Error Analysis and Instructional Strategy Adjustment in a Corpus of English Language Learners

Students’ linguistic errors are an inevitable phenomenon in the process of teaching English. And the role played by the correction of errors in English learning has not yet been finalized. In this paper, on the basis of recurrent neural network, we overcame the problem of long distance dependence of RNN by introducing three gating structures of LSTM, and constructed the Seq2Seq framework to model the text generation task through two structures of encoder and decoder. Collecting domestic and international learners’ corpus, setting up English error correction definition and evaluation standard, and establishing English grammar error correction model based on seq2seq. The attention mechanism is introduced on the model to deal with the grammar error correction problem. The n-gram grammar model is used to score the English sentences and filter the invalid suggested text. The P, R, F0.5 results were analyzed on the CoNLL-2014 test set, and the model after adding the attention mechanism was, in the index of F0.5, 25 more effective than without the attention mechanism. Meanwhile, after using the corpus for the error analysis, the students’ grammatical comprehension improved to different degrees, especially in the scoring rate of adverbs, which was increased from 86.93% to 100%, and the students basically mastered the use of adverbs.