Navigation auf uzh.ch
Translations play a significant part in our understanding of the Ancient World. In the 1970s, Norton (1968) noted a growing availability of translations of Latin and Greek classics. However, Homer’s Iliad, e.g., has been translated so often that Margon stated “that there can never be a definitive translation of any ancient work” (1970, p. 298). The questions of what happens to the original during and after the translation process, what the purpose of the translation is and how it represents (if at all) the intended meaning of the original were of particular interest to the classicists (George 1975) and also addressed by Jacques Derrida (1985).
In the digital age, the works of ancient authors are just one click away (see, e.g., Perseus and the Loeb Classics Library)—and with them, their (many) translations. This abundance of translations offers the possibility of building a diachronic corpus of translations for further analysis and comparisons. It is currently impossible to compare translations with the originals to answer questions like Are English translations different from German translations? or Are the differences grammatical, stylistic, or content-related?
Our case study will shed light on multilingual translation practices spanning several decades, exploiting large language models (LLMs) for analytical and interpretative tasks. We select a sample of Greek and Latin original texts on a spectrum from easy-to-read authors to more complex ones and build a translation database that allows users to view translation variants side-by-side (see Figure 1). We employ large LLMs like GPT (OpenAI 2023) to (i) highlight the differences (style, word choice, grammar, idiomaticity), (ii) provide preferences depending on the intended use of the translation (e.g., a layperson might be satisfied with a more accessible translation), and (iii) generate automatic translations. The findings will also improve the automatic translation of hitherto untranslated works.
Further relevant literature that is yet to be mentioned in the abstract could include Gillespie (2011), Bandina, Hadley, and McElduff (2024), and Lianeri and Zajko (2009).