Word Sense Disambiguation for Automatic Translation of Medical Dialogues into Pictographs

Proceedings of Recent Advances in Natural Language Processing, pages 803–812 Varna, Sep 4–6, 2023 https://doi.org/10.26615/978-954-452-092-2_087 803 Word Sense Disambiguation for Automatic Translation of Medical Dialogues into Pictographs Magali Norr´e1,2, R´emi Cardon1, Vincent Vandeghinste3, Thomas Franc¸ois1 1CENTAL, UCLouvain, Belgium 2FTI/TIM, Universit´e de Gen`eve, Switzerland 3Instituut voor de Nederlandse Taal, The Netherlands Centre for Computational Linguistics, Leuven.AI, KU Leuven, Belgium {magali.norre,remi.cardon,thomas.francois}@uclouvain.be vincent.vandeghinste@ivdnt.org Abstract Word sense disambiguation is an NLP task em- bedded in different applications. We propose to evaluate its contribution to the automatic translation of French texts into pictographs, in the context of communication between doc- tors and patients with an intellectual disabil- ity. Different general and/or medical language models (Word2Vec, fastText, CamemBERT, FlauBERT, DrBERT, and CamemBERT-bio) are tested in order to choose semantically cor- rect pictographs leveraging the synsets in the French WordNets (WOLF and WoNeF). The results of our automatic evaluations show that our method based on Word2Vec and fastText significantly improves the precision of medical translations into pictographs. We also present an evaluation corpus adapted to this task. 1 Introduction Dialogue between doctors and patients is essen- tial, as it enhances the patients’ health status, their medication adherence, and their overall quality of life (Riedl and Sch¨ußler, 2017). However, this dialogue can be impaired by misunderstandings, in particular for patients with an Intellectual Dis- ability (ID). Various Augmentative and Alternative Communication (AAC) systems are used by people with disabilities (Beukelman and Mirenda, 1998), including automatic translation tools from text into pictographs (Vandeghinste et al., 2015). One of the main issues that those systems face is polysemy. For example, in the French sentence to be translated “avez-vous appliqu´e une cr`eme sur la l´esion ?” (did you put cream on the lesion?), “cr`eme” (cream) can be interpreted as OINTMENT or LIQUID CREAM. A translation system has to be able to produce the correct pictograph, here one that would represent OINTMENT. In this article, we focus on Word Sense Dis- ambiguation (WSD) of French polysemous words that can be used orally by doctors in questions and instructions for anamnesis in emergency set- tings (Norr´e et al., 2022). The Text-to-Picto system we use translates French into Arasaac,1 Sclera2 or Beta3 pictograph sets, designed for AAC users with an ID (Norr´e et al., 2021). In order to provide a bet- ter semantic understanding of the input sentence, we test various language models (static, contextual, trained on general and/or medical data), and dif- ferent French sense inventories. In addition, we present an evaluation corpus adapted to this task. Section 2 describes existing work on WSD and text-to-pictograph systems. Section 3 introduces our methodology and the language models we used, while section 4 presents the Text-to-Picto system, the evaluation corpus, and the results. Our evalua- tions with Word2Vec and fastText show significant improvements over the baseline with the Text-to- Picto tool. We discuss the results in section 5. 2 Related Work WSD has already been used in automatic text-to- pictograph systems, in order to improve the transla- tion of polysemous words for the general language. …

Content truncated to 500 words