Introduction

There is a broad concensus among researchers working within the paradigm of HPSG that argument-inheritance (Hinrichs and Nakazawa [3,4]) is an essential operation in the analysis of so-called verb clusters in German. An argument-inheritance verb subcategorizes for an unsaturated verbal complement, and for all the complements on the COMPS-list of this verbal complement. In German, auxiliaries and modals are often treated as argument-inheritance verbs.

Questions concerning the constituency of phrases headed by such argument-inheritance verbs, however, have not been answered uniformly. Most analyses of the German VP distinguish between the verb cluster and the verb phrase, where the verb cluster is a subconstituent of the VP containing verbal signs only. The internal structure of the verb cluster has been argued to be left-branching (Hinrichs and Nakazawa [3,4]), but arguments for a right-branching structure have been given as well [8]. The VP itself is sometimes analyzed as a contoured structure, in which a verbal projection (a partial VP) combines with one complement at a time (Hinrichs and Nakazawa [3,4], [8]). Nerbonne [12] and [1], on the other hand, argue for a `flat' VP in which a verb combines with all its complements at once, and in which, in Baker's case, the verb cluster has been eliminated as well.

In this paper, we explore how the account of Dutch verb clusters developed in [15] might carry over to German. For Dutch, we have argued for an analysis in which a `flat' structure is assigned to the VP and the verb cluster is not a separate constituent. In this respect, our account is similar to Baker's analysis of German. Our analysis rests on the assumption that a single HEAD-COMPLEMENT schema exists, which licences phrases in which a lexical head (i.e a head of type word) combines with an arbitrary number of its complements. Two considerations led us to this proposal. First, the distinction between verb clusters and (partial) VP's, which is crucial to any account based on verb clusters, requires the introduction of features such as NPCOMPS, and mechanisms (i.e. additional rule schemata) for assigning a value to these features on complex phrases, that are superfluous in an account without verb clusters. Second, phenomena such as partial VP fronting and the Dutch `third construction' (partial VP extraposition), which have been used as arguments for partial VP's in the Mittelfeld, are not incompatible with our approach. While the HEAD-COMPLEMENT schema excludes partial VP's as heads, it does not exclude such phrases to be built altogether. In [15], we present an analysis of partial VP fronting and the `third construction' that is fully compatible with a `flat' analysis of the VP.¹

The most important challenge for a `flat' analysis of German is to formulate word order constraints that will allow not only the standard, nested dependency, word order (1a), but also the word orders known as Oberfeldumstellung or auxiliary flip (1b) and Zwischenstellung (1c). Hinrichs and Nakazawa [3,4] argue that (1a) and (1b) can best be accounted for if the verb cluster bestehen können wird is a separate constituent, with a binary, left-branching, structure. Given this assumption, examples such as (1b) are a matter of word order variation, with the head wird preceding its complement rather than following it. The word order shown in (1c) is problematic for a left-branching analysis, however, as in this case the head wird appears in the middle of what is seen as one constituent (i.e. bestehen können):

For a `flat' analysis, in which both bestehen and können are complements of the head wird, this problem does not arise, and both Oberfeldumstellung and Zwischenstellung can be treated as a matter of word order variation. However, if no word order constraints are imposed, the `flat' analysis will also admit many ungrammatical word orders.

For Dutch, [15] assumed a word order constraint that had the effect of ordering more oblique complements always closer to the head than less oblique complements. This basically has the effect of allowing nested dependency word orders only. To account for the various word orders allowed in German, a different type of constraint is necessary. Below, we adopt a proposal by [7], in which complements are ordered relative to their governor.

In section 2, we present lexical entries for German modal and auxiliary verbs, and the HEAD-COMPLEMENT schema introduced in [15]. An overview of previous analyses and the critical data that need to be accounted for is given in section 3. In section 4, we discuss Kathol's analysis of word order in German verb clusters. In section 5, we demonstrate that a combination of a `flat' analysis of the VP and the word order constraints proposed by Kathol accounts for the various word order patterns in the German verb cluster.