Word Order in German: A Formal Dependency Grammar Using a Topological Hierarchy Kim Gerdes Sylvain Kahane Lattice, Université Paris 7 Lattice, Université Paris 7 75251 Paris Cedex 05 75251 Paris Cedex 05 France France [email protected] [email protected] pursue this problem any further, but have Abstract limited our description to the link between This paper proposes a description of dependency and topology. Note that it is fun- German word order including phe- damental to our approach that syntactic nomena considered as complex, structure does not include word order. such as scrambling, (partial) VP To get the words in order, we group them in a fronting and verbal pied piping. Our hierarchy of phrases. The nature and the posi- description relates a syntactic de- tion of these phrases are constrained by our pendency structure directly to a topological model. For instance, a non-finite topological hierarchy without re- verb can open two kinds of topological sorting to movement or similar phrases, either a phrase, which we call domain, mechanisms.1 with positions for all of its dependents, or a restricted phrase, which forms the verb cluster, with no positions for dependents other than 1 Introduction predicative elements. These two kinds of phrases must be placed in very different The aim of this article is to describe the word topological positions. order of German verbs and their comple- The fact that we pass through a (topological) ments. German word order is not free, but phrase structure in order to relate dependency based on fairly simple rules, forming what is and word order distinguishes our approach usually called topological model, which sub- from usual dependency grammars (Mel'cuk & divides the sentence into a hierarchy of topo- Pertsov, 1987; Bröker, 1998; Kahane et al., logical domains that are themselves composed 1998; Duchier & Debusmann, 2001). The of fields (Vorfeld, Mittelfeld, right bracket…) description of German word order closest to (Drach, 1937; Bech, 1955). our analysis is the HPSG grammar of Kathol We start from a syntactic dependency tree, i.e. (1995; see also Reape 1994), who proposes an unordered tree whose nodes are labeled linearization rules exclusively based on a for- with the words of the sentence, and whose malization of the topological structure. How- branches are labeled with syntactic relations ever, as required by the formalism he uses, a among the words (subject, direct object…). regular phrase structure, which we do not need The syntactic dependency structure only en- in our analysis, still underlies the structures codes subcategorization and modification and obtained. must be completed by the communicative Our work constitutes a syntactic module which structure (partition into theme/rheme, focus…), links (unordered) syntactic structures with which plays a fundamental role in word order. topological phrase structures. Syntactic struc- It permits us to choose among all the different tures are related to semantic structures, possible orders corresponding to a given de- whereas topological phrase structures are re- pendency structure. In this paper we do not lated to phonological structures. In other words, our work lies within the scope of the 1 We would like to thank Werner Abraham, Tilman general framework of Meaning-Text-Theory Becker, Ralph Debusmann, Denys Duchier, and (Mel'cuk 1988), which considers the modeling Stefan Müller for fruitful discussions. Particular of a language as a modular (bi-directional) thanks to Igor Mel'cuk for the inspiration of the correspondence between meaning and text. It particular status we give to the phrase structure. must be clear that, in contrast to X-bar syntax, our topological phrase structure does not rep- resent the syntactic structure of the sentence. 2.2 Topological model Although the dependency information is es- sential in its construction, the phrase structure The internal structure of a domain is a se- only represents topology, i.e. the surface quence of fields. For example, the main do- grouping of the words. Topological phrases main is the underlying pattern of a declarative can be directly related to prosodic groups, and sentence, and it consists of the following se- topology represents an intermediate level be- quence of five fields: [Vorfeld, left bracket, tween dependency and phonology. Mittelfeld, right bracket, Nachfeld]. A domain In Section 2, the results of our findings are resembles a box whose ordered compartments, presented, without recourse to any mathemati- called fields, can themselves accommodate cal formalism, in the usual terminology of new boxes. In addition to the rules listing the traditional German grammars. In Section 3, a fields of each type of box, we propose two mathematical formalism is proposed to state further types of rules: the rules and the grammar fragment described • rules that indicate into which field a word in Section 2. can go–depending on the position of its governor; • 2 Description rules that indicate which type of box a word can create when it is placed into a given field. Word order in German is much freer than in The hierarchy of boxes forms the phrase English. The dependency tree of Fig. 1, which structure we construct. will be our reference example, has a few dozen linearizations: 2.3 Word order rules (1) a. Niemand hat diesem Mann das Buch zu lesen versprochen We have established the following rules for the b. Diesem Mann hat das Buch niemand linear order of verbs and their dependents: zu lesen versprochen • The finite verb takes the second position c. Das Buch zu lesen hat diesem Mann of the main domain, the left bracket. This niemand versprochen verb is also called V2. • A non-finite verb depending on V2 can d. Diesem Mann hat niemand verspro- 2 chen, das Buch zu lesen go into the right bracket. As a result, it e. Diesem Mann hat, das Buch zu lesen, opens a reduced phrase with only one po- niemand versprochen sition for a verbal dependent (see Section f. Zu lesen hat diesem Mann das Buch 2.8 for another possibility). If a subse- niemand versprochen quent third verb joins the verb already in g. Das Buch hat niemand diesem Mann the right bracket, it will again open a versprochen zu lesen phrase with a position to its left, and so on. ‘Nobody promised this man to read the The verbal constituent occupying the right book.’ bracket is called the verb cluster. • Some non-verbal dependents, such as hat ‘has’ separable verbal prefixes (for example the subj aux an of anfangen ‘begin’), predicative ad- versprochen jectives, and nouns governed by a copular ‘promised’ verb or a support verb, can go into the niemand right bracket (the prefix even forms one ‘noboby’-NOM iobj inf zu lesen word with its following governor). In con- diesem Mann ‘to read’ trast to verbs, these elements do not usu- ‘this man’-DAT dobj ally open up a new position for their de- pendents, which consequently have to be placed somewhere else.3 das Buch ‘the book’-ACC 2 We consider that in a compound verb form such as Fig. 1. Dependency tree of the sentences in (1) hat gelesen ‘has read’ the past participle depends syntactically on the auxiliary, which is the finite verb In this paper, we do not attempt to characterize form (cf. Tesnière 1959, Mel'cuk 1988). The V2 is well-formed German dependency trees al- thus always the root of the syntactic dependency tree. though we recognize that such a characteriza- 3 In examples such as (i), the separable verbal prefix tion is essential if we attempt to describe the an behaves like a subordinated verb intervening be- acceptable sentences of German. tween the ‘main’ verb and its dependent: • One dependent (verbal or non-verbal) of are restrictions that weigh more heavily than any of the verbs of the main domain (V2, the hierarchical position: pronominalization, any verb in the right bracket or even an focus, new information, weight, etc. embedded verb) has to occupy the first position, called the Vorfeld (VF, pre- hat ‘has’ field). • aux All the other non-verbal dependents of the subj versprochen verbs in the domain (V2 or part of the ‘promised’ verbal cluster) can go in the Mittelfeld niemand (MF, middle-field). ‘noboby’ iobj inf • zu lesen Some phrases, in particular sentential ‘to read’ complements (complementizer and rela- diesem Mann ‘to this man’ tive clauses), prepositional phrases, and dobj even some sufficiently heavy noun das Buch phrases, can be positioned in a field right ‘the book’ of the right bracket, the Nachfeld (NF, af- ter-field). Like the Mittelfeld, the Nachfeld can accommodate several dependents. Fig. 2. A phrase structure without embedded do- • When a verb is placed in any of the Major mains for (1a,b) Fields (Vor-, Mittel-, Nachfeld), it opens a The fact that a verbal projection (i.e. the verb new embedded domain. and all of its direct and indirect dependents) In the following section we illustrate our rules does not in general form a continuous phrase, with the dependency tree of Fig. 1 and show unlike in English and French, is called scram- how we describe phenomena such as scram- bling (Ross, 1967). This terminology is based bling and (partial) VP fronting. on an erroneous conception of syntax that supposes that word order is always an immedi- 2.4 Non-embedded construction and ate reflection of the syntactic hierarchy (i.e. every projection of a given element forms a “scrambling” phrase) and that any deviation from this con- Let us start with cases without embedding, i.e. stitutes a problem. In fact, it makes little sense where the subordinated verbs versprochen to form a phrase for each verb and its depend- ‘promised’ and zu lesen ‘to read’ will go into ents.