Nghiên cứu đánh giá độ chính xác bản dịch của công cụ dịch Google Translate và Naver Papago đối với văn bản báo chí

The rapid advancement of Artificial Intelligence (AI) has positioned machine translation (MT) tools, such as Google Translate and Naver Papago, as indispensable resources for accessing multilingual information. While these tools offer significant utility, particularly in foreign language education, their accuracy and naturalness for specialized or culturally nuanced texts remain a critical challenge. Existing research has explored MT performance across various language pairs and text types, but a comprehensive assessment for Korean-Vietnamese news articles, especially those rich in cultural and social context, is notably limited. This study addresses this gap by systematically evaluating the translation accuracy of Google Translate and Naver Papago for Korean news articles rendered into Vietnamese, aiming to identify their specific advantages and limitations, and offering practical recommendations for their effective deployment in academic and professional settings.

Researchers in machine translation, educators and students of Korean-Vietnamese translation, professional translators, and developers of MT tools.

This study conducted a comparative analysis of the translation accuracy between Google Translate (GT) and Naver Papago (NP) for Korean news articles translated into Vietnamese, focusing on cultural and social topics. The methodology involved translating 10 news articles (5 cultural, 5 social) sourced from 'Koreana' magazine using both machine translation tools. The output was then rigorously evaluated against a human-made reference translation, employing a 0-4 point scale to assess the fidelity of content delivery, grammatical correctness, and overall semantic coherence at the sentence level. The findings reveal that Google Translate achieved a significantly higher average accuracy of 90.69% compared to Naver Papago's 80.81%. GT demonstrated greater consistency across different articles and superior capability in handling complex sentences, specialized vocabulary, and minimizing grammatical and semantic errors. However, GT sometimes lacked naturalness in its output. Conversely, Naver Papago performed adequately for simple and shorter sentences but exhibited considerable limitations with complex linguistic structures, cultural nuances, and specialized terminology, often resulting in more grammatical and semantic inaccuracies, including omissions or additions of unnecessary information. Based on these findings, the research proposes a structured workflow for optimizing machine translation tool usage, encompassing tool selection, initial translation, thorough review, editing, and finalization. Furthermore, it suggests concrete improvements for MT developers, such as integrating advanced text classification, expanding training datasets with culturally specific Vietnamese-Korean content, and enhancing language models. The implications of this research are substantial for researchers in machine translation, educators and students in Korean-Vietnamese translation programs, and professional translators seeking to leverage these tools more effectively, contributing to a more nuanced understanding of MT capabilities and limitations in specific linguistic and textual contexts.