한국인 학습자 영어 말하기 자동 평가를 위한 문법 다양성 척도 연구

초록

This study investigates various measures of grammatical diversity for automated speaking assessment (ASA) with Korean learners of English. Data were extracted from the Korean monologue set in the International Corpus of Asian Learners of English and classified into two proficiency groups. Using two NLP toolkits, i.e., the Biber tagger and the argument structure construction annotator, we measured the grammatical diversity of the speaking data based on six features over three levels (i.e., word, phrase, and clause), and conducted correlation and binomial logistic regression analyses. The results reveal significant correlations among the features, particularly between those related to part-of-speech and the others. It is also found that only the clause-level feature of the subordination type frequency significantly predicts proficiency levels. These findings provide insights into the potential of grammatical diversity as a valuable metric in ASA systems for Korean learners of English.

키워드