Computer technology has developed at a rapid pace in recent decades. The combination of artificial intelligence and writing teaching has become an important development trend. For better utilization of Automated Writing Evaluation System in teaching, the quality of its feedback on various types of errors in compositions is one of the issues worth exploring. In order to investigate this issue, this study compares the diagnosis of EFL learners' compositions by Automated Writing Evaluation System with the actual errors in the compositions, and calculates the Precision and Recall of the iWrite system's feedback. Unlike previous studies, this study synthesizes macro discourse elements such as propositional logic, across-sentence logic, and discourse structure, and categorizes the actual errors in learners' compositions into seven kinds. The results show that the iWrite system performs excellent in detecting three kinds of errors, which are lexical errors, grammatical errors, and other detail-type errors, and is moderately effective in detecting pragmatic errors, while it fails to provide feedback on propositional logic, discourse cohesion and coherence, and paragraph structure, which are the high-level error types, indicating that iWrite system is completely unable to capture errors in terms of macro-discourse elements. These findings provide a strong basis for improving Automated Writing Evaluation System, and at the same time provide English teachers with a reference for writing teaching, which helps to better serve the field of foreign language education.
Read full abstract