblematic in German, (2a). With unaccusative or passive clauses, on the other hand, weak object pronouns can precede the subject in Dutch, (4), as parametrized for their canonical directionality), (iii) mutual minimal c-command as a general condition on identification of selected elements, and, additionally, (vi) asymmetric merger (merge is always to the left, cf. Haider's 1993 Branching Constraint). For reasons of space, the reader is referred to Haider's work for details. well as in German (2b-c). That is, there seems to be an effect of verb (phrase) type in Dutch. The weak object pronouns are set in bold type in (2)-(4). (2) (a) dass sie der Jan gestern geküsst hat that her-acc.weak the Jan-nom. yesterday kissed has 'that Jan kissed her yesterday' (b) dass ihr Unglücke zustießen that her-acc.weak calamities-nom. happened 'that calamities happened to her' (c) dass ihr das/ein Buch gegeben wurde that her-acc.weak the/a book-nom. given was 'that the/a book was given to her' (transitive) (unaccusative) (passive) (3) (a) *dat 'r Jan gisteren gekust heeft (b) dat Jan 'r gisteren gekust heft 'that Jan kissed her yesterday' (transitive) (4) (a) dat 'r rampen overkwamen dat rampen 'r overkwamen 'that calamities happened to her' (unaccusative) (b) dat 'r deze ramp overkwam dat deze ramp 'r overkwam 'that this calamity happened to her' (c) dat 'r het/een boek gegeven worden dat het/een boek 'r gegeven worden 'that the/a book was given to her' (unaccusative) (passive) In short, I have claimed in Kainhofer (to-appear) that the relative ordering of subject DPs and weak object pronouns in the Germanic OV languages is a function of the specific structural organisation of the verb phrase, i.e. it reflects the organization of the base projection of the verb in the respective language. Stated differently, differences in the distribution of weak object pronouns in relation to non-pronominal subjects in German and Dutch reflect the different status of transitive verb phrases in the two languages: non-layered in German vs. layered in Dutch. As already indicated, I basically follow Haider's reasoning concerning argument identification, cf. (1b) above. Most importantly, I refer to directionality of argument identification (VO vs. OV). In some important respects, however, I depart from Haider's approach. According to Haider's proposal to argument identification (as summarized in the tree structure in (1b) above) right-headed languages are predicted to behave uniformly with respect to the structural projection of clause structure. In more concrete terms, they are expected to behave uniformly with respect to syntactic phenomena which are structure-sensitive. According to Haider, all identification of nominal and pronominal arguments is done solely within the verbal projection, no licensing form outside the verb phrase (i.e. from a higher functional head) is needed in order for the arguments to be directionally licensed. In VO languages, on the contrary, the subject of a transitive verb (which is merged in the specifier of the verb phrase, be it VP as in GB times or vP as in minimalism) will never be included in the directional licensing domain of a verbal head within the verb phrase, given the fact that the licensing direction is uniformly in VO languages. In order for the subject argument to be directionally licensed, licensing from a verbal head position outside the verb phrase (TP) is necessary. This is not the case in OV languages where all arguments are within the directional licensing domain of the verbal head in each and every instantiation of a verb phrase (given licensing proceeds in right-headed languages). Haider's approach in the form outlined above does not predict a different syntactic structure for German and Dutch. Nevertheless, Haider & Rosengren (2003: 247) state that the "lack of overt case in Dutch necessitates a positional system of identification; consequently, arguments must keep their relative order (see Neeleman & Weerman, 1999: 78 for Dutch)". Even though they may be right in assuming this correlation, this does not follow from Haider's structure building mechanism, nor from his proposal to argument identification. Therefore, I suggest an additional factor which allows for a differentiation between German and Dutch, viz. the relevance of morpho-syntactic case features for argument licensing. This means that the morpho-syntactic nature of a language (rich vs. poor morphological case) 6 has important implications for the organization of its syntactic architecture. To put it in more concrete terms, I argue that languages with rich case morphology own a subfeature, which I call "m-mark" for expository purposes, on the uninterpretable case feature uCase of DPs.7 "m-mark" is short for "to be marked morphologically", and it is (per stipulation) present on each DP in languages which show rich case morphology. That is, m-mark is nothing more than the explicit inclusion of the parameter deciding between languages using rich morphological case-marking and languages which don't.8 (7) below shows the feature structure of DPs in a language with rich morphological case like German. The feature structure of DPs in languages with poor case morphology like Dutch, on the other hand, does not include the subfeature m-mark, cf. (9) below. The relevance of the subfeature m-mark, however, goes far beyond controlling morphological case marking only. It is a decisive factor in argument identification and structure building as well. The presence of m-mark on uCase in the feature structure of DPs has important syntactic consequences inasmuch as it interacts with structure building, as I argue. The relevant condition which controls argument identification is Unambiguity of argument identification, (5), as I claim. This condition needs to be added as a further factor which is crucial for argument identification in addition to the factors noted by Haider (2004, 2005) (in 6 Note that there is some disagreement in the literature as to how to establish the distinction between rich and poor morphology, cf. Haeberli 2002; Müller 2005 for some recent proposals. 7 Alternatively, m-mark could be an interpretable feature of DPs of its own instead of being a subfeature of uCase. It is not clear, however, how to capture the connection between m-mark and the case feature (uCase) in that case. 8 Therefore, m-mark is still rather descriptive in nature, and further research is needed in order to clarify its theoretical status. Up to that point, however, m-mark is used as a technical device included in order to avoid invoking look-ahead. particular, directionality of licensing). By including condition (5) the differences between German and Dutch fall out directly. (5) Unambiguity of argument identification: Arguments must be unambiguously identified within the minimal structural projection domain of their first merge.9 "Unambiguously identified within the minimal structural projection domain of first merge" is defined as in (6).10,11 (6) An argument is unambiguously identified within the minimal structural projection domain of its first merge iff (i) and (ii) holds: (i) it is within the directional licensing domain of the verbal head, and (ii) it is unambiguously identifiable, which means that (ii.a) and/or (ii.b) holds: (ii.a) it is identified by virtue of its morpho-syntactic features (i.e. by the presence of m-mark on uCase) and/or (ii.b) it is the only argument within the minimal structural projection domain of its first merge. The licensing of argumental DPs is a two-step process which consists of (i) the identification of the argument in its position of first merge. Argument identification applies simultaneously with external (first) merge and inspects the minimal structural domain for unambiguity of argument identification (i.e. for the presence of subfeature m-mark on uCase of DPs). As a second step, (ii) the case feature is licensed (matched) in a probe-goal relation with a functional head. The crucial point differentiating between German and Dutch is (6ii). In Dutch DPs do not carry a subfeature m-mark in their feature structure, therefore (6ii.a) does not apply. In order for Unambiguity of argument identification to be respected nonetheless, therefore, every argument has to constitute the only argument within the minimal structural projection domain of its first merge (= (6ii.b)). This requirement is only fulfilled if there is one minimal structural projection domain of first merge per argument. Therefore, Dutch is forced to project a layered verb phrase structure which distinguishes the arguments on purely structural grounds. 9 The "minimal structural projection domain of first merge" is to be defined as limited by the next higher verbal head position. The proper formal formulation, however, is not worked out yet. This will be done in Kainhofer (in-prep.), however. 10 Note, however, that this is only a preliminary definition which is in need of reformulation. A modification is essential because the present formulation does not capture, for example, subjects of transitive verbs in VO languages. The inclusion of VO languages, however, lies beyond the scope of this paper. The technical details of the mechanism and its implementation are not yet fully worked out. At present it is only a more technical formulation of a descriptive statement and functions as a guiding line for future research. 11 In languages with rich case morphology like German (6ii.a) and (6ii.b) are simultaneously fulfilled in intransitive verb phrases. 2.1. German: Non-layered Transitive Verb Phrases (vP/VP) German, on the other hand, exhibits rich morphological case which means that m-mark is present as a subfeature of uCase on every DP in German. The feature structure of DPs receiving structural case in German is illustrated in (7). (7) feature structure of DPs receiving structural case in German: DP -features (i.e. number, person, gender) uCase[ ] m-mark In German, therefore, arguments are unambiguously identifiable by virtue of their morphosyntactic features according to (6ii.a) above. Consequently, no structural differentiation of subject and object(s) is necessary; therefore, the verbal projection is non-layered for reasons of economy, (8). (8) internal structure of the transitive verb phrase in German: … vP/VP 2 subj 2 obj V/v Notice, however, that the structural organization of (8) coincides with the structure Haider (2004, 2005) proposes for OV languages in general. The basic reasoning underlying here is the following: Transitive V and v are basically nondistinct, that is they are of the same category and transitive V includes v (indicated by the notion "V/v"). V and v are split up only when necessary for argument identification, in accordance with condition (5) above. Whenever Unambiguity of argument identification can be established without distinguishing different structural domain by splitting V/v into two separate layers, on the other hand, a splitting is avoided for reasons of economy (as in the German case). In VO languages, transitive verbal projection obligatorily show a layered organization due to their directionality of licensing (). In OV languages, on the other hand, it depends on the feature structure of DPs, with the crucial factor being the presence vs. absence of the subfeature m-mark on the case feature of DPs. The mechanism is rather simple: While the verb phrase is projected, argument identification in accordance with condition (5) proceeds simultaneously. 2.2. Dutch: Layered Transitive Verb Phrases (vP-VP) Dutch differs from German in only showing poor morphological Case. 12 Therefore, arguments are not unambiguously identifiable by virtue of their morpho-syntactic features due to the absence of the subfeature m-mark on uCase in Dutch. (9) illustrates the feature structure of DPs receiving structural case in Dutch. (9) feature structure of DPs receiving structural case in Dutch: DP -features (i.e. number, person, gender) uCase[ ] By parity of reasoning, therefore, a vP-VP layering is forced by the need to unambiguously identify subject and object(s) in the absence of rich morphological case by distinguishing different structural domains (due to condition (6ii.b) above). This leads to a vP-VP structure for transitive verb phrases in Dutch, as illustrated in (10). If V and v are indeed non-distinct as assumed above one would expect the vP layer to be rightheaded in OV languages (Hubert Haider, p.c.). See Kainhofer (in-prep.) for discussion. (8) internal structure of the transitive verb phrase in Dutch: ... vP 2 subj uCase 2 obj uCase 2 VP v (+ verb clustering) V Recall that movement of weak object pronouns targets the edge of the verb phrase (i.e. it is movement to the outer specifier of the verb phrase or adjunction to the verb phrase). Once the weak object pronoun has moved to the edge of vP it is within the minimal structural projection domain of the subject argument, thereby violating Unambiguity of argument identification. This time, however, no further layering of the transitive verb phrase occurs because the weak object pronoun is remerged in this position rather than first merged (Recall the formulation of Unambiguity of argument identification in terms of first merge!). Weak object pronouns, however, do not precede the subject of a transitive verb in Dutch. Therefore, I hypothesized in Kainhofer (to-appear) that there is an obligatory EPP-feature on T for structural reasons, whenever there is a structural distinction between subject and object(s) within the verb phrase (i.e. a distinction by layering) because of the need to distinguish subject and object structurally.13 The EPP feature on T is checked by movement of the subject DP to the specifier of TP rather than by merge of a subject expletive (despite the familiar preference for merge over move). This is due to the fact that there is no subject expletive in Dutch (cf. Bennis 1986; Koeneman 2000 for argumentation that Dutch er is 12 13 See Kainhofer (in-prep., 2006) for discussion. The exact formalism of the mechanism has not been clarified yet. adverbial in nature and does not occupy spec,TP).14 That is, informally spoken, the language sticks to the strategy to distinguish subject and object structurally. After raising of the transitive subject to the specifier of TP subject and object(s) again occupy different structural domains despite the application of weak pronoun movement.15 For reasons of space we cannot go into the details here. Note, however, that the clarification of various points is still outstanding, so additional research is needed. 3. On some Consequences of Unambiguity of Argument Identification Modifying the syntactic architecture of the verb phrase in the way discussed in section 2 has important consequences for the organization of the syntactic system, of course. This approach to the clause structure of OV languages makes a number of predictions, both empirical as well as theory-internal ones. In the following subsections we will comment on some of them, however, mainly discussing circumstantial evidence only. As far as empirical phenomena are concerned, our approach predicts for example that Dutch and German should not behave alike with respect to phenomena which are correlated to the existence of a vP layer in transitive clauses. In subsection §6.1 we will discuss evidence for the different behaviour of Dutch and German with respect to one of these properties. Our treatment of Dutch presupposes that the parameter distinguishing between languages with rich vs. poor case morphology is set negatively throughout in Dutch. That is we have to guarantee that it is justified to assume "poor morphological case" even for the pronoun system of Dutch which shows some case marking. This is discussed in Kainhofer (2006) where it is argued that the pronoun system of Dutch is by far not rich enough to be classified as "rich case morphology". Supporting circumstantial evidence comes from the diachrony of Dutch; this will be discussed in subsection §6.2. below. In subsection §6.3. we will then be concerned with a theory-internal problem, viz. the question of how accusative matching works if there is no vP layer in German. Of course, additional predictions will have to be tested in order for the approach to argument identification and clausal architecture advocated in this paper to be justified indeed. At the current stage of research many questions are still unresolved. 3.1. On the Phase Status of Transitive Verb Phrases in German and Dutch In this section I am going to discuss one of the crucial consequences of this kind within a minimalist framework assuming phases (following Chomsky 2001a). The proposal outlined in the previous sections makes the testable prediction that properties associated with the phase status of vP should not turn up in German since there the vP layer is absent as I claim. If the vP-layer is indeed crucial for phase status, the transitive verb phrase in German should not be a (strong) phase, and, as a consequence, not exhibit (strong) phase 14 The existence of Transitive Expletive Constructions (TECs) in Dutch must receive a different explanation. Importantly, however, even in TECs weak object pronouns do not precede a subject DP. 15 Possibly, a formulation in terms of an obligatory EPP feature on T is not the best way of speaking; rather, what is needed might be a condition which is to some extent similar to Müller's (2000) Shape Conservation, though they would have to differ in important respects. The exact nature of this condition has to be left open for future research. properties. In Dutch, on the other hand, we should find evidence for phase properties of the vP. This is indeed borne out as the difference between the data from Dutch in (11) and German in (12) illustrate. Unfortunately, however, the absence of evidence for phase properties is not a proof of nonlayered transitive verb projections in German, but rather only circumstantial evidence. In the minimalist literature assuming phases (following Chomsky 2001a), a whole range of properties have been attributed to the phase status of vP: for example, reconstruction to the edge of vP, providing a target for quantifier raising, licensing of parasitic gaps, and isoliability of PF, among others. However, many of those properties are controversial and problematic from a conceptual and/or empirical point of view (cf. especially the discussion in Boeckx & Grohmann 2004 and references cited there). Therefore, they cannot be used as decisive evidence for our concerns. Fortunately, there is one property of (strong) phases which seems to be uncontroversial, however. This is stranding under successive-cyclic movement through phase edges. Stranded elements are taken to be evidence for intermediate landing sites of successive-cyclic movement through phase edges because in the minimalist phase model successive-cyclic movement is assumed to targets strong phases (CP, vP) only. In his (1995) dissertation, Barbiers presents evidence for focus particles occurring at the edge of vP in Dutch which must have been stranded under successive-cyclic wh-movement. The relevant examples are replicated in (11a,b). What is crucial here is the fact that they are ambiguous in interpretation. The focus particle maar 'just' surfaces within the main clause, which is particularly obvious in (11b), nevertheless it can be interpreted as part of the topicalized DP (reading 1). This is the stranding reading, i.e. maar must have been stranded in the specifier of vP of the matrix clause in the course of successive-cyclic movement of maar twee vogels 'just two birds' through phase edges on its way up to its topicalization site within the CP domain of the matrix clause. (11) (a) [TWEE vogels]i zei Jan maar [CP dat ie tigezien had] two birds said John just-Focprt that he seen had reading 1: 'John said that he had seen just two birds' reading 2: 'John just said that he had seen two birds' (Barbiers 1995: 84) (b) [TWEE vogels]i had Jan maar gezegd [CP dat ie ti gezien had] two birds had John just-Focprt said that he seen had (Barbiers 1995: 85) reading 1: 'John had said that he had seen just two birds' reading 2: 'John had just said that he had seen two birds' In German, on the other hand, no such ambiguity arises. The German focus particle nur 'just' cannot be interpreted as part of the topicalized DP, (12); that is reading 1 is unavailable in German. Note that nur precedes the matrix clause subject Jan in (12), in contrast to the ordering in Dutch. This is due to the fact that subjects do not move to a intermediate subject position (spec,TP) in German. Rather, they stay in their base position within the verb phrase (i.e. German is [-EPP] in the traditional sense). For independent reasons even reading 2 is unavailable here. The only possible reading of (12) is reading 3. This is the non-stranded reading, however. (12) [ZWEI Vögel]i hat nur Jan gesagt [CP dass er ti gesehen hatte] two birds had just-Focprt John said that he seen had (Barbiers 1995: 84-85) *reading 1: 'John had said that he had seen just two birds' *reading 2: 'Jan had just said that he had seen two birds' reading 3: 'Only Jan had said that he had seen two birds' To conclude, for German there is no positive evidence from the stranding of focus particles that there should be a vP phase. As mentioned before, however, this cannot be taken as a proof that the vP layer is absent in German. Rather it is only a piece of circumstantial evidence which points into that very direction, at best. 3.2. Supporting Circumstantial Evidence from Diachrony: Shannon (2000) on Dutch Shannon (2000) offers a study dealing with the ordering of nominal and pronominal arguments in the so-called midfield in Dutch and German. He basically examines the word order frequencies in a corpus of modern German and Dutch prose texts, comparing and contrasting the two languages. From a synchronic perspective, his findings once again confirm the descriptive statements reported in the literature: The unmarked ordering of clause-internal nominal subjects and pronominal objects is the pronominal object preceding the nominal subject in German, while we find the reverse situation in Dutch. Shannon, then, compares his results for the ordering of object pronouns in relation to nonpronominal subjects for the modern languages with 16th century texts, the Middle Dutch and Early New High German versions of the Eulenspiegel/Uhlenspieghel. To briefly sketch the overall picture: In the Middle Dutch Ulenspieghel, preposed object pronouns predominate. Consider the tables 1 to 3, which illustrate the frequencies (repeated from Shannon 2000: 174, 175, and 152); those which are particularly relevant for our concerns are set in bold type. Middle Dutch Modern Dutch N-subject + P-object 16 (18.2%) 58 (85%) P-object + N-subject 72 (81.8%) 10 (15%) Total 88 (100%) 68 (100%) Table 1: Order of nominal subject and pronominal object in the Middle Dutch Ulenspieghel (n = 88) and its modern Dutch translation (n = 68) Reflexive Personal Demonstrative Total N-subject + P-object 4 (20%) 7 (11.5%) 5 (71.4%) 16 (18.2%) P-object + N-subject 16 (80%) 54 (88.5%) 2 (81.8%) 72 (81.8%) Total 20 (100%) 61 (100%) 7 (100%) 88 (100%) Table 2: Order of nominal subject and pronominal object in the Middle Dutch Ulenspieghel (n = 88) by pronoun type (definite pronouns only) Pronoun Type Reflexive Personal Demonstrative Indefinite Negative Total N-subject + P-object 130 (85%) 221 (90.6%) 15 (93.75%) 39 (100%) 15 (100%) 420 (89.9%) P-object + N-subject 23 (15%) 23 (9.4%) 1 (6.25%) 0 (0%) 0 (0%) 47 (10.1%) Total 153 (100%) 244 (100%) 16 (100%) 39 (100%) 15 (100%) 88 (100%) Table 3: Word order frequencies in original Modern Dutch works (n = 467) by pronoun type Note that in Middle Dutch, the inflectional morphology was considerably richer than it is in modern Dutch.16 During the Middle Dutch period we observe a decline of the case system of Dutch: distinct case endings collapsed or disappeared. This decline of pronoun case markers, together with the complete loss of morphological case in nouns increased dramatically the number of instances where ambiguity could arise as to what is subject and what object. We cannot go into the details here, see Shannon (2000) and Kainhofer (2006) for discussion. Shannon's (2000) findings pattern nicely with what we expect for languages with rich case morphology: According to our argumentation in the previous sections, languages with rich case morphology own a sub-feature m-mark which is represented in the feature-matrix of the DPs. According to Unambiguity of argument identification, an additional vP layer is only projected when necessary for unambiguous argument identification, i.e. the vP/VP is only split up into two separate layers when unavoidable. Middle Dutch, therefore, does not split up its vP/VP layer since arguments can be unambiguously identified due to their sub-feature mmark; they do not have to be structurally differentiated. For Middle Dutch, thus, we expect object pronouns to be able to precede a nominal subject, just as in German. This prediction is borne out, as it seems. Some reservations are in place, however. One has to be careful in generalizing Shannon's (2000) findings for Middle Dutch since they are not extracted from a representative sample of Middle Dutch texts but rather based on a single text only: the Middle Dutch Ulenspieghel which comprises no more than 88 relevant examples. However, Shannon (2000) refers to a paper which was in the process of formation (Shannon in prep.) arguing that preliminary examination of other Middle Dutch texts points into the same direction. A second point is more serious, however. Unfortunately, Shannon (2000) does not distinguish between strong and weak forms of object pronouns in his analysis, nor does he distinguish different verb phrase types. Therefore, his findings are only weak circumstantial evidence for our concerns, at best. 3.3. How can Accusative be Assigned (Matched) if there is no vP Layer in German? Among the theory-internal consequence of our proposal there is a rather crucial and farreaching one. We have argued that German completely lacks any vP layer due to the fact that 16 Note that we are only concerned with the richness of case morphology; we do not make any claims about the richness of the verbal morphology of Dutch and German. unambiguous argument identification in German works without distinguishing different structural domains within the transitive verb projection. In recent minimalism it is the standard approach, however, that v crucially participates in case matching: it probes for a DP to match accusative case. If there is no vP layer in German it stays mysterious how accusative matching should work. So, we have to comment on the question how accusative can ever be assigned (matched) to the object in German in our approach. It is standardly assumed within recent minimalist work that accusative on DPs is universally matched by establishing an Agree relation between the v head (probe) and the respective DP (goal). As soon as one takes the step of questioning the universality of the presence of a vP layer, however, standard minimalist reasoning concerning matching of the accusative does not work any more, at least not for those languages which do not project a vP layer as we have claimed for German. On first sight, our approach seems to predict that there should be no accusative on objects of transitive verbs in German at all. This is obviously not true, however. (13a) shows a regular transitive clause with nominative on the subject and accusative on the object DP. In the passive construction, (13b), the subject of the clause (i.e. the argument corresponding to the object in the transitive clause) shows up in the nominative. This behaviour indicates that the accusative on the object DP in the (a) sentence is indeed a structural case. (13) (a) Peter las den Brief. Peter-nom. read-impf. the letter-acc. 'Peter read the letter' (b) Der Brief wurde gelesen. the letter-nom was-impf. read-pple 'the letter was read' This does not necessarily mean that our approach is on the wrong track, however, since there is an alternative available; and this is what we would like to propose: The structural accusative is not matched by the v head generally, but rather it is matched by establishing an Agree relation with the next higher functional head available. In languages which possess a vP layer the accusative is matched in an Agree relation between the v head and the DP because v is the next higher functional head. If there is no vP layer, like in German, the accusative is matched in an Agree relation between the next higher functional head (T) and the DP. T, then, is in an Agree relation with both the subject DP as well as the object DP; that is, it probes "all the way down" within its locality domain (phase). This presupposes that matching features of the probe T are not immediately set inactive and deleted from the computation as soon as they are valued but, rather, stay active until all Agree relations within the locality domain (phase) have been established. T does not immediately "assign" nominative and accusative to subject and object, respectively, but rather it "assigns" structural case which is later on interpreted as nominative and accusative by the morphological component. Consequently, the relation between the T head and nominative and the v head and accusative is not an original one but rather follows from the syntactic architecture of the respective language. As for the technical details of a mechanism of multiple case matching by a single probe, a promising suggestion has been made by Hiraiwa (2001, forthcoming). Hiraiwa (2001, forthcoming) suggested the inclusion of an operation called Multiple Agree into the minimalist framework, (14). In Chomsky (2004), Multiple Agree is mentioned as an integral part of the computational system of natural languages. (14) Multiple Agree (Hiraiwa 2001: 69) MULTIPLE AGREE (multiple feature checking) with a single probe is a single simultaneous syntactic operation; AGREE applies to all the matched goals at the same derivational point derivationally simultaneously. Hiraiwa invented Multiple Agree mainly for to capture the intricacies of the Icelandic agreement system (see Hiraiwa forthcoming for discussion). The examples in (15a-c) show transitive expletive constructions with dat.-nom. raising complement clauses from Icelandic (they are repeated from Hiraiwa forthcoming: 48 and are originally due to Hróarsdóttir, p.c.). (15) (a) það virðist/*virðast einhverjum stúdent lika hestarnir expl seem-dflt./*-pl. some student-dat.sg. to-like horses-def.nom.pl. 'It seems that some student likes the horses' (b) það virðist/virðast mörgum stúdentum lika hestarnir expl seem-dflt./-pl. many students-dat.pl. to-like horses-def.nom.pl. 'It seems that many students like the horses' (c) það virðist/*virðast mörgum stúdentum lika hestar expl seem-dflt./*-pl. many students-dat.pl. to-like horse-nom.sg. 'It seems that many students like a horse' Hiraiwa's line of reasoning is along the following lines: T is in an agreement relation with both the subject as well as the object (Multiple Agree) (with T a probe by inheritance from C). Both the subject and the object participate in establishing the actual morphological form of the finite verb (verbal inflectional morphology; i.e. valuation of u-features of the probe), resulting in default agreement on the finite verb when subject and object do no coincide with respect to their number feature. Note that agreement in Icelandic is nominative-controlled in general. In parallel fashion, we can make use of Multiple Agree for establishing agreement relations between T and the subject as well as the object DP in German transitive clauses: Since there is no vP layer in German, as we propose, feature matching starts as soon as T is merged. T establishes an Agree relation with both DPs within its locality domain (phase) and values their features. Note, however, that German does not show comparable overt reflexes of (failure of) multiple agreement relations (Multiple Agree) in the verbal agreement system. Again, the details are not fully clarified, yet; see Kainhofer (in-prep.) for a thorough discussion. References Baker, Mark. (1996). The Polysynthesis Parameter. Oxford: Oxford University Press. ---. (2001). The Atoms of Language: the Mind's Hidden Rules of Grammar. New York: Basic Books. Barbiers, Sjef. (1995). Syntax of Interpretation. Den Haag: Holland Academic Graphics. Bennis, Hans. (1986). Gaps and Dummis. Dordrecht: Foris. Boeckx, Cedric & Kleanthes Grohmann. (2004). Putting Phases into Perspective. Ms. Harvard University and University of Cyprus. Cardinaletti, Anna & Michal Starke. (1994). 'The typology of structural deficiency: on the three grammatical classes', University of Venice Working Papers in Linguistics 4: 41-109. ---. (1996). 'Deficient pronouns: a view from Germanic: a study in the unified description of Germanic and Romance', in H. Thráinsson, S. Epstein & St. Peter (edd.), Studies in Comparative Germanic Syntax. Dordrecht: Kluwer, 21-65. Chomsky, Noam. (2001a). 'Derivation by phase', in M. Kenstowicz (ed.), Ken Hale: a Life in Language. Cambridge: MIT Press, 1-52. ---. (2001b). 'Beyond explanatory adequacy', MIT Occasional Papers in Linguistics 20: 1-28. ---. (2004). On Phases. Ms. MIT. Haeberli, Eric. (2002). Features, Categories and the Syntax of A-Positions: Cross-linguistic Variation in the Germanic Languages. Dordrecht: Kluwer. Haider, Hubert. (1993). Deutsche Syntax – Generativ. Tübingen: Narr. ---. (1997). 'Projective economy: on the minimal functional structure of the German clause', in W. Abraham & E. van Gelderen (edd.), German: Syntactic Problems – Problematic Syntax. Tübingen: Niemeyer, 83-103. ---. (2004). Was es heißt, OV zu sein. Paper presented at the 32. Österreichische Linguistiktagung, Salzburg. ---. (2005). 'How to turn German into Icelandic – and derive the OV-VO contrasts', Journal of Comparative Germanic Linguistics 8: 1-56. Haider, Hubert & Inger Rosengren. (1998). 'Scrambling', Sprache und Pragmatik 49. ---. (2003). 'Scrambling: nontriggered chain formation in OV-languages', Journal of Germanic Linguistics 15: 203-267. Harley, Heidi. (1995). 'Abstracting away from abstract case', in J. Beckman (ed.), Proceedings of NELS 25. Amherst: GLSA, 207-221. Hawkins, J.A. (1986). A Comparative Typology of English and German. London : Croom Helm. Hiraiwa, Ken. (2001). 'Multiple agree and the defect intervention constraint', in O. Matushansky & E. Gurzoni (edd.), The Proceedings of the HUMIT2000, MIT Working Papers in Linguistics 40: 67-80. ---. (forthcoming). 'Dimensions of agreement' to appear in Cedric Boeckx (ed.): Complex Agreement Systems. Oxford. Jespersen, O. (1922). Language: Its Nature, Development and Origin. London: G. Allen and Unwin. Kainhofer, Judith. (2004). Über schwache Pronomina, Subjekte und Minimalismus: ein Blick auf das Deutsche und Niederländische. Paper presented at the 32. Österreichische Linguistiktagung, Salzburg. ---. (2006). Argumentidentifizierung und Strukturaufbau in den germanischen OV Sprachen. Paper presented at GeSuS 2006, Wrocław. ---. (in-prep.). Distribution of weak pronouns in the Continental West Germanic and North Germanic Languages. Doctoral dissertation. Salzburg. ---. (to-appear). 'On weak pronouns, subjects, and verbal projections. A view from continental West Germanic', in Proceedings of the 3rd Athens Postgraduate Conference in Linguistics. Kayne, Richard. (1994). The Antisymmetry of Syntax. Cambridge: MIT Press. ---. (1998). 'Overt vs. covert movement', Syntax 1: 128-191. Koeneman, Olaf. (2000). The Flexible Nature of Verb Movement. Ph.D.thesis. Utrecht: LOT Publications. Lenerz, Jürgen. (1977). Zur Abfolge nominaler Satzglieder im Deutschen. Tübingen: Narr. Marantz, Alec. (1991). 'Case and licensing', in G. Westphal, B. Ao & H.-R. Chae (edd.), Proceedings of ESCOL 91. Ithaca: CLC Publications, 234-253. Republished in Eric Reuland (ed.). (2000). Arguments and Case: explaining Burzio's Generalization. Philadelphia: John Benjamins, 11-30. McFadden, Thomas. (2004). The Position of Morphological Case in the Derivation: a Study on the Syntax-Morphology Interface. Ph.D.thesis. Philadelphia: University of Pennsylvania. Meillet, Antoine. (1921). Linguistique historique et linguistique générale. Paris: Librairie Ancienne Honoré Champion. Müller, Gereon. (2000). 'Shape conservation and remnant movement', in M. Hirotani, A. Coetzee, N. Hall & J.-Y. Kom (edd.), Proceedings of NELS 30. Amherst: GLSA Publications, 525-539. Revised and extended version appeared 2002 in A. Alexiadou, E. Anagnostopoulou, S. Barbiers & H.-M. Gärtner (edd.), Dimensions of Movement. Amsterdam: Benjamins, 209241. ---. (2005). Φ-feature Impoverishment in German Verb Inflection and the Theory of Pro-Drop. Paper presented at GGS, Tübingen. Neeleman, Ad & Fred Weerman. (1999). Flexible Syntax: A Theory of Case and Arguments. Dordrecht: Kluwer. Sapir, Edward. (1921). Language: An Introduction to the Study of Speech. New York: Harcourt, Brace and World. Shannon, Thomas F. (2000). 'On the order of (pro)nominal arguments in Dutch and German', in Th. F. Shannon & J. P. Snapper (edd.), The Berkeley Conference on Dutch Linguistics 1997. Lanham, MD.: University Press of America, 145-196. Sigurðsson, Halldór Ármann. (2003). 'Case: abstract vs. morphological', in E. Brandner & H. Zinsmeister (edd.), New Perspectives on Case Theory. Stanford: CSLI Publications, 223268. Vicente, Luis. (2004). 'Derived vs. base generated OV', Leiden Papers in Linguistics 1: 83-96.