Effiziente Verarbeitung deutscher Konstituentenstellung mit der Combinatorial Categorial Grammar: 2447

摘要

Combinatory Categorial Grammar provides a high degree of flexibility for modelling both simple linguistic phenomena and discontinuities as well as elliptic structures. However, the mechanisms required for these latter cases are problematic with respect to their efficiency. Unrestricted type raising enlarges the search space dramatically, while rules which enable incremental processing lead to spurious ambiguities. Moreover, to cope with German word order each verb form must be represented in several ways, which in turn increases the number of locally ambiguous derivations. Slight modifications of the rule set and the representation of categories allows efficient and constituent-based incremental processing while avoiding the problems mentioned above. Explicit type raising is made obsolete by integrating it into specialized rules. Free word order is achieved by use of a constituent counter, rendering multiple categories unnecessary. Besides main clauses in the active and passive voice, this approach also processes subordinate clauses, questions and modal verbs. The approach has been successfully applied in a speech processing system for instructing a two-arm-robot system in assembly tasks.