Using Coreference Chains for Text Summarization

Summary

The paper tests whether coreference chains can drive generic extractive summarization, on the assumption that a text is largely about one central entity. The authors add a module to LaSIE, the Sheffield information extraction system built for MUC coreference tasks. The module takes every chain in LaSIE's discourse model and picks one "best" chain using heuristics: how many mentions it has (length), how much of the text it covers (spread), where it starts, and focus information from a focus-based pronoun resolution algorithm. The summary is simply the source sentences that mention the chosen entity, in their original order. Unlike lexical-chain methods (Barzilay and Elhadad), this approach relies on true identity of reference, including pronouns. For evaluation, the authors reuse MUC-6 materials as an automatic substitute for SUMMAC-style manual judgments. They test how well summaries of 30 MUC-6 texts support template filling, a question-answering-like task. Topic-filtered single-chain summaries were very short and gave very low recall. Generic summaries were longer and reached higher recall. The work is a preliminary step toward coreference-based summarization and automated summary evaluation.

Key points

System and pipeline. The method runs on LaSIE (MUC-6/MUC-7). LaSIE processes text in three stages: lexical preprocessing, two-pass chart parsing with semantic interpretation, and discourse interpretation. Each coreference chain is stored as merged instances with character offsets.
Selection criteria. Chains are chosen by length, spread (byte distance between first and last mention), chain start, and focus-based variants. These criteria can be used alone or combined, and ties are broken with a secondary criterion.
Evaluation design. The authors summarized 30 MUC-6 evaluation texts and measured how much MUC-6 template-filling performance survived. The MUC-6 scenario involves three kinds of entity: an organisation, a managerial position, and the individual who holds it.
Topic-linked results. Summaries restricted to chains linked to the topic description scored roughly 1% recall and 80% precision. They kept about 6.1% of the original word count. The authors attribute the low recall to a single chain rarely covering all three entity types.
Generic results. Coreference-chain summaries scored about 30–33% recall and 65–67% precision. Focus-chain summaries scored about 19% recall and 73% precision, at roughly 18–20% of the original length. Some figures, such as word-count ratios, are garbled in the extracted PDF. The authors caution that these scores mostly reflect overlap between text topics and the query topic.

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

13 papers of our library cite it (of the 140673 whose bibliographies we read); 4 of them have a page here, 9 not yet.

Cite

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

@inproceedings{azzam1999using,
  title = {{Using Coreference Chains for Text Summarization}},
  author = {Azzam, Saliha and Humphreys, Kevin and Gaizauskas, Robert},
  booktitle = {Coreference and Its Applications},
  year = {1999},
  url = {https://aclanthology.org/W99-0211/}
}

References

11 references, 2 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

S. Azzam, K. Humphreys, and R. Gaizauskas. 1998. Evaluating a focus-based approach to anaphora resolution. In Proceedings of COLING-ACL'98, pages 74-78. find our card
A. Bagga and B. Baldwin. 1998. Entity-based cross-document coreferencing using the vector space model. In Proceedings of the COLING-ACL'98 Joint Conference (the 17th International Conference on Computational Linguistics and 36th Annual Meeting of the Association for Computational Linguisities, Montr… find
B. Baldwin and T.S. Morton. 1998. Dynamic coreference-based summarization. In Proceedings of the Conference on Empirical Methods in Natural Language Processing (EMNLP'98). find our card
R. Barzilay and M. Elhadad. 1997. Using lexical chains for text summarization. In Proceedings of the ACL Workshop on Intelligent Scalable Text Summarization, pages 10-17, Madrid, July. find
DARPA: Defense Advanced Research Projects Agency. 1998. Proceedings of the Seventh Message Understanding Conference (MUC-7). Available at http;//www .muc. s a i c . com. find
R. Gaizanskas and K. Humphreys. 1997. Quantitative Evaluation of Coreference Algorithms in an Information Extraction System. Technical report CS - 97 - 19, Department of Computer Science, University of Sheffield. find
R. Gaizauskas, T. Wakao, K Humphreys, H. Cunningham, and Y. Wilks. 1995. Description of the LaSIE system as used for MUC-6. In Proceedings of the Sixth Message Understanding Conference (MUC-6), pages 207-220. Morgan Kaufmann. find
M.A.K. Halliday and R. Hasan. 1976. Cohesion in English. Longman, London. find
I. Mani, D. House, G. Klein, L. Hirschman, L. Obrst, T. Firmin, M. Chrzanowski, and B. Sundheim. 1998. The TIPSTER SUMMAC text summarization evaluation: Final report. MITRE Technical Report MTR 98W0000138, MITRE. find
J. Morris and G. Hirst. 1994. Lexical cohesion computed by thesanral relations as an indicator of the structure of text. Computational Linguistics, 17(1):21-45. find
K. Sparck Jones. 1993. What might be in summary? In Knorz, Krause, and Womser-Hacker, editors, Information Retrieval 93: Von der Mod-ellie~ung zur Anwendung, pages 9-26. Universit-atsverlag Konstanz. find

↑ back to top

📄 Using Coreference Chains for Text Summarization