How did the idea for the Isovote project come about?
Samuel Ferey: The inception of the Isovote project was sparked by the availability of some rare materials, particularly the minutes from the Constitutional Council, which we were able to access thanks to a 2008 law pertaining to the dissemination of the Council’s deliberations. This unprecedented access allowed us to leverage all deliberations recorded between 1959 and 1995. Combining that with the powerful capabilities of Natural Language Processing (NLP) tools and AI led to the actualization of a highly scientific project.
Maxime Amblard: Prior to this, researchers were limited in their scope; they could analyze only a few deliberations at any given time due to resource constraints. Our significant innovation was to undertake an exhaustive, mostly automated analysis, establishing trends over multiple years or even decades. This initiative attracted professionals from diverse fields, including computer science, economics, public law, history, political science, and political philosophy, enabling us to better comprehend how the Constitutional Council has influenced the evolution of the French government since the establishment of the 5th Republic.
Concretely, how did you organise the work within the team?
S. F.: Initially, we had over 300 minutes stored in PDF format. The real challenge was transforming these documents into a corpus that could be critically analyzed across various disciplines. We engaged in collaborative discussions to determine our specific research objectives, acceptable error rates, and how to structure the content to clearly differentiate various components like formal language, dialogue among councillors, votes, and preliminary reports.
M. A.: By 2021, various software solutions existed for converting PDFs to text, but the error rates were often too high to yield a reliable corpus. We painstakingly reduced the error margin from 10% to 0.5%. This meticulous process involved integrating NLP tools, AI derived from NLP, and performing manual corrections for certain tasks—like accurately attributing fragments of dialogue to the correct councillor, which could significantly alter the interpretation of debates.
What does having this standardised and structured corpus available bring to the table?
M. A.: The main takeaway from our project is the establishment of a standardized corpus that serves as a foundation for researchers in various fields. In the realm of NLP, achieving a technical corpus replete with legal references and containing both formatted text and dialogue is a rarity. Specific methodologies were necessary to develop this corpus, and AI has played a crucial role in automating numerous tasks, such as evaluating word frequency, analyzing arguments, and detecting shifts over longer periods.
At this point, has there already been research using the corpus?
S. F.: Indeed. One of the first studies evaluated whether the political affiliations of councillors influenced their voting patterns. The findings indicated that the deliberative nature of the Constitutional Council itself inherently minimizes the likelihood of partisan votes. This research is set to be published by Springer. Additionally, we executed a more complex analysis on voting paradoxes, examining potential influences of doctrinal paradoxes on the Council’s votes. Also, our corpus and tools are being utilized by other researchers, such as Dominique Rémy-Granger, whose work titled Les Mots de Robert Badinter is scheduled for release in 2025.
M. A.: Beyond these initial steps, numerous articles are in progress that will probe deeper into the deliberation processes. Key inquiries include understanding how the Constitutional Council reaches its decisions, the legal arguments utilized, which councillors are most vocal, potential coalitions, and the influence of the constitutional principle of equality on decision-making. These insights will eventually enable additional teams of researchers to engage with the corpus once it’s made publicly available.
What do you already see as the impact of Isovote?
S. F.: The project has generated significant interest at the Constitutional Council, especially given its commitment to transparency and sharing internal documents since Robert Badinter’s presidency (1986-1995). The Council is eager to explore the implications of AI within the realm of constitutional law, leading us to discussions with their departments this autumn.
M. A.: The Isovote initiative has created a platform for collaboration among researchers from various disciplines, fostering mutual learning. Additionally, it has enhanced Tacteo, our collaborative platform for transcription and corpus annotation of digital sources. This is a significant step forward for humanities and social sciences researchers who invariably contend with vast documents. Moreover, Tacteo assures strict adherence to research data regulations, an aspect not always guaranteed by proprietary software.
* ANR-21-CE41-0008, Isovote 2021-2025.
The ANR Isovote project at a glance
• Duration: 48 months (October 2021 – October 2025)
• ANR funding: €125,000
• Partner: Maison des Sciences de l’Homme Lorraine
The team members, from a range of disciplines
- Samuel Ferey, project coordinator, professor of Economics in the Faculty of Law, Economics and Management at the University of Lorraine in Nancy, researcher at the Bureau d’Économie Théorique et Appliquée (BETA), specialised in legal economics and the history of liberalism.
- Maxime Amblard, professor of Computer Science at the University of Lorraine and the IDMC, specialist in AI and computational linguistics, head of the international Master’s of Natural Language Processing and associate professor in the Semagramme project team, a joint undertaking involving the CNRS, Inria and University of Lorraine, at the University of Lorraine Inria Centre and Loria.
- Youssef Kejji, work-study student in Cognitive Science, University of Lorraine. He worked on coding scripts to improve the structuring of files for the Isovote project.
- Sam Couqueberg, PhD student in History, University of Lorraine. He worked on structuring files, analysing votes and the biographies of the members of the Constitutional Council.
- Antoine Pouzet, postdoctoral researcher in Public Law, University of Lorraine. He was involved in structuring files and analysing the references produced by the corpus on the spirit of the Constitution.
- Katia Blairon, professor of Public Law, University of Lorraine. Isovote reference person on the internal functioning of the Constitutional Council, its interlocutors and its procedures.
- Gildas Renou, associate professor in Political Science, University of Lorraine. Reference person on the Constitutional Council’s historic approach.
- Anna Zielinska, associate professor in Philosophy, University of Lorraine. Reference person on the theme of “Equality”, specialist in the interpretation of the corpus from legal and political philosophy perspectives.
- Johanna Noël, associate professor in Public Law, University of Lorraine. She was involved in structuring files and the reference person on gender equality issues.
- Mathieu Disant, professor of Public Law, Paris 1 University. Reference person on the Constitutional Council’s deliberation process, its internal organisation, its departments and sub-departments.
- Antoine Chopplet, associate professor in Public Law, University of Reims. Reference person on the Constitutional Council’s internal organisation and its successive presidents.
- Herrade Igersheim, professor of Economics, University of Strasbourg. Specialist in economic voting and vote analysis.