<![CDATA[The swift progress of artificial intelligence (AI), particularly in Neural Machine Translation (NMT) systems, has substantially transformed the landscape of translation practices. While AI-powered applications such as Monica AI (Powered by ChatGPT) demonstrate a high degree of linguistic accuracy, there remains a notable research gap regarding the specific translation challenges encountered by professional translators when working with AI-generated outputs, especially in speech texts. The theoretical novelty of this study lies in integrating Nord’s text-typological translation problem framework with Molina and Albir’s translation technique model to evaluate NMT output, an approach rarely applied to spoken oratory texts. Empirically, the study provides a fine-grained error analysis of a ChatGPT-powered NMT system on formal political speech, quantifying problem types and mapping them to specific post-editing strategies. Utilizing a qualitative content analysis approach, this research examines a formal English-language speech text translated using Monica AI. A total of 282 source sentences, along with their AI-generated and post-edited versions, served as the corpus. Findings reveal that 91.1% of the translated sentences were accurate, while 8.9% contained identifiable issues, predominantly within pragmatic, conventional, and text-specific domains. This study emphasizes the indispensable role of human translators in ensuring cultural and contextual appropriateness in AI-assisted translation.]]>
Copyrights © 2026
