Application Of The Centering Local Coherence Model In Second Language Writing Evaluation

Coherence is a subjective measure in second language writing evaluation. Though various coherence theories in linguistics have proposed several quantitative ways of analyzing and measuring coherence in texts, effective and computable measures are yet to be explored, which makes objective essay scoring difficult. Therefore, textual coherence remains to be a linguistic feature which is largely neglected in current automated essay scoring systems.Centering theory might provide a solution to the problems mentioned above. The theory is concerned with relationships among focus of attention, choice of referring expression, and perceived coherence of utterances within a discourse segment and, presents a framework of centering intended to model the local component of attentional state. Since the theory was proposed in early 1980s, it has been widely used in anaphora resolution and text processing. In recent years, researchers abroad have begun to do research on its application in textual coherence evaluation and text generation by exploring its constraints and rules, especially the centering transition model stipulated in Rule 2.However, apart from the original centering transition model, there are two other centering models of coherence proposed by Poesio et al. (2004) and Kibble (2001). Poesio et al. (2004) argues that the possible number of centering transitions in texts can be seven rather than four, while Kibble (2001) replaces rule 2 with four rules, including continuity, salience, cheapness and cohesion. As to the applicability of the three models in coherence evaluation, there is not yet adequate research on what measure the three models can offer for second language writing assessment or automated essay scoring. Moreover, though the local centering coherence model might be useful in evaluating second language writing quality, its relative effectiveness compared with other language features, such as lexical and syntactic features, is still unknown.To solve these problems, we carried out an empirical study on the applicability of the three centering local coherence models in coherence evaluation and their relative effectiveness in second language writing evaluation compared with lexical and syntactic measures. The study is done on the basis of CEM Corpus for English Majors through statistical analysis with the help of the statistics software SPSS 16.0. Statistical analysis shows that Kibble's (2001) model is a more comprehensive and effective one than the other two, especially the continuity rule, which, according to both correlation and regression analyses, can be an effective measure of local coherence of student essays and a predictory factor in essay scoring. Though, compared with the other two predictory factors of the student essays'quality, i.e. language accuracy and syntactic complexity, the continuity rule is relatively weak, it can be a rather objective coherence measure for second language writing evaluation, and an effective measure for improving the accuracy of automated essay scoring.