Analysis And Application Of Grammatical Error Correction Model For English Learner

To improve the learning efficiency of English learner and to help them to figure out the impact brought by their mother tongue,a research on the high-frequency grammatical error made by the learners is needed by iWen,an app developed by ADAPT lab of Shanghai Jiaotong University.During the research,two problems were raised:How to get the fresh text data made by English learners?How to correct automatically the errors in the text?As to the data collection,we automate the process of searching online English learners,collecting their Tweets and deducting their mother tongue,we use these data to create a dataset named ESL-TWEET so as to get the knowledge on the real errors made by English learners in their daily usage.As to Grammatical Error Correction(GEC),we start with the stateof-the-art models,and break down the GEC task into two tasks:give the best edit that should be made on certain words,find the best word to replace the origin word or to insert between two words.We propose a new GEC model,GET-MF,which keeps a competitive performance while improving the capacity of detecting errors like word collocation and making the model more adaptive to situations where the best wording is different according to the subjects.Finally,we apply GET-MF on ESL-TWEET,we analysis the common grammatical errors made by English learners of different mother tongue,and show the effect of one’s mother tongue on English learning.