Development of an instructional model for Korean translation in multilingual classroom contexts

Abstract

Multilingual classroom contexts pose significant pedagogical challenges, particularly when students have diverse native languages and varying levels of Korean proficiency. This study introduces a data-driven instructional model for Korean translation education that employs machine learning to address learner diversity. The model evaluates translation outputs, identifies learner-specific error patterns, and personalises instruction based on three key variables: native language influence, historical translation accuracy, and individual learning progression. A dataset of Korean translation tasks was collected from university students representing six L1 backgrounds – Chinese, Vietnamese, Arabic, Russian, Japanese, and Spanish. Texts were pre-processed through tokenisation, lemmatisation, and POS tagging, with Word2Vec embeddings used for feature extraction. The proposed Sparrow Search Optimiser Tuned Attention-based Sequence-to-Sequence (SSO-Attn-Seq2Seq) model demonstrated substantial improvements, achieving 88–91% across accuracy, precision, recall, and F1-score. Results highlight its adaptability in handling idiomatic expressions and syntactic variation, providing a scalable solution for multilingual Korean language education.