Elamur Research⌕
☰
home / papers / 2010 / BART: A Multilingual Anaphora Resolution System

BART: A Multilingual Anaphora Resolution System

Summary

BART is a modular coreference resolution toolkit, first built for English, which the authors adapt to German and Italian for SemEval-2010 task 1. The system splits the work into a preprocessing pipeline, a mention factory, feature extraction, a decoder that trains a pairwise classifier, and an encoder that applies it and builds coreference chains. A separate LanguagePlugin holds all language-dependent knowledge. Retraining on a new corpus gives a language-agnostic baseline. This baseline is then improved with language-specific mention detection and features: GermaNet relatedness and morphological stem matching for German, hand-built aliasing rules for Italian, and an entity-mention model based on Semantic Trees for English. German and Italian use a pairwise maximum entropy classifier. On the SemEval data, BART ranked first among regular-track systems for German in both mention detection and coreference. It also performed well on English across metrics and gave reasonable Italian results despite head-alignment mismatches. The work is an early step toward a freely available, easily portable coreference resolver that keeps linguistic knowledge apart from the learning machinery.

Key points

  • The architecture has five components plus the LanguagePlugin. The preprocessing output is stored as separate XML layers.
  • English mention detection: base noun phrases are taken from parses produced by the Petrov et al. (2006) constituent parser.
  • English entity model: each discourse entity gets a Semantic Tree "filecard" with types, attributes (premodifiers) and relations (prepositional postmodifiers). Filecards are merged when mentions are linked, and incompatible ones block wrong links. StrudelRelatedness adds a cosine score from the Strudel distributional model.
  • German: SemEval dependency trees are converted to constituents, and non-referring positions such as expletive "es" are skipped. Added features are NodeDistance, PartialMorphMatch (stems via SMOR/Morphisto) and GermanetRelatedness (Pathfinder library, Wu & Palmer measure).
  • Italian: mentions are detected with Typhoon and DEMention, and heads come from MaltParser trained on TUT. The mention detector uses affixes (1–4 characters), TextPro output, phone-book and Wikipedia gazetteers, and WordNet within a ±2-word window. AliasItalian covers designators such as "S.p.a". The paper reports no numeric scores, only rankings relative to other systems.

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); 3 of them have a page here, 10 not yet.

Cite

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

@inproceedings{broscheit2010bart,
  title = {{BART: A Multilingual Anaphora Resolution System}},
  author = {Broscheit, Samuel and Poesio, Massimo and Ponzetto, Simone Paolo and Rodriguez, Kepa Joseba and Romano, Lorenza and Uryupina, Olga and Versley, Yannick and Zanoli, Roberto},
  booktitle = {Proceedings of the 5th International Workshop on Semantic Evaluation},
  year = {2010},
  url = {https://aclanthology.org/S10-1021/}
}

References

12 references, 6 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Marco Baroni, Brian Murphy, Eduard Barbu, and Massimo Poesio. 2010. Strudel: A corpus-based semantic model based on properties and types. Cognitive Science, 34(2):222–254. find
  2. Silvana Marianela Bernaola Biggio, Claudio Giuliano, Massimo Poesio, Yannick Versley, Olga Uryupina, and Roberto Zanoli. 2009. Local entity detection and recognition task. In Proc. of Evalita-09. find
  3. Samuel Broscheit, Simone Paolo Ponzetto, Yannick Versley, and Massimo Poesio. 2010. Extending BART to provide a coreference resolution system for German. In Proc. of LREC ’10. find our card
  4. Marc Finthammer and Irene Cramer. 2008. Exploring and navigating: Tools for GermaNet. In Proc. of LREC ’08. find
  5. Joakim Nivre, Johan Hall, Jens Nilsson, Atanas Chanev, Gulsen Eryigit, Sandra Kübler, Svetoslav Marinov, and Erwin Marsi. 2007. Maltparser: A language-independent system for data-driven dependency parsing. Natural Language Engineering, 13(2):95–135. find
  6. Slav Petrov, Leon Barett, Romain Thibaux, and Dan Klein. 2006. Learning accurate, compact, and interpretable tree annotation. In Proc. of COLING-ACL-06. find our card
  7. Massimo Poesio, Olga Uryupina, and Yannick Versley. 2010. Creating a coreference resolution system for Italian. In Proc. of LREC ’10. find our card
  8. Helmut Schmid, Arne Fitschen, and Ulrich Heid. 2004. SMOR: A German computational morphology covering derivation, composition and inflection. In Proc. of LREC ’04. find our card
  9. Wee Meng Soon, Hwee Tou Ng, and Daniel Chung Yong Lim. 2001. A machine learning approach to coreference resolution of noun phrases. Computational Linguistics (Special Issue on Computational Anaphora Resolution), 27(4):521–544. find our card
  10. Yannick Versley, Simone Paolo Ponzetto, Massimo Poesio, Vladimir Eidelman, Alan Jern, Jason Smith, Xiaofeng Yang, and Alessandro Moschitti. 2008. BART: A modular toolkit for coreference resolution. In Proceedings of the Linguistic Coreference Workshop at the International Conference on Language Reso… find our card
  11. Zhibiao Wu and Martha Palmer. 1994. Verb semantics and lexical selection. In Proc. of ACL-94, pages 133– 138. find
  12. Roberto Zanoli, Emiliano Pianta, and Claudio Giuliano. 2009. Named entity recognition through redundancy driven classifier. In Proc. of Evalita-09. find

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/broscheit-2010-bart-multilingual-anaphora-resolution-system/