Hacer visible lo implícito: la construcción discursiva del acusado en sentencias sobre chemsex mediante marcos, presuposición e IA generativa

Abstract

This study analyses the discursive construction of the defendant in a corpus of twelve criminal court rulings on chemsex (39,856 words) issued by the Provincial Court of Madrid (2023–2025). Using Fillmore's frame semantics and Sbisà's taxonomy of presupposition triggers, two AI systems (Claude and ChatGPT) independently annotated eight frames across three functional levels: criminal act, institutional apparatus, and accused subject. Cohen's κ reveals a differentiated pattern: substantial for Substance, Commerce and Suspicion (κ = 0.66–0.68), moderate for Possession, Criminal_investigation, Being_at_risk and Body_description (κ = 0.49–0.58), and weak for Evidence (κ = 0.39). From a corpus-assisted Critical Discourse Analysis perspective, this gradient is interpreted as the quantitative trace of normalisation mechanisms operating through implicit presuppositional structures, within the Foucauldian framework of gouvernementalité différentielle and the analytical lens of queer criminology.