Machine Translation of Movie Subtitles: A Case Study of Genres
Phonlaphat Na Pompech, Wasan Na Chai, Peerachet Porkaew, Nattapol Kritsuthikul · 2024
This study explores the effectiveness of genre-specific machine translation (MT) in translating English movie subtitles into Thai, with a focus on the romance and horror genres. Standard MT models often fall short in capturing the unique nuances and context-specific language of different genres, potentially affecting translation quality. To address this, we collected parallel subtitle data for romance and horror films from the OpenSubtitles 2018 dataset. We systematically compared three MT approaches on the two genres. By fine-tuning the pretrained model on genres-specific dataset, we obtain +1.7 and +1.1 BLEU improvements on Romance and Horror, respectively. The results demonstrate that genre-specific models yield higher accuracy when applied to genre-matched content, underscoring the value of tailored MT models for enhancing subtitle translations.