Frederic Bechet

Frederic Bechet is a researcher affiliated with Aix-Marseille Université, CNRS, and the LIF UMR 7279 laboratory. His work spans Natural Language Processing, Computational Linguistics, and Machine Learning, focusing on task structure analysis, factual knowledge robustness, and semantic-driven evaluation methodologies.

• Research Interests: Semantic Parsing, Question Answering, Text Summarization, and Multitask Learning.
• Recent Trends: Empirical studies on task inclusion via statistical deficiency, POS-driven specialization in Mixture-of-Experts models, distractor-based factual evaluation, and temporal knowledge decay in LLMs. His work on WikiFactDiff introduces a realistic framework for atomic fact updates in causal language models.
• Collaborations: Co-authored with experts in semantic annotation (Géraldine Damnati), QA systems (Alexis Nasr), and adversarial learning (Gabriel Marzinotto).
• Labs/Teams: Involved in the DECODA corpus for call-center analysis and the MEDIA corpus for dialogue understanding.