In Search of Grammar Structures in Authentic Language Materials: Enhancing Semgrex for Dependency Trees

This study explores the potential of Natural Language Processing in the realm of second language education, particularly focusing on the effective use of authentic language materials – real-life linguistic resources. Despite the numerous advantages of these materials, they come with inherent challenges like the inadequate complexity level for the learner’s proficiency and the lack of structure and progression, which affect the efficiency of learning. To address these, we propose a novel approach to identify Grammar Learning Objectives (GLOs) in these materials and extract relevant fragments. We employ several methods based on semgrex pattern matching on a dependency tree and evaluate them against a set of basic metrics. Our experiments indicate that the best outcomes can be achieved through deep learning methods; however, the limitations of the study (small evaluation dataset and relatively basic nature of evaluation metrics) might affect the findings. Thus, we suggest that further research should focus on incorporating more diverse data and more advanced measurement tools.