Elamur Research⌕
☰
home / papers / 1993 / Text-Translation Alignment

Text-Translation Alignment

Summary

Kay and Röscheisen address the task of aligning a text with its translation, mainly at the sentence level. They use only evidence inside the two texts, with no dictionaries or other knowledge about the languages. Their central observation is that if two words reliably translate each other, the sentences containing them must also correspond. A sparse word-level alignment can therefore produce a much fuller sentence-level one.

The method is an iterative relaxation process:

  • It starts from a band of plausible sentence pairs near the diagonal.
  • It proposes word pairs whose occurrences across those candidate sentences match closely.
  • It uses the most reliable word pairs to commit to some sentence correspondences.
  • It narrows the candidate band and repeats.

Committed pairs are never withdrawn, so the process stops once nothing new appears. Heuristics for morphological variation are also added. Tested on a Scientific American article and its German version in Spektrum der Wissenschaft, the algorithm finds the correct sentence alignment within a few passes. This supports translator reference tools, lexicography and statistical MT built on bilingual corpora such as the Canadian Hansard.

Key points

  • Four core tables drive the algorithm: Word-Sentence Index (WSI), Alignable Sentence Table (AST), Word Alignment Table (WAT) and Sentence Alignment Table (SAT). The SAT from the final pass is essentially the output.
  • The initial AST is derived only from text lengths. A sentence near the middle may pair with up to O(√n) sentences on the other side, which gives overall cost of about O(n√n). The authors note that Gale and Church or Brown, Lai, and Mercer length-based methods could seed the AST and make this step roughly linear.
  • In the test data, sentences mostly align one-to-one and one-to-two pairings are uncommon. No one-to-three cases were observed.
  • In the English–German texts, one English character corresponds to roughly 1.2 German characters. Paragraph-length correlation is 0.952, compared with 0.991 reported by Gale and Church, which suggests these texts are relatively hard to align.
  • The approach depends on one-to-one links between technical terms and proper names. Proposed future work covers one-to-many links and recurring phrases, for example identified with Patricia trees.

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 · Natural language processing

Cited by

90 papers of our library cite it (of the 140673 whose bibliographies we read); 4 of them have a page here, 86 not yet.

Cite

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

@article{kay1993text,
  title = {{Text-Translation Alignment}},
  author = {Kay, Martin and Roscheisen, Martin},
  journal = {Computational Linguistics},
  year = {1993},
  url = {https://aclanthology.org/J93-1006/}
}

References

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

  1. Baayen, H. (1991). "A stochastic process for word frequency distributions." In Proceedings, 29th Annual Meeting of the Association for Computational Linguistics, Berkeley, CA. find
  2. Brown, P.; Lai, J. C.; and Mercer, R. L. (1991). "Aligning sentences in parallel corpora." In Proceedings, 29th Annual Meeting of the Association for Computational Linguistics, Berkeley, CA. find our card
  3. Brown, P.; Cocke, J.; Della Pietra, S.; Della Pietra, V.; Jelinek, F.; Lafferty, J.; Mercer, R.; and Roossin P. (1990). "A statistical approach to machine translation." Computational Linguistics, 16, 79-85. find
  4. Church, K. W., and Hanks, P. (1990). "Word association norms, mutual information, and lexicography." Computational Linguistics, 16(1), 22-29. find our card
  5. Drela, M., and Langford, J. S. (1985). "Human-powered flight." Scientific American, 253(5). find
  6. Drela, M., and Langford, J. S. (1986). "Fliegen mit Muskelkraft." Spektrum der Wissenschaft. find
  7. Fano, R. (1961). Transmission of Information. A Statistical Theory of Communications. MIT Press. find
  8. Gale, W. A., and Church, K. W. (1991). "A program for aligning sentences in bilingual corpora." In Proceedings, 29th Annual Meeting of the Association for Computational Linguistics. Berkeley, CA. find our card
  9. Kay, M., and R6scheisen, M. (1988). "Text-translation alignment." Technical Report, Xerox Palo Alto Research Center. find
  10. Knuth, D. E. (1973). The Art of Computer Programming. Vol. 3, Sorting and Searching. Addison-Wesley. find
  11. MacKeown, P. K., and Weekes, T. C. (1985). "Cosmic rays from Cygnus X-3." Scientific American, 253(5). find
  12. MacKeown, P. K., and Weekes, T. C. (1986). "Kosmische Strahlen von Cygnus X-3." Spektrum der Wissenschafl. van Rijsbergen, C. J. (1979). Information Retrieval. Butterworths. find
  13. Sato, S., and Nagao, M. (1990). "Toward memory-based translation." In Proceedings, I5th International Conference on Computational Linguistics (COLING-90). find

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/kay-1993-text-translation-alignment/