Automated Error Detection for Developing Grammar Proficiency of ESL Learners

Thanks to natural language processing technologies, computer programs are actively being used not only for holistic scoring, but also for formative evaluation of writing. CyWrite is one such program that is under development. The program is built upon Second Language Acquisition theories and aims to assist ESL learners in higher education by providing them with effective formative feedback to facilitate autonomous learning and improvement of their writing skills. In this study, we focus on CyWrite's capacity to detect grammatical errors in student writing. We specifically report on (1) computational and pedagogical approaches to the development of the tool in terms of students' grammatical accuracy, and (2) the performance of our grammatical analyzer. We evaluated the performance of CyWrite on a corpus of essays written by ESL undergraduate students with regards to four types of grammatical errors: quantifiers, subject-verb agreement, articles, and run-on sentences. We compared CyWrite's performance at detecting these errors to the performance of a well-known commercially available AWE tool, Criterion. Our findings demonstrated better performance metrics of our tool as compared to Criterion, and a deeper analysis of false positives and false negatives shed light on how CyWrite's performance can be improved.