Flexible finite-state lexical selection for rule-based machine translation

What it is about

The paper adds a lexical-selection module to the Apertium pipeline. When a dictionary offers several target words for one source word, a rule chooses one from the context. Rules are written in XML as context patterns, compiled into a finite-state transducer, and applied with a best-coverage dynamic program that prefers the fewest, longest matches. The rules can also be learned: a parallel corpus (Europarl, first 1.4 million lines) is passed through the lexical transfer, and for each ambiguous word the surrounding trigram and five-gram contexts of adjectives, nouns and verbs are counted against the aligned translation. Three filters are compared: no filter, dropping contexts seen once, and dropping rules whose translation is under half the word's frequency. On 434 News Commentary sentences, the filtered system had a lexical-selection error of about 18 to 25 percent (95 percent intervals), better than the most-frequent-aligned baseline and the linguist default. BLEU differences were small, and a paired bootstrap showed the filtered system was better in 95 percent of resamples.

Ideas

• I1. Use the paper's coarsest filter as a merge test. A meta-concept merge should be kept only if its dominant member is at least half of the word's total count. Then measure the reader and encoder bits with and without the test, which is the residual-driven lexicon of open problem 2.
• I2. Make every chosen chunk traceable to the context rule that chose it. The paper's requirement is that each translation can name its rules. The encoder in profile.rs should record which context level (concept with tag and neighbours, or concept alone) produced each chunk, so that a wrong word can be explained.
• I3. Replace fixed n-gram windows with the head-dependent context from the tree: the governor and the dependents of a meta token. The paper says fixed ordered contexts cannot see a word at any distance. Measure the encoder error on dev with tree contexts against n-gram contexts.

For Mova

The code is tools/translate: profile.rs (the encoder's choice among chunks by context) and meta.rs (the merge rule, open problem 2). The coverage algorithm is not needed, since Mova has no rule transducer, but the traceability requirement and the frequency filter are cheap to adopt. The n-gram rules are the baseline to beat, not the design to copy.

Topics: Machine translation

Cite

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

@inproceedings{tyers2012flexible,
  title = {{Flexible finite-state lexical selection for rule-based machine translation}},
  author = {Tyers, Francis M. and Sánchez-Martínez, Felipe and Forcada, Mikel L.},
  booktitle = {Proceedings of the 16th Annual Conference of the European Association for Machine Translation},
  year = {2012},
  url = {https://aclanthology.org/2012.eamt-1.54/}
}

References

24 references, 1 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Brandt, M. D., H. Loftsson, H. Sigurþórsson, and F. M. Tyers. 2011. Apertium-icenlp: A rule-based icelandic to english machine translation system. In Proceedings of the 16th Annual Conference of the European Association of Machine Translation, pages 217–224. find
2. Callison-Burch, C., P. Koehn, C. Monz, and O. F. Zaidan, editors. 2011. Proceedings of the Sixth Workshop on Statistical Machine Translation. Association for Computational Linguistics. find
3. Carpuat, M. and D. Wu. 2007. How phrase sense disambiguation outperforms word sense disambiguation for statistical machine translation. In Proceedings of the 11th International Conference on Theoretical and Methodological Issues in Machine Translation, pages 43–52. find
4. Cortés-Vaíllo, S. and S. Ortiz-Rojas. 2011. Using apertium linguistic data for tokenization to improve moses smt performance. In Proceedings of the International Workshop on Using Linguistic Information for Hybrid Machine Translation, LIHMT-2011, pages 29–35. System freq ling alig tlm all filt1 filt2 Total rules - 667 600 - 77,077 9,978 2,661 Called - 473 533 - 503 503 532 Table 2: Evaluation results for the seven systems on the news commentary test corpus. find
5. Dagan, I. and A. Itai. 1994. Word sense disambiguation using a second language monolingual corpus. Computational Linguistics, 40(4):563–596. find
6. Efron, B. and R. J. Tibshirani. 1994. An introduction to the Bootstrap. CRC Press. find
7. Federico, M., N. Bertoldi, and M. Cettolo. 2008. Irstlm: an open source toolkit for handling large scale language models. In Proceedings of Interspeech, Brisbane, Australia. find
8. Forcada, M. L., M. Ginestí-Rosell, J. Nordfalk, J. O’Regan, S. Ortiz-Rojas, J. A. Pérez-Ortiz, F. Sánchez-Martínez, G. Ramírez-Sánchez, and F. M. Tyers. 2011. Apertium: a free/open-source platform for rule-based machine translation. Machine Translation, 25(2):127–144. find
9. Ide, N. and J. Véronis. 1998. Word sense disambiguation: The state of the art. Computational Linguistics, 24(1):1–41. find
10. Karlsson, F., A. Voutilainen, J. Heikkilä, and A. Anttila. 1995. Constraint Grammar: A language independent system for parsing unrestricted text. Mouton de Gruyter. find
11. Koehn, P. 2004. Statistical significance tests for machine translation evaluation. In Proceedings of the Conference on Empirical Methods in Natural Language Processing, pages 388–395. find
12. Koehn, P. 2005. Europarl: A parallel corpus for statistical machine translation. In Proceedings of the 10th MT Summit, pages 79–86. find
13. Koehn, P. 2010. Statistical Machine Translation. Cambridge University Press, United Kingdom. find
14. Levenshtein, V. I. 1965. Binary codes capable of correcting deletions, insertions, and reversals. Doklady Akademii Nauk SSSR, 163(4):845–848. English translation in Soviet Physics Doklady, 10(8), 707– 710. find
15. Melero, M., A. Oliver, T. Badia, and T. Suñol. 2007. Dealing with bilingual divergences in mt using target language n-gram models. In Proceedings of the METIS-II Workshop: New Approaches to Machine Translation, CLIN17, pages 19–26. Error [42.8, 50.3] [25.4, 30.7] [19.3, 25.8] [37.0, 44.9] [21.3, 28.2] [20.3, 26.9] [17.9, 24.7] BLEU [0.1687, 0.1794] find
16. Och, F. J. and H. Ney. 2003. A systematic comparison of various statistical alignment models. Computational Linguistics, 29(1):19–51. Papineni, K., S. Roukos, T. Ward, and W-J. find
17. Zhu. 2002. Bleu: A method for automatic evaluation of machine translation. In Proceedings of the 40th Annual Meeting of the Assoc. Comp. Ling., pages 311–318. find our card
18. Sánchez-Martínez, F., J. A. Pérez-Ortiz, and M. L. Forcada. 2007. Integrating corpus-based and rule-based approaches in an open-source machine translation system. In Proceedings of the METIS-II Workshop: New Approaches to Machine Translation, CLIN17, pages 73–82. find
19. Sánchez-Martínez, F., J. A. Pérez-Ortiz, and M. L. Forcada. 2008. Using target-language information to train part-of-speech taggers for machine translation. Machine Translation, 22(1-2):29–66. find
20. Sánchez-Martínez, F., M. L. Forcada, and A. Way. 2009. Hybrid rule-based – example-based MT: Feeding apertium with sub-sentential translation units. In Proceedings of the 3rd Workshop on EBMT, pages 11–18. find
21. Venkatapathy, S. and S. Bangalore. 2007. Three models for discriminative machine translation using global lexical selection and sentence reconstruction. In Proceedings of SSST, NAACL-HLT/AMTA Workshop on Syntax and Structure in Statistical Translation, pages 152–159. find
22. Vickrey, D., L. Biewald, M. Teyssier, and D. Koller. 2005. Word-sense disambiguation for machine translation. In Proceedings of Human Language Technology Conference and Conference on Empirical Methods in Natural Language Processing, pages 771– 778. find
23. Wiechetek, L., F. M. Tyers, and T. Omma. 2010. Shooting at flies in the dark: Rule-based lexical selection for a minority language pair. LNAI, 6233:418–429. find
24. Zhang, Y. and S. Vogel. 2004. Measuring confidence intervals for the machine translation evaluation metrics. In Proceedings of The 10th International Conference on Theoretical and Methodological Issues in Machine Translation. find

This page in Markdown → https://md.mova.elamur.ai/free/en/docs/papers/tyers-2012-flexible-finite-state-lexical-selection/