Elamur Research⌕
☰
home / papers / 1998 / An IR Approach for Translating New Words from Nonparallel, Comparable Texts

An IR Approach for Translating New Words from Nonparallel, Comparable Texts

Summary

The paper addresses how to find translations for new words, such as recent proper names or freshly coined terms, that a machine translation lexicon lacks and that no parallel corpus yet contains. Fung and Yee use nonparallel but comparable English–Chinese newspaper text downloaded from the Web: Hong Kong Standard and Mingpao issues from 12–31 December 1997. Their method borrows from information retrieval. Each word is represented as a vector over the "seed words" of an existing bilingual MT lexicon that co-occur with it in sentences. Dimensions are weighted with TF/IDF and down-weighted when a seed translation is ambiguous. Candidate translations are then ranked with cosine- and Dice-style similarity scores and their products. Applied to unknown Chinese words, mostly names, almost every unambiguous one has its correct English translation within the top 100 candidates, and six are ranked first. The work shows that comparable, continuously updated monolingual sources can support lexicon expansion without costly parallel data.

Key points

  • Corpus: about 3 Mb of English (Hong Kong Standard) and 8.8 Mb of Chinese (Mingpao) from the same three-week period. The authors treat the two papers as comparable in size and topic.
  • The seed bridges come from the 20,000-entry lexicon of a broad-coverage English–Chinese MT system. The motivating unknown word was the Chinese term for flu (liougan).
  • The context of liougan shares 233 seed words with the context of flu but only 121 with Africa. IDF lowers the weight of the ubiquitous Hong Kong relative to virus.
  • The paper defines similarity measures S0–S7. These include plain cosine, cosine with IDF, a Dice variant, versions with 1/k confidence weighting for the k-th lexicon candidate, and products of measures. The measures with IDF and confidence weighting place flu first for liougan, while plain S0 does not.
  • Errors include Lunar (not actually unknown in English) and Yeltsin (correct answer ranked 4th). Precision is described as good, but recall is limited by segmentation and word ambiguity.

Interesting ideas

  • I1. Translate words the way IR matches documents. Treating a word's sentence-level context as a "document" makes standard IR machinery, such as TF/IDF weighting and cosine or Dice similarity, directly reusable for cross-lingual lexicon induction. This is handy for any language pair with only comparable corpora, for example Ukrainian–English news.
  • I2. Discount ambiguous bridges. Dividing a seed pair's weight by the rank of that translation in the lexicon is a cheap way to reduce noise from one-to-many dictionary entries. It is relevant wherever the bilingual dictionary is rich but ambiguous.
  • I3. Combine complementary similarity scores. The authors reason that comparing context segments sits between query-to-document and document-to-document matching. They therefore multiply measures designed for each case, a simple ensemble trick worth trying in other matching tasks.
  • I4. Use time-aligned web news as a corpus. Downloading same-period newspapers in both languages makes the corpus timely and topically aligned, which is exactly what new names and terms require. Collection can also run automatically in the background.
  • I5. Bootstrap the seed lexicon. The authors propose adding high-confidence new pairs back into the seed list so that each iteration can uncover more translations. This is a natural self-training loop for growing low-resource dictionaries.

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: Natural language processing

Cited by

126 papers of our library cite it (of the 140884 whose bibliographies we read); 7 of them have a page here, 119 not yet.

Cite

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.3115/980845.980916

@inproceedings{fung1998ir,
  title = {{An IR Approach for Translating New Words from Nonparallel, Comparable Texts}},
  author = {Fung, Pascale and Yee, Lo Yuen},
  booktitle = {36th Annual Meeting of the Association for Computational Linguistics and 17th International Conference on Computational Linguistics, Volume 1},
  year = {1998},
  doi = {10.3115/980845.980916},
  url = {https://doi.org/10.3115/980845.980916}
}

References

31 references, 7 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. A. Bookstein. 1983. Explanation and generalization of vector models in information retrieval. In Proceedings of the 6th Annual International Conference on Research and Development in Information Retrieval, pages 118-132. find
  2. P. Brown, J. Lai, and R. Mercer. 1991. Aligning sentences in parallel corpora. In Proceedings of the P9th Annual Conference of the Association for Computational Linguistics. find our card
  3. Table 6: English words most similar to ~,~/liougan SO 0.181114 0.088879 0.085886 0.081411 Lei flu Tang b'-~,~ ~,l~ Ap ~'~ $4 flu Lei Beijing poultry ~,~ ~,~ ~r~ ~,r~, 0.120879 0.097577 0.068657 0.065833 $5 0.086287 0.040090 0.028157 0.024500 $6 0.010430 0.001854 0.001840 0.001682 $7 0.007669 0.00195… find
  4. Kenneth Church. 1993. Char.align: A program for aligning parallel texts at the character level. In Proceedings of the 31st Annual Conference of the Association for Computational Linguistics, pages 1-8, Columbus, Ohio, June. find
  5. W. Bruce Croft. 1984. A comparison of the cosine correlation and the modified probabilistic model. In Information Technology, volume 3, pages 113-114. find
  6. Ido Dagan and Kenneth W. Church. 1994. Termight: Identifying and translating technical terminology. In Proceedings of the 4th Conference on Applied Natural Language Processing, pages 34-40, Stuttgart, Germany, October. find
  7. Ido Dagan and Alon Itai. 1994. Word sense disambiguation using a second language monolingual corpus. In Computational Linguistics, pages 564-596. find
  8. William B. Frakes and Ricardo Baeza-Yates, editors. 1992. Information Retrieval: Data structures Algorithms. Prentice-Hall. find
  9. Pascale Fung and Kenneth Church. 1994. Kvec: A new approach for aligning parallel texts. In Proceedings of COL-ING 9J, pages 1096-1102, Kyoto, Japan, August. find
  10. Pascale Fung and Kathleen McKeown. 1997. Finding terminology translations from non-parallel corpora. In The 5th Annual Workshop on Very Large Corpora, pages 192-202, Hong Kong, Aug. find
  11. Pascale Fung and Dekai Wu. 1994. Statistical augmentation of a Chinese machine-readable dictionary. In Proceedings of the Second Annual Workshop on Very Large Corpora, pages 69-85, Kyoto, Japan, June. find
  12. Pascale Fung. 1995a. Compiling bilingual lexicon entries from a non-parallel English-Chinese corpus. In Proceedings of the Third Annual Workshop on Very Large Corpora, pages 173-183, Boston, Massachusettes, June. find
  13. Pascale Fung. 1995b. A pattern matching method for finding noun and proper noun translations from noisy parallel corpora. In Proceedings of the 33rd Annual Conference of the Association for Computational Linguistics, pages 236-233, Boston, Massachusettes, June. find
  14. William Gale and Kenneth Church. 1991. Identifying word correspondences in parallel text. In Proceedings of the Fourth Darpa Workshop on Speech and Natural Language, Asilomar. find
  15. William A. Gale and Kenneth W. Church. 1993. A program for aligning sentences in bilingual corpora. Computational Linguistics, 19(1):75-102. find our card
  16. William A. Gale and Kenneth W. Church. 1994. Discrimination decisions in 100,000 dimensional spaces. Current Issues in Computational Linguisitcs: In honour of Don Walker, pages 429-550. find
  17. W. Gale, K. Church, and D. Yarowsky. 1992. Estimating upper and lower bounds on the performance of word-sense disambiguation programs. In Proceedings of the 30th Conference of the Association for Computational Linguistics. find
  18. W. Gale, K. Church, and D. Yarowsky. 1993. A method for disambiguating word senses in a large corpus. In Computers and Humanities, volume 26, pages 415-439. find
  19. K. Sparck Jones. 1979. Experiments in relevance weighting of search terms. In Information Processing and Management, pages 133-144. find
  20. Martin Kay and Martin R6scheisen. 1993. Text-Translation alignment. Computational Linguistics, 19(1):121-142. find our card
  21. Robert Korfhage. 1995. Some thoughts on similarity measures. In The SIGIR Forum, volume 29, page 8. find
  22. Julian Kupiec. 1993. An algorithm for finding noun phrase correspondences in bilingual corpora. In Proceedings of the 31st Annual Conference of the Association for Computational Linguistics, pages 17-22, Columbus, Ohio, June. find
  23. Reinhard Rapp. 1995. Identifying word translations in non-parallel texts. In Proceedings of the 35th Conference of the Association of Computational Linguistics, student session, pages 321-322, Boston, Mass. find our card
  24. G. Salton and C. Buckley. 1988. Term-weighting approaches in automatic text retrieval. In Information Processing and Management, pages 513-523. find
  25. G. Salton and C. Yang. 1973. On the specification of term values in automatic indexing, volume 29. find
  26. Hinrich Shiitze. 1992. Dimensions of meaning. In Proceedings of Supercomputing '92. find
  27. M. Simard, G Foster, and P. Isabelle. 1992. Using cognates to align sentences in bilingual corpora. In Proceedings of the Forth International Conference on Theoretical and Methodological Issues in Machine Translation, Montreal, Canada. find our card
  28. Frank Smadja, Kathleen McKeown, and Vasileios Hatzsivas-siloglou. 1996. Translating collocations for bilingual lexicons: A statistical approach. Computational Linguistics, 21(4):1-38. find our card
  29. Howard R. Turtle and W. Bruce Croft. 1992. A comparison of text retrieval methods. In The Computer Journal, volume 35, pages 279-290. find
  30. Dekai Wu and Xuanyin Xia. 1994. Learning an English-Chinese lexicon from a parallel corpus. In Proceedings of the First Conference of the Association for Machine Translation in the Americas, pages 206-213, Columbia, Maryland, October. find
  31. D. Yarowsky. 1995. Unsupervised word sense disambiguation rivaling supervised methods. In Proceedings of the 33rd Conference o.f the Association for Computational Linguistics, pages 189-196. Association for Computational Linguistics. find our card

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/fung-1998-ir-approach-translating-new-words/