Data extraction from corpora as a tool to classify predicates by their subcategorisation properties

This paper describes an approach to the classification of different predicate types based on their subcategorisation. German predicates (verbs, nouns, multiwords and adjectives) which subcategorize for a subordinate clause (that, if, wh-clause) are semi-automatically extracted from text corpora and classified according to their subcategorisation properties. Our approach can be applied in natural language processing, e.g. in lexicon creation.