Traduction automatique du français vers les pictogrammes pour améliorer la communication entre médecins et patients avec une déficience intellectuelle en milieu hospitalier
(2025)
Files
Thesis_Magali_Norre.pdf
Open Access - Adobe PDF
- 19.54 MB
Details
- Authors
- Supervisors
- Abstract
- (en) The objective of this thesis is to investigate the ways in which communication between a doctor and a patient with an Intellectual Disability (ID) can be improved using a pictograph-based system. Communication between doctors and patients is essential because it improves adherence to treatment, patients’ health status and quality of life. However, this type of interaction can lead to various complications that are rooted in problems of understanding, especially in people with ID. We study the use of a pictograph-based communication system in the context of medical consultations. In order to promote self-determination of adults with ID, we evaluate their understanding of the French medical language in the form of its translation into pictographs. We propose an original use case: we adapt Text-to-Picto to French, an automatic translation system from text to pictographs, and we extend it to the Arasaac pictographs. We then analyze the system’s errors on a corpus of medical sentences, explore various automatic simplification and disambiguation strategies, in order to propose medical explanations efficiently translated in pictographs. Simplification and word sense disambiguation for automatic translation from French into pictographs are studied through an approach involving Natural Language Processing (NLP): with rules, linguistic resources, and general and/or medical language models (neural, pre-trained, generative). Our results are similar to the state of the art for other languages and show some improvements over our first versions of Text-to-Picto. As part of these developments, we created several evaluation corpora, available for research. Finally, we propose two user studies, both quantitative and qualitative, which evaluate the understanding of isolated pictographs and the pictograph sentences among adults with ID. The first study exploits translations generated by two different systems, Text-to-Picto and PictoDr, while the second focuses on the translation of anamnesis in the context of COVID-19. The results show that several factors have a significant impact on the comprehension. They also highlight the problem of compositionality: isolated pictographs are better understood than pictograph sentences.
- Affiliations
Citations
Norré, M. (2025). Traduction automatique du français vers les pictogrammes pour améliorer la communication entre médecins et patients avec une déficience intellectuelle en milieu hospitalier. https://hdl.handle.net/2078.5/241403