Research And Implementation Of Grammatical Error Correction

In the context of globalization, the growing popularity and development of English has increased the number of ESL (English as Second Language) learners steadily every year. During listening,speaking, reading and writing four basic English skills, Writing is considered to be one of the most difficult skills for ESL learners. At the same time, due to the lack of grammar knowledge and the influence of mother tongue, Grammatical errors become one of the most common errors in writing. Grammatical Error Correction(GEC) means use computer to correct grammatical errors automatically.Aiming at common grammatical errors, we propose a n-gram voting strategy to correct grammatical errors. Aiming at article and preposition errors, a two layers GEC method based on ESL and news corpus is proposed, due to the limited form and closed Confusion Set.Aiming at prepositon errors, a GEC method based on the feedforward network is proposed, due to the general F1 score based on two layer GEC method.Experimental results on CoNLL-2013's GEC data show that our method of article-error correction obtains a F1 score of 34.01%, higher than UIUC's F1 score (33.40%). The F1 measure of Preposition errors reaches 10.08%, exceeding UIUC's F1 measure (7.22%)?The main contributions of this paper are as follows:(1)we propose a n-gram voting strategy to correct grammatical errors.In the past time, GEC method based on n-gram always use long n-gram fragment to correct errors, without consider different length n-gram's influence. So we propose a n-gram voting strategy which Solved this problem, aiming at different length of n-gram, setting different voting weight, longer n-gram have greater weight, finally use these n-gram to vote and choose the candidate with the highest of votes as the correct result.(2) Aiming at article and preposition errors, a two layers GEC method based on ESL and news corpus is proposed.When use n-gram vote strategy to correct grammatical errors, existing a phenomenon which transfer right sentences to wrong sentences. So we propose this method to solve this problem, Firstly, we use ESL corpus to train the recognition model. Then, we use the model to judge whether the sentence has grammatical errors. Any grammatical error, if found, be corrected by using news corpus and n-gram vote strategy.(3) Expand the knowledge base and establish the efficient n-gram search engine. Expand the verb and noun form, Effectively improve the result of verb and noun errors. If sequential search the n-gram, efficiency is very low, so we build index for n-gram and design more reasonable search strategy to make the system more rapid and flexible to retrieve,indirectly improve the efficiency of the GEC system.(4) Developed an efficient GEC system.Aiming at the method proposed in this paper, design and realize an efficient GEC system. .System consists of an interface, an input text box, an output text box and five buttons, respectively corresponding to the five kinds of error correction function.