Coreference Semantics from Web Features

TitleCoreference Semantics from Web Features
Publication TypeConference Paper
Year of Publication2012
AuthorsBansal, M., & Klein D.
Other Numbers3400
Abstract

To address semantic ambiguities in coreferenceresolution, we use Web n-gram featuresthat capture a range of world knowledge in adiffuse but robust way. Specifically, we exploitshort-distance cues to hypernymy, semanticcompatibility, and semantic context, aswell as general lexical co-occurrence. Whenadded to a state-of-the-art coreference baseline,ourWeb features give significant gains onmultiple datasets (ACE 2004 and ACE 2005)and metrics (MUC and B3), resulting in thebest results reported to date for the end-to-endtask of coreference resolution.

Acknowledgment

We would like to thank Nathan Gilbert, Adam Pauls,and the anonymous reviewers for their helpful suggestions.This research is supported by Qualcommvia an Innovation Fellowship to the first authorand by BBN under DARPA contract HR0011-12-C-0014.

URLhttps://www.icsi.berkeley.edu/pubs/ai/conferencesemantics12.pdf
Bibliographic Notes

Proceedings of the 50th Annual Meeting of the Association of Computational Linguistics (ACL 2012), Jeju, Korea

Abbreviated Authors

M. Bansal and D. Klein

ICSI Research Group

AI

ICSI Publication Type

Article in conference proceedings