Using Structured Health Information for Controlled Generation of Clinical Cases in French

Hugo Boulanger (1, 2) , Nicolas Hiebel (3, 4, 5, 6) , Olivier Ferret (1, 2) , Karën Fort (7, 8) , Aurélie Névéol (6, 5)

1 DIASI (CEA, LIST) - Département Intelligence Ambiante et Systèmes Interactifs
2 LASTI - Laboratoire Analyse Sémantique Textes et Images
3 Université Paris-Saclay
4 CNRS - Centre National de la Recherche Scientifique
5 LISN - Laboratoire Interdisciplinaire des Sciences du Numérique
6 STL - Sciences et Technologies des Langues - LISN
7 SEMAGRAMME - Semantic Analysis of Natural Language
8 SU - Sorbonne Université

Text generation opens up new prospects for overcoming the lack of open corpora in fields such as healthcare, where data sharing is bound by confidentiality. In this study, we compare the performance of encoder-decoder and decoder-only language models for the controlled generation of clinical cases in French. To do so, we fine-tuned several pre-trained models on French clinical cases for each architecture and generate clinical cases conditioned by patient demographic information (gender and age) and clinical features. Our results suggest that encoder-decoder models are easier to control than decoder-only models, but more

Mots clés: Controlled Generation; Clinical Texts; Synthetic Texts; French

Domaines: Informatique [cs]

Hugo Boulanger, Nicolas Hiebel, Olivier Ferret, Karën Fort, Aurélie Névéol. Using Structured Health Information for Controlled Generation of Clinical Cases in French. The 6th Clinical Natural Language Processing Workshop At NAACL 2024 (ClinicalNLP 2024), Jun 2024, Mexico city, Mexico. ⟨hal-04558890⟩