| Latent Semantic Analysis (LSA) has been used to develop an operational Automated Essay Scoring (AES) system - the Intelligent Essay Assessor (IEA), finding that its scores typically correlate as well with human raters as the raters do with each other, when IEA automatically scores English essays written by speakers of English in the USA. By contrast, we manage to apply the LSA approach in this study to develop on our own an LSA-based EFL Essay Scorer (LEES) to validate scoring College English Test Band 4 and Band 6 (CET-4/6) essays by non-native speakers of English in Chinese EFL context, only to find that LSA's performance (with human-computer correlation r = 0.154) is not so good as that reported in the previous studies cited above. The correlation between LEES' scores and CET-4/6 human readers' is positive but low and small. We fail to provide strong evidence in this study that LSA is appropriate to be applied to automated scoring of CET-4/6 essays in Chinese EFL context. We draw a conclusion that LSA's application in Chinese EFL context is only to a limited extent at the moment, based on the results which are made manifest in this study. We then discuss the possible reasons for this failure to match the results, such as: LSA focuses solely on content assessment, totally blind to syntax, grammar, and other relevant factors, for instance, Chinese EFL learners use a limited vocabulary for guided writing in the timed writing test; human raters are grammar-oriented while grading the essays, and potential drawbacks of the LEES system. Finally, we describe the limitations to the current study, mention the applications in LSA research and AES studies, and sketch the implications for future plans for developing an ideal AES system in Chinese EFL context. (292 words)...
|
PDF Full Text Request