<![CDATA[Constraints and guidelines on subtitle translation impose demanding tasks on machine translation systems, hence the urge to control translation output from machine translators. Unlike conventional machine translation, text-generating artificial intelligence such as ChatGPT can produce controlled translations that satisfy subtitle translation guidelines. Thus, this study aims to maintain the quality of subtitle translation while limiting character length to keep subtitles compact. The study employed a descriptive-comparative analysis to examine the differences between a plain prompt and a properly designed prompt. The prompts used in this study included details such as contextual information, target audience specifications, formal language requirements, character-length limitations, prohibition on altering sentence order, and illustrative examples. This study utilized 50 subtitle segments from the film Mufasa: The Lion King (2024) and implemented four prompt types: Plain Prompt, Zero-Shot Prompt, One-Shot Prompt, and Few-Shot Prompt. Translation quality was measured using COMET, BERTScore, and chrF++. Scores from these metrics showed improvement, with the Few-Shot Prompt reaching 89.920, 88.830, and 59.904, respectively, while the Plain Prompt yielded only 89.000, 85.806, and 55.767. The findings also revealed a reduction in character count, with the Plain Prompt generating 1,834 characters and the Few-Shot Prompt generating 1,377. These results indicate that the Few-Shot Prompt produces higher-quality translations and maintains greater compactness compared to the Plain Prompt.]]>
Copyrights © 2026
