The Investigation of Machine Learning in Grammar Correction

Abstract

As English continues to be the dominant language acquired by the majority of the global population, the demand for learning it is on the rise. Within English education, grammar plays a vital role and can now benefit from machine learning-based correctional applications. This paper explores various methods employed in prior research for grammar correction. Notably, the feature extraction method proves to be effective in capturing essential information from the text, resulting in more precise revisions of grammatical errors. The feedback filtering module will select valid improvement advice from users to develop more efficient application. Recurrent Neural Network (RNN) which is a widely-used model can also be adopted to grammar correction due to its memorizing ability. In previous studies, these methods are tested to see their validity in English grammar correction. Results of feedback filtering module show that it can sort out users advice into useful and useless so that the modification of the application can be more accurate. In another experiment, the F-0.5 score of RNN is measured with several other models and RNN has apparent advantage over the majority in grammar error detection and correction. Admittedly, however, these methods still have space for further enhancement to provide high precision in correction. Means to eliminate possible errors and inaccuracy are urged to be found, but probably the only way out is the innumerable data fed to computers. This paper offers a comprehensive view of current study progress in the field and encourage new evolution.