Research

COMPLEX SYSTEMS AND COMPUTATIONAL SOCIAL SCIENCE

COMPUTATIONAL SOCIAL SCIENCE FOR POLICY

I have always been convinced of the importance of informing policymakers. Computational methods could help us reach general enough and sufficiently results in a timeframe adapted to actual political decision. I have for instance proposed an agenda for a "computational diplomacy" framework, or evaluated in real-time the impact of the "leaks" organized by the French government regarding future measures the opinion and behaviors of the French population during COVID-19.

SOCIAL NETWORKS AND VACCINE CONTROVERSIES

In the past twenty years or so, doubts towards vaccines have become an increasingly important global public health issue - especially in France, where vaccine hesitancy is among the highest in the western countries. Through various types of data, ranging from the results of the French citizen consultation on vaccination to contents posted on websites or Twitter, I am studying the dynamics of criticism or defense of vaccines. 

CONSPIRACY THEORIES: DEFINiTION, DETERMINANTS, VARIATIONS

Using wide-scale surveys or analyzing online behaviour, we examine the social and demographic determinants of conspiracy beliefs.

COMPUTATIONAL LINGUISTICS AND DIGITAL HUMANITIES

WHO WROTE WHAT ? AUTHORSHIP ATTRIBUTION FROM THE TROUBADOURS TO QANON

Since its first steps in the 1960s, the study of linguistic properties specific to a certain author, also known as stylometry, has made tremendous progress. Under certain conditions, it can help to settle debates about the purported author of a text. It for instance confirms historians' claim that Molière is actually the author of his plays, and did not use a ghostwriter such as Pierre Corneille, as a century-old theory suggested. It can also  give clues about texts whose authors are unknown - as in the case of the Troubadour's vidas. Without being able to name the author of each text, these techniques helped us understanding which texts could have been written by the same person or during the same period.

ORIGINS OF OUR LANGUAGE, ORIGINS OF OUR TEXTS

The birth of the French language

For a long time, latin and "vulgar" languages have been considered as two very distinct worlds, evolving their own separate ways. In an ongoing research projet with Rémy Verdo, we investigate the transformation from late latin to early romance in France. Inspired by works in sociolinguistics by Michel Banniard, we show how vulgar language and latine actually co-evolved and form a continuum rather than two separate entities. We study the variation of language registers from text to text, but also inside each text.  We try at systematizing our study through machine learning tools.

Reconstructing manuscripts

Most ancient texts or scores we read today have been copied numerous times before arriving to us. In the process, voluntary changes by the copyists, or mistakes they have made, have altered the work's original version. To understand these errors and changes' history and get a better idea of what the original texts or scores looked like, Jean-Baptiste Camps and I elaborated on works by William Poole to develop an algorithm dedicated to the task. We also wrote  a package for the software R, stemmatology to help implement our method.

WHEN WRITING IS FEELING : ADVANCES IN COMPUTATIONAL PSYCHOLINGUISTICS

Using techniques derived from stylometry, I have experimented with colleagues met at Geneva University on our capacity to detect phenomena described by clinical psychology, only relying on the text written by subjects diagnosed with a particular condition. We first focused on Attention Deficit with or without Hyperactivity Disorders (ADHD), a complicated syndrom to diagnose, which draws more and more attention in the public sphere. Our first results give promising results, both for analyzing what ADHD does to linguistic expression, and to help diagnose ADHD.


Communications

Conferences (selection 2017 - * )

Invited talks (selection 2018 - * )

Professional service


Invitations