Traditional learning-based coreference resolvers operate by training themention-pair model for determining whether two mentions are coreferent or not.Though conceptually simple and easy to understand, the mention-pair model islinguistically rather unappealing and lags far behind the heuristic-basedcoreference models proposed in the pre-statistical NLP era in terms ofsophistication. Two independent lines of recent research have attempted toimprove the mention-pair model, one by acquiring the mention-ranking model torank preceding mentions for a given anaphor, and the other by training theentity-mention model to determine whether a preceding cluster is coreferentwith a given mention. We propose a cluster-ranking approach to coreferenceresolution, which combines the strengths of the mention-ranking model and theentity-mention model, and is therefore theoretically more appealing than bothof these models. In addition, we seek to improve cluster rankers via twoextensions: (1) lexicalization and (2) incorporating knowledge of anaphoricityby jointly modeling anaphoricity determination and coreference resolution.Experimental results on the ACE data sets demonstrate the superior performanceof cluster rankers to competing approaches as well as the effectiveness of ourtwo extensions.
展开▼
