Reconciling OntoNotes: Unrestricted Coreference Resolution in OntoNotes with Reconcile

摘要

This paper describes our entry to the 2011 CoNLL closed task (Pradhan et al., 2011) on modeling unrestricted coreference in OntoNotes. Our system is based on the Reconcile coreference resolution research platform. Reconcile is a general software infrastructure for the development of learning-based noun phrase (NP) coreference resolution systems. Our entry for the CoNLL closed task is a configuration of Reconcile intended to do well on OntoNotes data. This paper describes our configuration of Reconcile as well as the changes that we had to implement to integrate with the OntoNotes task definition and data formats. We also present and discuss the performance of our system under different testing conditions on a withheld validation set.