Elamur Research⌕
☰
home / papers / 2008 / Who is Who and What is What: Experiments in Cross-Document Co-Reference

Who is Who and What is What: Experiments in Cross-Document Co-Reference

Summary

Linking mentions of the same real-world person or organization across many documents involves two problems. One entity may appear under several names, such as Mahmoud Abbas and Abu Mazen. One name may also belong to several different entities. Baron and Freedman (BBN) present a language-independent system that handles both problems. They evaluated it in the ACE 2008 global entity task on English and Arabic collections of about 10K documents each.

The SERIF information extraction engine first detects document-level entities. A name variation module then scores pairs of corpus name strings using aliases, string similarity and translation resources. For each name and its accepted variants, a complete-link agglomerative clustering step separates the candidate entities. Because training data was scarce, features are combined with manually tuned heuristics.

The system beats an exact-match baseline in both languages, and the gains are larger on entities that occur in several documents. Clustering small subsets keeps runtime manageable. Feeding cross-document results back also improved within-document extraction. This suggests that corpus-level knowledge can strengthen document-level IE.

Key points

  • Results:
    • English: 8.4% relative error reduction over exact name match, rising to 14.2% for entities seen in more than one document.
    • Arabic: 7% reduction, rising to 12.4% on the same subset.
  • Evaluation data: ACE 2008 GEDR covered named persons and organizations, with over 10K documents per language and roughly 400 annotated reference documents per language.
  • Name variation features:
    • web-mined and corpus-mined aliases;
    • character edit distance;
    • token subset trees;
    • SERIF relation and coreference statistics;
    • translation dictionaries drawn from MT training data and Wikipedia cross-language links.
  • Scalability on English:
    • about 135K document entities and 62,516 unique name strings, each starting one clustering instance;
    • 36% of instances were skipped, and 99.7% of runs finished in under 1 second;
    • the largest instance ("Xinhua", 1848 entities) took 2.6 hours.
  • Persons vs. organizations: organizations scored noticeably worse than persons. In the name-only local task, ORG B3 was 82.6 versus 90.1 for PER, partly because the name-versus-nominal distinction is fuzzier for organizations.

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); 2 of them have a page here, 11 not yet.

Cite

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

@inproceedings{baron2008who,
  title = {{Who is Who and What is What: Experiments in Cross-Document Co-Reference}},
  author = {Baron, Alex and Freedman, Marjorie},
  booktitle = {Proceedings of the 2008 Conference on Empirical Methods in Natural Language Processing},
  year = {2008},
  url = {https://aclanthology.org/D08-1029/}
}

References

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

  1. Artiles, Javier, Julio Gonzalo. & Felisa Verdejo.. 2005. A Testbed for People Searching Strategies. In the WWW. find
  2. SIGIR 2005 Conference. Salvador, Brazil. find
  3. Artiles, Javier, Julio Gonzalo. & Satochi Sekine.. 2007. The SemEval-2007 WePS Evaluation: Establishing a benchmark for the Web People Search Task. Proceedings of the 4th International Workshop on Semantic Evaluations (SemEval-2007), pages 64–69, Prague, Czech. find
  4. Bagga, Amit & Breck Baldwin. 1998a. Algorithms for Scoring Coreference Chains. In Proceedings of the Linguistic Coreference Workshop at the First International Conference on Language Resources and Evaluation (LREC'98), pages 563-566. find our card
  5. Bagga, Amit & Breck Baldwin. 1998b. Entity-Based Cross-Document Coreferencing Using the Vector Space Model. In Proceedings of the 36th Annual Meeting of the Association for Computational Linguistics and the 17th International Conference on Computational Linguistics (COLING-ACL’98), pages 79-85. find
  6. Day, David.,Jason Duncan, Claudio Guiliano, Rob Hall, Janet Hitzeman,Su Jian, Paul McNamee, Gideon Mann, Stanley Yong & Mike Wick. 2007. CDC Features. Johns Hopkins Summer Workshop on Cross-Document Entity Disambiguation. http://www.clsp.jhu.edu/ws2007/groups/elerfed/docu ments/fullCDED.ppt Gooi, Ch… ↗ our card
  7. Lloyd, Levon., Andrew Mehler & Steven Skiena 2006. find
  8. Identifying Co-referential Names Across Large Corpora. Combinatorial Pattern Matching. 2006, pages 12-23, Barcelona, Spain. find
  9. Linguistic Data Consortium 2008a. ACE (Automatic Content Extraction) English Annotation Guidelines for Entities Version 6.6 2008.06.13. . Linguistic Data Consortium, Philadelphia. http://projects.ldc.upenn.edu/ace/docs/English-Entities-Guidelines_v6.6.pdf ↗
  10. Linguistic Data Consortium, 2008b. find
  11. ACE 2008 XDOC Pilot Data V2.1. LDC2007E64. Linguistic Data Consortium, Philadelphia. find
  12. Mann, Gideon S. & Yarowsky, David. 2003. Unsupervised Personal Name Disambiguation In Proceedings of the seventh conference on Natural language learning at HLT-NAACL, pages 33-40. find
  13. NIST Speech Group. 2008. find
  14. The ACE 2008 evaluation plan: Assessment of Detection and Recognition of Entities and Relations Within and Across Documents. http://www.nist.gov/speech/tests/ace/2008/doc/ace08 -evalplan.v1.2d.pdf Ramshaw, Lance, E. Boschee, S. Bratus, S. Miller, R. Stone, R. Weischedel and A. Zamanian: “Experiments… ↗
  15. Sista, S, R. Schwartz, T. Leek, and J. Makhoul. An Algorithm for Unsupervised Topic Discovery from Broadcast News Stories. In Proceedings of ACM HLT, San Diego, CA, 2002. find

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/baron-2008-who-is-who-what-is/