Large Language Models in metaphor identification -- The case of presuicidal interaction

Gábor Simon, Eötvös Loránd University and Tímea Borbála Bajzát, Eötvös Loránd University and Natabara Máté Gyöngyössy, Eötvös Loránd University and Péter Gergő Molnár, Eötvös Loránd University and Noémi Prótár, Eötvös Loránd University and Balázs Indig, Eötvös Loránd University (2026) Large Language Models in metaphor identification -- The case of presuicidal interaction. ACTA LINGUISTICA ACADEMICA. ISSN 2559-8201 (In Press)

Preview

Text Acta_paper_revised_after editorial notes.pdf - Accepted Version Available under License Creative Commons Attribution. Download (662kB) | Preview

Abstract

This paper addresses the challenge of automating metaphor identification in Hungarian using generative AI technology. Our tool uses a dictionary-augmented chain-of-thought prompting method to perform metaphor identification and implements a widely used protocol for manual annotation. The study employs a curated corpus of online forum posts about suicidal ideation, with manually annotated linguistic metaphors. It describes the challenges of automating metaphor annotation, discusses alternative methodological approaches to address these issues, outlines the project’s materials, and assesses the performance of the models (reporting on ~89% of general accuracy and ~79% of balanced accuracy in the case of the best-performing model, GPT-5), concluding with suggestions for further development.

Actions (login required)

View Item