Elamur Research⌕
☰
home / papers / 2016 / Patrons sémantiques pour l'extraction de relations entre termes - Application aux comptes rendus radiologiques (Here the title in English)

Patrons sémantiques pour l'extraction de relations entre termes - Application aux comptes rendus radiologiques (Here the title in English)

Paper · 2016 · Lionel Ramadier and Mathieu Lafourcade

Summary

The paper addresses extracting typed semantic relations between medical terms in French radiology reports, a genre where verbs are often dropped and short connectors such as du, avec or en carry most of the relational meaning. The authors work on a corpus of 35,000 reports covering MRI, CT, X-ray, ultrasound and interventional radiology. Their system, IMAIOS, first matches hand-written linguistic patterns derived from a subset of the corpus. Because generic connectors fit many relation types, candidate matches are then filtered by semantic constraints: rules check the categories of both arguments in the crowdsourced lexical-semantic network JeuxDeMots (JDM). For example, if both sides are anatomical locations, du yields holonymy, but if the left side is a disease and the right side is a location, it yields a location relation. According to the authors, adding these type checks significantly raises precision. The method needs no syntactic parser and no POS tagger. This matters because a rich typed lexicon can resolve ambiguous surface patterns cheaply and transparently.

Key points

  • The corpus contains 35,000 French radiology reports across several imaging modalities. Relation types were chosen with advice from radiologists.
  • Ten target relations are used: characteristic, location, target (affected population), holonymy, part-of, sign/symptom, cause, consequence, treatment and accompaniment.
  • JDM serves as the knowledge base. It is a general-purpose French network built through a game with a purpose plus the Diko contribution tool, and it has been enriched with radiology knowledge. It offers POS, lemmas, spelling variants and about 80 relation types. For instance, it has 25,580 incoming relations for anatomie and 15,000 for maladie.
  • The first version uses 42 semantic patterns, selected carefully so that recall would not drop too much once constraints were added.
  • The characteristic relation is found differently, through repeated noun+adjective co-occurrence. For example, multifocal appears 5 times with carcinome hépatocellulaire.

Interesting ideas

  • I1. Each ambiguous connector gets a small rule table keyed on the semantic classes of its two arguments, so one surface pattern can map to different relation types depending on what fills its slots.
  • I2. A single phrase may legitimately carry two relations at once (lobe caudé du foie is both location and holonymy), so the output does not force one label.
  • I3. Precision is gained by type-checking pattern candidates against a lexical-semantic network, rather than by adding parsing machinery.
  • I4. The authors deliberately limit the rule set so that recall stays acceptable, which treats the precision–recall trade-off as an explicit design budget.
  • I5. Repeated adjacent noun+adjective pairs in the corpus become evidence of a typical characteristic.

For Mova

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

  • I1 (frames): Rules of this kind could disambiguate the roles of prepositions like "of", "with" and "in" inside noun phrases. Each decision could be logged with the class check that fired, which keeps it explainable.
  • I2 (world): The world model could store several compatible relations for one connector instead of picking one arbitrarily. Contradictions could then be reserved for incompatible readings only.
  • I3 (world): Mova's ontology from WordNet/Wikidata could act as a selectional-restriction filter on candidate relations. Candidates whose argument types fail the check would be rejected with a stated reason.
  • I4 (learn): When self-induced repair rules are evaluated, each new constraint could be scored on how much recall it costs as well as on what it fixes.
  • I5 (world): Frequency thresholds over parsed amod edges could seed default properties of entity types in the knowledge graph.

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, topics and concepts

📚 Shelves 6The field of science this work belongs to (OpenAlex fields).Computer Science · 18,614 worksArtificial Intelligence · 17,965 worksArts and Humanities · 17,450 worksLanguage and Linguistics · 17,296 worksBiochemistry, Genetics and Molecular Biology · 263 worksMolecular Biology · 260 works

🧭 Topics 1Research subjects and languages the work is about.English · 2,379 works

Cite

Cite as Lionel Ramadier and Mathieu Lafourcade. 2016. Patrons sémantiques pour l'extraction de relations entre termes - Application aux comptes rendus radiologiques (Here the title in English). Actes de la conférence conjointe JEP-TALN-RECITAL 2016. volume 2 : TALN (Posters).

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

@inproceedings{ramadier2016patrons,
  title = {{Patrons sémantiques pour l'extraction de relations entre termes - Application aux comptes rendus radiologiques (Here the title in English)}},
  author = {Lionel Ramadier and Mathieu Lafourcade},
  booktitle = {Actes de la conférence conjointe JEP-TALN-RECITAL 2016. volume 2 : TALN (Posters)},
  year = {2016},
  url = {https://aclanthology.org/2016.jeptalnrecital-poster.26/}
}

References

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

  1. A BACHA A. B. & Z WEIGENBAUM P. (2011). Automatic extraction of semantic relations between medical entities : a rule based approach. J. Biomedical Semantics, 2(S-5), S4. find
  2. A UGER A. & B ARRIÈRE C. (2008). Pattern-based approaches to semantic relation extraction : A state-of-the-art. Terminology, 14(1), 1–19. find
  3. C HKLOVSKI T. & P ANTEL P. (2004). Verbocean : Mining the web for fine-grained semantic verb relations. In EMNLP, volume 4, p. 33–40. find
  4. C HOWDHURY M. F. M. & L AVELLI A. (2012). Combining tree structures, flat features and patterns for biomedical relation extraction. In Proceedings of the 13th Conference of the European Chapter of the Association for Computational Linguistics, p. 420–429 : Association for Computational Linguistics. find
  5. C IMINO J., B ARNETT G. et al. (1993). Automatic knowledge acquisition from medline. Methods of information in medicine, 32, 120–120. find
  6. D ELBECQUE T., J ACQUEMART P. & Z WEIGENBAUM P. (2005). Utilisation du réseau sémantique de l’umls pour la définition de types d’entités nommées médicales. In CORIA, p. 101–118. find
  7. E MBAREK M. & F ERRET O. (2008). Learning patterns for building resources about semantic relations in the medical domain. In LREC. find
  8. F ADER A., S ODERLAND S. & E TZIONI O. (2011). Identifying relations for open information extraction. In Proceedings of the Conference on Empirical Methods in Natural Language Processing, p. 1535–1545 : Association for Computational Linguistics. find
  9. G IRJU R., B ADULESCU A. & M OLDOVAN D. (2003). Learning semantic constraints for the automatic discovery of part-whole relations. In Proceedings of the 2003 Conference of the North American Chapter of the Association for Computational Linguistics on Human Language Technology-Volume 1, p. 1–8 : Association for Computational Linguistics.
  10. H EARST M. A. (1992). Automatic acquisition of hyponyms from large text corpora. In Proceedings of the 14th conference on Computational linguistics-Volume 2, p. 539–545 : Association for Computational Linguistics.
  11. H INDLE D. (1990). Noun classification from predicate-argument structures. In Proceedings of the 28th annual meeting on Association for Computational Linguistics, p. 268–275 : Association for Computational Linguistics.
  12. H OGENBOOM F., IJ NTEMA W. & F RASINCAR F. (2012). Text-based information extraction using lexico-semantic patterns. In 24th Benelux Conference on Artificial Intelligence (BNAIC 2012), p. 293–294. find
  13. J ELIER R., J ENSTER G., D ORSSERS L. C., VAN DER E IJK C. C., VAN M ULLIGEN E. M., M ONS B. & K ORS J. A. (2005). Co-occurrence based meta-analysis of scientific texts : retrieving biological relationships between genes. Bioinformatics, 21(9), 2049–2058. find
  14. L AFOURCADE M. (2007). Making people play for lexical acquisition with the jeuxdemots prototype. In SNLP’07 : 7th international symposium on natural language processing, p.7̃. find
  15. L EE C.-H., K HOO C. S. & N A J.-C. (2004). Automatic identification of treatment relations for medical ontology learning : An exploratory study. find
  16. M ENG F. & M ORIOKA C. (2015). Automating the generation of lexical patterns for processing free text in clinical documents. Journal of the American Medical Informatics Association, p. ocv012. find
  17. N AZAR R., V IVALDI J. & W ANNER L. (2012). Co-occurrence graphs applied to taxonomy extraction in scientific and technical corpora. Procesamiento del lenguaje natural, 49, 67–74. find
  18. R AMADIER L., Z ARROUK M., L AFOURCADE M. & M ICHEAU A. (2014). Spreading relation annotations in a lexical semantic network applied to radiology. In Computational Linguistics and Intelligent Text Processing, p. 40–51. Springer. find
  19. R INK B., H ARABAGIU S. & R OBERTS K. (2011). Automatic extraction of relations between medical concepts in clinical texts. Journal of the American Medical Informatics Association, 18(5), 594–600. find
  20. S NOW R., J URAFSKY D. & N G A. Y. (2006). Semantic taxonomy induction from heterogenous evidence. In Proceedings of the 21st International Conference on Computational Linguistics and the 44th annual meeting of the Association for Computational Linguistics, p. 801–808 : Association for Computational Linguistics. find
  21. S ONG M., K IM W. C., L EE D., H EO G. E. & K ANG K. Y. (2015). Pkde4j : Entity and relation extraction for public knowledge discovery. Journal of biomedical informatics, 57, 320–332. find
  22. S UCHANEK F. M., I FRIM G. & W EIKUM G. (2006). Combining linguistic and statistical analysis to extract relations from web documents. In Proceedings of the 12th ACM SIGKDD international conference on Knowledge discovery and data mining, p. 712–717 : ACM. find
  23. V ON A HN L. (2006). Games with a purpose. Computer, 39(6), 92–94. find
  24. Z ARROUK M., L AFOURCADE M. & J OUBERT A. (2014). About inferences in a crowdsourced lexical-semantic network. EACL 2014, p. 174. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/ramadier-2016-patrons-semantiques-pour-l-extraction/