Using Knowledge and Constraints To Find the Best Antecedent

Proceedings of COLING 2012: Technical Papers, pages 1327–1342, COLING 2012, Mumbai, December 2012. Using Knowledge and Constraints To Find the Best Antecedent Prateek Jindal Dan Roth Dept. of Computer Science, UIUC 201 N. Goodwin Ave., Urbana, IL - 61801 {jindal2, danr}@illinois.edu Abstract Coreference resolution is the problem of clustering mentions into entities and is very critical for natural language understanding. This paper studies the problem of coreference resolution in the context of the newly emerging domain of Electronic Health Records (EHRs). The commonly used “best-link” model for coreference resolution considers only the scores from a pairwise classifier in selecting the best antecedent. In this paper, we extend this model to include several constraints derived from surface-form of the mentions and the context in which they appear. Another major contribution of this paper is to show the use of domain-specific knowledge sources, mention parsing and clinical descriptors in deriving features which contribute to improved coreference resolution performance. We present experiments on 4 different clinical datasets illustrating that our approach outperforms a strong baseline and a state-of-the-art system by a wide margin. Keywords: Natural Language Processing, Information Extraction, Coreference Resolution, Elec- tronic Health Records, Knowledge Based Systems. 1327 1 Introduction The HITECH (Health Information Technology for Economic and Clinical Health) Act, part of the 2009 economic stimulus package (American Recovery and Reinvestment Act) passed by the US Congress, aims at inducing more physicians to adopt Electronic Health Records (EHRs). An EHR is an evolving concept defined as a systematic collection of electronic health information about individual patient. Ability to automatically extract information from EHRs lies at the heart of several applications. This paper addresses the task of coreference resolution for EHRs. Coreference resolution is the task of finding referring expressions in a text that refer to the same entity, i.e., finding expressions that corefer. The set of coreferring expressions is called as a coreference chain. Consider the following text sampled from one of the EHRs in the corpus used by us: This 63-year-old man had [malignant fibrous histiocytoma of duodenum], discovered in 02/95. Other than [a mass in the duodenum], the patient was also diagnosed with anemia. [leiomyosarcoma] was resected after embolization of the splenic artery. However, [it] could notA be completely excised; moreover [the tumor] metastasized to the liver as was discovered on follow up scan in 06/95. In the above text, all the phrases which are shown in brackets refer to the same entity and hence form a coreference chain. It is clear that identifying such coreference chains requires a lot of medical knowledge. For example, we need to know that “mass” can refer to a “malignant histiocytoma”. Most of the work on coreference resolution has focussed on the news text. Several different architectures have been proposed for coreference resolution. Recently, entity-based models for coreference resolution have been proposed. Such approaches try to directly model the entities in the text and usually involve some kind of global inference and tend to be quite complex. However, most of the best results …

Content truncated to 500 words