Research On The Automated Grammatical Error Correction Model Of English Composition

The rapid growth of society has caused the explosive growth of information.In the process of information exchange,language is an extremely important carrier.Among all kinds of languages,English is one of the most commonly used languages and plays an important role in our daily life.Therefore,the significance of English education cannot be ignored.With the large-scale coverage of the Internet,English teaching has not relied on teachers’ blackboard writing,and English exam has gradually based on computer.With the help of technical methods in natural language processing,designing an automatic correction model for grammar error in English essays can greatly reduce the workload of teachers,assist teachers to improve the quality of teaching.As far as students are concerned,they can also get checking result in time that improving their efficiency of English learning.In addition,compared with manual correction of essays,the correction model gets rid of some subjective factors.The evaluation criteria are completely unified,so the correction results have a high degree of objectivity.This research takes English essays such as English compositions written by Chinese English learners as the study object,designs an automatic grammatical errors correction model for English essays.This model can automatically correct most of the grammatical errors in the students’ compositions,and at the same time can give more fluent expressions,which can more objectively reflect the students’ writing level.The main contents of this research are as follows:1.An automatic grammatical error correction model to reference long-distance context information is studied and designed.The model is based on the encoder-decoder structure and adopts a dual encoder structure.The transformer encoder is used to extract the context information of the sentence.Bidirectional gated recurrent unit encoder extracts the information of the source sentence,and integrates the input information through the gating structure into the decoder.Each part is matched with different attention mechanisms to achieve adaptation,which improves the model’s ability to extract relevant features in the sentence.The experimental results show that the precision of grammatical error correction in this model is up to 81.08%.2.A data augmentation method based on fluency is proposed.Combined with the model in this paper,it can generate grammatical errors imitating wrong sentences written by Chinese English learners,and merge with fluent sentences to form sentence pairs.By adjusting the fluency ratio between the target sentence and the source sentence,sentences with different fluency levels can be selected,which can better adapt to the performance of different error correction systems.At the same time,compared with the error sentences generated by statistic model,the sentences generated by our model contain more grammatical errors frequently made by Chinese students.3.A dynamic beam search method for decoding is studied and designed.This method based on beam search method and combined with nuclear sampling technology to achieve dynamic extraction for decoder output,and add a penalty factor to reduce the probability of repeated words,and suppress the model to generate shorter sentences.