Comparative Analysis of AI-Generated Translations of Ancient Greek Texts

Abstract: 

This paper examines differences between AI-generated and human translations of Ancient Greek texts to assess the reliability of generative AI models. Focusing on verb semantics, it compares English translations produced by ChatGPT to those by established scholars, W.H.S. Jones (Pausanias) and Charles Darwin Adams (Hippocrates), analyzing two context-independent passages. Semantic fields of verbs are assigned and compared across three versions (original Greek, human translation, and AI output). Divergences reveal that human translators often employ interpretive or stylistic choices that can shift meanings, while AI models consistently produce literal translations, sometimes missing cultural context or subtle nuances—in one case, rendering “painting” as “inscription.” The findings indicate that, although AI may provide swift translations, it frequently fails to capture interpretive depth and historical context needed for classical studies. As a result, AI translations are less suitable for scholarly and pedagogical use compared to authoritative human versions. The paper concludes that generative AI tools for classical languages should include clear warnings about their limitations, ensuring that researchers and educators use them critic ally and with caution. This study highlights the ongoing need for human expertise in interpreting ancient texts and suggests guidelines for responsible use of AI in the digital humanities and classics fields.