Opening the Black Box: Interpretable Machine Learning Reveals the Relationship Between Lexical Diversity and Writing Quality

Machine learning serves as the core component of artificial intelligence (AI), however, there is scant interdisciplinary research between AI and applied linguistics. Very few studies touched upon the aspect of how to leverage interpretable machine learning to solve the problem in language learning, teaching, and assessment. This study aims to lift the veil and fill this gap by employing interpretable machine-learning approaches to explore the relationship between lexical diversity and writing quality. The findings contributed to providing a finer-grained understanding of the CET-4 writing construct and its rating scale. Furthermore, it may lend empirical evidence for the Dynamic Systematic Theory (DST) in language acquisition, verify the validity of automatic essay scoring, and advance teachers' language assessment literacy in the long run.