Abstract
Against the problems which can’t be solved by the word-level based local coherence analysis model, we propose a new discourse coherence quality analysis model (abbreviated RST-DCQA) by analyzing the full hierarchical discourse structure of English essays. Under the framework of rhetorical structure theory (RST), firstly, we design an RST-style discourse relations parser to capture the deep hierarchical discourse structure of essays; secondly, we transform the discourse relation information into a discourse relation matrix; finally, we design an algorithm to analyze the discourse coherence quality of student’s English essays. The experimental results show that the average error of our model’s score and teacher’s score is only 2.63, and the Pearson correlation coefficient is 0.71. Compared with the other models, our RST-DCQA model has a higher accuracy and better practicality in the field of students’ essays assessment.
Highlights
In recent years, due to the continuous development of research in artificial intelligence and its subdivided fields, many machine output results will be provided to people
Among the researchers’ studies, the most typical ones are the latent semantic analysis (LSA) method proposed by Foltz et al [1] and the entity grid model proposed by Barzilay & Lapata[2,3]
In the following part of this section: firstly, we introduce the discourse relation tree which is generated by the parser; secondly, we describe the discourse relation matrix; we analyze the discourse coherence quality of essays by the rhetorical structure theory (RST)-DCQA model
Summary
Due to the continuous development of research in artificial intelligence and its subdivided fields, many machine output results will be provided to people. If the existing entity grid model is used to coherence assessment, the result will be incoherent. Aiming at this problem, Lin et al [7] use an end-to-end Penn Discourse Treebank style (PDTB-style) discourse method to encode the discourse relations of text. PDTB-style method only encodes very shallow discourse structures, the relations are mostly locality and adjacency, within a single sentence or between two adjacent sentences. They can’t find the discourse relations existing in higher levels.
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Free) 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a call
