Elamur Research⌕
☰
home / papers / 2010 / Automatically Identifying the Source Words of Lexical Blends in English

Automatically Identifying the Source Words of Lexical Blends in English

Paper · 2010 · Paul Cook and Suzanne Stevenson

Summary

NLP systems stumble on freshly coined words because their lexicons have no entries for them, and such words are too rare for distributional learning. Cook and Stevenson address lexical blends such as brunch or gaydar. They focus on simple two-word sequential blends, where the start of one word joins the end of another, sometimes with overlap.

Method. For each blend, the model enumerates every plausible source-word pair from a lexicon. It then ranks the pairs with statistical features grounded in how people recognize blend sources:

  • frequency and neighborhood frequency,
  • co-occurrence,
  • length, contribution and phonology,
  • semantics,
  • syllable structure.

A syllable-based filter prunes unlikely candidates, and a machine-learning combiner merges the features.

Data. The authors annotated 1,186 recent coinages from a neologism website, of which 515 are blends. Evaluation uses a 324-blend subset plus a set of dictionary-established blends.

Results. On novel blends the method picks the correct source pair 40% of the time, which cuts the error rate by 39% relative to an informed baseline. The same features also help separate blends from other neologisms. This is an early step toward automatically extending lexicons with new word formations.

Key points

  • The annotated resource has 1,186 recent coinages, about 43% of them blends (515). By contrast, Algeo's earlier study found only about 5% blends among new words.
  • Candidate generation works in three steps:
    • Split the blend into every prefix–suffix pair, each part at least two characters long.
    • Pair all lexicon words starting with the prefix against all words ending with the suffix.
    • Remove duplicate pairs.
  • There are ten frequency features. They include each word's raw frequency, the frequency of a word divided by the total frequency of words sharing its prefix or suffix, and six new co-occurrence features that replace the earlier PMI.
  • On the 324 novel blends, accuracy is 40%, a 39% error-rate reduction over an informed baseline. The approach also clearly improves on Cook and Stevenson (2007).
  • Novel blends behave differently statistically from dictionary-established blends. A syllable-based filter shrinks candidate sets and gives modest gains, even though it sometimes discards the correct pair.

Interesting ideas

  • I1. Enumerate, then rank: The model lists every way a word could split into two lexicon pieces and then scores each analysis, with a minimum fragment length to keep the search manageable; this turns an open-ended guess into a ranking over a finite set.
  • I2. Neighborhood-normalized frequency: A candidate's frequency is weighed against all words sharing the same fragment, which captures how easily a reader would recover that word rather than how common it is overall.
  • I3. Quality of the best explanation as a classifier: If no good source pair exists, the word probably is not a blend, so the same scores serve both decomposition and detection.
  • I4. Lossy pruning can still pay off: The syllable filter sometimes discards the correct answer yet improves overall accuracy, because it removes far more noise than signal.
  • I5. Test on truly new items: Established dictionary blends gave a misleading picture, so evaluation needs genuinely novel coinages.

For Mova

How these ideas could serve Mova, our small open language model — its files, stages and tests:

  • I1 (tokens): An unknown-word fallback in crate en could propose blend or compound decompositions and borrow lemma and POS hints from the best-scoring pair, recording the chosen split as the explanation.
  • I2 (tokens): This ratio could rank candidate expansions for clipped or truncated forms during morphology lookup, preferring analyses a human reader would actually recover.
  • I3 (learn): Self-induced repair rules could fire only when the best analysis clears a threshold, with non-blend coinages used as negative controls to check for false triggers.
  • I4 (tree): Cheap linguistically motivated pruning of arc or label candidates in the perceptron parser could be accepted and justified by measured net gains rather than perfect recall.
  • I5 (learn): Evaluation suites could include a held-out set of recent neologisms, kept separate from dictionary words, so that tokenizer and lexicon scores reflect real unseen-word behavior.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Institutions

🎓University of TorontoToronto · Canada96 works · 142,007 citations

 Fields, topics and concepts

📚 Shelves 4The field of science this work belongs to (OpenAlex fields).Computer Science · 16,667 worksArtificial Intelligence · 16,194 worksArts and Humanities · 16,039 worksLanguage and Linguistics · 15,891 works

🧭 Topics 1Research subjects and languages the work is about.English · 2,263 works

Cite

Cite as Paul Cook and Suzanne Stevenson. 2010. Automatically Identifying the Source Words of Lexical Blends in English. Computational Linguistics.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.1162/coli.2010.36.1.36104

@article{cook2010automatically,
  title = {{Automatically Identifying the Source Words of Lexical Blends in English}},
  author = {Paul Cook and Suzanne Stevenson},
  journal = {Computational Linguistics},
  year = {2010},
  doi = {10.1162/coli.2010.36.1.36104},
  url = {https://doi.org/10.1162/coli.2010.36.1.36104}
}

References

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

  1. Alex, Beatrice. 2008. Comparing corpus-based to Web-based lookup techniques for automatic English inclusion detection. In Proceedings of the Sixth International Language Resources and Evaluation Conference (LREC’08), pages 2693–2697, Marrakech.
  2. Algeo, John. 1980. Where do all the new words come from? American Speech, 55(4):264–277. ↗
  3. Algeo, John, editor. 1991. Fifty Years Among the New Words. Cambridge University Press, Cambridge. find
  4. Ayto, John, editor. 1990. The Longman Register of New Words, volume 2. Longman, London. find
  5. Ayto, John. 2006. Movers and Shakers: A Chronology of Words that Shaped our Age. Oxford University Press, Oxford. find
  6. Baayen, R. Harald, Richard Piepenbrock, and Leon Gulikers. 1995. The CELEX Lexical Database (release 2) [CD-ROM]. Philadelphia, PA: Linguistic Data Consortium, University of Pennsylvania [distributor]. find
  7. Baker, Kirk and Chris Brew. 2008. Statistical identification of English loanwords in Korean using automatically generated training data. In Proceedings of the Sixth International Language Resources and Evaluation Conference (LREC’08), pages 1159–1163, Marrakech. find
  8. Bartlett, Susan, Grzegorz Kondrak, and Colin Cherry. 2008. Automatic syllabification with structured SVMs for letter-to-phoneme conversion. In Proceedings of the 46th Annual Meeting of the Association for Computational Linguistics (ACL-08): Human Language Technologies, pages 568–576, Columbus, OH. find
  9. Bauer, Laurie. 1983. English Word-formation. Cambridge University Press, Cambridge. ↗
  10. Brants, Thorsten and Alex Franz. 2006. Web 1T 5-gram Corpus version 1.1. Linguistic Data Consortium, Philadelphia, PA. find
  11. Brill, Eric. 1994. Some advances in transformation-based part of speech tagging. In Proceedings of the Twelfth National Conference on Artificial Intelligence, pages 722–727, Seattle, WA. ↗
  12. Cardie, Claire. 1993. A case-based approach to knowledge acquisition for domain-specific sentence analysis. In Proceedings of the Eleventh National Conference on Artificial Intelligence, pages 798–803, Washington, DC. find
  13. Cook, Paul and Suzanne Stevenson. 2007. Automagically inferring the source words of lexical blends. In Proceedings of the Tenth Conference of the Pacific Association for Computational Linguistics (PACLING-2007), pages 289–297, Melbourne. find
  14. Delbridge, Arthur, editor. 1981. The Macquarie Dictionary. Macquarie Library, Sydney. find
  15. Granger, Richard H. 1977. FOUL-UP: A program that figures out the meanings of words from context. In Proceedings of the Fifth International Joint Conference on Artificial Intelligence, pages 172–178, Cambridge, MA. find
  16. Gries, Stefan Th. 2004. Shouldn’t it be breakfunch? A quantitative analysis of the structure of blends. Linguistics, 42(3):639–667. find
  17. Gries, Stefan Th. 2006. Cognitive determinants of subtractive word-formation processes: A corpus-based perspective. Cognitive Linguistics, 17(4):535–558. find
  18. Hastings, Peter M. and Steven L. Lytinen. 1994. The ups and downs of lexical acquisition. In Proceedings of the Twelfth National Conference on Artificial Intelligence, pages 754–759, Seattle, WA. find
  19. Hindle, Donald. 1990. Noun classification from predicate-argument structures. In Proceedings of the 28th Annual Meeting of the Association for Computational Linguistics, pages 268–275, Pittsburgh, PA.
  20. Jiang, Jay J. and David W. Conrath. 1997. Semantic similarity based on corpus statistics and lexical taxonomy. In Proceedings of the International Conference on Research in Computational Linguistics (ROCLING X), pages 19–33, Taiwan. ↗
  21. Joanis, Eric, Suzanne Stevenson, and David James. 2008. A general feature space for automatic verb classification. Natural Language Engineering, 14(3):337–367. find
  22. Kelly, Michael H. 1998. To “brunch” or to “brench”: Some aspects of blend structure. Linguistics, 36(3):579–590. ↗
  23. Knowles, Elizabeth and Julia Elliott, editors. 1997. The Oxford Dictionary of New Words. Oxford University Press, New York. find
  24. Kubozono, Haruo. 1990. Phonological constraints on blending in English as a case for phonology-morphology interface. Yearbook of Morphology, 3:1–20. ↗
  25. Lapata, Mirella and Chris Brew. 2004. Verb class disambiguation using informative priors. Computational Linguistics, 30(1):45–73. ↗
  26. Lauer, Mark. 1995. Designing Statistical Language Learners: Experiments on Noun Compounds. Ph.D. thesis, Macquarie University, Sydney. ↗
  27. Lehrer, Adrienne. 2003. Understanding trendy neologisms. Italian Journal of Linguistics, 15(2):369–382. find
  28. Means, Linda G. 1988. Cn yur cmputr raed ths? In Proceedings of the Second Conference on Applied Natural Language Processing, pages 93–100, Austin, TX. find
  29. Mikheev, Andrei. 1997. Automatic rule induction for unknown-word guessing. Computational Linguistics, 23(3):405–423.
  30. Mohammad, Saif and Graeme Hirst. 2006. Distributional measures of concept-distance: A task-oriented evaluation. In Proceedings of the 2006 Conference on Empirical Methods in Natural Language Processing (EMNLP-2006), pages 35–43, Sydney. find
  31. Nadeau, David and Peter D. Turney. 2005. A supervised learning approach to acronym identification. In Proceedings of the Eighteenth Canadian Conference on Artificial Intelligence (AI’2005), pages 319–329, Victoria. find
  32. Okazaki, Naoaki and Sophia Ananiadou. 2006. A term recognition approach to acronym recognition. In Proceedings of the 21st International Conference on Computational Linguistics and the 44th Annual Meeting of the Association for Computational Linguistics (Coling-ACL 2006), pages 643–650, Sydney. find
  33. Pedersen, Ted, Siddharth Patwardhan, and Jason Michelizzi. 2004. Wordnet::Similarity—Measuring the relatedness of concepts. In Demonstration Papers at the Human Language Technology Conference of the North American Chapter of the Association for Computational Linguistics (HLT-NAACL 2004), pages 38–41, Boston, MA. find
  34. Plag, Ingo. 2003. Word-formation in English. Cambridge University Press, Cambridge. ↗
  35. Ratnaparkhi, Adwait. 1996. A maximum entropy model for part-of-speech tagging. In Proceedings of the Conference on Empirical Methods in Natural Language Processing, pages 133–142, Philadelphia, PA.
  36. Schwartz, Ariel S. and Marti A. Hearst. 2003. A simple algorithm for identifying abbreviation definitions in biomedical texts. In Proceedings of the Pacific Symposium on Biocomputing (PSB 2003), pages 451–462, Lihue, HI. find
  37. Shen, Libin and Aravind K. Joshi. 2005. Ranking and reranking with perceptron. Machine Learning, 60(1):73–96. ↗

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/cook-2010-automatically-identifying-source-words-lexical/