Exploring the Applicability of Deep Learning Models in Korean Language Education: Focusing on a BERT-based Predication Model for Korean NP Complement Clauses

This study developed a KR-BERT-based prediction model for Korean NP complement clauses, addressing three research questions. First, the model successfully learned both short and long NP complement clause types. Second, it demonstrated strong classification performance, with metrics including 0.9565 accuracy, 0.9216 precision, 0.9451 recall, and 0.9333 F1-score, though results varied depending on corpus sources. Third, attention analysis revealed a tendency to focus on head nouns over embedded predicates, suggesting the model’s dependence on noun modifiers for accurate classification. Despite limitations in data variety and component interaction, this research highlights the model’s potential for supporting Korean language education and grammar error correction.