Elamur Research⌕
☰
home / papers / 2006 / Transferring Coreference Chains through Word Alignment

Transferring Coreference Chains through Word Alignment

Summary

Coreference-annotated corpora exist for only a few languages, so the authors examine whether English coreference annotation can be carried over to Romanian through a parallel text. Their data are the first chapter of Orwell's "1984" in English and in its Romanian translation, split into three parts and annotated with referential expressions (REs) and discourse entities (DEs) on both sides. English annotations were built from FDG-tagger output and then corrected by hand. Romanian was annotated manually and serves only as the gold standard. The method has three steps. The COWAL aligner links words across the two texts. Each English RE is then mapped to the Romanian span between its first and last aligned words, with the head required to be a noun, pronoun or numeral. Finally, the English clusters are copied over unchanged. Projected heads are very precise (about 95%) but miss roughly 30% of Romanian REs. Chains built on correctly found REs reach B-cubed F above 90%. Since precision matters most when annotators correct a draft, the approach works well as a pre-annotation step.

Key points

  • Corpus size: about 12.9k English and 13.1k Romanian words, 638 aligned sentences, 3,610 English REs and 3,422 gold Romanian REs. Of the English REs, 2,491 were transferred to Romanian, yielding 1,552 projected DEs against 1,818 English DEs.
  • Word alignment: COWAL (Tufiş et al.) scores 83.30% F-measure. An earlier version ranked first among 37 systems in the ACL2005 word-alignment shared task.
  • Detection results:
    • RE heads: precision about 95–96%, recall about 68–70%, F about 80–81%.
    • Full RE spans: F about 72–74%.
    • Spans of correctly detected REs: overlap about 90–92%.
  • Coreference chains:
    • Against all gold REs: MUC F of 62–65% and B-cubed F of 81–85%.
    • On correctly detected REs only: MUC F of 85–89% and B-cubed F of 92–95%.
  • Sources of error:
    • Faulty alignments.
    • Part-of-speech shifts in translation, such as an English adjective or verb rendered as a Romanian NP.
    • Material added by the translator.
    • English noun premodifiers turned into Romanian postmodifiers.
    • Translation choices that break English coreference links.

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

17 papers of our library cite it (of the 140673 whose bibliographies we read); 10 of them have a page here, 7 not yet.

Cite

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

@inproceedings{postolache2006transferring,
  title = {{Transferring Coreference Chains through Word Alignment}},
  author = {Postolache, Oana and Cristea, Dan and Orasan, Constantin},
  booktitle = {Proceedings of the Fifth International Conference on Language Resources and Evaluation (LREC'06)},
  year = {2006},
  url = {https://aclanthology.org/L06-1122/}
}

References

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

  1. ACE. 2003. Entity Detection and Tracking - Phase 1. www.itl.nist.gov/iaui/894.01/tests/ace/phase1/index.htm. find
  2. A. Bagga and B. Baldwin. 1998. Algorithms for scoring coreference chains. In Proceedings of the Linguistic Conference Workshop at LREC’98, pages 563–566. find our card
  3. S. M. Harabagiu and S. J. Maiorano. 2000. Multilingual coreference resolution. In Proceedings of the ANLP-NAACL2000, Seattle, WA. find our card
  4. L. Hirschman and N. Chinchor. 1997. MUC-7 Coreference Task Definition, version 3.0. In Proceedings of MUC-7. See also: http://www.muc.saic.com. ↗ our card
  5. R. Hwa, P. Resnik, A. Weinberg, C. Cabezas, and O. Kolak. 1994. Bootstrapping parsers via syntactic projection across. Computational Linguistics, 20:535–561. find
  6. T. Järvinen and P. Tapanainen. 1997. A Dependecy Parser for English. Technical report, University of Helsinky. find
  7. R. Mitkov and C. Barbu. 2000. Mutual enhancement of performance: Bilingual pronoun resolution for English and French. In Proceedings of DAARC2000. find
  8. S. Pado and M. Lapata. 2005. Cross-linguistic projection of role-semantic information. In Proceedings of HLT/EMNLP2005, pages 859–866, Vancouver, Canada. find
  9. O. Postolache and C. Forăscu. 2004. A coreference model on excerpt from a novel. In Proceeding of The European Summer School in Logic Language and Information - ESSLLI’2004, pages 202–213, Nancy, France. find
  10. O. Postolache. 2004. RARE - Robust Anaphora Resolution Engine. Master’s thesis, University of Iaşi. find
  11. D. Tufiş, R. Ion, A. Ceauşu, and D. Stefănescu. 2005. Combined aligners. In ACL2005 Workshop on “Building and Using Parallel Corpora: Data-driven Machine Translation and Beyond”, Ann Arbor, Michigan. find
  12. D. Tufiş, R. Ion, A. Ceauşu, and D. Stefănescu. 2006. Improved lexical alignment by combining multiple reified alignments. In Proceedings of the EACL2006. find
  13. R. Vieira and M. Poesio. 2000. An empirically based system for processing definite descriptions. Computational Linguistics, 26(4):539 – 593. find
  14. M. Vilain, J. Burger, J. Aberdeen, D. Connolly, and L. Hirshman. 1995. A model-theoretic coreference scoring scheme. In Proceedings of the MUC6, pages 45–52. find our card
  15. D. Yarowsky and G. Ngai. 2001. Inducing multilingual POS taggers and NP bracketers via robust projection across aligned corpora. In Proceedings of the NAACL2001. find

↑ back to top