Multi-Perspective Coherence Assessment Method for English Texts

Coherence is an important component of text quality assessment and plays an indispensable role in many natural language processing tasks, such as automated essay scoring, text translation, and generative text quality assessment. Most existing coherence evaluation methods focus on local coherence between adjacent sentences or on discourse coherence based on the document, lacking a comprehensive evaluation from multiple perspectives. To address these issues, a multi-perspective coherence assessment model for English texts (MPCA) has been proposed. The MPCA model includes a text semantic extraction module based on attention mechanisms, a text local coherence extraction module based on Bi-linear layers, and a text discourse coherence extraction module based on lightweight convolution (LCNN). It comprehensively evaluates the coherence of English texts from three perspectives: semantic consistency, local coherence, and discourse coherence. Experiments have shown that compared to baseline models, the MPCA model achieves optimal performance in terms of QWK values for coherence feature scores on eight subsets of the ASAP++ public English essay datasets, and reaches the best overall average QWK value of 72.15 %, demonstrating the effectiveness of the MPCA model in the task of English text coherence assessment.