Subtitling Jordanian culture-bound expressions into English: ChatGPT vs. YouTube

This study aims to compare and contrast YouTube’s auto-generated captioning and ChatGPT subtitling of Jordanian culture-bound expressions into English. The data consists of 100 examples from six episodes of the Jordanian series ‘Female’. The collected data were categorized into four categories, namely vocatives (10 instances), idioms and proverbs (52 instances), Jordanian culture-bound expressions (18 instances), and offensive language and swearing (20 instances). The findings showed that culture-bound expressions were rendered using five translation strategies: literal translation (38% by YouTube and 6% by ChatGPT), paraphrase (3% by YouTube and 53% by ChatGPT), cultural substitution (1% by YouTube and 25% by ChatGPT), transliteration (13% by YouTube and 4% by ChatGPT), and omission (7% by YouTube and 8% by ChatGPT). Some incidents of mistranslations were also found (38% by YouTube and 4% by ChatGPT). The results also show that YouTube relies heavily on literal translation and omission, which frequently result in mistranslations and loss of pragmatic meaning. ChatGPT tends to employ strategies that maintain functional and contextual aspects, utilizing adaptation and pragmatic equivalence, resulting in more accurate and acceptable cultural renderings. The study concludes that human post-editing is necessary to ensure accuracy, especially when relying on auto-generated subtitles. This study recommends that future researchers examine the performance of other large language models (LLMs) and AI-powered subtitle generators in subtitling different Arabic dialects.