Analysis of Coreference Relations in the Biomedical Literature

Analysis of Coreference Relations in the Biomedical Literature Miji Choi1,2 Karin Verspoor1 Justin Zobel1 jooc1@student.unimelb.edu.au {karin.verspoor,jzobel}@unimelb.edu.au 1Department of Computing and Information Systems, The University of Melbourne 2National ICT Australia Abstract In this study, we perform an investigation of coreference resolution in the biomed- ical 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, perfor- mance of the system is still modest, with recall a particular problem (80% precision and 24% recall). Through analysis of fea- tures of coreference, organised by type of anaphors, we identify that differentiated strategies for each type could be applied to achieve further improvement. 1 Introduction The peer-reviewed scientific literature is a vast repository of authoritative knowledge. However, with around 40,000 new journal papers every month, manual discovery or annotation is infea- sible, and thus it is critical that document process- ing techniques be robust and accurate, to enable not only conventional search, but automated dis- covery and assessment of knowledge such as in- teracting relationships (events and facts) between biomolecules such as proteins, genes, chemical compounds and drugs. Biological molecular path- ways, for example, integrated with knowledge of relevant protein-protein interactions, are used to understand complex biological processes. Coreference resolution is an essential task in information extraction, because it can automati- cally provide links between entities, and as well can facilitate better indexing for medical infor- mation search with rich semantic information. A key obstacle is the low detection reliability of hidden or complex mentions of entities involving coreference expressions in natural language texts (Kim et al., 2011a; Miwa et al., 2010). Such anaphoric coreference expressions such as pro- nouns are mostly ignored by event extraction sys- tems, and are not considered as term occurrences in information retrieval systems. For example, the following passage includes an interacting relation; the binding event between the anaphoric mention the protein and a cell entity CD40 is implied in the text. The mention, the protein, refers to the specific protein name, TRAF2, previously introduced in the same text. (1) ...The phosphorylation appears to be re- lated to the signalling events ... be phosphorylated significantly less thanto the wild-type protein. Furthermore, the phosphorylation status of TRAF2 had sig- nificant effects on the ability of the protein to bind to CD40, as evidenced by our ... [PMID:10080948] In this paper, we investigate the challenges of biomedical coreference resolution, and provide an evaluation of general domain coreference reso- lution system on biomedical texts. Prior work demonstrated the importance of domain-specific knowledge for coreference (Choi et al., 2014). We extend that work with a detailed analysis of fea- tures of coreference relations with respect to the type of the anaphor defined by a previously pro- posed framework (Nguyen and Kim, 2008), and propose an efficient strategy towards improved anaphoric coreference resolution in the biomedi- cal literature building on that framework. 2 Background Related Work For general coreference resolution, several strate- gies and methodologies have been developed since …

Content truncated to 500 words