A discourse coherence model for analyzing Chinese students' essay

Abstract

Many attempts on improving entity grid model have been made and boost the research on text coherence. However, only a few of them applied it to noisy data domain and students' essay. Thus, we proposed a novel discourse coherence model, which is based on entity grid model, to evaluate coherence on Chinese students' essay. In allusion to the feature of Chinese students' essay, frequently using lexical repetition and coreference methods in coherence essay, we designed a coreference module, instead of clustering algorithm or knowledge base search methods, to integrate with the enhanced coherence model. In addition, we fully merged coreference feature into similarity assessment of adjacent sentences, and the semantic coherence. Experiments show that our model outperforms entity-based model and LSA methods and has an ideal effect on students' essay automatic assessment.