Large-scale Investigation of French Interrogative Structures Variation on Twitter: a Pilot Study

International audience This paper presents the foundation of a work which aims at studying direct interrogative structures in French varieties on Twitter in the European and African French-speaking areas within the framework of variationist sociolinguistics. French speakers have a wide range of interrogative structures for producing questions: we can count up to four different structures for Yes-no questions and more than fifteen structures for Wh- questions depending on the presence, the position and the combination of the different elements which compose them. Numerous studies within the framework of variationist sociolinguistics have shown that French interrogative structures are distributed and correlate differently with several variation dimensions such as diatopy (for instance concerning the frequency distribution of Yes-no question vs Wh-question between spoken metropolitan, Belgian and Canadian French (De Cat, 2007)), diastraty (for instance between colloquial bourgeois and worker French (Behnstedt, 1973)), diaphasy (for instance concerning the distribution of formal and less formal questions in SMS (Cougnon, 2015) or more broadly across registers (Coveney, 2011)), and diachrony (Druetta, 2011). The lack of agreement of structures distribution calls into question the scope and the validity of the results which often rely on small datasets which cannot permit to isolate one dimension from another (for instance diatopy from diastraty) or lacks from methodological rigor (for instance extralinguistic factors not distinguished from linguistic ones) was pointed out by Boutin & Rossi-Gensane (2015), Guryev & Delafontaine (2015) or Dagnac (2017). Taking Twitter as a corpus offers a promising new ground for a robust sociolinguistic interrogative structures study based on a large-scale research survey. Under this scope this paper investigates how we can automatically extract interrogative structures from tweets users contents and how we can rely on sociodemographic users information and topologic properties of networks in order to explore how interrogative structures distribution produced by European and African French speakers on Twitter correlate with sociolinguistic and/or linguistic dimensions.