Speaker
Description
In recent years, ChatGPT used for translation tasks has been increasingly popular although it still has some limits when translating Vietnamese sources, especially in genres requiring cultural sensitivity, contextual awareness and pragmatic nuance. Therefore, the aim of the study is to investigate the underlying causes of ChatGPT’s deficiencies in Vietnamese English translation, focusing on linguistic and computational perspectives. Employing a qualitative error analysis and comparative evaluation approach between ChatGPT translations and human-generated reference translations across literary, academic and conversational texts grounded in key dimensions such as lexical accuracy, syntactic fidelity, idiomaticity, stylistic consistency, cultural reference handling, and contextual meaning interpretation, the research indicates that errors including literal translation, unnatural phrasing, misinterpretation of pronouns, and loss of culturally embedded meanings recured in ChatGPT-based translation texts. These shortcomings were attributable to the linguistic complexity of Vietnamese such as its pro-drop nature and context-dependent structures and computational constraints including insufficient high-quality training data and English-centric model bias. It also reveals that there were disparities in translation quality for different genres and the necessity of human oversight were emphasized when ChatGPT was applied for tasks involving high-context language. Moreover, the findings will contribute to the advancement of machine translation research and offer insights for enhancing AI performance in low-resource language contexts.
Keywords: ChatGPT, Vietnamese-English translation, linguistic complexity, computational limitations
