AN E-TUTOR TOOL FOR GRAMMATICAL ERROR CORRECTION

Abstract

Intelligent language learning tools have evolved into an inevitable aid for ESL/EFL (English as second or foreign language) learners to improve their linguistic skills. By functioning as a machine translation task, automatic grammatical error correction (GEC) has made significant progress with the help of deep neural networks, but its accuracy and coverage rates on different error types have not been fully satisfactory in practice. This paper designs an e-Tutor tool for GEC to help ESL/EFL learners automatically inspect and correct their grammatical errors in writing. One of the core research tasks in GEC is how to improve its generalizability while dealing with more complex error types. In the paper, we propose a novel data augmentation method to add artificial noise to native English corpora during training a neural translation model for GEC. To improve the output quality of GEC, we also design a re-editing module, which mainly consists of a statistical language model, along with a grammatical error detection classifier, in validating each sentence generated by GEC. Experiment results on GEC show that our e-Tutor tool can achieve state-of-the-art performance on the CoNLL-2014 dataset.