Elamur ResearchβŒ•
☰
home / papers / 2012 / Using Sense-labeled Discourse Connectives for Statistical Machine Translation

Using Sense-labeled Discourse Connectives for Statistical Machine Translation

Summary

Phrase-based SMT systems often mistranslate discourse connectives such as since, while or though. The correct target-language word depends on which discourse relation (contrast, concession, cause, time, etc.) the connective signals in its context. The authors test whether resolving this ambiguity before translation helps English-to-French SMT.

They first build one Maximum Entropy classifier per connective for 13 ambiguous English connectives, trained on Penn Discourse Treebank (PDTB) data using second-level sense labels. Features include syntactic information, WordNet relations, temporal expressions and candidate translations. The classifiers reach state-of-the-art F1 of about 0.6–0.8.

They then feed the sense labels into Moses in three ways:

  • editing the phrase table,
  • training on Europarl with hand-labeled connectives,
  • training on Europarl with automatically labeled connectives.

The modified systems translate connectives more accurately without lowering BLEU. A combined system that applies labels only when the classifier is highly confident improves BLEU by 0.2–0.4 points. The authors describe this as the first attempt to join connective disambiguation and SMT, showing that discourse-level information can benefit translation.

Key points

  • Disambiguation data: PDTB (sections 02–21 for training, section 23 for testing). The 13 connectives covered are while, since, although, after, however, meanwhile, then, still, yet, indeed, rather, nevertheless and nonetheless.
  • Europarl annotation: Five connectives (although, even though, since, though, while) were manually labeled on Europarl via translation spotting. Annotators marked each connective's French equivalent, and these equivalents were then grouped into sense labels.
  • SMT setup: Moses phrase-based decoder with a 5-gram French language model built on Europarl v5 (346,803 training sentences). Systems were tuned with MERT on the News Commentary 2009 development set and tested on News Commentary and UN corpus sentences.
  • Evaluation: Gains in connective translation were counted manually on small test sets of 35 and 62 sentences.
  • Threshold-based combination: For since, a confidence threshold of 0.85 gave +0.39 BLEU (p = 0.005). For although, the difference had p = 0.12.

Our summary of the paper, in our own words β€” not its text. The work belongs to its authors; read it at the links above.

Fields: Linguistics & languages

Topics: Discourse analysis Β· Machine translation

Cited by

34 papers of our library cite it (of the 140673 whose bibliographies we read); 13 of them have a page here, 21 not yet.

Cite

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

@inproceedings{meyer2012using,
  title = {{Using Sense-labeled Discourse Connectives for Statistical Machine Translation}},
  author = {Meyer, Thomas and Popescu-Belis, Andrei},
  booktitle = {Proceedings of the Joint Workshop on Exploiting Synergies between Information Retrieval and Machine Translation (ESIRMT) and Hybrid Approaches to Machine Translation (HyTra)},
  year = {2012},
  url = {https://aclanthology.org/W12-0117/}
}

References

25 references, 7 of them with a card of ours. β†— leads to the work, Β«findΒ» searches for one we have no id for yet.

  1. Satanjeev Banerjee and Ted Pedersen. 2002. An Adapted Lesk Algorithm for Word Sense Disambiguation Using WordNet. In Alexander Gelbukh, editor, Computational Linguistics and Intelligent Text Processing, LNCS 2276, pages 117–171. Springer, Berlin/Heidelberg. find
  2. Marine Carpuat and Dekai Wu. 2007. Improving Statistical Machine Translation using Word Sense Disambiguation. Proc. of EMNLP-CoNLL, pages 61– 72, Prague. find
  3. Bruno Cartoni, Sandrine Zufferey, Thomas Meyer, and Andrei Popescu-Belis. 2011. How Comparable are Parallel Corpora? Measuring the Distribution of General Vocabulary and Connectives. Proc. of the 4th Workshop on Building and Using Comparable Corpora (BUCC), pages 78–86, Portland, OR. find
  4. Pi-Chuan Chang, Dan Jurafsky, and Christopher D. Manning. 2009. Disambiguating β€˜DE’ for Chinese-English Machine Translation. Proc. of the Fourth Workshop on Statistical Machine Translation at EACL-2009, Athens. find
  5. Eugene Charniak and Mark Johnson. 2005. Coarse-to-fine n-best Parsing and MaxEnt Discriminative Reranking. Proc. of the 43rd Annual Meeting of the ACL, pages 173–180, Ann Arbor, MI. find our card
  6. Laurence Danlos and Charlotte Roze. 2011. Traduction (Automatique) des Connecteurs de Discours. Actes 18e Conférence sur le Traitement Automatique des Langues Naturelles (TALN), Montpellier. find
  7. Robert Elwell and Jason Baldridge. 2008. Discourse Connective Argument Identification with Connective Specific Rankers. Proc. of the 2nd IEEE International Conference on Semantic Computing (ICSC), pages 198–205, Santa Clara, CA. find
  8. Hugo Hernault, Helmut Prendinger, David A. duVerle, and Mitsuru Ishizuka. 2010. HILDA: A Discourse Parser using Support Vector Machine classification. Dialogue and Discourse, 3(1):1–33. find our card
  9. Philipp Koehn, et al. 2007. Moses: Open Source Toolkit for Statistical Machine Translation. Proc. of 45th Annual Meeting of the ACL, Demonstration Session, pages 177–180, Prague. find
  10. Philipp Koehn. 2005. Europarl: A Parallel Corpus for Statistical Machine Translation. Proc. of MT Summit X, pages 79–86, Phuket. find
  11. Ziheng Lin, Hwee Tou Ng, and Min-Yen Kan. 2010. A PDTB-styled End-to-end Discourse Parser. Technical Report TRB8/10, School of Computing, National University of Singapore. find
  12. Jianjun Ma, Degen Huang, Haixia Liu, and Wenfeng Sheng. 2011. POS Tagging of English Particles for Machine Translation. Proc. of MT Summit XIII, pages 57–63, Xiamen. find
  13. Christopher Manning and Dan Klein. 2003. Optimization, MaxEnt Models, and Conditional Estimation without Magic. Tutorial at HLT-NAACL and 41st ACL conferences, Edmonton, AB and Sapporo. find
  14. Daniel Marcu. 2000. The Theory and Practice of Discourse Parsing and Summarization. A Bradford Book. The MIT Press, Cambridge, MA. find
  15. George A. Miller. 1995. WordNet: A Lexical Database for English. Communications of the ACM, 38(11):39–41. find our card
  16. Eleni Miltsakaki, Nikhil Dinesh, Rashmi Prasad, Aravind Joshi, and Bonnie Webber. 2005. Experiments on Sense Annotations and Sense Disambiguation of Discourse Connectives. Proc. of the 4th Workshop on Treebanks and Linguistic Theories (TLT), Barcelona. find
  17. Franz Josef Och and Hermann Ney. 2003. A Systematic Comparison of Various Statistical Alignment Models. Computational Linguistics, 29(1):19–51. find
  18. Franz Josef Och. 2003. Minimum Error Rate Training in Statistical Machine Translation. Proc. of the 41st Annual Meeting of the ACL, pages 160–167, Sapporo. find
  19. Kishore Papineni, Salim Roukos, Todd Ward, and Wei-Jing Zhu. 2002. BLEU: A method for Automatic Evaluation of Machine Translation. Proc. of 40th Annual Meeting of the ACL, pages 311–318, Philadelphia, PA. find our card
  20. Emily Pitler and Ani Nenkova. 2009. Using Syntax to Disambiguate Explicit Discourse Connectives in Text. Proc. of the 47th Annual Meeting of the ACL and the 4th International Joint Conference of the AFNLP (ACL-IJCNLP), Short Papers, pages 13–16, Singapore. find our card
  21. Rashmi Prasad, Nikhil Dinesh, Alan Lee, Eleni Miltsakaki, Livio Robaldo, Aravind Joshi, and Bonnie Webber. 2008. The Penn Discourse Treebank 2.0. Proc. of the 6th International Conference on Language Resources and Evaluation (LREC), pages 2961–2968, Marrakech. find our card
  22. Marc Verhagen and James Pustejovsky. 2008. Temporal Processing with the TARSQI Toolkit. Proc. of the 22nd International Conference on Computational Linguistics (COLING), Demonstrations, pages 189–192, Manchester, UK. find
  23. Ben Wellner, James Pustejovsky, Catherine Havasi, Roser Sauri, and Anna Rumshisky. 2006. Classification of Discourse Coherence Relations: An Exploratory Study using Multiple Knowledge Sources. Proc. of the 7th SIGdial Meeting on Discourse and Dialog, pages 117–125, Sydney. find our card
  24. Ben Wellner. 2009. Sequence Models and Ranking Methods for Discourse Parsing. PhD thesis, Brandeis University, Waltham, MA. find
  25. Ying Zhang and Stefan Vogel. 2010. Significance Tests of Automatic Machine Translation Evaluation Metrics. Machine Translation, 24(1):51–65. find

↑ back to top

This page in Markdown β†’