Elamur Research
☰
home / papers / 1998 / Automatic Word Sense Discrimination

Automatic Word Sense Discrimination

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/J98-1004.pdf

Summary

The paper addresses word sense discrimination: deciding which occurrences of an ambiguous word share a meaning, without naming that meaning. Treating this separately from sense labeling removes the need for dictionaries, thesauri, bilingual corpora or hand-tagged examples. Schütze proposes context-group discrimination. Each word gets a vector in Word Space, a high-dimensional real-valued space, built from the words it co-occurs with in a training corpus. A context of the ambiguous word is represented by summing the vectors of its surrounding words. Similarity therefore rests on second-order co-occurrence, which is less sparse and more robust than direct co-occurrence. Training contexts are clustered, and each cluster's centroid becomes a sense vector. A new occurrence is assigned to the nearest centroid. Several representations are compared, some reduced with SVD. The method is fully unsupervised and was tested on natural and artificial ambiguous words from the New York Times News Wire, where it performed well. This matters for information access, such as sense-based retrieval and search interfaces that account for ambiguity. The supplied extract omits the exact accuracy figures.

Key points

  • Disambiguation is split into discrimination and labeling. Only discrimination is tackled, so no external knowledge source is needed in training or application.

  • The method uses three kinds of vectors:

    • word vectors, from neighbors within a sentence or larger window;
    • context vectors, the sum of the word vectors in a context;
    • sense vectors, the cluster centroids.

    Variants with SVD dimensionality reduction are also tested.

  • Cluster granularity is adjustable. With few clusters, space splits into "outer space" and "spatial extent". With more clusters (e.g., n = 10), finer uses such as office vs. exhibition space appear.

  • The evaluation data comes from the New York Times News Wire and covers both natural and artificial ambiguous words.

  • Background and motivation:

    • The idea was first described in Schütze (1992b).
    • Schütze and Pedersen (1995) found that sense-based document–query similarity improved relevance ranking.
    • Miller and Charles (1991) support the link between contextual and semantic similarity.

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

Topics: Word sense disambiguation

Cited by

223 papers of our library cite it (of the 140673 whose bibliographies we read); 5 of them have a page here, 218 not yet.

Cite

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

@article{schutze1998automatic,
  title = {{Automatic Word Sense Discrimination}},
  author = {Schütze, Hinrich},
  journal = {Computational Linguistics},
  year = {1998},
  url = {https://aclanthology.org/J98-1004/}
}

References

15 references, 1 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Sanderson, Mark. 1994. Word sense disambiguation and information retrieval. In Proceedings of SIGIR "94, pages 142-151. find
  2. Schiitze, Hinrich. 1992a. Context space. In Robert Goldman, Peter Norvig, Eugene Charniak, and Bill Gale, editors, Working Notes of the AAAI Fall Symposium on Probabilistic Approaches to Natural Language, pages 113-120, AAAI Press, Menlo Park, CA. find
  3. Schiitze, Hinrich. 1992b. Dimensions of meaning. In Proceedings of Supercomputing "92, pages 787-796, Minneapolis, MN. find
  4. Schtitze, Hinrich. 1997. Ambiguity Resolution in Language Learning. CSLI Publications, Stanford, CA. find
  5. Schiitze, Hinrich and Jan O. Pedersen. 1995. Information retrieval based on word senses. In Proceedings of the Fourth Annual Symposium on Document Analysis and Information Retrieval, pages 161-175, Las Vegas, NV. find
  6. Schiitze, Hinrich and Jan O. Pedersen. 1997. A cooccurrence-based thesaurus and two applications to information retrieval. find
  7. Sparck-Jones, Karen. 1986. Synonymy and Semantic ClassiJication. Edinburgh University Press, Edinburgh. (Publication of Ph.D. thesis, University of Cambridge, 1964.) find
  8. Sparck-Jones, Karen. 1991. Notes and references on early classification work. ACM SIGIR Forum, 25(1):10-17. van Rijsbergen, C. J. 1979. Information Retrieval. Second edition. Butterworths, London. find
  9. Voorhees, Ellen M. 1993. Using WordNet to disambiguate word senses for text retrieval. In Proceedings of SIGIR "93, pages 171-180. find
  10. Walker, Donald E. and Robert A. Amsler. 1986. The use of machine-readable dictionaries in sublanguage analysis. In Ralph Grishman and Richard Kittredge, editors, Analyzing Language in Restricted Domains: Sublanguage Description and Processing. L. Erlbaum Associates, Hillsdale, NJ, pages 69-84. find
  11. Wilks, Yorick A., Dan C. Fass, Cheng Ming Guo, James E. McDonald, Tony Plate, and Brian M. Slator. 1990. Providing machine tractable dictionary tools. Journal of Computers and Translation, 2. find
  12. Willett, Peter. 1988. Recent trends in hierarchic document clustering: A critical review. Information Processing & Management, 24(5):577-597. find
  13. Winer, B. J. 1971. Statistical Principles in Experimental Design. Second edition. McGraw-Hill, New York, NY. find
  14. Yarowsky, David. 1992. Word-sense disambiguation using statistical models of Roget's categories trained on large corpora. In Proceedings of Coling-92, pages 454-460, Nantes, France. find
  15. Yarowsky, David. 1995. Unsupervised word sense disambiguation rivaling supervised methods. In Proceedings of the 33rd Annual Meeting, Cambridge, MA. Association for Computational Linguistics. find our card