Abstract
The study contributes to current research on the application of AI to classical languages and offers
insights into how far Machine Translation (MT) systems have progressed in the translation of ancient
texts. In this paper, we present a comparative investigation of MT from Ancient Greek to Italian, focusing
on the performance of ChatGPT as an AI language model applied to a low-resource classical language. In
this case, Ancient Greek poses substantial challenges for computational processing due to a lower
availability of digital corpora, as well as its linguistic complexity: highly inflectional morphology, flexible
word order, extensive use of particles, and pragmatic ambiguity. Within this context, the study aims to
assess to what extent ChatGPT can produce translations that are not only formally well-structured, but
also philologically plausible. The analysis is conducted on a selected corpus of Ancient Greek passages
drawn from different genres and historical periods, with authors such as Homer, Hippocrates, and
Sappho. The methodology is the same for the analysis of each passage: the Italian translation generated by
ChatGPT is systematically compared with authoritative published human translations, so that the pros and
cons of the result emerge along with any similarities or differences. The assessment of outputs is both
quantitative and qualitative. In fact, both the frequency and types of errors encountered are analysed, as
well as a linguistic interpretation based on morphosyntax and, of course, a lexical check that also observes
the correspondence of culturally significant terms. The results indicate that ChatGPT generally produces a
fluent Italian output and shows a noteworthy ability to capture the global meaning of many passages; at
the same time, the system often opts for oversimplified or generic solutions. Performance also varies
according to genre, with greater instability in poetic and fragmentated texts. These findings suggest that,
although ChatGPT in its current state cannot be considered a reliable autonomous translator for Ancient
Greek, it can function as a supportive tool within a post-editing framework, potentially reducing the initial
workload of human translators.
Keywords: Ancient Greek, Italian, Machine Translation, AI
