中央研究院歷史語言研究所

講   者：James L. Zainaldin 博士（美國范德堡大學古典與地中海研究助理教授）

主   題：Can AI Read Galen? Evaluating Machine Translation for Ancient Greek Medical Texts

主持人：張谷銘 博士（本所副研究員）

時    間：2026年4月17日（五）10:00−12:00

地    點：本所研究大樓七樓701會議室

主辦單位：本院關鍵突破計劃「人工智慧時代的創作——跨領域的探索」

協辦單位：本所文化思想史研究室

演講摘要：

The rapid adoption of large language models like ChatGPT in academic contexts raises urgent questions for humanistic disciplines working with ancient languages. How good are these tools, really? Can classicists and historians responsibly use them? This paper explores these questions through a case study in translating the Greek physician Galen of Pergamum (129–216 CE), whose vast and varied corpus—much of it still untranslated—exemplifies the challenges and opportunities AI presents for scholarship on technical ancient texts. We examine translations produced by three major commercial LLMs (ChatGPT, Claude, and Gemini), subjecting them to both computational metrics standard in natural language processing and close philological analysis by domain experts. Our investigation foregrounds methodological problems: What does translation “quality” mean when expert translators legitimately disagree? How do we evaluate accuracy for texts with no existing English translation? And does AI actually “translate,” or merely regurgitate human translations absorbed during training? By bringing computational and humanistic approaches into dialogue, we aim to develop frameworks for responsible AI integration in philological research—neither dismissing these tools nor adopting them uncritically, but understanding precisely what they can and cannot do for scholars of the ancient world.

（本演講以英文進行，無須事先報名）