“文心一言”与“ChatGPT-4.0”近代文言文汉韩翻译效果对比研究 — 以《女界钟》西源人名地名音译词为例 = A Comparative Study of Classical Chinese-to-Korean Translation between “Ernie Bot” and “ChatGPT-4.0”: Focusing on Transliteration of Foreign Names in Nüjie Zhong

다국어 초록 (Multilingual Abstract)

This study systematically evaluates the Korean translations of personal and place names in Nüjie Zhong by Ernie Bot and ChatGPT, focusing on three types of transliterated words: ‘phonetic loanwords retaining the same form across historical periods(古今同形音译词),’ ‘phonetic loanwords with partially altered forms across historical periods(古今部分同形部分异形音译词),’ and ‘phonetic loanwords with entirely different forms across historical periods(古今形态全异音译词).’ The results indicate that for ‘phonetic loanwords retaining the same form across historical periods,’ the two AI systems achieved high accuracy rates of 90% and 96.3%, respectively, demonstrating consistent adherence to standardized translation methods. However, accuracy significantly declined for the more complex ‘phonetic loanwords with partially altered forms across historical periods’ and ‘phonetic loanwords with entirely different forms across historical periods.’ For the former, the accuracy rates of ChatGPT and Ernie Bot were 58.3% and 41.7%, respectively, and for the latter, 40% and 33.3%, respectively. These findings highlight the limitations of both AI systems in providing standardized translations for unfamiliar or historically nuanced terms, revealing discrepancies in translation accuracy. This study sheds light on the differences exhibited by contemporary generative AI tools in transliterating phonetic loanwords in modern Chinese classical literature and offers insights for improving future machine translation technologies.