A deep learning-based model for automatic syntactic complexity assessment in L2 English writing: development and...

Area of Science:

• Natural Language Processing
• Second Language Acquisition
• Computational Linguistics

Background:

• Syntactic complexity is key to L2 writing proficiency.
• Traditional assessment methods lack scalability and consistency.
• Processability theory and usage-based L2A inform current research.

Purpose of the Study:

• Propose a deep learning architecture for automatic syntactic complexity assessment in L2 English writing.
• Integrate BERT and graph attention networks for hierarchical structure analysis.
• Utilize multi-task learning for predicting multiple complexity dimensions.

Main Methods:

• Developed a deep learning model combining BERT and GAT.
• Implemented a multi-task learning framework for syntactic complexity.
• Validated the model on learner corpora using five-fold cross-validation.
• Conducted a quasi-experimental study with 186 Chinese university students.

Main Results:

• The model achieved a Pearson correlation of 0.923 with expert ratings.
• Outperformed rule-based and baseline neural network approaches.
• Automated feedback group showed significant writing development (effect sizes 0.71-0.89).

Conclusions:

• Deep learning-based assessment shows potential for L2 syntactic development.
• Automated syntactic feedback can support writing improvement in educational settings.
• Further research is needed to isolate the model's specific impact.