Elamur Research⌕
☰
home / papers / 2006 / Semantic Role Labeling for Coreference Resolution

Semantic Role Labeling for Coreference Resolution

Summary

Most machine-learning coreference resolvers depend on surface cues such as string overlap, sentence distance and grammatical form, so they often fail when linking pronouns or common nouns requires semantic knowledge. Ponzetto and Strube test whether automatically produced semantic role labels can fill this gap. They reimplemented the Soon et al. (2001) pairwise system with 12 features and a Maximum Entropy classifier. They then ran the ASSERT parser, an SVM-based tagger that outputs PropBank arguments, and added two features: the argument–predicate pairs of each mention in a candidate pair (I SEMROLE, J SEMROLE).

The system was prototyped on MUC-6/MUC-7 and evaluated on the ACE 2003 training corpus (BNEWS and NWIRE, split 60/20/20). Adding SRL raised the MUC F1 on merged ACE data from 67.3 to 68.8, mainly through better recall. The largest gains were for pronouns and common nouns, while proper names saw no benefit. Feature analysis ranked the SRL features just below string matching and alias. The study shows that shallow event-level semantics can help coreference where morpho-syntactic features are insufficient.

Key points

  • The duplicated Soon baseline matched or exceeded the original system on MUC. On MUC-6 F1 was 65.3 versus 62.3, and on MUC-7 it was 61.1 versus 60.4.
  • On ACE 2003 merged data, adding SRL changed the scores as follows:
    • Recall rose from 54.5 to 56.4, and precision went from 88.0 to 88.2.
    • Pronoun accuracy rose from 34.7 to 40.3, and common-noun accuracy from 20.4 to 22.0.
    • Proper-name accuracy fell from 53.1 to 52.1.
  • F1 also improved within each section: BNEWS went from 60.6 to 64.0, and NWIRE from 69.0 to 69.8.
  • ASSERT labels covered 11,406 of 32,502 extracted noun phrases, using 2,801 distinct argument–predicate pairs. A label was attached to a mention only when the argument and the noun phrase shared a head.
  • In the ablation, removing both SEMROLE features cost 1.50 F1, the 5th-largest drop of the 12 removals listed. Removing the noisy WN CLASS feature cost only 0.56 F1.

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 · Semantic role labeling

Cited by

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

Cite

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

@inproceedings{ponzetto2006semantic,
  title = {{Semantic Role Labeling for Coreference Resolution}},
  author = {Ponzetto, Simone Paolo and Strube, Michael},
  booktitle = {Demonstrations},
  year = {2006},
  doi = {10.3115/1608974.1608993},
  url = {https://doi.org/10.3115/1608974.1608993}
}

References

15 references, 8 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Berger, A., S. A. Della Pietra & V. J. Della Pietra (1996). A maximum entropy approach to natural language processing. Computational Linguistics, 22(1):39–71. find our card
  2. Carreras, X. & L. Màrquez (2005). Introduction to the CoNLL-2005 shared task: Semantic role labeling. In Proc. of CoNLL-05, pp. 152–164. find our card
  3. Charniak, E. (1973). Jack and Janet in search of a theory of knowledge. In Advance Papers from the Third International Joint Conference on Artificial Intelligence, Stanford, Cal., pp. 337–343. find
  4. Chinchor, N. (2001). Message Understanding Conference (MUC) 7. LDC2001T02, Philadelphia, Penn: Linguistic Data Consortium. find
  5. Chinchor, N. & B. Sundheim (2003). Message Understanding Conference (MUC) 6. LDC2003T13, Philadelphia, Penn: Linguistic Data Consortium. find
  6. Gildea, D. & D. Jurafsky (2002). Automatic labeling of semantic roles. Computational Linguistics, 28(3):245–288. find our card
  7. Ji, H., D. Westbrook & R. Grishman (2005). Using semantic relations to refine coreference decisions. In Proc. HLT-EMNLP ’05, pp. 17–24. find our card
  8. Kehler, A., D. Appelt, L. Taylor & A. Simma (2004). The (non)utility of predicate-argument frequencies for pronoun interpretation. In Proc. of HLT-NAACL-04, pp. 289– 296. find
  9. Klebanov, B. & P. Wiemer-Hastings (2002). The role of wor(l)d knowledge in pronominal anaphora resolution. In Proceedings of the International Symposium on Reference Resolution for Natural Language Processing, Alicante, Spain, 3–4 June, 2002, pp. 1–8. find
  10. Mitchell, A., S. Strassel, M. Przybocki, J. Davis, G. Doddington, R. Grishman, A. Meyers, A. Brunstain, L. Ferro & B. Sundheim (2003). TIDES Extraction (ACE) 2003 Multilingual Training Data. LDC2004T09, Philadelphia, Penn.: Linguistic Data Consortium. find
  11. Ng, V. & C. Cardie (2002). Improving machine learning approaches to coreference resolution. In Proc. of ACL-02, pp. 104–111. find our card
  12. Palmer, M., D. Gildea & P. Kingsbury (2005). The proposition bank: An annotated corpus of semantic roles. Computational Linguistics, 31(1):71–105. find our card
  13. Pradhan, S., W. Ward, K. Hacioglu, J. H. Martin & D. Jurafsky (2004). Shallow semantic parsing using support vector machines. In Proc. of HLT-NAACL-04, pp. 233–240. find
  14. Soon, W. M., H. T. Ng & D. C. Y. Lim (2001). A machine learning approach to coreference resolution of noun phrases. Computational Linguistics, 27(4):521–544. find our card
  15. Vilain, M., J. Burger, J. Aberdeen, D. Connolly & L. Hirschman (1995). A model-theoretic coreference scoring scheme. In Proceedings of the 6th Message Understanding Conference (MUC-6), pp. 45–52. find our card

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/papers/ponzetto-2006-semantic-role-labeling-coreference-resolution/