Elamur Research⌕
☰
home / papers / 2018 / A Rule-Based System for Disambiguating French Locative Verbs and Their Translation into Arabic

A Rule-Based System for Disambiguating French Locative Verbs and Their Translation into Arabic

Paper · 2018 · Safa Boudhina and Héla Fehri

Summary

The paper addresses lexical ambiguity in French verbs, focusing on locative verbs, and how resolving it improves verb translation into Arabic. The authors rely on the LVF (Dubois and Dubois-Charlier's French verb dictionary), which links each verb sense to a syntactic construction and to semantic features of its arguments, such as human, animal or inanimate. They converted the locative class of LVF into a machine-usable NooJ dictionary, LocativeVerbs.dic. Each sense becomes its own entry with an inflection model, construction code, sense label and Arabic equivalent. From the construction codes they derived disambiguation rules and encoded them as NooJ transducers that match sentence patterns. A second set of transducers then translates the chosen sense into Arabic while keeping the source tense and inflection. Disambiguation reached a precision of 0.75, a recall of 0.85 and an F-measure of 0.79. Translation precision was 87%, and the authors compare their output favourably with Google and Babylon on example sentences. The work shows that a carefully formalized, linguist-built lexicon can drive explainable WSD and translation without annotated corpora.

Key points

  • LVF holds 25,609 entries for 12,310 verbs, grouped into 14 classes. Class L (locative) contains 1,524 verbs, and all of them were formalized into LocativeVerbs.dic in NooJ.
  • LVF distinguishes four construction types, each marked by a letter: A for intransitive, N for transitive with an indirect object, T for transitive with a direct object, and P for pronominal. Arguments carry semantic traits such as human, inanimate or animal. For example, apparaître has two constructions: A11 and A31.
  • Disambiguation results were P = 0.75, R = 0.85 and F = 0.79. Translation of the disambiguated verbs reached 87% precision.
  • The authors identify three main error sources:
    • implicit or underspecified subjects (e.g., "most" with no head noun), whose semantic type cannot be determined;
    • metonymy (e.g., "embassy" acting as a human agent);
    • metaphor.
  • The related-work baselines cited are Lesk at 50–70%, Brun et al. at 58% for verbs, and Nasreddine et al.'s annotation-transfer Bayesian approach at about 52%.

Interesting ideas

  • I1. Each dictionary sense of a verb is listed as a separate lexicon entry carrying its own construction code and target-language equivalent, so picking a sense also picks its translation.
  • I2. Sense selection is driven by the pairing of a syntactic frame (intransitive, transitive, pronominal…) with coarse semantic types of the subject and object, rather than by context word overlap.
  • I3. Translation is a separate stage applied only after disambiguation, and it carries over tense and inflection from the source form. Meaning choice and surface realization therefore stay decoupled.
  • I4. The error analysis singles out underspecified subjects and metonymic arguments ("embassy" as people) as places where purely lexical semantic types break.

For Mova

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

  • I1 (world): Verb-class entries from VerbNet could likewise be split per sense, each with its own frame and gloss. That way the sense chosen during parsing directly fixes the downstream frame and any generated paraphrase.
  • I2 (frames): After the UD tree is built, prag could choose a VerbNet sense by checking the dependency pattern plus animacy or human features of the nsubj and obj heads. This keeps every sense decision traceable to a rule.
  • I3 (back): Generation from the meaning graph could first fix the lexical sense, then realize morphology from stored tense and aspect features. Errors can then be attributed either to sense choice or to inflection.
  • I4 (learn): These cases make good negative-control and stress-test sets, built from sentences with metonymic or headless subjects. They can check whether the frame stage wrongly commits to a sense instead of flagging ambiguity.

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

📚 Shelves 4The field of science this work belongs to (OpenAlex fields).Computer Science · 9,086 worksArtificial Intelligence · 8,889 worksArts and Humanities · 9,296 worksLanguage and Linguistics · 9,206 works

🧭 Topics 2Research subjects and languages the work is about.Semitic languages · 1,484 worksFrench · 65 works

Cite

Cite as Safa Boudhina and Héla Fehri. 2018. A Rule-Based System for Disambiguating French Locative Verbs and Their Translation into Arabic. Proceedings of the First Workshop on Linguistic Resources for Natural Language Processing.

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

@inproceedings{boudhina2018rule,
  title = {{A Rule-Based System for Disambiguating French Locative Verbs and Their Translation into Arabic}},
  author = {Safa Boudhina and Héla Fehri},
  booktitle = {Proceedings of the First Workshop on Linguistic Resources for Natural Language Processing},
  year = {2018},
  url = {https://aclanthology.org/W18-3806/}
}

References

16 references, 1 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Marianna Apidaniaki. 2008. Acquisition automatique de sens pour la désambiguïsation et la sélection lexicale en traduction. PhD thesis, PARIS University. DIDEROT (Paris 7). find
  2. Laurent Audibert. 2003. Outils d'exploration de corpus et désambiguïsation lexicale automatique. Autre [cs.OH]. University of Provence - Aix-Marseille I, French. find
  3. Fatiha Boubekeur-Amirouche. 2008. Contribution à la définition de modèles de recherche d'information flexibles basés sur les CP-Nets. PhD thesis in computer science, University of Toulouse. Caroline Brun. find
  4. Bernard Jacquemin, and Frédérique Segond. 2001. Exploitation de dictionnaires électroniques pour la désambiguïsation sémantique lexicale. Automatic Language Processing, ATALA, 42 (3): 677-691. find
  5. Vimal Dixit, Kamlesh Dutta and Pardeep Singh. 2015. Word Sense Disambiguation and Its Approaches. CPUH-Research Journal: 2015, 1(2): 54-58. find
  6. Héla Fehri. 2012. Reconnaissance automatique des entités nommées arabes et leurs traductions vers le français. PhD thesis in computer science, University of Sfax. find
  7. Héla Fehri, Kais Haddar, and Abdelmajid Ben Hamadou. 2011. Recognition and Translation of Arabic Named Entities with NooJ Using a New Representation Model. Proceedings of the 9 th International Workshop on Finite State Methods and Natural Language Processing, Blois, France, pages 134–142. find
  8. Héla Fehri, Mohamed Zaidi, and Kamel Boudhina. 2016. Création d’un dictionnaire des verbes NooJ open source pour la langue Arabe. Report of the end of studies project. Higher Institute of Management of Gabes. find
  9. Gaston Gross and André Clas. 1997. Synonymie, polysémie et classes d'objets. Meta 421 (1997.DOI: 10.7202/002977ar): 147–154. ↗
  10. Mohammad Nessereddine, Andon Tchechmedjiev, Hervé Blanchon, and Didier Schwab. 2015. Création rapide et efficace d’un système de désambiguïsation lexicale pour une langue peu dotée. 22 nd Automatic Processing of Natural Languages, Caen. http://talnarchives.atala.org/TALN/TALN-2015/taln-2015-long-008.pdf. ↗
  11. Hinrich Schütze. 1998. Automatic word sense discrimination. Computational Linguistics: Special Issue on Word Sense Disambiguation, 24 (1): 97–123.
  12. Morgane Sénéchal and Dominique Willems. 2007. Classes verbales et régularités polysémiques : le cas des verbes trivalenciels locatifs. French language (no. 153): 92-110. ↗
  13. Max Silberztein. 2010. La formalisation du dictionnaire LVF avec NooJ et ses applications pour l'analyse automatique de corpus. Languages (No. 179-180), DOI 10.3917 / lang.179.0221: 221-241. ↗
  14. Andon Tchechmedjiev. 2012. État de l’art : mesures de similarité sémantique locales & algorithmes globaux pour la désambiguïsation lexicale à base de connaissances. Proceedings of the joint JEP-TALN-RECITAL conference volume 3: RECITAL, Grenoble, pages 295–308. find
  15. D.Yu Turdakov. 2010. Word Sense Disambiguation Methods. Programming and Computer Software. Vol. 36, No. 6: 309–326. find
  16. Florentina Vasilescu. 2003. Désambiguïsation de corpus monolingues par des approches de type Lesk. Master of Science in Computer Science, University of Montréal. ↗

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/boudhina-2018-rule-based-system-disambiguating-french/