Annotation of Predicate-argument Structure on Molecular Biology Text

Annotated corpora are essential resources for natural language processing. This paper describes our approach for building a corpus annotated with predicateargument structure on research abstracts in molecular biology domain. Observation of the records in a database of cell signaling events and corresponding research abstracts showed that extracting predicateargument structure is a useful intermediate step for extracting reaction events but analysis of verb phrases is not sufficient because reactions and relations are often expressed in nominal phrases. Based on the observation, we try to annotate the predicate-argument structure in verbs including their nominalized forms in order to help develop extraction systems based on predicate-argument analysis.