Elamur Research⌕
☰
home / papers / 2010 / Learning Recursive Segments for Discourse Parsing

Learning Recursive Segments for Discourse Parsing

Summary

Most earlier work on discourse segmentation was developed on the RST corpus, where elementary discourse units (EDUs) never nest. That allowed segmentation to be treated as binary boundary detection. SDRT-style annotation, however, permits embedded EDUs, for example for appositions and non-restrictive relatives. The authors address this setting using the first validated portion of the French Annodis corpus: 47 documents of Wikipedia and L'est républicain text, containing 1,445 EDUs, about 10% of them nested.

Following the clause boundary identification approach from CoNLL-2001, each token is labelled as opening a segment, closing one, forming a one-token segment, or none of these. A regularized MaxEnt model trained with MegaM makes these predictions using lexical markers and morpho-syntactic features from the Macaon chunker and the Syntex dependency parser. Chunk boundaries guide a resampling step. A left-to-right and right-to-left repair heuristic then adds or removes brackets so the output is well formed.

With repair, the system reaches roughly 73% F-score on complete EDUs, close to CBI results. This makes it a usable base for SDRT parsing and for pre-annotation.

Key points

  • Data: 47 validated Annodis documents (14,384 tokens, about 33 EDUs per text). The full corpus is planned at 100–150 texts.
  • Features include lemma, POS, chunk tags, dependency paths (Syntex), outward chunk sequences, context n-grams (1 < n ≤ 6), discourse connectives and reporting verbs.
  • Resampling relies on Macaon output: tokens inside chunks and sentence-boundary tokens are dropped from training. This cuts nothing instances from about 12,000 to about 9,200.
  • Without repair, EDU F-score is 0.575 (recall 0.427). With repair it rises to 0.733 (P 0.748, R 0.719), and 98% of segmentations are well formed. Left and right boundary F-scores are about 0.88 and 0.89.
  • Adding previous labels as features with Viterbi decoding reduced performance. The learning curve suggests more data would bring only modest gains.

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: Discourse analysis

Cited by

15 papers of our library cite it (of the 140673 whose bibliographies we read); 4 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. arXiv 1003.5372

@inproceedings{afantenos2010learning,
  title = {{Learning Recursive Segments for Discourse Parsing}},
  author = {Afantenos, Stergos and Denis, Pascal and Muller, Philippe and Danlos, Laurence},
  booktitle = {Proceedings of the Seventh International Conference on Language Resources and Evaluation (LREC'10)},
  year = {2010},
  eprint = {1003.5372},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/1003.5372}
}

References

19 references, 5 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. N. Asher. 1993. Reference to Abstract Objects in Discourse. Kluwer. find
  2. J. Baldridge, N. Asher, and J. Hunter. 2007. Annotation for and robust parsing of discourse structure on unrestricted texts. Zeitschrift fur Sprachwissenschaft, 26:213–239. find
  3. A. Berger, S. Della Pietra, and V. Della Pietra. 1996. A maximum entropy approach to natural language processing. Computational Linguistics, 22(1):39–71. find our card
  4. D. Bourigault, C. Fabre, C. Frérot, M.-P. Jacques, and S. Ozdowska. 2005. Syntex, analyseur syntaxique de corpus. In Actes des 12èmes journées sur le Traitement Automatique des Langues Naturelles. find
  5. Lynn Carlson, Daniel Marcu, and Mary Ellen Okurowski. 2003. Building a discourse-tagged corpus in the framework of rhetorical structure theory. In Jan van Kuppevelt and Ronnie Smith, editors, Current Directions in Discourse and Dialogue, pages 85–112. Kluwer Academic Publishers. 0.8 + ♦ + ♦ + ♦ 0.6… find our card
  6. Figure 3: Learning curve for the series of experiments including chunking. find
  7. Xavier Carreras and Luís Màrquez. 2001. Boosting trees for clause splitting. In Walter Daelemans and Rémi Zajac, editors, Proceedings of CoNLL-2001, pages 73–75. Toulouse, France. find
  8. Eva Ejerhed. 1996. Finite state segmentation of discourse into clauses. In Proceedings of the ECAI ’96 Workshop on Extended finite state models of language, Budapest. find
  9. Seeger Fisher and Brian Roark. 2007. The utility of parse-derived features for automatic discourse segmentation. In ACL. find
  10. Robert Malouf. 2002. A comparison of algorithms for maximum entropy parameter estimation. In Proceedings of the Sixth Workshop on Natural Language Learning, pages 49–55, Taipei, Taiwan. find our card
  11. W. Mann and S. Thompson. 1987. Rhetorical structure theory : a theory of text organization. Technical report, Information Science Institute. find
  12. Daniel Marcu. 2000. The rhetorical parsing of unrestricted texts: a surface-based approach. 26(3):395–448. find our card
  13. Alexis Nasr and Alexandra Volanschi, 2006. Finite-State Methods and Natural Language Processing, volume 4002 of Lecture Notes in Computer Science, chapter Integrating a Part of Speech Tagger and a Chunker Implemented as Weighted Finite-State Machines, pages 167– 178. Springer. find
  14. Rashmi Prasad, Eleni Miltsakaki, Aravind Joshi, and Bonnie Webber. 2004. Annotation and + ♦ + ♦ + ♦ ♦ + ♦ + ♦ + F-measure on Left F-measure on Right ♦ + 30 25 20 find
  15. Number of files data mining of the penn discourse treebank. In Proceedings of the ACL Workshop on Discourse Annotation, Barcelona, Spain. find
  16. Radu Soricut and Daniel Marcu. 2003. Sentence level discourse parsing using syntactic and lexical information. In HLT/NAACL, Edmonton, Canada. find our card
  17. Caroline Sporleder and Mirella Lapata. 2005. Discourse chunking and its application to sentence compression. In HLT/EMNLP. find
  18. Milan Tofiloski, Julian Brooke, and Maite Taboada. 2009. A syntactic and lexical-based discourse segmenter. In Proceedings of the ACL-IJCNLP 2009 Conference Short Papers, pages 77–80, Suntec, Singapore, August. Association for Computational Linguistics. find
  19. Bonnie L. Webber. 2004. D-ltag: extending lexicalized tag to discourse. Cognitive Science, 28(5):751–779. find

↑ back to top

This page in Markdown →