Extraposition: A Case Study in German Sentence Realization

Extraposition: A case study in German sentence realization Michael GAMON†, Eric RINGGER†, Zhu ZHANG‡, Robert MOORE†, Simon CORSTON-OLIVER† †Microsoft Research Microsoft Corporation Redmond, WA 98052 {mgamon, ringger, bobmoore, simonco}@microsoft.com ‡University of Michigan Ann Arbor, MI 48109 zhuzhang@umich.edu Abstract We profile the occurrence of clausal extraposition in corpora from different domains and demonstrate that extraposition is a pervasive phenomenon in German that must be addressed in German sentence realization. We present two different approaches to the modeling of extraposition, both based on machine learned decision tree classifiers. The two approaches differ in their view of the movement operation: one approach models multi-step movement through intermediate nodes to the ultimate target node, while the other approach models one-step movement to the target node. We compare the resulting models, trained on data from two domains and discuss the differences between the two types of models and between the results obtained in the different domains. Introduction Sentence realization, the last stage in natural language generation, derives a surface string from a more abstract representation. Numerous complex operations are necessary to produce fluent output, including syntactic aggregation, constituent ordering, word inflection, etc. We argue that for fluent output from German sentence realization, clausal extraposition needs to be included. We show how to accomplish this task by applying machine learning techniques. A comparison between English and German illustrates that it is possible in both languages to extrapose clausal material to the right periphery of a clause, as the following examples show: Relative clause extraposition: English: A man just left who had come to ask a question. German: Der Mann ist gerade weggegangen, der gekommen war, um eine Frage zu stellen. Infinitival clause extraposition: English: A decision was made to leave the country. German: Eine Entscheidung wurde getroffen, das Land zu verlassen. Complement clause extraposition: English: A rumor has been circulating that he is ill. German: Ein Gerücht ging um, dass er krank ist. Unlike obligatory movement phenomena such as Wh-movement, extraposition is subject to pragmatic variability. A widely-cited factor influencing extraposition is clausal heaviness; in general, extraposition of heavy clauses is preferred over leaving them in place. Consider the following example from the technical domain: German: Es werden Datenstrukturen verwendet, die für die Benutzer nicht sichtbar sind. English: Data structures are used which are not visible to the user. This perfectly fluent sentence contains an extraposed relative clause. If the relative clause is left in place, as in the following example, the result is less fluent, though still grammatical: ? Es werden Datenstrukturen, die für die Benutzer nicht sichtbar sind, verwendet. Data structures which are not visible to the users are used. Table 1 presents a quantitative analysis of the frequency of extraposition in different corpora in both English and German. This analysis is based on automatic data profiling using the NLPWin system (Heidorn 2000). The technical manual corpus consists of 100,000 aligned English-German sentence pairs from Microsoft technical manuals. The Encarta corpora consist of 100,000 randomly selected sentences from the Encarta encyclopedia in both English …

Content truncated to 500 words