Implementation of Deep Learning Model-based Korean Sentence Syntactic Complexity Assessment Model

Abstract

This study developed a method to assess the text level automatically regarding syntactic complexity. The new method was developed by improving the method of measuring the syntactic complexity of large-scale texts with various types. We implemented a Korean sentence syntactic complexity assessment model based on the deep learning models, especially the Korean BERT models. In particular, the KcBERT-based model, fine-tuned through the “National Institute of Korean Language Dependency-Parsed Corpus (v.2.0)”, showed excellent performance with an accuracy of 0.949. This model is expected to contribute to establishing an integrated model to assess the text level as the sub-factor model. By segmenting the text assessment model by factors, it could overcome the limitations of the existing research using unexplainable deep learning models to provide a direction for more sophisticated educational treatment.