Multi Head Attention Based Confusion Module for English Grammer Error Correction Model

The Grammar Error Correction (GEC) is main task in Natural Language Processing (NLP) that is currently attracted high attention among researchers. Referred to satisfactory outcomes of universal language in GEC task and research lack on GEC task in less resource languages. The research proposed a Multi Head Attention based Confusion Module (MHA based CM) method for grammar error correction model. The proposed MHA method has copy mechanism for addressing unknown and particular words which appeared in sentences of sources. Additionally, to over come the data sparsity in sentences here proposed semi supervised confusion module method and that trained the data from monolingual corpus. The dataset used for evaluation in the research are synthesized data. The performance of proposed method is evaluated in terms of precision, recall, fl-score and BELU score. The proposed MHA based CM method attained highest precision 84.32%, recall 85.67%, fl-score 85.51 % and BLEU 66.14% which is superior than other neural networks like Recurrent Neural Network (RNN) and Convolutional Neural Network (CNN).