Lost in Punslation:A comparative study of AI and Google Translate in handling Humor

This paper attempts to explore and compare the challenges of translating puns by each of AI and Google Translate. Puns are considered to be a type of humor, and humor basically relies on culture; therefore, translating it is challenging and needs a more comprehensive background, as sometimes it is not translatable at all. It is hypothesized that using AI to translate puns is better than Google Translate, as AI language models have enormous data and can generate more fluent and natural-sounding translations. Besides, the ability to learn makes AI models perform better in translating puns compared to Google Translate. The data of this study were derived from the first season of the Friends sitcom, whereby 240 pun samples were collected and compared. The samples were first categorized into four types, following Delabastita's (1996) categorization. Second, they were translated from English into Kurdish by AI once and by Google Translate another time. Third, the translation strategies are denoted, following Delabastita's (1996) eight strategies for translating puns. The two versions of the translations were compared to identify the similarities and differences, frequencies, and challenges of translating puns through AI and machine translation. Only central Kurdish is considered to be tested for the TL. For accuracy, the study used the Deep Seek-v3, which is more updated, and its Kurdish version is believed to be more accurate than the other models. The study results show that meaning and humor are often lost when translating puns ‘punslation’ in both cases. AI models and machine translation both face challenges in dealing with puns when translating from English into Kurdish.