Abstract:

This talk will first briefly describe information extraction systems --- natural language understanding systems that take as input a collection of unrestricted texts and ``summarize'' each text with respect to a prespecified topic or domain of interest. We then will focus on the problem of noun phrase coreference, one of the most cited underlying problems for information extraction and for many other practical natural language processing tasks. The goal for noun phrase coreference algorithms is to determine which noun phrases in a text or dialogue refer to the same real-world entity. We will introduce an algorithm for noun phrase coreference resolution that differs from existing methods in that it views coreference resolution as a clustering task. In initial evaluations on a standard coreference resolution corpus, our results are extremely encouraging. The coreference clustering algorithm appears to provide a flexible mechanism for combining context-independent coreference constraints and context-dependent preferences for accurate partitioning of noun phrases into coreference equivalence classes.