Comparing the performance of AI translators in handling rhetorical devices in French-to-Korean speech.

This study aims to compare the performance of various AI-based translators in handling rhetorical devices deployed to enhance the persuasive effect of speeches. Different presidents employ their preferred rhetorical techniques, among whom French President Emanuel Macron is known for frequent use of repetition. For example, in his 2024 New Year's address, Macron utilized several rhetorical devices (anaphora, antithesis, anadiplosis, amplification), including repetition at the sentence structure level. Since translation should take into account the text's type and function, it is crucial that these rhetorical elements, which amplify the speech’s persuasive function, are accurately conveyed, even in machine translations. In this context, this study analyzes Macron's 2024 New Year's speech, focusing on phrases employing sentence-level rhetoric. These phrases were translated from French to Korean using three artificial neural network-based translators (Google, DeepL, and Papago) and a generative AI (ChatGPT) to assess their ability to handle rhetorical devices. The analysis revealed that the machine translators varied in their effectiveness at processing rhetorical elements, with ChatGPT performing the best, followed by DeepL. ChatGPT produced fewer semantic errors, even when omitting certain words or expressions, and generally handled rhetorical translations more effectively. However, most translators struggled with longer sentences containing multiple rhetorical devices, such as omissions.