Elamur Research⌕
☰
home / papers / 2005 / Using Semantic Relations to Refine Coreference Decisions

Using Semantic Relations to Refine Coreference Decisions

Summary

Most coreference resolvers depend on local lexical and syntactic cues such as string matching, agreement and distance. These cues fail on harder cases where world knowledge would help. One example is two "officials" who work for different organizations. Ji, Westbrook and Grishman address this by adding document-level semantic context through a Relational Coreference Model (RCM). They train a relation tagger on data annotated with the ACE 2004 relation ontology. For a candidate mention pair, they then check whether both mentions take part in relations, whether those relations share type, subtype and direction, and whether the relation partners corefer. From this they derive three heuristic rules. Two lower the likelihood of coreference to prune wrong links. One raises it to recover missed links. Each rule instance is kept only if it proves reliable under cross-validation, and the kept rules rescore the output of a baseline maxent coreference classifier. The approach improves F-measure in both English and Chinese. It is cheap and independent of domain, which suggests that lightweight semantics can strengthen pipelines for information extraction.

Key points

  • The ACE 2004 relation ontology has 7 main relation types (e.g., EMP-ORG, GPE-AFF, PER-SOC, PHYS) and 23 subtypes.
  • Rule instances are retained only at a reliability of 70% or higher. This yields 58 type-specific instances of Rules 1 and 3, the general Rule 2 (97% accurate), and 24 more rules built by adding up to 3 boolean tests (e.g., head match, number agreement, no "former"/"current").
  • With perfect mentions and relations, F rises from 80.1 to 82.4 for English and from 85.3 to 86.7 for Chinese. With system mentions and relations, F rises from 81.9 to 83.7 for English and from 75.6 to 76.3 for Chinese. All gains are significant by a sign test at 95%.
  • The pipeline resolves coreference, tags relations, and then resolves coreference again. The authors see this as a step toward iterative joint architectures.
  • Errors come partly from ignoring time. For example, one person can be "president" of different organizations at different times.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

Fields: Linguistics & languages

Topics: Coreference resolution

Cited by

20 papers of our library cite it (of the 140673 whose bibliographies we read); 8 of them have a page here, 12 not yet.

Cite

The paper itself, not this page. BibTeX from the library's record of the work.

@inproceedings{ji2005using,
  title = {{Using Semantic Relations to Refine Coreference Decisions}},
  author = {Ji, Heng and Westbrook, David and Grishman, Ralph},
  booktitle = {Proceedings of Human Language Technology Conference and Conference on Empirical Methods in Natural Language Processing},
  year = {2005},
  doi = {10.3115/1220575.1220578},
  url = {https://doi.org/10.3115/1220575.1220578}
}

References

13 references, 7 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. David Bean, Ellen Riloff. 2004. Unsupervised learning of contextual role knowledge for coreference resolution. find our card
  2. Daniel M. Bikel, Scott Miller, Richard Schwartz, and Ralph Weischedel. 1997. Nymble: A high-performance learning name-finder. Proc. Fifth Conf. on Applied Natural Language Processing, Washington, D.C., pp. 194-201. find
  3. Carbonell, Jaime and Ralf Brown. 1988. Anaphora resolution: A multi-strategy approach. find our card
  4. Eugene Charniak. 1972. Toward a model of children's story comprehension. Ph.D. thesis, Massachusetts Institute of Technology, Cambridge, MA. find
  5. Niyu Ge, John Hale and Eugene Charniak. 1998. A statistical approach to anaphora resolution. Proc. the Sixth Workshop on Very Large Corpora. find our card
  6. Jerry Hobbs, Mark Stickel, Douglas Appelt and Paul Martin. 1993. Interpretation as abduction. Artificial Intelligence, 63, pp. 69-142. find
  7. Ruslan Mitkov. 2000. Towards a more consistent and comprehensive evaluation of anaphora resolution algorithms and systems. Proc. 2nd Discourse Anaphora and Anaphora Resolution Colloquium, pp. 96-107 find
  8. Vincent Ng and Claire Cardie. 2002. Improving machine learning approaches to coreference resolution. find our card
  9. Wee Meng Soon, Hwee Tou Ng, and Daniel Chung Yong Lim. 2001. A machine learning approach to coreference resolution of noun phrases. Computational Linguistics, Volume 27, Number 4, pp. 521-544 find our card
  10. Joel R. Tetreault. 2001. A corpus-based evaluation of centering and pronoun resolution. Computational Linguistics, Volume 27, Number 4, pp. 507-520 find our card
  11. Joel R. Tetreault and James Allen. 2004. Semantics, Dialogue, and Pronoun Resolution. Proc. CATALOG '04 Barcelona, Spain. find
  12. Marc Vilain, John Burger, John Aberdeen, Dennis Connolly, and Lynette Hirschman. 1995. A model-theoretic coreference scoring scheme. Proc. the 6th Message Understanding Conference (MUC-6). San Mateo, Cal. Morgan Kaufmann. find our card
  13. Robert Wilensky. 1983. Planning and Understanding. Addison-Wesley. find

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/papers/ji-2005-semantic-relations-refine-coreference-decisions/