Elamur Research⌕
☰
home / papers / 2006 / Automatic Identification of Non-Compositional Multi-Word Expressions using Latent Semantic Analysis

Automatic Identification of Non-Compositional Multi-Word Expressions using Latent Semantic Analysis

Summary

Katz and Giesbrecht ask whether local context alone can show if a German multi-word expression (MWE) carries an idiomatic, non-compositional meaning. They frame the problem like word-sense disambiguation, with two parts. One is discrimination: does an expression have an idiomatic reading at all? The other is selection: which reading does a particular occurrence use?

They built an LSA Word Space model with the Infomap package on a Süddeutsche Zeitung newspaper corpus and gave each MWE its own context vector. An estimated "compositional" vector was formed by adding the vectors of the component words. Its cosine similarity to the MWE's observed vector then served as a compositionality score, where low similarity points to idiomatic use.

Tested on preposition-noun-verb collocation candidates from Krenn's (2000) database, this score separated idiomatic from literal combinations. It outperformed the head/non-head comparison proposed by Baldwin et al. (2003). Two contributions stand out:

  • It offers an unsupervised, raw-text way to grade compositionality.
  • It argues for evaluating against hand-annotated compositionality judgments rather than idiom dictionaries or WordNet.

Key points

  • Corpus and model: Süddeutsche Zeitung 2003 corpus of about 42 million words, processed with Infomap.
    • 1000 most frequent non-stop-list words served as dimensions; the stop list had 102 words.
    • Rows covered the 20,000 most frequent content words, with co-occurrences counted in a 30-word window.
    • SVD reduced the space to 100 dimensions.
  • Test set: 81 German PNV candidates from Krenn (2000), each occurring more than 30 times in the corpus. Krenn's manual annotation was the gold standard.
  • Score range: Cosines between composed and actual MWE vectors ran from 0.01 to 0.80. Among items scoring below 0.1, slightly more than half were idiomatic.
  • Best threshold: F-score peaked at a cosine threshold of about 0.2; the conclusion places the optimum between 0.1 and 0.2.
  • Example and baseline comparison:
    • For ins Wasser fallen, the composed vector was closer to the literal-use vector (0.2937) than to the idiomatic-use vector (0.1733).
    • In the Baldwin et al.-style comparison, the expected low similarity held only for the noun constituent, not the verb, and was generally weaker than the composed-vector method.

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

Topics: Word sense disambiguation

Cited by

37 papers of our library cite it (of the 140673 whose bibliographies we read); 6 of them have a page here, 31 not yet.

Cite

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

@inproceedings{katz2006automatic,
  title = {{Automatic Identification of Non-Compositional Multi-Word Expressions using Latent Semantic Analysis}},
  author = {Katz, Graham and Giesbrecht, Eugenie},
  booktitle = {Proceedings of the Workshop on Multiword Expressions: Identifying and Exploiting Underlying Properties},
  year = {2006},
  doi = {10.3115/1613692.1613696},
  url = {https://doi.org/10.3115/1613692.1613696}
}

References

26 references, 4 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Ricardo A. Baeza-Yates and Berthier A. Ribeiro-Neto. 1999. Modern Information Retrieval. ACM Press / Addison-Wesley. find
  2. Timothy Baldwin, Colin Bannard, Takaaki Tanaka, and Dominic Widdows. 2003. An empirical model 11 One such example from the SZ corpus: Der Auftakt wäre allerdings fast ins Wasser gefallen, weil ein geplatzter Hydrant eine fünfzehn Meter hohe Wasserfontäne in die Luft schleuderte. find
  3. ‘The prelude almost didn’t occur, because a burst hydrant shot a fifteen-meter high fountain into the sky.’ cos \< 0.1 0.53 0.42 0.47 cos \< 0.2 0.39 0.63 0.48 Precision Recall F-measure Table 4: Evaluation of Various Similarity Thresholds cos \< 0.1 0.21 0.28 cos \< 0.2 0.16 0.51 Verb F-measure Noun F-… find
  4. Colin Bannard, Timothy Baldwin, and Alex Lascarides. 2003. A statistical approach to the semantics of verb-particles. In Proceedings of the ACL-2003 Workshop on Multiword Expressions: Analysis, Acquisition and Treatment, pages 65–72, Sapporo, Japan. find
  5. Michael W. Berry, Zlatko Drmavc, and Elisabeth R. Jessup. 1999. Matrices, vector spaces, and information retrieval. SIAM Review, 41(2):335–362. find
  6. Jean Carletta. 1996. Assessing agreement on classification tasks: The kappa statistic. Computational Linguistics, 22(2):249–254. find our card
  7. Scott Cederberg and Dominic Widdows. 2003. Using LSA and noun coordination information to improve the precision and recall of automatic hyponymy extraction. In In Seventh Conference on Computational Natural Language Learning, pages 111–118, Edmonton, Canada, June. find
  8. Scott C. Deerwester, Susan T. Dumais, Thomas K. Landauer, George W. Furnas, and Richard A. Harshman. 1990. Indexing by latent semantic analysis. Journal of the American Society of Information Science, 41(6):391–407. find
  9. Stefan Evert and Hannah Kermes. 2003. Experiments on candidate data for collocation extraction. In Companion Volume to the Proceedings of the 10th Conference of The European Chapter of the Association for Computational Linguistics, pages 83–86, Budapest, Hungary. find
  10. Stefan Evert and Brigitte Krenn. 2001. Methods for the qualitative evaluation of lexical association measures. In Proceedings of the 39th Annual Meeting of the Association for Computational Linguistics, pages 188–195, Toulouse, France. find
  11. Stefan Evert. 2004. The Statistics of Word Cooccurrences: Word Pairs and Collocations. Ph.D. thesis, University of Stuttgart. cos \< 0.3 0.29 0.84 0.43 cos \< 0.4 0.22 0.89 0.35 cos \< 0.5 0.21 0.95 0.34 cos \< 0.3 0.29 0.43 cos \< 0.4 0.26 0.39 cos \< 0.5 0.27 0.33 find
  12. Christiane Fellbaum. 1998. WordNet, an electronic lexical database. MIT Press, Cambridge, MA. find
  13. Nancy Ide and Jean Véronis. 1998. Word sense disambiguation: The state of the art. Computational Linguistics, 14(1). find
  14. Walter Kintsch. 2001. Predication. Cognitive Science, 25(2):173–202. find
  15. Brigitte Krenn. 2000. The Usual Suspects: Data-Oriented Models for Identification and Representation of Lexical Collocations. Dissertations in Computational Linguistics and Language Technology. German Research Center for Artificial Intelligence and Saarland University, Saarbrücken, Germany. find
  16. Thomas K. Landauer and Susan T. Dumais. 1997. A solution to plato’s problem: The latent semantic analysis theory of the acquisition, induction, and representation of knowledge. Psychological Review, 104:211–240. find
  17. Thomas K. Landauer and Joseph Psotka. 2000. Simulating text understanding for educational applications with latent semantic analysis: Introduction to LSA. Interactive Learning Environments, 8(2):73– 86. find
  18. Dekang Lin. 1999. Automatic identification of non-compositional phrases. In Proceedings of the 37th Annual Meeting of the Association for Computational Linguistics, pages 317–324, College Park, MD. find our card
  19. Christopher D. Manning and Hinrich Schütze. 1999. Foundations of Statistical NaturalLanguage Processing. The MIT Press, Cambridge, MA. find
  20. Ivan A. Sag, Timothy Baldwin, Francis Bond, Ann A. Copestake, and Dan Flickinger. 2002. Multiword expressions: A pain in the neck for NLP. In Proceedings of the 3rd International Conferences on Intelligent Text Processing and Computational Linguistics, pages 1–15. find our card
  21. Patrick Schone and Daniel Jurafsky. 2001. Is knowledge-free induction of multiword unit dictionary headwords a solved problem? In Proceedings of Empirical Methods in Natural Language Processing, Pittsburgh, PA. find
  22. Hinrich Schütze. 1998. Automatic word sense discrimination. Computational Linguistics, 24(1):97– 124. find our card
  23. Begoña Villada Moirón and Jörg Tiedemann. 2006. Identifying idiomatic expressions using automatic word-alignment. In Proceedings of the EACL 2006 Workshop on Multiword Expressions in a Multilingual Context, Trento, Italy. find
  24. Dominic Widdows and Stanley Peters. 2003. Word vectors and quantum logic: Experiments with negation and disjunction. In Eighth Mathematics of Language Conference, pages 141–150, Bloomington, Indiana. find
  25. Chengxiang Zhai. 1997. Exploiting context to identify lexical atoms – a statistical view of linguistic context. In Proceedings of the International and Interdisciplinary Conference on Modelling and Using Context (CONTEXT-97), pages 119–129. find
  26. APPENDIX Similarity (cosine) values for the combined and the MWE vector. Uppercase entries are those hand-annotated as being MWEs which have an idiomatic interpretation. find

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/papers/katz-2006-automatic-identification-non-compositional-multi/