Including a knowledge base into Haghighi & Klein's coreference resolution system
Haghighi and Klein (2010) proposed a system for coreference resolution that learned distributions over the noun phrases used in several types of nominal mentions of entities. This system learned these distributions from large amounts of unlabeled (but parsed) text, and used them to improve performance on the common coreference resolution task. Other work has attempted to use information contained in knowledge bases in order to improve performance on coreference resolution. We hypothesize that combining the two approaches, using not just distributions over noun phrases for each entity, but also facts about them obtained from a knowledge base, will perform even better.