Elamur Research⌕
☰
home / papers / 2022 / Potential Idiomatic Expression (PIE)-English: Corpus for Classes of Idioms

Potential Idiomatic Expression (PIE)-English: Corpus for Classes of Idioms

Paper · 2022 · Tosin Adewumi, Roshanak Vadoodi, Aparajita Tripathy et al.

Summary

Idioms are hard for NLP tasks such as machine translation, WSD and dialogue. Earlier English idiom corpora mostly separate only literal from figurative use, and many are small. Adewumi and colleagues built PIE-English, a token-based corpus that labels each potential idiomatic expression with a finer figure-of-speech class. The ten classes consist of literal use plus nine figure types: irony, hyperbole, paradox, oxymoron, personification, parallelism, euphemism, simile and metaphor.

Idioms were taken alphabetically from an online idiom dictionary. Example sentences were then drawn manually from the BNC, with UKWaC as a fallback, by four L2-English contributors, and a near-native speaker reviewed the whole collection. Two independent annotations agreed on 88.89% of cases. Disagreements were resolved using the taxonomy of Alm-Arvius. The corpus includes NLTK PoS tags, the idiom meanings, and baseline classifiers, among them BERT, which the authors report perform well.

The resource matters because it is large and gives class-level senses. A system can then react to the kind of figure used, for example responding with sympathy to a euphemism for death, instead of only noting that an idiom is present.

Key points

  • The corpus has 20,174 samples covering 1,197 idiom cases and 10 classes. Roughly 96.9% of samples come from the BNC and 3.1% from UKWaC.
  • Each case has a fixed quota of samples. Cases with a literal reading get 22 samples, of which 6 are literal, and figurative-only cases get 16.
  • The classes are very unbalanced: metaphor 14,666, euphemism 2,384, simile 1,232, literal 1,140, personification 448, paradox 112, parallelism 64, hyperbole 48, oxymoron 48, irony 32.
  • Collection was done by hand rather than by crowdsourcing or automatic extraction. This avoids crowd cheating and fatigue, duplicate sentences, and false hits.
  • The data is released under CC-BY 4.0, with code in the idesk GitHub repository. The paper includes a data statement describing contributor and annotator demographics.

Interesting ideas

  • I1. Labelling the type of figure, not just "idiomatic vs literal", lets a downstream system choose a fitting reaction, such as treating "kicked the bucket" as a gentle way of reporting a death.
  • I2. Every idiom case gets a fixed quota of literal counter-examples alongside its figurative ones, which gives built-in contrast pairs for the same surface string.
  • I3. When the corpora lacked enough literal examples, the authors made more by re-inflecting real sentences, for instance changing person and pronouns. This expands the data without inventing new contexts.
  • I4. Annotator disagreement was sometimes read as one label being more specific than the other (e.g. euphemism versus a broader metaphor), and a fixed taxonomy settled the remaining conflicts.
  • I5. Each idiom case is stored with its meaning, which the authors suggest could let MT substitute the literal sense.

For Mova

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

  • I1 (frames): prag could attach a figure-type tag (euphemism, hyperbole, irony) to recognised idioms, so speech-act and implicature reasoning adjusts tone and literal strength.
  • I2 (learn): an evaluation set for idiom handling could pair each idiom with literal uses of the same words, acting as negative controls that catch a reader that always fires on the phrase.
  • I3 (learn): silver data for rare constructions could be grown by morphologically varying agreed sentences, keeping the UD tree shape while testing lemma and agreement robustness.
  • I4 (learn): when witness parsers or LLM teachers disagree on a label, first check whether one label subsumes the other before counting it as a conflict, then adjudicate with a written guideline.
  • I5 (world): idiom spans could be mapped to their literal paraphrase before building entities and states, so "got cold feet" yields a hesitation state rather than a temperature fact.

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

🎓Luleå University of TechnologyLuleå · Sweden4 works · 682 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 2Research subjects and languages the work is about.English · 2,263 worksIdioms & multiword expressions · 217 works

Cite

Cite as Tosin Adewumi, Roshanak Vadoodi, Aparajita Tripathy, Konstantina Nikolaido, Foteini Liwicki and Marcus Liwicki. 2022. Potential Idiomatic Expression (PIE)-English: Corpus for Classes of Idioms. Proceedings of the Thirteenth Language Resources and Evaluation Conference.

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

@inproceedings{adewumi2022potential,
  title = {{Potential Idiomatic Expression (PIE)-English: Corpus for Classes of Idioms}},
  author = {Tosin Adewumi, Roshanak Vadoodi, Aparajita Tripathy, Konstantina Nikolaido, Foteini Liwicki and Marcus Liwicki},
  booktitle = {Proceedings of the Thirteenth Language Resources and Evaluation Conference},
  year = {2022},
  doi = {10.63317/4fxm7wjczqyt},
  url = {https://doi.org/10.63317/4fxm7wjczqyt}
}

References

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

  1. Adewumi, T. P., Liwicki, F., and Liwicki, M. (2019). Conversational systems in machine learning from the point of view of the philosophy of science—using alime chat and related studies. Philosophies, 4(3):41. ↗
  2. Adewumi, T., Brännvall, R., Abid, N., Pahlavan, M., Sabry, S. S., Liwicki, F., and Liwicki, M. (2022a). Småprat: Dialogpt for natural language generation of swedish dialogue by transfer learning. In 5th Northern Lights Deep Learning Workshop, Tromsø, Norway, volume 3. Septentrio Academic Publishing. find
  3. Adewumi, T., Liwicki, F., and Liwicki, M. (2022b). Word2vec: Optimal hyperparameters and their impact on natural language processing downstream tasks. Open Computer Science, 12(1):134–141. ↗
  4. Alm-Arvius, C. (2003). Figures of speech. Studentlitteratur. find
  5. Bender, E. M. and Friedman, B. (2018). Data statements for natural language processing: Toward mitigating system bias and enabling better science. Transactions of the Association for Computational Linguistics, 6:587–604. ↗
  6. Birke, J. and Sarkar, A. (2006). A clustering approach for nearly unsupervised recognition of nonliteral language. In 11th Conference of the European Chapter of the Association for Computational Linguistics. find
  7. Bizzoni, Y., Chatzikyriakidis, S., and Ghanimifard, M. (2017). “deep” learning: Detecting metaphoricity in adjective-noun pairs. In Proceedings of the Workshop on Stylistic Variation, pages 43–52. find
  8. Cook, P., Fazly, A., and Stevenson, S. (2007). Pulling their weight: Exploiting syntactic forms for the automatic identification of idiomatic expressions in context. In Proceedings of the workshop on a broader perspective on multiword expressions, pages 41–48.
  9. Cowie, A. P. and Mackin, R. (1983). Oxford dictionary of current idiomatic english v. 2:phrase, clause & sentence idioms. find
  10. Devlin, J., Chang, M.-W., Lee, K., and Toutanova, K. (2018). Bert: Pre-training of deep bidirectional transformers for language understanding. arXiv preprint arXiv:1810.04805. ↗
  11. Diab, M. and Bhutada, P. (2009). Verb noun construction mwe token classification. In Proceedings of the Workshop on Multiword Expressions: Identification, Interpretation, Disambiguation and Applications (MWE 2009), pages 17–22. find
  12. Drew, P. and Holt, E. (1998). Figures of speech: Figurative expressions and the management of topic transition in conversation. Language in society, pages 495–522. ↗
  13. Ferraresi, A., Zanchetta, E., Baroni, M., and Bernardini, S. (2008). Introducing and evaluating ukwac, a very large web-derived corpus of english. In Proceedings of the 4th Web as Corpus Workshop (WAC-4) Can we beat Google, pages 47–54. find
  14. Grant, L. and Bauer, L. (2004). Criteria for re-defining idioms: Are we barking up the wrong tree? Applied linguistics, 25(1):38–61. ↗
  15. Haagsma, H., Bos, J., and Nissim, M. (2020). Magpie: A large corpus of potentially idiomatic expressions. In Proceedings of The 12th Language Resources and Evaluation Conference, pages 279–287.
  16. Korkontzelos, I., Zesch, T., Zanzotto, F. M., and Biemann, C. (2013). Semeval-2013 task 5: Evaluating phrasal semantics. In Second Joint Conference on Lexical and Computational Semantics (* SEM), Volume 2: Proceedings of the Seventh International Workshop on Semantic Evaluation (SemEval 2013), pages 39–47. find
  17. Lakoff, G. and Johnson, M. (2008). Metaphors we live by. University of Chicago press. find
  18. Li, L. and Sporleder, C. (2009). Classifier combination for contextual idiom detection without labelled data. In Proceedings of the 2009 Conference on Empirical Methods in Natural Language Processing, pages 315–323. find
  19. Loper, E. and Bird, S. (2002). Nltk: the natural language toolkit. arXiv preprint cs/0205028. find
  20. Peng, J., Feldman, A., and Jazmati, H. (2015). Classifying idiomatic and literal expressions using vector space representations. In Proceedings of the International Conference Recent Advances in Natural Language Processing, pages 507–511. find
  21. Quinn, A. and Quinn, B. R. (1993). Figures of speech: 60 ways to turn a phrase. Psychology Press. find
  22. Raffel, C., Shazeer, N., Roberts, A., Lee, K., Narang, S., Matena, M., Zhou, Y., Li, W., and Liu, P. J. (2020). Exploring the limits of transfer learning with a unified text-to-text transformer. Journal of Machine Learning Research, 21(140):1–67. ↗
  23. Roh, Y., Heo, G., and Whang, S. E. (2019). A survey on data collection for machine learning: a big data-ai integration perspective. IEEE Transactions on Knowledge and Data Engineering. ↗
  24. Saxena, P. and Paul, S. (2020). Epie dataset: A corpus for possible idiomatic expressions. In International Conference on Text, Speech, and Dialogue, pages 87–94. Springer. find
  25. Sporleder, C., Li, L., Gorinski, P., and Koch, X. (2010). Idioms in context: The idix corpus. In LREC. Citeseer. Acronyms BERT Bidirectional Encoder Representations from Transformers. 1, 2, 4, 5 BNC British National Corpus. 2–5, 8 IAA inter-annotator agreement. 1, 3, 8 mNB multinomial Naive Bayes. 5 MT Machine Translation. 1 MWE Multi-Word Expression. 1–3 NLP Natural Language Processing. 1, 2, 4, 5 PIE Potential Idiomatic Expression. 1, 4, 5, 8 PoS part of speech. 1, 4 SoTA state-of-the-art. 1 SVM Support Vector Machine. 2, 5 UKWaC UK Web Pages. 2–5, 8 WSD word sense disambiguation. 1 Data statement for the PIE-English idioms corpus for idiom identification. Details Curation rationale Due to the unavailability of idioms dataset with more than the 2classes of literal & general figurative speech classification, this dataset was created. Dataset language English Demographics of contributors No of contributors 4 Age 42 — - — - — - Gender Male — Female — Female — Female Language L2 — L2 — L2 — L2 Demographics of annotators No of annotators 2 Annotator 1 Age - Gender Male Language L2 Annotator 2 Age - Gender - Language L2 Data characteristics Total samples 20,174 Number of classes 10 Number of cases 1,197 (e.g. “the nick of time”, “a laugh a minute”) Total samples of euphemism (2,384), literal (1,140), metaphor (14,666), personification (448), simile (1,232), parallelism (64), paradox (112), hyperbole (48), oxymoron (48), and irony (32) Base data BNC and UKWaC. Others IAA 88.89% (raw percentage) Licence CC-BY 4.0. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/adewumi-2022-potential-idiomatic-expression-pie-english/