Un traitement hybride du vague textuel : du système expert VAGO à son clone neuronal
Benjamin Icard, Vincent Claveau, Ghislain Auguste Atemezing, Paul Égré · HAL (Le Centre pour la Communication Scientifique Directe) · 2023
The VAGO tool is an expert system for lexical vagueness detection that also measures the degree of subjectivity of the speech, as well as its level of detail. In this paper, we build a neural clone of VAGO, based on a BERT-like architecture, trained on symbolic VAGO scores on a French press corpus (FreSaDa). The qualitative and quantitative analysis shows the fidelity of the neural version. By exploiting explainability tools (LIME), we then show the interest of this neural version for the enrichment of the lexicons of the symbolic version, and for the production of versions in other languages.