Elamur Research⌕
☰
home / papers / 2024 / Multi-word expressions in biomedical abstracts and their plain English adaptations

Multi-word expressions in biomedical abstracts and their plain English adaptations

Paper · 2024 · Sergei Bagdasarov and Elke Teich

Summary

The paper asks whether plain-language rewrites of biomedical abstracts rely on multi-word expressions (MWEs) differently from the original expert abstracts, and whether those MWEs are cheaper to process. The authors used the PLABA dataset, which pairs PubMed abstracts with human-written lay adaptations. They tokenized it with TreeTagger and parsed it with Stanza. MWEs were found in two ways:

  • by reading off UD relations that mark multi-word units, together with their heads;
  • by matching entries from the Academic Formulas List (AFL).

They compared MWE distributions with Kullback-Leibler Divergence and estimated per-MWE surprisal with a four-gram model. Contrary to their hypothesis, plain abstracts contain slightly fewer MWEs. However, the two registers differ clearly in which MWEs they use:

  • Expert texts favour statistical and study-design terminology.
  • Lay texts lean on phrasal verbs and everyday compounds.

Compounds and AFL formulas, and marginally phrasal verbs, carry lower surprisal in the plain versions. This suggests that MWE choice helps reduce reading effort, which matters for health communication and automatic simplification.

Key points

  • Data: PLABA, with 749 complex abstracts (199,851 tokens) and 919 plain adaptations (249,301 tokens). The plain side has fewer types (13,117 vs 17,425).
  • MWE extraction used the UD labels compound, compound:prt, fixed, flat and flat:foreign (the last did not occur), plus AFL core and written formulas. No frequency cutoff was applied.
  • Plain abstracts have fewer unique MWEs (6,155 vs 6,700) and a marginally lower rate per million words (62,976 vs 63,802). Phrasal verbs are far more varied in plain texts (114 vs 26 unique), while proper names (flat) drop (29 vs 67).
  • KLD between the two registers is 3.35 bits (complex vs plain) and 3.18 bits (plain vs complex). Typical of the complex side are terms like odds ratio and cohort study; typical of the plain side are find out and make up.
  • Wilcoxon rank sum tests show significantly lower surprisal in plain texts for compounds and AFL formulas, and a marginal effect for phrasal verbs. Paired examples:
    • blood glucose levels (4.73 bits) vs blood sugar levels (3.24 bits)
    • the expert term neurodegenerative disease scores 12.20 bits, while its lay counterpart brain disorder scores only 3.65 bits

Interesting ideas

  • I1. MWEs are taken directly from dependency labels (compound, compound:prt, fixed, flat) by joining each dependent with its head, so the inventory is grammar-driven rather than threshold-driven.
  • I2. Per-feature KLD contributions are used to rank which items most separate two corpora, turning one divergence number into an inspectable list.
  • I3. Surprisal is averaged over all occurrences of an MWE to obtain a per-expression processing-cost score, then compared across paired expert/lay variants.
  • I4. The parallel complex/plain corpus yields concept-preserving substitutions, sometimes changing structure, such as one word becoming a compound (placebo → dummy treatment) or an adjective-noun term becoming a noun compound (renal cancer → kidney cancer).
  • I5. The study separates "how many MWEs" from "how informative they are" and finds the frequency hypothesis wrong while the informativity one partly holds.

For Mova

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

  • I1 (tree): After parsing, the same arcs could be collapsed into single lexical units, so later stages see follow up or according to as one item rather than two words.
  • I2 (learn): Comparing parser or frame outputs on two text sources this way would show exactly which constructions drive a mismatch, giving explainable diagnostics for silver-data drift.
  • I3 (back): During generation for simple output, candidate phrasings of the same concept could be scored this way and the lower-cost one preferred.
  • I4 (world): Such pairs could be stored as equivalence links in the ontology, so expert and lay mentions resolve to the same entity.
  • I5 (learn): Evaluations of simplification or paraphrase should report count-based and information-based measures separately, since they can point in opposite directions.

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

🎓Saarland UniversitySaarbrücken · Germany148 works · 18,287 citations

 Fields, topics and concepts

📚 Shelves 5The 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 worksLiterature and Literary Theory · 519 works

🧭 Topics 2Research subjects and languages the work is about.English · 2,263 worksIdioms & multiword expressions · 217 works

Cite

Cite as Sergei Bagdasarov and Elke Teich. 2024. Multi-word expressions in biomedical abstracts and their plain English adaptations. Proceedings of the 4th International Conference on Natural Language Processing for Digital Humanities.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.18653/v1/2024.nlp4dh-1.46

@inproceedings{bagdasarov2024multi,
  title = {{Multi-word expressions in biomedical abstracts and their plain English adaptations}},
  author = {Sergei Bagdasarov and Elke Teich},
  booktitle = {Proceedings of the 4th International Conference on Natural Language Processing for Digital Humanities},
  year = {2024},
  doi = {10.18653/v1/2024.nlp4dh-1.46},
  url = {https://doi.org/10.18653/v1/2024.nlp4dh-1.46}
}

References

19 references, 3 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Britt Erman and Beatrice Warren. 2000. The idiom principle and the open choice principle. Text & Talk, 20(1):29–62. ↗
  2. Diego Alves, Stefania Degaetano-Ortlieb, Elena Schmidt, and Elke Teich. 2024a. Diachronic analysis of multi-word expression functional categories in scientific English. In Proceedings of the Joint Pauline Foster. 2001. Rules and routines: A consideration of their role in the task-based language production of native and non-native speakers. In Martin Bygate, Peter Skehan, and Merrill Swain, editors, Researching Pedagogic Tasks: Second Language Learning, Teaching, and Testing, pages 75–93. Longman, Harlow.
  3. Claude Shannon. 1948. A mathematical theory of communication. The Bell System Technical Journal, 27:379–423, 623–656. ↗
  4. Rita Simpson-Vlach and Nick C. Ellis. 2010. An Academic Formulas List: New Methods in Phraseology Research. Applied Linguistics, 31(4):487–512. ↗
  5. Dmitriy Genzel and Eugene Charniak. 2002. Entropy rate constancy in text. In Proceedings of the 40th ACL, pages 199–206, Philadelphia, PA, USA.
  6. John Sinclair. 1991. Corpus, Concordance, Collocation. Oxford University Press, Oxford. find
  7. M.A.K. Halliday. 1988. On the language of physical science. In Mohsen Ghadessy, editor, Registers of written English: Situational factors and linguistic features, pages 162–177. Pinter, London. find
  8. Steven A. Hauck. 2019. Sharing planetary science in plain language. Journal of Geophysical Research: Planets, 124(10):2462–2464. ↗
  9. Solomon Kullback and Richard A. Leibler. 1951. On information and sufficiency. The Annals of Mathematical Statistics, 22(1):79–86. ↗
  10. Hui Li, Kate L. Warrington, Ascensión Pagán, Kevin B. Paterson, and Xinchun Wang. 2021. Independent effects of collocation strength and contextual predictability on eye movements in reading. Language, Cognition and Neuroscience, 36(8):1001–1009. ↗
  11. Anna Siyanova-Chanturia, Kathy Conklin, Sendy Caffarra, Edith Kaan, and Walter J.B. van Heuven. 2017. Representation and processing of multi-word expressions in the brain. Brain and Language, 175:111– 122. ↗
  12. Anna Siyanova-Chanturia, Kathy Conklin, and Norbert Schmitt. 2011. Adding more fuel to the fire: An eye-tracking study of idiom processing by native and non-native speakers. Second Language Research, 27(2):251–272. ↗
  13. Antoine Tremblay, Bruce Derwing, Gary Libben, and Chris Westbury. 2011. Processing advantages of lexical bundles: Evidence from self-paced reading and sentence recall tasks. Language learning, 61(2):569– 613. ↗
  14. Andrew Pawley and Frances H. Syder. 1983. Two puzzles for linguistic theory: Nativelike selection and nativelike fluency. In Jack C. Richards and Richard W. Schmidt, editors, Language and communication, pages 191–225. Longman, London, UK. find
  15. Peng Qi, Yuhao Zhang, Yuhui Zhang, Jason Bolton, and Christopher D. Manning. 2020. Stanza: A Python Natural Language Processing Toolkit for Many Human Languages. In Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics: System Demonstrations, pages 101–108, Online. Association for Computational Linguistics. R Development Core Team. R: A language and environment for statistical computing. R Foundation for Statistical Computing. ISBN 3-900051-07-0.
  16. C. Sophie Rammell, Diana Van Lancker Sidtis, and David B. Pisoni. 2017. Perception of formulaic and novel expressions under acoustic degradation. Ment Lex, 12(2):234–262. ↗
  17. Helmut Schmid. 1994. Probabilistic part-of-speech tagging using decision trees. In Proceedings of the International Conference on New Methods in Language Processing, Manchester, UK. find
  18. Helmut Schmid. 1995. Improvements in part-of-speech tagging with an application to German. In Proceedings of the ACL SIGDAT-Workshop, Kyoto, Japan. find
  19. Cassie Sedgwick, Laura Belmonte, Amanda Margolis, Patricia Osborn Shafer, Jennifer Pitterle, and Barry E. Gidal. 2021. Extending the reach of science – talk in plain language. Epilepsy Behavior Reports, 16:100493. ↗

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/bagdasarov-2024-multi-word-expressions-biomedical-abstracts/