Unsupervised Coreference Resolution

Coreference resolution is the task of partitioning the noun phrase mentions according to their underlying referent. For instance consider,

where mentions have been given colors to indicate entity clusters. Of course, some entities are mentioned in several documents (e. g. the US in the paragraph above).

We approach this problem as a clustering problem where the number of entities, or clusters, are not known in advance. We assume there is a global distribution over entities in a document corpus and that each document draws its own distribution over entities mentioned in the document. Our model addresses the problem of global entity resolution as well as the problem of sequential anaphoric resolution within a document. This model can be represented pictorially:

Publications

!IS_A_LIST Unsupervised Coreference Resolution in a Nonparametric Bayesian Model, Aria Haghighi and Dan Klein, In proceedings of ACL 2007. [pdf] [slides] [bib]