Elamur Research⌕
☰
home / papers / 1997 / Factors in anaphora resolution: they are not the only things that matter. A case study based on two different approaches

Factors in anaphora resolution: they are not the only things that matter. A case study based on two different approaches

Summary

Mitkov asks whether good anaphora resolution depends only on picking reliable factors, or also on how those factors are applied. He tests this by giving two of his own pronoun resolvers the same core set of factors. These include gender and number agreement, c-command, semantic consistency, syntactic and semantic parallelism, topicalisation, and several empirically motivated preferences such as section heading and object preference signalled by verbs or nouns. The first system, the Integrated Approach (IA), uses constraints to remove impossible candidates and then ranks the rest by preferences, with the strongest weight on the previous clause's center. The second, the Uncertainty Reasoning Approach (URA), never discards candidates. It treats every factor as a weighted preference and accepts or rejects each candidate through certainty-factor arithmetic. The texts come mainly from the computer-science sublanguage, as in the original systems. The study finds that the two strategies perform differently even with identical factors, and that combining them as a two-engine system works best. The broader lesson is that system builders should choose the resolution strategy as carefully as the factors.

Key points

  • Integrated Approach (IA). It is built from rule-based syntactic, semantic, domain, discourse and heuristic modules, plus a filter for non-anaphoric "it". Its discourse module estimates the center of each sentence, including the first, using an inference engine based on Bayes' formula.
  • Uncertainty Reasoning Approach (URA). It uses a modified version of van Melle's formula (from Buchanan & Shortliffe 84) to update a candidate's certainty factor until it crosses an acceptance threshold or falls below a rejection limit. Candidates are tested starting from the noun phrase closest to the anaphor. In the paper's worked example, a candidate at 0.5 that meets a symptom worth 0.45 rises to 0.725.
  • Main finding. The paper's own results report that combining IA and URA into a two-engine strategy outperformed either system alone. This partly agrees with Carter's earlier comparison of SPAR and the SRI Core Language Engine.
  • Factor dependence. The two object-preference factors (indicated by verbs and by nouns) are mutually dependent. Mitkov partly compensates by giving the noun-based factor a lower value, but argues that proper modelling requires dependence data from large annotated corpora.
  • Order of application. Applying preferences in descending versus ascending order of value made no essential difference to the outcome, but the ascending order made the search longer. Mitkov leans towards preference-only designs, because strict agreement constraints can wrongly exclude correct antecedents, for example collective nouns referred to by "they".

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

11 papers of our library cite it (of the 140673 whose bibliographies we read); 6 of them have a page here, 5 not yet.

Cite

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

@inproceedings{mitkov1997factors,
  title = {{Factors in anaphora resolution: they are not the only things that matter. A case study based on two different approaches}},
  author = {Mitkov, Ruslan},
  booktitle = {Operational Factors in Practical, Robust Anaphora Resolution for Unrestricted Texts},
  year = {1997},
  url = {https://aclanthology.org/W97-1303/}
}

References

14 references, 4 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. H. Alshawi - Resolving quasi logical forms. Computational Linguistics, 16:3, 1990 find
  2. B. Buchanan, FEd. Shortliffe Rule-based expert systems. Addison-Wesley, 1984 find
  3. J. Carbonell, R. Brown - - Anaphora resolution: a multi-strategy approach. Proceedings of the 12. International Conference on Computational Linguistics COLING'88, Budapest, August, 1988 find our card
  4. D. Carter - Interpreting anaphora in natural language texts. Chichester: Ellis Horwood, 1987 find
  5. David M. Carter - Control issues in anaphor resolution. Journal of Semantics, 7, 1990 find
  6. I. Dagan - Multilingual statistical approaches for natural language disambiguation (in Hebrew). PhD dissertation. Technion-Israel Institute of Technology, Haifa find
  7. Ido Dagan, John Justeson, Shalom Lappin, Hergert Leass and Amnon Ribak - Syntax and lexical statistics in anaphora resolution. Applied Artificial Intelligence, 9, 1995 find
  8. Sh. Lappin, H. Leass - An algorithm for pronominal anaphora resolution. Computational Linguistics, 20(4), 1994 find our card
  9. Mitkov R. - An integrated model for anaphora resolution. Proceedings of the 15th International Conference on Computational Linguistics COLING'94, Kyoto, Japan, 5-9 August 1994 find our card
  10. Mitkov R. - A new approach for tracking center. In Proceedings of the International Conference "New Methods in Language Processing", UMIST, Manchester, UK, 13-16 September 1994 find
  11. R. Mitkov - An uncertainy reasoning approach to anaphora resolution. Proceedings of the Natural Language Pacific Rim Symposium, 4-7 December 1995, Seoul, Korea find
  12. Mitkov R. - Anaphor resolution in Natural Language Processing and Machine Translation. Proceedings of the International Colloquium on Discourse Anaphora and Anaphora Resolution. Lancaster, 17-19 July 1996 (keynote speech) find
  13. C. Rico P6rez - Resolucidn de la an6fora discursiva mediante una estrategia de in-spiraci6n vectoral. Proceedings of the SEPLN'94 conference, Cordoba 20-22 July 1994 [Preul3 94 et all Preul3 S., Schmitz B., Hauenschild C., Umbach C. - Anaphora Resolution bz Machine Translation. In W. Ramm (ed): Stud… find
  14. E. Rich, S. LuperFoy - An architecture for anaphora resolution. Proceedings of the Second Conference on Applied Natural Language Processing, Austin, Texas, 9-12 February 1988 find our card

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/papers/mitkov-1997-factors-anaphora-resolution-they-are/