Analysis of coreference relations in the biomedical literature

Abstract

In this study, we perform an investigation of coreference resolution in the biomedical literature. We compare a state-of-the-art general system with a purpose-built system, demonstrating that the use of domain-specific knowledge results in dramatic improvement. However, performance of the system is still modest, with recall a particular problem (80% precision and 24% recall). Through analysis of features of coreference, organised by type of anaphors, we identify that differentiated strategies for each type could be applied to achieve further improvement.