Elamur Research⌕
☰
home / papers / 2005 / A Twin-Candidate Model of Coreference Resolution with Non-Anaphor Identification Capability

A Twin-Candidate Model of Coreference Resolution with Non-Anaphor Identification Capability

Summary

The twin-candidate approach to coreference resolution compares pairs of antecedent candidates and selects the preferred one. It is good at choosing antecedents, but it always returns some antecedent, even when the noun phrase is not anaphoric at all. Earlier systems handled this with a separate anaphoricity determination step or a confidence threshold. Yang, Su and Tan instead change how the model is trained. Non-anaphoric NPs are paired with a randomly chosen anchor candidate and every other candidate, and these training instances get a third label, "00", meaning that neither candidate is preferred. During round-robin resolution, a "00" verdict penalizes both candidates. If no candidate ends with a positive score, the NP is left unresolved. On the MUC-6 and MUC-7 newswire data, this approach (TC_NEW) beat a single-candidate baseline and both alternative twin-candidate setups. Its behaviour was also much less sensitive to threshold settings. The result matters because the twin-candidate model can now perform full coreference resolution without any extra module.

Key points

  • Labels: Training instances have the form i{C1, C2, M}. The labels "10" and "01" express which candidate is preferred, and the new "00" label marks NPs that are not anaphors.
  • Systems compared:
    • SC: single-candidate baseline.
    • TC_AD: twin-candidate with an anaphoricity determination filter. The authors note such filters reach only about 80% accuracy. This system had high precision but low recall.
    • TC_THRESH: twin-candidate with a confidence threshold. Recall exceeded 70%, but precision was the lowest of all systems.
    • TC_NEW: the proposed framework.
  • Main result: TC_NEW reached an F-measure of 67.3% on MUC-6 and 67.2% on MUC-7. This was significantly better than the other twin-candidate systems (sign test, p ≤ 0.05), and its precision was the highest of the four systems.
  • Training data: The default setup used 30 training documents. With 100 additional annotated documents, TC_NEW kept improving and peaked at about 90 documents, while SC showed little change.
  • Threshold sensitivity: TC_THRESH gave acceptable results only with thresholds above 95; at lower values its precision stayed under 40%. In contrast, TC_NEW remained stable across thresholds from 65 to 100 and works well with no threshold at all.

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

5 papers of our library cite it (of the 140673 whose bibliographies we read); 2 of them have a page here, 3 not yet.

Cite

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.1007/11562214_63

@inproceedings{yang2005twin,
  title = {{A Twin-Candidate Model of Coreference Resolution with Non-Anaphor Identification Capability}},
  author = {Yang, Xiaofeng and Su, Jian and Tan, Chew Lim},
  booktitle = {Second International Joint Conference on Natural Language Processing: Full Papers},
  year = {2005},
  doi = {10.1007/11562214\_63},
  url = {https://doi.org/10.1007/11562214_63}
}

References

11 references, 5 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. McCarthy, J., Lehnert, Q.: Using decision trees for coreference resolution. In: Proceedings of the 14th International Conference on Artificial Intelligences. (1995) 1050–1055 find
  2. Connolly, D., Burger, J., Day, D. New Methods in Language Processing. In: A machine learning approach to anaphoric reference. (1997) 133–144 find
  3. Soon, W., Ng, H., Lim, D.: A machine learning approach to coreference resolution of noun phrases. Computational Linguistics 27 (2001) 521–544 find our card
  4. Ng, V., Cardie, C.: Improving machine learning approaches to coreference resolution. In: Proceedings of the 40th Annual Meeting of the Association for Computational Linguistics, Philadelphia (2002) 104–111 find our card
  5. Yang, X., Zhou, G., Su, J., Tan, C.: Coreference resolution using competition learning approach. In: Proceedings of the 41st Annual Meeting of the Association for Computational Linguistics, Japan (2003) find our card
  6. Iida, R., Inui, K., Takamura, H., Matsumoto, Y.: Incorporating contextual cues in trainable models for coreference resolution. In: Proceedings of the 10th Conference of EACL, Workshop “The Computational Treatment of Anaphora”. (2003) find
  7. Ng, V., Cardie, C.: Identifying anaphoric and non-anaphoric noun phrases to improve coreference resolution. In: Proceedings of the 19th International Conference on Computational Linguistics (COLING02). (2002) find
  8. MUC-6: Proceedings of the Sixth Message Understanding Conference. Morgan Kaufmann Publishers, San Francisco, CA (1995) find
  9. MUC-7: Proceedings of the Seventh Message Understanding Conference. Morgan Kaufmann Publishers, San Francisco, CA (1998) find
  10. Quinlan, J.R.: C4.5: Programs for machine learning. Morgan Kaufmann Publishers, San Francisco, CA (1993) find our card
  11. Vilain, M., Burger, J., Aberdeen, J., Connolly, D., Hirschman, L.: A model-theoretic coreference scoring scheme. In: Proceedings of the Sixth Message understanding Conference (MUC-6), San Francisco, CA, Morgan Kaufmann Publishers (1995) 45–52 find our card

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/yang-2005-twin-candidate-model-coreference-resolution/