Generative Artificial Intelligence and Historical Research: Challenges, Potentials, and Limitations. Application of RAG to French Parliamentary Debates of the Third Republic (1881-1940)

This work focuses on the analysis of parliamentary debates from the Third Republic (1870-1940), utilizing recent advances in text generation models based on information retrieval. Our objective is to evaluate the capabilities and limitations of this approach in modeling and interrogating this vast textual corpus, in order to better understand complex issues such as political dynamics, recurring themes, and factors influencing legislative decision-making. By combining automatic information extraction techniques and generative models, our aim is to structure the debates while ensuring the traceability of sources. This approach enables an in-depth exploration of parliamentary interactions and offers new perspectives for the historical study of political debates. In this paper, we also detail a sub-problem closely related to these issues, namely the strategies for segmenting long documents, with an evaluation method based on the automatic generation of question-answer pairs.

Découvrez d'autres productions de l'École sur les mêmes thématiques.