A Discriminative Latent Variable Model for Clustering of Streaming Data with Application to Coreference Resolution

Rajhans Samdani, Kai-Wei Chang, Dan Roth

Decision: conferencePoster

Abstract: We present a latent variable structured prediction model, called the Latent Left-linking Model (L3M), for discriminative supervised clustering of items that follow a streaming order. L3M admits efficient inference and we present a learning framework for L3M that smoothly interpolates between latent structural SVMs and hidden variable CRFs. We present a fast stochastic gradient-based learning technique for L3M. We apply L3M to coreference resolution, which is a well known clustering task in Natural Language Processing, and experimentally show that L3M outperforms several existing structured prediction-based techniques for coreference as well as several state-of-the-art, albeit ad hoc, approaches.