Skip to main navigation Skip to search Skip to main content

Confidence measures and thresholding in coreference resolution

  • John Chen
  • , Laurie Crist
  • , Len Enyon
  • , Cassandre Creswell
  • , Amit Mhatre
  • , Rohini Srihari
  • Janya Inc.

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

Abstract

Coreference resolution is an important component of information extraction systems. Machine learning methods have been found to perform quite well for this task, leading to research in a variety of such methods. In our current work, we explore the approach of increasing the performance of an existing pairwise coreference system by using a confidence measure in order to filter out low-scoring classifications. We explore several ways to define a confidence measure. Subsequently, we use a confidence measure in conjunction with thresholding. We find that a multiple threshold system, with the thresholds defined the right way, outperforms both the baseline and a single threshold system. We also discover that basing a threshold as close as possible to the evaluation metric is a good idea, and explore reasons why this might be so.

Original languageEnglish
Title of host publicationInternational Conference Recent Advances in Natural Language Processing, RANLP 2007 - Proceedings
EditorsGalia Angelova, Kalina Bontcheva, Ruslan Mitkov, Nicolas Nicolov, Nikolai Nikolov
PublisherAssociation for Computational Linguistics (ACL)
Pages121-127
Number of pages7
ISBN (Electronic)9789549174373
StatePublished - 2007
EventInternational Conference Recent Advances in Natural Language Processing, RANLP 2007 - Borovets, Bulgaria
Duration: Sep 27 2007Sep 29 2007

Publication series

NameInternational Conference Recent Advances in Natural Language Processing, RANLP
Volume2007-January

Conference

ConferenceInternational Conference Recent Advances in Natural Language Processing, RANLP 2007
Country/TerritoryBulgaria
CityBorovets
Period09/27/0709/29/07

Keywords

  • Confidence estimation
  • Coreference resolution
  • Machine learning

Fingerprint

Dive into the research topics of 'Confidence measures and thresholding in coreference resolution'. Together they form a unique fingerprint.

Cite this