Automated template discovery for information extraction fiom biomedical literature

We propose a method to automatically extract templates from biomedical literature without background knowledge. The proposed method automatically extracts verbs and templates indicating interactions between biomolecules with a large dictionary called an extensional ontology. We applied our method to two datasets: one comprised 299 full texts from Cell (1998-2000) and 13,818 entries from OMIM (Online Mendelian Inheritance in Man); the other included 33,622 abstracts from Medline (2000). Experimental results showed that our method could extract verbs and templates that had been manually collected in related works. For extracting templates, our method only needs to prepare ontology (or dictionary) and a large body of texts. Consequently, it can be applied to those of other fields as well as the biomedical literature.