Le prochain séminaire du D4 aura lieu le lundi 11 décembre à 14h dans la salle C005.
Tom Bourgeade (Université de Turin – Département d’Informatique), présentera un exposé intitulé “Understanding and Detecting: from Explainability in Natural Language Processing to Abusive Language Detection”.
In the rapidly evolving field of Natural Language Processing (NLP), where neural architectures are becoming increasingly more complex, explainability of models is becoming essential to maintain users’ trust. At the same time, the rise in usage of social media is making manual moderation of content increasingly difficult, accentuating the need for automated abusive language detection systems, based on these same architectures. This presentation will focus on two main areas of my research: first, explainability in NLP, in particular, the detection of biases in datasets and the creation of understandable natural language explanations. Second, I will present my work on abusive language detection, through the creation and analysis of a contextualized corpus, annotated for the presence of anti-immigrant stereotypes. I will also discuss the generalizability of these detection models, and the importance and difficulties of taking conversational context into account.