AI 문학번역에서 프롬프트 엔지니어링이 번역 오류와 창의성에 미치는 영향
Effects of prompt engineering on error reduction and creativity in AI literary translation
1
조회
0
다운로드
0
인용
0
참고문헌
키워드
초록
This study explores how highly calibrated prompts influence the translation of Hyŏn Chin’gŏn’s 1921 short story “Sul kwŏnhanŭn sahoe” (A Society That Drives You to Drink). It draws comparisons between English translations generated with a Role-Goal-Context (RGC) prompt and those produced with a baseline control prompt (“Please translate the Korean text into English, preserving its structure and details as much as possible.”). The results show no significant difference in the overall frequency of translation errors between the two prompt types. However, translations created with the RGC prompt exhibit a higher rate of over-translation, whereas rates of omission and mistranslation remain relatively stable across both conditions. Moreover, RGC-prompted translations demonstrate a greater use of creative strategies, such as paraphrasing and elaboration. These findings suggest that while RGC prompts may not reduce the overall incidence of translation errors, they enhance the contextual sensitivity and readability of the output.
Abstract
This study explores how highly calibrated prompts influence the translation of Hyŏn Chin’gŏn’s 1921 short story “Sul kwŏnhanŭn sahoe” (A Society That Drives You to Drink). It draws comparisons between English translations generated with a Role-Goal-Context (RGC) prompt and those produced with a baseline control prompt (“Please translate the Korean text into English, preserving its structure and details as much as possible.”). The results show no significant difference in the overall frequency of translation errors between the two prompt types. However, translations created with the RGC prompt exhibit a higher rate of over-translation, whereas rates of omission and mistranslation remain relatively stable across both conditions. Moreover, RGC-prompted translations demonstrate a greater use of creative strategies, such as paraphrasing and elaboration. These findings suggest that while RGC prompts may not reduce the overall incidence of translation errors, they enhance the contextual sensitivity and readability of the output.