Elamur Research⌕
☰
home / papers / 2011 / Fill-up versus interpolation methods for phrase-based SMT adaptation

Fill-up versus interpolation methods for phrase-based SMT adaptation

Summary

The paper addresses domain adaptation for phrase-based SMT when only a small in-domain parallel corpus is available alongside large general-purpose corpora for the same language pair. The test case is the IWSLT 2011 TED talks task, which has roughly 100K in-domain sentence pairs plus out-of-domain news, UN and European Parliament data. The authors study phrase table fill-up. A separate model is trained on each corpus. The in-domain table is kept exactly as it is, and background entries are added only for phrase pairs it lacks. A binary provenance feature marks the borrowed entries, and its MERT-learned weight effectively discounts out-of-domain scores. The same merging is applied to the lexicalized reordering table and extended to a cascade over several ranked background corpora. Pruning filters restrict which background pairs are added. Fill-up is compared with linear interpolation (uniform weights) and log-linear table combination. It reaches similar BLEU and NIST scores while producing more compact models with fewer weights to tune. This matters because fill-up needs no mixture-weight estimation beyond standard MERT.

Key points

  • Data: The IWSLT11 TED training set is small (about 100K parallel sentences), and the background corpora are news, UN and Europarl. The topical variety of TED makes in-domain coverage insufficient.
  • Model structure: The filled-up table is the union of the in-domain and background tables. Each pair has five standard features (two phrase probabilities, two lexical weights, phrase penalty) plus one provenance feature. This feature takes exp(0) for in-domain pairs and exp(1) for background pairs, unlike Nakov's 0.5/1 variant.
  • Pruning filters tested:
    • NewSourceMaxLength caps the source length of added pairs.
    • OnlyNewSourcePhrases adds background pairs only for uncovered source phrases.
    • OnlyNewSourceWords adds them only for pairs containing source words unknown to the in-domain vocabulary.
  • Baselines:
    • Linear interpolation uses uniform weights, since optimizing mixture weights has no agreed best method.
    • Log-linear combination adds a separate feature set per table. It can produce duplicate translation options that interfere with beam and option-limit pruning.
  • Release: Fill-up and cascaded fill-up were announced for open-source release in the Moses toolkit under GNU LGPL.

Interesting ideas

  • I1. Trust hierarchy instead of mixing: Keep scores from the most reliable source untouched and use secondary sources only to fill coverage gaps. This is attractive for low-resource settings like Ukrainian, where a small curated corpus should not be diluted by large noisy data.
  • I2. Provenance feature as a learned penalty: A single indicator on borrowed entries lets the standard tuner decide how much to discount them. The same trick works for any merged lexicon or knowledge base where source reliability differs.
  • I3. Cascading by relevance: With several background resources ranked by relevance, fill them in one after another, each with its own indicator weight. This gives per-source control without multiplying the full feature set.
  • I4. Selective import filters: Admit background entries only when they cover new source phrases or unknown words. This keeps models compact and is a cheap, reusable recipe for vocabulary-driven augmentation of dictionaries or translation memories.
  • I5. Shared alignment, separate scoring: Word alignment can be computed on all corpora concatenated, while phrase extraction and scoring stay per corpus. This improves alignment quality without losing information about which corpus each pair came from.

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 · Machine translation

Topics: Machine translation

Cited by

61 papers of our library cite it (of the 140673 whose bibliographies we read); 2 of them have a page here, 59 not yet.

Cite

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

@inproceedings{bisazza2011fill,
  title = {{Fill-up versus interpolation methods for phrase-based SMT adaptation}},
  author = {Bisazza, Arianna and Ruiz, Nick and Federico, Marcello},
  booktitle = {Proceedings of the 8th International Workshop on Spoken Language Translation: Evaluation Campaign},
  year = {2011},
  url = {https://aclanthology.org/2011.iwslt-evaluation.18/}
}

References

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

  1. S. Besling and H. Meier, “Language model speaker adaptation,” in Proceedings of the 4th European Conference on Speech Communication and Technology, vol. 3, Madrid, Spain, 1995, pp. 1755–1758. find
  2. P. Nakov, “Improving English-Spanish Statistical Machine Translation: Experiments in Domain Adaptation, Sentence Paraphrasing, Tokenization, and Recasing. ,” in Workshop on Statistical Machine Translation, Association for Computational Linguistics, 2008. find
  3. M. Federico, L. Bentivogli, M. Paul, and S. Stüker, “Overview of the IWSLT 2011 Evaluation Campaign,” in International Workshop on Spoken Language Translation (IWSLT), San Francisco, CA, 2011. find
  4. M. Eck, S. Vogel, and A. Waibel, “Language model adaptation for statistical machine translation based on information retrieval,” in Proceedings of the International Conference on Language Resources and Evaluation (LREC), Lisbon, Portugal, 2004, pp. 327–330. find
  5. G. Foster and R. Kuhn, “Mixture-model adaptation for SMT,” in Proceedings of the Second Workshop on Statistical Machine Translation. Prague, Czech Republic: Association for Computational Linguistics, June 2007, pp. 128–135. [Online]. Available: http://www.aclweb.org/anthology/W/W07/W07-0217 ↗ our card
  6. F. J. Och, “Minimum Error Rate Training in Statistical Machine Translation,” in Proceedings of the 41st Annual Meeting of the Association for Computational Linguistics, E. Hinrichs and D. Roth, Eds., 2003, pp. 160–167. [Online]. Available: http://www.aclweb.org/anthology/P03-1021.pdf ↗
  7. P. Koehn and J. Schroeder, “Experiments in Domain Adaptation for Statistical Machine Translation,” in Proceedings of the Second Workshop on Statistical Machine Translation. Prague, Czech Republic: Association for Computational Linguistics, June 2007, pp. 224–227. [Online]. Available: http://www.aclw… ↗ our card
  8. K. Yasuda, R. Zhang, H. Yamamoto, and E. Sumita, “Method of Selecting Training Data to Build a Compact and Efficient Translation Model,” in International Joint Conference on Natural Language Processing, 2008. find
  9. A. Axelrod, X. He, and J. Gao, “Domain Adaptation via Pseudo In-Domain Data Selection,” in Conference on Empirical Methods in Natural Language Processing,, 2011, pp. 355–362. find our card
  10. J. Niehues and A. Waibel, “Domain adaptation in statistical machine translation using factored translation models,” in Proceedings of the 14th Annual Conference of the European Association for Machine Translation (EAMT), St. Raphael, France, 2010. [Online]. Available: http://www.mt-archive.info/EAMT… ↗
  11. P. Koehn, F. J. Och, and D. Marcu, “Statistical phrase-based translation,” in Proceedings of HLT-NAACL 2003, Edmonton, Canada, 2003, pp. 127– 133. [Online]. Available: http://aclweb.org/anthology-new/N/N03/N03-1017.pdf ↗
  12. C. Tillmann, “A unigram orientation model for statistical machine translation,” in Proceedings of the Joint Conference on Human Language Technologies and the Annual Meeting of the North American Chapter of the Association of Computational Linguistics (HLT-NAACL), 2004. find our card
  13. P. Koehn, A. Axelrod, A. B. Mayne, C. Callison-Burch, M. Osborne, and D. Talbot, “Edinburgh system description for the 2005 IWSLT speech translation evaluation,” in Proc. of the International Workshop on Spoken Language Translation, October 2005. find our card
  14. M. Galley and C. D. Manning, “A simple and effective hierarchical phrase reordering model,” in EMNLP ’08: Proceedings of the Conference on Empirical Methods in Natural Language Processing. Morristown, NJ, USA: Association for Computational Linguistics, 2008, pp. 848–856. find
  15. P. Koehn, H. Hoang, A. Birch, C. Callison-Burch, M. Federico, N. Bertoldi, B. Cowan, W. Shen, C. Moran, R. Zens, C. Dyer, O. Bojar, A. Constantin, and E. Herbst, “Moses: Open Source Toolkit for Statistical Machine Translation,” in Proceedings of the 45th Annual Meeting of the Association for Computa… find
  16. M. Diab, K. Hacioglu, and D. Jurafsky, “Automatic Tagging of Arabic Text: From Raw Text to Base Phrase Chunks,” in HLT-NAACL 2004: Short Papers, D. M. Susan Dumais and S. Roukos, Eds. Boston, Massachusetts, USA: Association for Computational Linguistics, May 2 - May 7 2004, pp. 149–152. find
  17. H. Johnson, J. Martin, G. Foster, and R. Kuhn, “Improving translation quality by discarding most of the phrasetable,” in In Proceedings of EMNLP-CoNLL 07, 2007, pp. 967–975. find our card
  18. F. Och and H. Ney, “A systematic comparison of various statistical alignment models,” Computational Linguistics, vol. 29, no. 1, pp. 19–51, 2003. find
  19. S. Riezler and J. T. Maxwell, “On some pitfalls in automatic evaluation and significance testing for MT,” in Proceedings of the ACL Workshop on Intrinsic and Extrinsic Evaluation Measures for Machine Translation and/or Summarization. Ann Arbor, Michigan: Association for Computational Linguistics, Ju… find
  20. E. Hasler, B. Haddow, and P. Koehn, “Margin Infused Relaxed Algorithm for Moses,” The Prague Bulletin of Mathematical Linguistics, vol. 96, pp. 69–78, 2011. find

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/bisazza-2011-fill-up-versus-interpolation-methods/