Word order in German verb final sentences : A linguistic and psycholinguistic problem

This paper deals with linearization of complements of German verbs. In German all permutations of the subject, the indirect, and the direct object do occur. Yet, they are supposed to differ regarding their degree of acceptability. Uszkoreit (1987) proposed a set of rules which aimed at representing such preferences as the product of different factors. This theoretical account leads to a predicted ranking of the possible syntactic forms. In a set of experiments we tested some of these predictions by application of different methods for tapping into the actual processing of the sentences. In particular, the predictions were (a) that sentences are more acceptable if subjects precede objects than vice versa and (b) that sentences are more acceptable if indirect objects precede direct objects than vice versa. Both comprehension and production experiments were carried out. The methods we used included a ranking task, delayed sentence matching, delayed articulation, rapid serial visual presentation and a sentence generation task. The findings yielded a very consistent picture concerning the position of the subject. Sentences were particularly easy to process if the subject was in initial position and particularly hard to process in subject-final constructions. Furthermore, there is somewhat weaker evidence for the assumption that sentences are easier to process if direct objects are preceded by indirect objects. Since these results were obtained by rather different methods they can be regarded as particularly reliable. Moreover, the data did provide evidence for a gradual increase or decrease of acceptability and no evidence for a jump function, sharply separating grammatical from ungrammatical forms. One of the principal aims of this first phase of our investigations which is reported in the present paper was to find experimental methods which consistently differentiate between the various permutations of verb complements as predicted by theoretical assumptions. This aim could be achieved. The next step will be to include pragmatic factors which are supposed to play a significant role in determining the acceptability of the sentences we are studying. Word order in modern linguistics Natural languages differ significantly with regard to the degree of variation in their word order. On one end of the scale, we find languages such as English that exhibit a rather strict order, at least among heads and complements. On the opposite end of the scale there are languages such as the Australian language Warlpiri for which only very few ordering constraints can be observed. For some time it was claimed that this language allows for any permutation of the words of a sentence (cf. .i.Hale, 1983);. Although this extreme claim turned out to be untenable, Warlpiri remains to be a language which allows variation of word order to a very large extent. The majority of languages, including German, mix fixed and free word order in numerous exciting ways. Though it might be regarded a core concept of syntax, word order variation has not played the role it deserves in modern linguistics. This might in parts be due to the anglocentric view of the disciplin. However, it also reflects the lack of a good understanding of the interface between syntax and pragmatics. In grammars based on the notion of constituency, the prevalent descriptive means has been the phrase-structure tree. In Chomsky's Generative Transformational Grammar (TG; cf. .i.Chomsky, 1957, 1965); mappings from one phrase structure tree to another were introduced, but each derivation started from an ordered tree. The implication from transformational grammar was that of a default or basic word order for a given language. Obligatory deviations could be achieved by obligatory transformations, stylistic variations by stylistic transformations. In the theory of Government and Binding (GB; .i.Chomsky, 1981); which is the historical child of TG, word order is still explained by instances of alphamovement applied to a basic word order. However, in European linguistics grammars based on the TG model never succeeded in completely reigning the field. Particularly for Slavic languages (cf. .i.Melcuk, 1974;; .i.Sgall, Hajicov & Panevov , 1986); alternative models always kept the floor, in particular Dependency Grammar (.i.TesniŠre, 1959);. The same holds for German (.i.Kunze, 1975); and Romanic languages (.i.TesniŠre, 1959);. Dependency Grammar permits a more intuitive treatment of word order variation, for it provides a level of description at which heads and their arguments and adjuncts are not ordered with respect to each other. From this level one can define rules for the linearization at the surface. In feature-oriented declarative grammar formalisms, commonly referred to as unification grammars, word order is often described by a method which was first proposed by Gazdar and Pullum (.i.Gazdar & Pullum, 1982); for the framework of Generalized Phrase Structure Grammar (GPSG). The descriptive tool is called the ID/LP format. It is based on the separation of the immediate dominance (ID) concept and the linear precedence (LP) concept in the phrase structure component of a grammar. ID-rules determine the hierarchical structure of a sentence. LPrules restrict the sequence of sister-nodes in a tree. A grammar without LPrules generates a language in which all sister-nodes are free to permute.; ID-rules in GPSG take the form X--->Y1, Y2, ...,Y3 where X is an element of VN and Yi is an element of the union of VN and VT. The commas between the symbols on the right-hand-side of the rule indicate that the rule does not specify the sequence of these symbols. LP-rules take the form Y1<Y2 saying that each constituent Y1 must precede each sister constituent Y2 if constituents of both categories are present in a local tree. By applying the ID/LP format it became possible to chunk the regularities of word order for one grammar which had previously required a large number of phrase structure rules. Sequential variants could simply be described by the absence of certain LP-rules. By using a feature-oriented representation of categories various word order regularities could be described by, for instance, morphological features, the syntactic category of constituents, or thematic roles (cf. .i.Gazdar, Klein, Pullum & Sag, 1985;; .i.Uszkoreit, 1987);. The following three rules may serve as examples : [cat:np] < [cat:pp] [case:dat] < [case:acc] [th-role:goal] < [th-role:theme] The mixture of fixed and free word order in German could be handled by the application of LP-rules to certain phenomena, but not to others. LP-rules could be applied, for instance, to the linear sequence of articles and nouns or to the position of the verb, but not to the position of adverbials. The positions of finite verb forms could be represented as equivalent variants of linearization without the assumption of a basic word order (cf. .i.Uszkoreit, 1987);. This method was adopted in various forms by other declarative grammar formalisms. The variants differ with respect to the expressive power of the LPrules (cf. .i.Pollard & Sag, 1987,; concerning HPSG) or with respect to the domain of their application (e.g. .i.Reape, 1990);. However, the problem of partially free word order cannot be solved by simple LPrules, since these do not express preferences, but strict regularities. If the sequence of two constituents A and B is not restricted by an LP-rule, both orderings, A<B and B<A, are equally grammatical. Yet, if an LP-rule exists such that A<B, then only this one sequence is grammatical. The 'Mittelfeld' (middle field) in German verb final sentences as a critical case A revealing case regarding partially free word order and one to which the framework just mentioned has already been applied concerns the variation of verb complements in the so-called 'Mittelfeld' (middle field) in German verb final sentences (cf. .i.Thiersch, 1982;; .i.Koster, 1987;; .i.Grewendorf, 1988;; .i.Fanselow, 1990;; .i.Sternefeld, 1990);. The middle field may consist, for instance, of the subject and two objects: (1) Dann wird der Richter dem Anwalt den Beweis liefern. Then will the judge the lawyer the proof present. In German, any permutation of the subject, the indirect object, and the direct object is considered to be grammatical and acceptable. However, this is not to say that the sequence of the verb complements is not rule-governed. There are clear preferences which depend upon the thematic roles, the discourse functions, and the pronominalization of subject and objects (cf. .i.Lenerz, 1977);. Sometimes the factors mentioned above strongly exclude some sequences. In order to formally represent such preferences as the product of different factors, .i.Uszkoreit (1986); has proposed complex LP-rules consisting of a number of sequential rules which must not all be fulfilled in a particular case. Different weights can be attached to single principles. A decrease in grammaticality then results from the sum of violations of single principles. For the German middle field .i.Uszkoreit (1987); has proposed the following complex LP-rule which is based on morphological marking (case; the dichotomy of noun/pronoun) as well as on discursive roles (focus): [case:nom] < [case:acc] [case:nom] < [case:dat] [case:dat] < [case:acc] [cat:pron] < [cat:noun] [focus:-] < [focus:+] These principles are not supposed to have the same weight. In particular, the tendency to place the subject (nominative) before the indirect (dative) or direct (accusative) object, for instance, is assumed to be considerably stronger than the tendency to have the accusative object preceded by the dative object. Also, the inclination to put personal pronouns in front of other elements is rather strong in German. Another ordering principle that has been observed, concerns the length or syntactic heaviness of the permuting elements. Shorter and less complex elements tend to precede longer and more comple