Elamur Research⌕
☰
home / papers / 2012 / Zero Pronoun Resolution can Improve the Quality of J-E Translation

Zero Pronoun Resolution can Improve the Quality of J-E Translation

Summary

Spoken Japanese often omits subjects, objects and possessors. A statistical Japanese-to-English system then tends to produce English output that is also missing these pronouns. The authors asked whether filling in these gaps before translation helps a phrase-based system. They took the tourism-oriented BTEC corpus from IWSLT07 and manually annotated the dropped pronouns and their antecedents, using the English references as a guide. They also built a simple rule-based resolver that inserts "I", "you" or "it" depending on predicate type and sentence modality. They trained and tuned Moses on three versions of the source (original, rule-completed, human-completed) and compared them. Since BLEU is a poor measure of pronoun quality, they also proposed an antecedent precision/recall/F-measure. With gold completions, BLEU barely changed, but antecedent F-measure rose noticeably, mainly through better recall. The size of the gain depended on pronoun person and case. The findings indicate that zero pronoun resolution is useful for J-E SMT, provided sentence modality is recognised reliably.

Key points

  • Data: BTEC (IWSLT07). The splits were train (39,953 segments, 1 reference), devset1-3 for tuning (1,512 segments, 16 references) and test (489 segments, 16 references). devset4 and devset5 were left out because their reference counts differ.
  • Pipeline: MeCab segmented the Japanese and the Moses tokenizer handled the English. Moses was used with default settings, tuned with MERT on the completed sentences.
  • Baseline resolver: it targets sentences that have a predicate but no "wa", "mo" or "ga". It adds "watashi wa" (declarative verb) or "anata wa" (question verb) at the start of the sentence, or "sore wa" for adjective and copula predicates. Imperatives are skipped.
  • BLEU was nearly flat: 45.4% for the baseline resolver and 45.6% for the human completion, both only slightly above the uncompleted input. Antecedent F-measure improved clearly with the human completion, and every person category (1st, 2nd, 3rd) beat the original.
  • Error analysis showed two patterns. First-person gains were smaller than expected, while second-person and possessive pronouns benefited more. The rule system sometimes mistook imperatives and wrongly inserted "I", which shows that modality detection is critical.

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 · Machine translation

Cited by

10 papers of our library cite it (of the 140673 whose bibliographies we read); 2 of them have a page here, 8 not yet.

Cite

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

@inproceedings{taira2012zero,
  title = {{Zero Pronoun Resolution can Improve the Quality of J-E Translation}},
  author = {Taira, Hirotoshi and Sudoh, Katsuhito and Nagata, Masaaki},
  booktitle = {Proceedings of the Sixth Workshop on Syntax, Semantics and Structure in Statistical Translation},
  year = {2012},
  url = {https://aclanthology.org/W12-4213/}
}

References

18 references, 3 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Kohji Dohsaka. 1990. Identifying the referents of zero-pronouns in japanese based on pragmatic constraint interpretation. In Proceedings of ECAI, pages 240–245. find
  2. C.S. Fordyce. 2007. Overview of the iwslt 2007 evaluation campaign. In Proceedings of the International Workshop on Spoken Language Translation, pages 1– 12. find
  3. M. Furuichi, J. Murakami, M. Tokuhisa, and M. Murata. 2011. The effect of complement subject in japanese to english statistical machine translation (in Japanese). In Proceedings of the 17th Annual Meeting of The Parallel Corpus for Training Japanese Shoushou ukagai tai koto ga ari masu ga. 。 ---- --… find
  4. Zero pronoun annotation by hand or baseline system Parallel Corpus for Test Japanese honkon ryokou ni tsuite siri tain desu ga. I would like to know about the Hong Kong trip. Figure 4: Outline of the experiment Association for Natural Language Processing (NLP-2012). find
  5. C. Hardmeier, M. Federico, and F.B. Kessler. 2010. Modelling pronominal anaphora in statistical machine translation. In Proceedings of the seventh International Workshop on Spoken Language Translation (IWSLT), pages 283–289. find our card
  6. H. Isozaki, K. Sudoh, H. Tsukada, and K. Duh. 2010. Head finalization: A simple reordering rule for sov languages. In Proceedings of the Joint Fifth Workshop on Statistical Machine Translation and Metrics-MATR, pages 244–251. Association for Computational Linguistics. find
  7. Mark Johnson. 2002. A simple pattern-matching algorithm for recovering empty nodes and their an-English I have some questions to ask . ---- ---- Tuning Parallel Corpus for Tuning English Japanese ---- ---- Zero pronoun annotation by hand or baseline system Completed Sentences ---- ---- Decoding Syst… find
  8. Evaluation tecedents. In Proceedings of 40th Annual Meeting of the Association for Computational Linguistics, pages 136–143, Philadelphia, Pennsylvania, USA, July. Association for Computational Linguistics. find
  9. G. Kikui, E. Sumita, T. Takezawa, and S. Yamamoto. 2003. Creating corpora for speech-to-speech translation. In Proceedings of EUROSPEECH, pages 381– 384. find
  10. Philipp Koehn, Hieu Hoang, Alexandra Birch, Chris Callison-Burch, Marcello Federico, Nicola Bertoldi, Brooke Cowan, Wade Shen, Christine Moran, Richard Zens, Chris Dyer, Ondrej Bojar, Alexandra Constantin, and Evan Herbst. 2007. Moses: open source toolkit for statistical machine translation. In Proc… find
  11. Can you send it by this evening ? (Ref) Can you deliver them by this evening? It by this evening ? worse hurry up Isoi-de . Hurry up Isoi-de . I-TOP watashi-wa Hurry up . I ‘m in a hurry . (Ref) Hurry up. Unchanged map-BY point_out would QUES chizu-de sashi-te morae-masu ka. You-TOP anata-wa Would y… find
  12. H. Nakaiwa and S. Yamada. 1997. Automatic identification of zero pronouns and their antecedents within aligned sentence pairs. In Proc. of the 3rd Annual Meeting of the Association for Natural Language Processing. find
  13. S. Nariyama. 2003. Ellipsis and reference tracking in Japanese, volume 66. John Benjamins Publishing Company. find
  14. Franz Josef Och. 2003. Minimum error rate training for statistical machine translation. In Proc. of the ACL. find
  15. Kishore Papineni, Salim Roukos, Todd Ward, and Wei-Jing Zhu. 2002. BLEU: A method for automatic evaluation of machine translation. In Proc. of the 40th Annual Conference of the Association for Computational Linguistics (ACL-02). find our card
  16. K. Yamamoto, E. Sumita, O. Furuse, and H. Iida. 1997. Ellipsis resolution in dialogues via decision-tree learning. In Proc. of NLPRS, volume 97. Citeseer. find
  17. K. Yoshimoto. 1988. Identifying zero pronouns in japanese dialogue. In Proceedings of the 12th conference on Computational linguistics-Volume 2, pages 779–784. Association for Computational Linguistics. find our card
  18. Today’s evening by send would QUES kyou-no yuugata made-ni todoke-te morae-masu ka . you-TOP anata-wa map-BY point_out would QUES chizu-de sashi-te morae-masu ka. Table 2: The Type Distributions of Zero Pronouns in Test Set Type First personal pronoun Second personal pronoun Third personal pronoun a… find

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/papers/taira-2012-zero-pronoun-resolution-can-improve/