A Joint Inference Architecture for Global Coreference Clustering with Anaphoricity

摘要: We present an architecture for coreference resolution based on joint inference over anaphoricity and coreference, using Markov Logic Networks. Mentions are discriminatively clustered with discourse entities established by classifier. Our entity-based is realized in a setting to compensate erroneous classifications avoids local misclassifications through global consistency constraints. Defining pairwise features achieves efficient perspective. With small feature set we obtain performance of 63.56% (gold mentions) the official CoNLL 2012 data set.