Learning Global Features for Coreference Resolution

Proceedings of NAACL-HLT 2016, pages 994–1004, San Diego, California, June 12-17, 2016. c⃝2016 Association for Computational Linguistics Learning Global Features for Coreference Resolution Sam Wiseman and Alexander M. Rush and Stuart M. Shieber School of Engineering and Applied Sciences Harvard University Cambridge, MA, USA {swiseman,srush,shieber}@seas.harvard.edu Abstract There is compelling evidence that corefer- ence prediction would benefit from modeling global information about entity-clusters. Yet, state-of-the-art performance can be achieved with systems treating each mention prediction independently, which we attribute to the inher- ent difficulty of crafting informative cluster- level features. We instead propose to use re- current neural networks (RNNs) to learn la- tent, global representations of entity clusters directly from their mentions. We show that such representations are especially useful for the prediction of pronominal mentions, and can be incorporated into an end-to-end coref- erence system that outperforms the state of the art without requiring any additional search. 1 Introduction While structured, non-local coreference models would seem to hold promise for avoiding many com- mon coreference errors (as discussed further in Sec- tion 3), the results of employing such models in practice are decidedly mixed, and state-of-the-art results can be obtained using a completely local, mention-ranking system. In this work, we posit that global context is indeed necessary for further improvements in coreference resolution, but argue that informative cluster, rather than mention, level features are very difficult to de- vise, limiting their effectiveness. Accordingly, we instead propose to learn representations of mention clusters by embedding them sequentially using a re- current neural network (shown in Section 4). Our model has no manually defined cluster features, but instead learns a global representation from the indi- vidual mentions present in each cluster. We incor- porate these representations into a mention-ranking style coreference system. The entire model, including the recurrent neu- ral network and the mention-ranking sub-system, is trained end-to-end on the coreference task. We train the model as a local classifier with fixed context (that is, as a history-based model). As such, unlike several recent approaches, which may require complicated inference during training, we are able to train our model in much the same way as a vanilla mention- ranking model. Experiments compare the use of learned global features to several strong baseline systems for coref- erence resolution. We demonstrate that the learned global representations capture important underlying information that can help resolve difficult pronom- inal mentions, which remain a persistent source of errors for modern coreference systems (Durrett and Klein, 2013; Kummerfeld and Klein, 2013; Wise- man et al., 2015; Martschat and Strube, 2015). Our final system improves over 0.8 points in CoNLL score over the current state of the art, and the im- provement is statistically significant on all three CoNLL metrics. 2 Background and Notation Coreference resolution is fundamentally a clustering task. Given a sequence (xn)N n=1 of (intra-document) mentions – that is, syntactic units that can refer or be referred to – coreference resolution involves parti- tioning (xn) into a sequence of clusters (X(m))M m=1 such that all …

Content truncated to 500 words