ASynGEC: Adaptive  Syntax-Enhanced Grammatical Error Correction

Grammatical Error Correction (GEC) aims to generate a correct sentence from an erroneous one. However, due to the scarcity of high-quality datasets, it still faces challenges in achieving a deep semantic understanding of sentences. This paper proposes an Adaptive Syntax-Enhanced Grammatical Error Correction (ASynGEC) model that incorporates adaptive dependency syntax information into the GEC framework. During the syntax processing phase, word-level syntactic information is chosen as input. This information captures rich relationships between words. We employ a Graph Convolutional Network as the syntax encoder to capture syntactic information better. Additionally, considering that erroneous sentences can negatively impact the extraction of syntactic information, we propose a Part-Of-Speech guidance module to minimize the influence of incorrect syntactic cues further. By combining these two modules, the GEC model effectively integrates syntactic information, enhancing the model's ability to comprehend the input sentence deeply. Experiments on mainstream Chinese GEC datasets demonstrate that ASynGEC significantly outperforms strong baselines, achieving competitive performance.