GDR TAL Workshop: Explainability and Interpretability in Natural Language Processing

 

Place : Sorbonne Université, ISIR, Jussieu, Paris
Date : 17 March 2026
Time : 9h30 – 17h

LIUM participated in the GDR NLP workshop titled “Explainability and Interpretability in Natural Language Processing: Challenges, Methods, and Applications,” organized by the Interpretability Working Group, of which Marie Tahon is a member.

The LST team was well represented at this event: Théo Mariotte gave an oral presentation, Félix Saget presented a poster, and Nicolas Dugué delivered a full-length presentation.

Program The full program is available on the event website: https://gdr-tal-i.sciencesconf.org/program?lang=fr

Keynote Speaker: Learning, Evaluating, and Exploring Interpretable Lexical Embeddings by Nicolas Dugué

Context: With the entry into force of European regulations on AI, explainability and interpretability have become major challenges for Natural Language Processing (NLP). These approaches make it possible to understand how models work, identify their biases and limitations, and design more robust and responsible architectures.

 

Themes: The event will provide an opportunity to present exploratory contributions, feedback, and work in progress focused on Natural Language Processing (NLP), addressing in particular :

• Interpretability for bias detection : How do explanatory methods reveal biases in data or models, and how do they contribute to more equitable systems?
• Interpretability and Applications : The use of explanatory approaches in sensitive fields (medicine, education, social sciences) and their societal impact.
• Speech, Multimodality, and Interpretability : Specific challenges related to explaining language models or multimodal systems (text, images, audio, video).
• Interpretability by design : Incorporating explainability from the very beginning of model design, or as a catalyst for new architectures.