Grammar error diagnosis using graph convolutional networks with knowledge graph integration

Area of Science:

• Natural Language Processing
• Computational Linguistics
• Artificial Intelligence

Background:

• Automated grammar error diagnosis is complex due to intricate syntax and semantics.
• Existing methods struggle with nuanced grammatical structures.

Purpose of the Study:

• To propose a novel framework for enhanced English grammar error detection and correction.
• To integrate Graph Convolutional Networks (GCNs) with domain-specific knowledge graphs.

Main Methods:

• Construct sentence-level dependency graphs to model syntactic relationships.
• Utilize a multi-layered grammar knowledge graph for concepts, errors, and corrections.
• Employ multi-layer graph convolutions and attention mechanisms for contextual information and diagnostic relevance.

Main Results:

• Achieved F1-scores of 0.6484 (CoNLL-2014), 0.6719 (JFLEG), and 0.6367 (BEA-2019).
• Outperformed BERT+BiLSTM by 8.8% on CoNLL-2014 and GECToR by 4.4%.
• Demonstrated significant effectiveness in identifying syntactic errors like verb tense and subject-verb agreement issues.

Conclusions:

• The framework enhances grammar error detection and correction by combining data-driven learning with explicit grammatical knowledge.
• This approach advances graph-based NLP and offers potential for pedagogical tools in language education.