• Funding : ANR, Artois
  • Start year :
  • 2025

Computational argumentation theory provides essential tools for analyzing structured debates, with applications in online discussion platforms and AI-assisted decision-making systems. In this context, arguments are evaluated using acceptability semantics, grounded in principles that determine their strength and relevance. However, the alignment of these semantics with human intuition remains underexplored, limiting their explainability and acceptability. This PhD thesis aims to evaluate and improve this alignment through several objectives.

First, it will assess whether current principles are perceived as intuitive by users and identify potential divergences. Then, it will examine the explanatory power of these principles to clarify how acceptability semantics work. Where needed, new principles and semantics will be formalized to better reflect human reasoning. Finally, the project will evaluate the intuitiveness of impact measures, which are essential for understanding how arguments influence each other within debates.

In close collaboration with psychologists, this research will refine theoretical principles based on empirical results, contributing to the development of more explainable, intuitive, and responsible AI. The findings of this thesis will not only advance theoretical work in computational argumentation but also support real-world applications such as online debate moderation and decision support in critical domains like health and the environment.