Learning Domain-Specific Transfer Rules: An Experiment with Korean to English Translation

What it is about

An MT system that learns lexico-structural transfer rules from parsed bitexts, for domain-specific Korean-to-English translation. Source and target sentences are parsed, the parse nodes are aligned, and candidate rules are built from any sub-pattern of the parse features (labels, roles, tense, number, person). Rules are kept by statistical co-occurrence and then filtered by error: a rule stays only if it helps on the training data. Evaluation used a test set of 50 Korean sentences. The Bleu scores were 0.0802 for a commercial system (Babelfish), 0.0767 for the transfer dictionary alone and 0.0950 for the dictionary with induced rules. A human ranking preferred the induced-rule system about 20 percent more often than the commercial one. The authors say the number of rule candidates grows exponentially with the size of the parse pairs, so long sentences need to be cut into smaller units.

Ideas

• I1. Error-driven filtering of merges: keep a concept merge in the lexicon only if it lowers the reader and encoder bits on the training verses, and drop it otherwise. This is the paper's filter applied to open problem 2, with the existing rule of three or more tokens fixed as a second check.
• I2. Cut long verses into clauses before candidate generation, as the paper suggests, so the number of candidate pairs stays bounded. Count candidates per verse before and after the cut.
• I3. Parse features as match features: use the UD FEATS (number, person, tense) in the tree pattern match, and measure which feature, added on its own, reduces the residual. This is open problem 5 in a measurable form.

For Mova

The error-driven filter is the most direct use: it is the count-based version of the rule that a merge must pay for itself, and it is cheap to test on the Bible pairs. Clause cutting addresses the candidate growth that a verse-level reader will meet. The feature-match idea needs the UD features that Mova's trees already carry. The paper's Bleu numbers come from a small Korean test set and are not compared with Mova.

Topics: Machine translation

Cite

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

@inproceedings{lavoie2002learning,
  title = {{Learning Domain-Specific Transfer Rules: An Experiment with Korean to English Translation}},
  author = {Lavoie, Benoit and White, Michael and Korelsky, Tanya},
  booktitle = {COLING-02: Machine Translation in Asia},
  year = {2002},
  url = {https://aclanthology.org/W02-1610/}
}

References

21 references, 2 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Philip Clarkson and Ronald Rosenfeld. 1997. Statistical Language Modeling Using the CMU-Cambridge Toolkit. In Proceedings of Eurospeech’97. find
2. Bonnie Dorr. 1994. Machine translation divergences: A formal description and proposed solution. Computational Linguistics, 20(4):597–635. find our card
3. Ulrich Germann, Michael Jahr, Kevin Knight, Daniel Marcu and Kenji Yamada. 2001. Fast Decoding and Optimal Decoding for Machine Translation. In Proceedings of ACL’01, Toulouse, France. find
4. Chung hye Han, Benoit Lavoie, Martha Palmer, Owen Rambow, Richard Kittredge, Tanya Korelsky, Nari Kim, and Myunghee Kim. 2000. Handling Structural Divergences and Recovering Dropped Arguments in a Korean-English Machine Translation System. In Proceedings of the Fourth Conference on Machine Translation in the Americas (AMTA’00), Mision Del Sol, Mexico. find
5. Daniel Karp and Yves Schabes. 1992. A Wide Coverage Public Domain Morphological Analyzer for English. In Proceedings of the Fifteenth International Conference on Computational Linguistics (COLING’92), pages 950–955, Nantes, France. find
6. Benoit Lavoie and Owen Rambow. 1997. RealPro – A Fast, Portable Sentence Realizer. In Proceedings of the Conference on Applied Natural Language Processing (ANLP’97), Washington, DC. find
7. Benoit Lavoie, Richard Kittredge, Tanya Korelsky, and Owen Rambow. 2000. A framework for MT and multilingual NLG systems based on uniform lexico-structural processing. In Proceedings of ANLP/NAACL 2000, Seattle, Washington. find
8. Benoit Lavoie, Michael White, and Tanya Korelsky. 2001. Inducing Lexico-Structural Transfer Rules from Parsed Bitexts. In Proceedings of the ACL 2001 Workshop on Data-driven Machine Translation, pages 17–24, Toulouse, France. find
9. Christopher D. Manning and Hinrich Schutze. 1999. Foundations of Statistical Natural Language Processing. MIT Press. find
10. H. Maruyama and H. Watanabe. 1992. Tree cover Search Algorithm for Example-Based Translation. In Proceedings of the Fourth International Conference on Theoretical and Methodological Issues in Machine Translation (TMI’92), pages 173–184. find
11. Yuji Matsumoto, Hiroyuki Hishimoto, and Takehito Utsuro. 1993. Structural Matching of Parallel Texts. In Proceedings of the 31st Annual Meetings of the Association for Computational Linguistics (ACL’93), pages 23–30. find
12. Igor Mel’čuk. 1988. Dependency Syntax. State University of New York Press, Albany, NY. find
13. Adam Meyers, Roman Yangarber, Ralph Grishman, Catherine Macleod, and Antonio Moreno-Sandoval. 1998. Deriving Transfer Rules from Dominance-Preserving Alignments. In Proceedings of COLING-ACL’98, pages 843–847. find
14. Makoto Nagao. 1984. A framework of a mechanical translation between Japenese and English by analogy principle. In A. Elithorn and R. Banerji, editors, Artificial and Human Intelligence. NATO Publications. find
15. Alexis Nasr, Owen Rambow, Martha Palmer, and Joseph Rosenzweig. 1997. Enriching lexical transfer with cross-linguistic semantic features. In Proceedings of the Interlingua Workshop at the MT Summit, San Diego, California. find
16. Franz Josef Och and Hermann Ney. 2000. Improved Statistical Alignment Models. In Proceedings of ACL’00, pages 440– 447, Hongkong, China. find
17. Kishore Papineni, Salim Roukos, Todd Ward, and Wei-Jing Zhu. 2001. Bleu: a Method for Automatic Evaluation of Machine Translation. In IBM technical report, RC22176. find our card
18. Stephen D. Richardson, William B. Dolan, Arul Menezes, and Monica Corston-Oliver. 2001. Overcoming the Customization Bottleneck using Example-based MT. In Proceedings of the ACL 2001 Workshop on Data-driven Machine Translation, pages 9–16, Toulouse, France. find
19. Satoshi Sato and Makoto Nagao. 1990. Toward Memory-based Translation. In Proceedings of the 13th International Conference on Computational Linguistics (COLING’90), pages 247–252. find
20. Fei Xia and Martha Palmer. 2001. Converting Dependency Structures to Phrase Structures. In Notes of the First Human Language Technology Conference, San Diego, California. find
21. Juntae Yoon, Seonho Kim, and Mansuk Song. 1997. New parsing method using global association table. In Proceedings of the 5th International Workshop on Parsing Technology. find

This page in Markdown → https://md.mova.elamur.ai/free/en/docs/papers/lavoie-2002-learning-domain-specific-transfer-rules/