Nullius addictus iurare in verba artificialis intelligentiae oder fide, sed cui, vide

Miscellaneous notes on ‘Translating Latin and Ancient Greek with ChatGPT – Reloaded: On the Structure of Interpretative Decisions’ by Sylvia Kurowsky

Abstract

Starting from a close translation of Cic. Mur. 51 produced by ChatGPT, this note examines the lexicographical grounding of selected semantic decisions. Particular attention is given to the verbs purgare and induere, whose alleged dictionary attestations in the OLD, Lewis & Short, and Georges are critically reassessed. The analysis concludes that the model occasionally attributes meanings or cites authorities in ways that cannot be substantiated in the claimed form. This leads to a broader methodological question: how should probabilistic language models be evaluated in relation to traditional philological standards of evidence and semantic justification?