Finding Predominant Word Senses in Untagged Text

What it is about

The most-frequent (predominant) sense is a strong baseline for word sense disambiguation, but it normally needs a hand-tagged corpus such as SemCor, which is small (about 250,000 words) and fixed in genre. The paper asks whether the predominant sense of a noun can be ranked from raw text alone, and whether the ranking changes with domain. The method builds a thesaurus of distributional nearest neighbours (Lin's measure) from 90 million words of the British National Corpus, using grammatical relations (verb objects and subjects, noun-noun and adjective-noun modifiers, with a frequency floor of 10). Each neighbour votes for the WordNet senses it is semantically close to, weighted by its distributional similarity and by a WordNet similarity measure (lesk or jcn). On the polysemous nouns of SemCor the ranking is right about 54% of the time with either measure, against 32% for random choice; on the SENSEVAL-2 all-words nouns the automatic sense reaches 64% precision, against 69% for the SemCor sense and 92% for the test answer key. On two Reuters domain corpora (SPORTS and FINANCE) the top sense shifts toward the domain for several words, checked by hand on 10 words and on a 38-word set against domain labels.

Ideas

• I1. Test the paper's weighting in onto/src/wsd.rs, step 2. The current prior counts the neighbour votes; the paper also weights each vote by the distributional similarity. Compare the two on the same held-out polysemous nouns with gold senses (tests only), by two-fold cv, and run both twice. Negative control: shuffle the neighbour lists within each noun; the prior must fall toward the random baseline.
• I2. Print the neighbours behind each decision. For the chosen sense and the runner-up, the explanation lists the top neighbours that voted for each, with their similarity. The paper's own example is pipe: neighbours such as tube, cable and wire support the non-tobacco sense. Check 50 random decisions by hand and record whether the listed neighbours really fit the sense.
• I3. A domain prior as a list of flips. Run the same ranking over two corpora Mova has, the tale texts from tools/gutenberg and a general text corpus, and list the nouns whose predominant sense changes. The paper shows the effect only qualitatively; Mova should count the flips and check 20 of them by hand. The tale reader in world can then use the tale-domain sense for the nouns on that list.
• I4. A backoff threshold. The paper uses the predominant sense as a backoff when the context evidence is weak, and wsd.rs sums its terms instead. Test a rule: if the margin of the context score is below a threshold, the prior alone decides. Choose the threshold by cv on training nouns and report once on the test set. Do not tune it on the frozen 60 sentences.

For Mova

Step 2 of onto/src/wsd.rs already implements this paper's method, so the ideas refine it rather than add it. The grammatical relations for the thesaurus come from the parser output in en/src/parse.rs and en/src/ud.rs. The chapter is 19. The lexicon: vocabulary, multiword expressions and register, and the WSD v1 entry is in Smart annotator — layers, context, memory. Measures: the random and most-frequent-sense baselines on the same nouns, two identical runs before any gain is claimed, and the shuffled-neighbour control in I1. Licence: the paper is CC BY-NC-SA, so this card is a summary only. SemCor, the SENSEVAL-2 data and the Reuters corpus are not in the repo, and their licences are not given in the paper. Check each in Data and model licences; until then they are for tests only, and nothing from the paper's data goes into a release.

Topics: Word sense disambiguation

Cite

The paper itself, not this card. BibTeX from the library's record of the work. DOI 10.3115/1218955.1218991

@inproceedings{mccarthy2004finding,
  title = {{Finding Predominant Word Senses in Untagged Text}},
  author = {McCarthy, Diana and Koeling, Rob and Weeds, Julie and Carroll, John},
  booktitle = {Proceedings of the 42nd Annual Meeting of the Association for Computational Linguistics (ACL-04)},
  year = {2004},
  doi = {10.3115/1218955.1218991},
  url = {https://doi.org/10.3115/1218955.1218991}
}

References

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

1. Satanjeev Banerjee and Ted Pedersen. 2002. An adapted Lesk algorithm for word sense disambiguation using WordNet. In Proceedings of the Third International Conference on Intelligent Text Processing and Computational Linguistics (CICLing-02), Mexico City. find
2. Edward Briscoe and John Carroll. 2002. Robust accurate statistical annotation of general text. In Proceedings of the Third International Conference on Language Resources and Evaluation (LREC), pages 1499–1504, Las Palmas, Canary Islands, Spain. find
3. Paul Buitelaar and Bogdan Sacaleanu. 2001. Ranking and selecting synsets by domain relevance. In Proceedings of WordNet and Other Lexical Resources: Applications, Extensions and Customizations, NAACL 2001 Workshop, Pittsburgh, PA. find
4. Massimiliano Ciaramita and Mark Johnson. 2003. Supersense tagging of unknown nouns in Word-Net. In Proceedings of the Conference on Empirical Methods in Natural Language Processing (EMNLP 2003). find
5. Scott Cotton, Phil Edmonds, Adam Kilgarriff, and Martha Palmer. 1998. SENSEVAL -2. http://www.sle.sharp.co.uk/senseval2/. ↗
6. Jordi Daudé, Lluis Padró, and German Rigau. 2000. Mapping wordnets using structural information. In Proceedings of the 38th Annual Meeting of the Association for Computational Linguistics, Hong Kong. find
7. Véronique Hoste, Anne Kool, and Walter Daelemans. 2001. Classifier optimization and combination in the English all words task. In Proceedings of the SENSEVAL -2 workshop, pages 84–86. find
8. Jay Jiang and David Conrath. 1997. Semantic similarity based on corpus statistics and lexical taxonomy. In International Conference on Research in Computational Linguistics, Taiwan. ↗
9. Anna Korhonen. 2002. Semantically motivated subcategorization acquisition. In Proceedings of the ACL Workshop on Unsupervised Lexical Acquisition, Philadelphia, USA. find
10. Mirella Lapata and Chris Brew. 2004. Verb class disambiguation using informative priors. Computational Linguistics, 30(1):45–75. ↗
11. Beth Levin. 1993. English Verb Classes and Alternations: a Preliminary Investigation. University of Chicago Press, Chicago and London. find
12. Dekang Lin. 1998. Automatic retrieval and clustering of similar words. In Proceedings of COLING-ACL 98, Montreal, Canada. find
13. Bernardo Magnini and Gabriela Cavaglià. 2000. Integrating subject field codes into WordNet. In Proceedings of LREC-2000, Athens, Greece. find
14. Bernardo Magnini, Carlo Strapparava, Giovanni Pezzuli, and Alfio Gliozzo. 2001. Using domain information for word sense disambiguation. In Proceedings of the SENSEVAL -2 workshop, pages 111–114. find
15. Diana McCarthy, Rob Koeling, Julie Weeds, and John Carrolł. 2004. Using automatically acquired predominant senses for word sense disambiguation. In Proceedings of the ACL SENSEVAL -3 workshop. find
16. Diana McCarthy. 1997. Word sense disambiguation for acquisition of selectional preferences. In Proceedings of the ACL/EACL 97 Workshop Automatic Information Extraction and Building of Lexical Semantic Resources for NLP Applications, pages 52–61. find
17. Paola Merlo and Matthias Leybold. 2001. Automatic distinction of arguments and modifiers: the case of prepositional phrases. In Proceedings of the Workshop on Computational Language Learning (CoNLL 2001), Toulouse, France. find
18. George A. Miller, Claudia Leacock, Randee Tengi, and Ross T Bunker. 1993. A semantic concordance. In Proceedings of the ARPA Workshop on Human Language Technology, pages 303–308. Morgan Kaufman. find
19. Martha Palmer, Christiane Fellbaum, Scott Cotton, Lauren Delfs, and Hoa Trang Dang. 2001. English tasks: All-words and verb lexical sample. In Proceedings of the SENSEVAL -2 workshop, pages 21–24. find
20. Patrick Pantel and Dekang Lin. 2002. Discovering word senses from text. In Proceedings of ACM SIGKDD Conference on Knowledge Discovery and Data Mining, pages 613–619, Edmonton, Canada. find
21. Siddharth Patwardhan and Ted Pedersen. 2003. The cpan wordnet::similarity package. http://search.cpan.org/author/SID/WordNet-Similarity-0.03/. ↗
22. Siddharth Patwardhan, Satanjeev Banerjee, and Ted Pedersen. 2003. Using measures of semantic relatedness for word sense disambiguation. In Proceedings of the Fourth International Conference on Intelligent Text Processing and Computational Linguistics (CICLing 2003), Mexico City. find
23. Tony G. Rose, Mary Stevenson, and Miles Whitehead. 2002. The Reuters Corpus volume 1 - from yesterday’s news to tomorrow’s language resources. In Proc. of Third International Conference on Language Resources and Evaluation, Las Palmas de Gran Canaria. find
24. Julie Weeds, David Weir, and Diana McCarthy. 2004. Characterising measures of lexical distributional similarity. find
25. Yorick Wilks and Mark Stevenson. 1998. The grammar of sense: using part-of speech tags as a first step in semantic disambiguation. Natural Language Engineering, 4(2):135–143. ↗
26. David Yarowsky and Radu Florian. 2002. Evaluating sense disambiguation performance across diverse parameter spaces. Natural Language Engineering, 8(4):293–310. find

This page in Markdown → https://md.mova.elamur.ai/free/en/docs/papers/onto-2004-mccarthy-finding-predominant-word-senses-untagged/