Elamur Research
☰
home / papers / 1998 / Machine Transliteration

Machine Transliteration

Written for the Research library (claude-opus-5-5) from the paper text, in our own words: the license "CC BY-NC-SA 3.0 (ACL Anthology, before 2016)" does not allow redistributing the text itself. Original — aclanthology.org/J98-4003.pdf

Summary

Names and technical terms are often carried between languages with different scripts and sound systems by phonetic approximation. Knight and Graehl address the reverse problem: recovering the original English phrase from a Japanese katakana rendering. Katakana loanwords are the main source of phrases missing from bilingual dictionaries, and the conversion loses information (for example, "lamp" and "ramp" collapse into one form). Simple letter-to-letter mapping gives poor results, so the authors model the process as a chain of probabilistic stages: an English phrase, its English pronunciation, the adaptation to Japanese sounds, the katakana spelling, and optional OCR noise. Each stage is a weighted finite-state machine. The stages are composed and then searched in reverse with Bayes' rule, which yields a ranked list of English candidates. On names of US politicians, the system clearly beat native English-speaking human subjects. It also stayed fairly accurate on OCR-corrupted input. The modular design means several components can be reused for other language pairs, such as Arabic/English.

Key points

  • The model chains five distributions: P(w), P(e|w), P(j|e), P(k|j) and P(o|k). P(w) is a WFSA, and the other four are WFSTs combined by composition, following Pereira and Riley (1997).
  • Decoding uses Dijkstra's shortest path for the single best answer and Eppstein's k-shortest-paths algorithm for top-k lists. No pruning is applied, and the only approximation is Viterbi.
  • Resources:
    • a 262,000-entry unigram list built from the Wall Street Journal corpus, an English name list and a place-name gazetteer;
    • the CMU Pronunciation Dictionary (110,000 words, 40 sounds without stress marks).
  • On 100 US politician names, the machine got 64% correct, 12% phonetically right but misspelled, and 24% wrong. The four human subjects scored 27%, 7% and 66% respectively.
  • Results did not change when word separators were removed. With OCR input, 7% of tokens were misread (affecting half of the strings), and accuracy fell only from 64% to 52%.

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

237 papers of our library cite it (of the 140673 whose bibliographies we read).

Cite

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

@article{knight1998machine,
  title = {{Machine Transliteration}},
  author = {Knight, Kevin and Graehl, Jonathan},
  journal = {Computational Linguistics},
  year = {1998},
  url = {https://aclanthology.org/J98-4003/}
}

References

12 references, 0 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Arbabi, Mansur, Scott M. Fischthal, Vincent C. Cheng, and Elizabeth Bart. 1994. Algorithms for Arabic name transliteration. IBM Journal of Research and Development, 38(2):183-193. find
  2. Baum, Leonard E. 1972. An inequality and associated maximization technique in statistical estimation of probabilistic functions of a Markov process. Inequalities, 3:1-8. find
  3. Brown, Peter E, Stephen A. Della Pietra, Vincent J. Della Pietra, and Robert L. Mercer. 1993. The mathematics of statistical machine translation: Parameter estimation. Computational Linguistics, 19(2):263-311. find
  4. Dempster, A. P., N. M. Laird, and D. B. Rubin. 1977. Maximum likelihood from incomplete via the EM algorithm. Journal of the Royal Statistical Society, 39(B):1-38. find
  5. Dijkstra, Edsgar W. 1959. A note on two problems in connexion with graphs. Numerische Mathematik, 1:269-271. find
  6. Divay, Michel and Anthony J. Vitale. 1997. Algorithms for grapheme-phoneme translation for English and French:Applications. Computational Linguistics, 23(4):495-524. find
  7. Eppstein, David. 1994. Finding the k shortest paths. In Proceedings of the 35thSymposium on the Foundations of Computer Science, pages 154-165. find
  8. Jorden, Eleanor H. and Hamako I. Chaplin. 1976. Reading Japanese. Yale University Press, New Haven. find
  9. Knoblock, Craig. 1996. Trends and controversies: Statistical versus knowledge-based machine translation. IEEE Expert, 11(2):12-18. find
  10. Pereira, Fernando C. N. and Michael Riley. 1997. Speech recognition by composition of weighted finite automata. In E. Roche and Y. Schabes, editors, Finite-State Language Processing, pages 431-453. MIT Press. find
  11. Weaver, Warren. 1955. Translation. In William N. Locke and A. Donald Booth, editors, Machine Translation of Languages. Technology Press of MIT and John Wiley & Sons, New York (1949 memorandum, reprinted, quoting a 1947 letter from Weaver to Norbert Wiener). find
  12. Yamron, Jonathan, James Cant, Anne Demedts, Taiko Dietzel, and Yoshiko Ito. 1994. The automatic component of the LINGSTAT machine-aided translation system. In Proceedings of the ARPA Workshop on Human Language Technology, pages 163-168. Morgan Kaufmann. find