Combining GCN and Transformer for Chinese Grammatical Error Detection

<p>This paper describes our system at a task: Chinese Grammatical Error Diagnosis (CGED). The task is held by the Natural Language Processing Techniques for Educational Applications (NLP-TEA) to encourage development of automatic grammatical error diagnosis in learning since 2014. goal CGED diagnose four types errors: word selection (S), redundant words (R), missing (M), and disordered (W). contains two parts including detection correction designed solve problem. Our built on three models: 1) BERT-based model leveraging syntactic information; 2) contextual embeddings; 3) lexicon-based graph neural network lexical information. We also design an ensemble mechanism improve single model’s performance. Finally, achieves highest F1 scores level identification among all teams participating 2020 task.</p> <p> </p>