L’analyse du discours et l’intelligence artificielle pour réaliser une écriture inclusive : le projet E- MIMIC

This paper presents the E- MIMIC project, an application that aims to eliminate non-inclusive, prejudiced language forms in administrative texts written in European countries, starting with those written in Romance languages. It presents a methodology based on discourse criteria inspired by French discourse analysis and used to label a corpus of institutional documents, which are used for the deep learning of neural networks. Deep Language Modelling architectures are exploited to automatically identify non-inclusive text snippets, suggest alternative forms, and produce inclusive text rephrasing. A preliminary evaluation conducted on a benchmark dataset in Italian shows promising results and encourages us to finalise the application and to implement it also for other languages, such as French.