Extracting Collocations from Syntactically Annotated Corpora
Oliver Sander, Ingrid Fischer, Harald Kirsch · 2002
In this paper the extraction of collocations from biomedical text sources is described. The extraction of uninterrupted collocation candidates is introduced and for interrupted candidates a new technique using suffix tries is developed. It reduces computational complexity, compared to previous approaches to this task. Extraction is further extended to annotated biomedical corpora, which enables the discovery of patterns with mixed attributes like base forms combined with syntactical categories. Finally, different common approaches for evaluating candidates are explained, and their applicability to interrupted and mixed attribute candidates is discussed.