베트남 한국어 학습자의 발화 오류와 말하기 자동채점 평가에 대한 탐색적 분석 = An Exploratory Study on Speech Errors and Automated Speaking Evaluation of Vietnamese Korean Language Learners

다국어 초록 (Multilingual Abstract)

Purpose: This study aimed to investigate the impact of structural differences between Vietnamese-speaking Korean learners' languages (e.g., tone-based prosody, absence of tense consonants, and differences in diphthong systems) on the fairness and validity of automated Korean speaking scoring through theory-driven exploratory analysis, and to suggest measures to improve scoring criteria and systems. Methods: A five-step analytical framework (① comparison of Korean and Vietnamese phonological structures, ② error typology, ③ diagnosis of conflicts with existing speaking scoring criteria, ④ analysis of factors hindering the reliability of the automated scoring pipeline, ⑤ suggestion of speech data collection, correction weights, and auxiliary rubrics) was applied to comprehensively analyze literature, policy, and technical reports. Results: (1) Differences in tone-based prosody in Vietnamese and hierarchical intonation in Korean predict systematic errors such as tense consonant underrepresentation, affricate substitution, confusion of front rounded vowels, and diphthong reduction. (2) The current holistic scoring criteria uniformly penalize language-specific errors, which risks underestimating proficiency. (3) The automated scoring pipeline is highly likely to misrecognize or miss certain errors due to data bias and undermodeling of prosody-segment interactions. Conclusion: We introduce language-specific auxiliary rubrics and structure-based correction weights, and present an implementation roadmap for app-based natural speech data accumulation, model refinement, and validation. This leads to a scoring design that restores fairness in assessment beyond technical accuracy.