Elamur Research⌕
☰
home / papers / 1998 / Using Corpus Statistics and WordNet Relations for Sense Identification

Using Corpus Statistics and WordNet Relations for Sense Identification

Summary

Supervised word sense identification depends on sense-tagged training text, which is expensive to produce. Leacock, Chodorow and Miller tackle this by combining corpus statistics with WordNet. They introduce TLC (Topical/Local Classifier), a statistical classifier that can use topical cues (open-class words co-occurring with a sense), local cues (words and function items in a narrow window around the target), or both. TLC was first trained on hand-tagged examples of three target words, the noun line, the verb serve and the adjective hard, taken from LDC newswire and other corpora. Local context proved the stronger signal. Adding topical context helped line substantially, helped serve somewhat, and did not help hard. The authors then gathered training data automatically by retrieving corpus examples of monosemous WordNet "relatives" of each sense. This unsupervised harvesting worked, although it fell short of hand-tagged training. With a confidence threshold, it yields high-precision labels. The work shows a practical route to building training sets for thousands of polysemous words without manual annotation.

Key points

  • Data and senses: Hand-tagged experiments used six senses of line, four of serve and three of hard.
    • Sources were the 1987–89 LDC Wall Street Journal, the American Printing House for the Blind corpus (25M words) and the LDC San Jose Mercury News.
    • Training sets contained 25, 50, 100 or 200 examples of the rarest sense. Each test set held 150 examples of it.
  • Comparison with a thesaurus-based approach: On one word, training on monosemous relatives gave 93.5% precision at 100% recall. A previously reported thesaurus-based method reached 96%.
  • Performance on eight other words: Accuracy was close to the most-frequent-sense baseline. When TLC was tuned for 95% precision, recall varied widely, from 95% for bank to 2% for work.
  • Predicting difficulty from sense overlap: Cosine correlation between sense context vectors flags hard cases. The two senses of work correlate at r = .49, versus a mean of .31 for the other words.
  • Nontopical senses: Some senses are not tied to any topic. This limits the "one sense per discourse" heuristic.

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

 Institutions

🎓Princeton UniversityPrinceton · United States45 works · 34,022 citations🤝Educational Testing ServicePrinceton · United States12 works · 1,801 citations🎓Hunter CollegeNew York · United States4 works · 1,077 citations

Cite

Cite as Claudia Leacock, Martin Chodorow and George A. Miller. 1998. Using Corpus Statistics and WordNet Relations for Sense Identification. Computational Linguistics.

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

@article{leacock1998using,
  title = {{Using Corpus Statistics and WordNet Relations for Sense Identification}},
  author = {Claudia Leacock, Martin Chodorow and George A. Miller},
  journal = {Computational Linguistics},
  year = {1998},
  doi = {10.5555/972719.972726},
  url = {https://doi.org/10.5555/972719.972726}
}

References

26 references, 5 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Brill, Eric. 1994. Some advances in rule-based part of speech tagging. In Proceedings of the Twelfth National Conference on Artificial Intelligence, Seattle. AAAI. find
  2. Bruce, Rebecca and Janyce Wiebe. 1994a. A new approach to word sense disambiguation. In Proceedings of the ARPA Workshop on Human Language Technology, San Francisco, CA, Morgan Kaufman. find
  3. Bruce, Rebecca and Janyce Wiebe. 1994b. Word-sense disambiguation using decomposable models. In Proceedings of the 32nd Annual Meeting, Las Cruces, NM. Association for Computational Linguistics. find
  4. Chiang, T-H., Y-C. Lin, and K-Y Su. 1995. Robust learning, smoothing, and parameter tying on syntactic ambiguity resolution. Computational Linguistics, 21 (3):321-349. find
  5. Dagan, Ido and Alon Itai. 1994. Word sense disambiguation using a second language monolingual corpus. Computational Linguistics, 20(4). our card
  6. Gale, William, Kenneth W. Church, and David Yarowsky. 1992a. A method for disambiguating word senses in a large corpus. Computers and the Humanities, 26. find
  7. Gale, William, Kenneth W. Church, and David Yarowsky. 1992b. One sense per discourse. In Proceedings of the Speech and Natural Language Workshop, San Francisco, CA, Morgan Kaufmann. find
  8. Golding, Andrew. 1995. A Bayesian hybrid method for context-sensitive spelling correction. In Proceedings of the Third Workshop on Very Large Corpora, Cambridge, MA. ACL. find
  9. Good, I. E 1953. The population frequencies of species and the estimation of population parameters. Biometrica, 40:237-264. find
  10. Hearst, Marti A. 1991. Noun homograph disambiguation using local context in large text corpora. In Proceedings of the Seventh Annual Conference of the UW Centre for the New OED and Text Research: Using Corpora, pages 1-22, Oxford. find
  11. Hirst, Graeme. 1987. Semantic Interpretation and the Resolution of Ambiguity. Cambridge University Press, Cambridge, MA. find
  12. Landes, Shari, Claudia Leacock, and Randee Tengi. 1998. Building semantic concordances. In Christiane Fellbaum, editor, WordNet: A Lexical Reference System and its Application. MIT Press, Cambridge, MA. find
  13. Leacock, Claudia and Martin Chodorow. 1998. Combining local context with WordNet similarity for word sense identification. In Christiane Fellbaum, editor, WordNet: A Lexical Reference System and its Application. MIT Press, Cambridge, MA. find
  14. Leacock, Claudia, Geoffrey Towell, and Ellen M. Voorhees. 1993. Corpus-based statistical sense resolution. In Proceedings of the ARPA Workshop on Human Language Technology, San Francisco, CA, Morgan Kaufman. find
  15. Leacock, Claudia, Geoffrey Towell, and Ellen M. Voorhees. 1996. Towards building contextual representations of word senses using statistical models. In Branimir Boguraev and James Pustejovsky, editors, Corpus Processing for Lexical Acquisition. MIT Press, Cambridge, MA. find
  16. Miller, George A., editor, 1990. WordNet: An On-Line Lexical Database. Volume 3(4) of the International Journal of Lexicography. Oxford University Press. find
  17. Miller, George A. 1995. WordNet: An on-line lexical database. Communications of the ACM, 38(11). find
  18. Resnik, Philip. 1992. WordNet and distributional analysis: A class-based approach to lexical discovery. In Workshop on Statistically-Based Natural-Language-Processing Techniques, San Jose, July. find
  19. Schfitze, Hinrich. 1995. Ambiguity and Language Learning: Computational and Cognitive Models. Ph.D. thesis, Stanford University. find
  20. V~ronis, Jean and Nancy Ide. 1990. Word sense disambiguation with very large neural networks extracted from machine readable dictionaries. In Proceedings of COLING-90. find
  21. Weiss, Stephen. 1973. Learning to disambiguate. Information Storage and Retrieval, 9. find
  22. Wilks, Yorick, Dan Fass, Cheng Ming Guo, James E. McDonald, Tony Plate, and Brian M. Slator. 1993. Machine tractable dictionary tools. In James Pustejovsky, editor, Semantics and the Lexicon. Kluwer, Dordrecht. find
  23. Yarowsky, David. 1992. Word-sense disambiguation using statistical models of Roget's categories trained on large corpora. In Proceedings of COLING-92, Nantes, France. our card
  24. Yarowsky, David. 1993. One sense per collocation. In Proceedings of the ARPA Workshop on Human Language Technology, San Francisco, CA, Morgan Kaufman. our card
  25. Yarowsky, David. 1994. Decision lists for lexical ambiguity resolution: Application to accent restoration in Spanish and French. In Proceedings of the 32nd Annual Meeting. Las Cruces, NM. Association for Computational Linguistics. our card
  26. Yarowsky, David. 1995. Unsupervised word sense disambiguation rivaling supervised methods. In Proceedings of the 33rd Annual Meeting, Cambridge, MA. Association for Computational Linguistics. our card

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/leacock-1998-corpus-statistics-wordnet-relations-sense/