Year: 2025
Type: conference-paper
Abstract: Natural language explanations are attracting increasing attention in the eXplainable AI context given their ability to elucidate human and artificial decision making processes. These kinds of explanations are particularly relevant in sensitive scenarios like medicine, law and defence. However, these explanations may be constructed in different ways, depending on the overall purpose they are employed for. In this paper, we propose a novel approach to characterize natural language explanations from an argumentation perspective. The proposed architecture is composed of a first module addressing an end-to-end argument mining pipeline to identify argument components and relations, then, a second module analyzing the resulting argumentation graphs to identify patterns which caracterize explanations from an argumentation viewpoint. The experimental setting relies on the Casimedicos dataset of clinical cases. Obtained results show that the proposed approach is a promising solution towards applying AI in education with the aim to demonstrate to medical residents how to improve their explanations for a given diagnosis so as to increase patients' understanding of clinical deliberations.
Authors: Benjamin Molinet, Elena Cabrio, Serena Villata
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: bronze
OpenAlex ID: W4410357168