Exploiting large corpora: A circular process of partial syntactic analysis, corpus query and extraction of lexicographic information

Our approach follows the work of Eckle-Kohler (1999) who used a regular grammar to extract lexicographic information from text corpora. We employ a system that allows to improve her querybased grammar especially with respect to recall and speed without reducing accuracy. In contrast to Eckle-Kohler (1999), we do not attempt to parse a whole sentence or phrase at once during the extraction process, but build complex structures incrementally. The intermediate results are annotated in the corpus and used as input for following iterations. This concept enables us to accommodate new aspects such as agreement information and the use of annotated structures together with their features (for partial parsing as well as for extraction purposes).