Elamur Research⌕
☰
home / papers / 1994 / DECISION LISTS FOR LEXICAL AMBIGUITY RESOLUTION: Application to Accent Restoration in Spanish and French

DECISION LISTS FOR LEXICAL AMBIGUITY RESOLUTION: Application to Accent Restoration in Spanish and French

Summary

Yarowsky tackles lexical ambiguity resolution, a family of problems that includes word-sense disambiguation, homograph handling in speech synthesis, and capitalization recovery. He studies it through one concrete case: putting missing accents back into Spanish and French text. The method is a decision list in the sense of Rivest (1987). It is learned from accented corpora by measuring how unevenly different collocations are distributed across the competing accent patterns, using log-likelihoods. Features include adjacent words, words within a ±k window, and word pairs at fixed offsets. Unlike a Bayesian classifier, it does not sum all the evidence. It labels each instance using only the single strongest matching clue, so correlated features can be used without modeling their dependencies. Training and test data come for free: you remove the accents from correct text and keep the original as the answer key. Accuracy is above 99% on the full task and usually above 90% even for the hardest ambiguities. The paper argues the approach is cheap, transparent, and quick to move to new domains.

Key points

  • The core design choice is to classify by the single best piece of evidence present in the context. The author reports precision equal to, or slightly better than, combining all the evidence over the same features.
  • French training data came from the French Canadian Hansards (1986–1988, 19 million words) and Le Monde (1 million words). Spanish evaluation was compared against AP Newswire Spanish text.
  • Typical hard cases are:
    • Spanish -o/-ó, a present vs. preterite contrast.
    • Spanish -ara/-ará, a past subjunctive vs. future contrast that local trigram taggers handle poorly.
    • French -e/-é.
    • Purely semantic pairs such as secretaria/secretaría and traité/traite.
  • Corpus errors push measured agreement below true accuracy. A Spanish-speaking judge checked disagreements and found the corpus, not the system, was wrong in 46% of disputed mi/mí tokens and 56% of disputed anuncio/anunció tokens.
  • The same code was first used for homograph disambiguation in text-to-speech (Sproat et al., 1992), reaching 97% mean accuracy on 13 homographs. Porting it to French took days and needed no special lexical resources.

Interesting ideas

  • I1. Strip a signal to create labeled data. Deleting information that is reliably present (accents, capitalization, Hebrew vowels) yields unlimited, objective test material with no hand annotation. The same trick could work for Ukrainian tasks such as apostrophe or stress restoration.
  • I2. "Strongest single clue wins" instead of combining evidence. Ranking features by reliability and stopping at the first match avoids independence assumptions and gives a readable list of rules. This helps when features overlap heavily or when someone must audit and correct the model.
  • I3. One framework for many "conflated form" problems. Treating sense disambiguation, homophones, casing, and diacritics as the same problem lets one tool cover all of them with almost no code changes. That keeps engineering cost low across projects.
  • I4. Audit disagreements before trusting the metric. Having a human check whether the gold data or the model is wrong exposed noise in the reference corpus. This is worth doing whenever gold labels come from naturally produced text.
  • I5. Mix surface features of different reach. Combining adjacent-word patterns, word-pair templates, and wider context windows in one ranked list lets syntactic and topical clues compete fairly. This suits tasks where the decisive cue may be either local or distant.

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 · Natural language processing

Topics: Word sense disambiguation

Cited by

85 papers of our library cite it (of the 140673 whose bibliographies we read); 2 of them have a page here, 83 not yet.

Cite

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.3115/981732.981745

@inproceedings{yarowsky1994decision,
  title = {{DECISION LISTS FOR LEXICAL AMBIGUITY RESOLUTION: Application to Accent Restoration in Spanish and French}},
  author = {Yarowsky, David},
  booktitle = {32nd Annual Meeting of the Association for Computational Linguistics},
  year = {1994},
  doi = {10.3115/981732.981745},
  url = {https://doi.org/10.3115/981732.981745}
}

References

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

  1. Brill, Eric, "A Corpus-Based Approach to Language Learning," Ph.D. Thesis, University of Pennsylvania, 1993. find
  2. Brown, Peter, Stephen Della Pietra, Vincent Della Pietra, and Robert Mercer, "Word Sense Disambiguation using Statistical Methods," Proceedings of the 29th Annual Meeting of the Association for Computational Linguistics, pp. 264-270, 1991. find
  3. Chang, Jing-Shin, Yin-Fen Luo and Keh-Yih Su, "GPSM: A Generalized Probabilistie Semantic Model for Ambiguity Resolution," in Proceedings of the 30th Annual Meeting of the Association for Computational Linguistics, pp. 177-184, 1992. find
  4. Church, K.W., "A Stochastic Parts Program and Noun Phrase Parser for Unrestricted Text," in Proceedings of the Second Conference on Applied Natural Language Processing, ACL, 136-143, 1988. find
  5. Gale, W., K. Church, and D. Yarowsky, "A Method for Disambiguating Word Senses in a Large Corpus," Computers and the Humanities, 26,415-439, 1992. find
  6. Hearst, Marti, "Noun Homograph Disambiguation Using Local Context in Large Text Corpora," in Using Corpora, University of Waterloo, Waterloo, Ontario, 1991. find
  7. Jelinek, F., "Markov Source Modeling of Text Generation," in Impact of Processing Techniques on Communication, J. Skwirzinski, ed., Dordrecht, 1985. find
  8. Kupiec, Julian, "Probabilistic Models of Short and Long Distance Word Dependencies in Running Text," in Proceedings, DARPA Speech and Natural Language Workshop, Philadelphia, February, pp. 290-295, 1989. find
  9. Leacock, Claudia, Geoffrey Towell and Ellen Voorhees "Corpus-Based Statistical Sense Resolution," in Proceedings, ARPA Human Language Technology Workshop, 1993. find
  10. Marty, Fernand, "Trois syst~mes informatiques de transcription phon~tique et graph@mique", in Le Frangais Moderne, pp. 179-197, 1992. find
  11. Marty, F. and R.S. Hart, "Computer Program to Transcribe French Text into Speech: Problems and Suggested Solutions", Technical Report No. LLL-T-6-85. Language Learning Laboratory; University of Illinois. Urbana, Illinois, 1985. find
  12. Merialdo, B., 'Tagging Text with a Probabilistic Model," in Proceedings of the IBM Natural Language ITL, Paris, France, pp. 161-172, 1990. find
  13. Mosteller, Frederick, and David Wallace, Inference and Disputed Authorship: The Federalist, Addison-Wesley, Reading, Massachusetts, 1964. find
  14. Resnik, Philip, "Selection and Information: A Class-Based Approach to Lexical Relationships," Ph.D. Thesis, University of Pennsylvania, 1993. find
  15. Rivest, R. L., "Learning Decision Lists," in Machine Learning, 2, 229-246, 1987. find
  16. Sproat, Richard, Julia Hirschberg and David Yarowsky "A Corpus-based Synthesizer," in Proceedings, International Conference on Spoken Language Processing, Banff, Alberta, October 1992. find
  17. Tzoukermann, Evelyne and Mark Liberman, " A Finite-state Morphological Processor for Spanish," in Proceedings, COLING-90, Helsinki, 1990. find
  18. Yarowsky, David, "Word-Sense Disambiguation Using Statistical Models of Roget's Categories Trained on Large Corpora," in Proceedings, COLING-92, Nantes, France, 1992. find
  19. Yarowsky, David, "One Sense Per Collocation," in Proceedings, ARPA Human Language Technology Workshop, Princeton, 1993. find
  20. Yarowsky, David, "A Comparison of Corpus-based Techniques for Restoring Accents in Spanish and French Text," to appear in Proceedings, 2nd Annual Workshop on Very Large Text Corpora, Kyoto, Japan, 1994. find

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/yarowsky-1994-decision-lists-lexical-ambiguity-resolution/