Unsupervised learning of contextual role knowledge for coreference resolution

We present a coreference resolver called BABAR that uses contextual role knowledge to evaluate possible antecedents for an anaphor. BABAR uses information extraction patterns to identify contextual roles and creates four contextual role knowledge sources using unsupervised learning. These knowledge sources determine whether the contexts surrounding an anaphor and antecedent are compatible. BABAR applies a Dempster-Shafer probabilistic model to make resolutions based on evidence from the contextual role knowledge sources as well as general knowledge sources. Experiments in two domains showed that the contextual role knowledge improved coreference performance, especially on pronouns.