A Research On Semantic Misuse Detection Of English Verbs

English as the world's most widely used language has been widely studied and applied.In the writing of the English as second language(ESL),the misuse of verbs is one of the most frequent errors in all grammatical mistakes.The misuse of verbs is divided into two kinds:from the perspective of grammar,the error mainly occurs in the subject-verb agreement,tense and spelling;on the other hand,from the perspective of semantics,the ESL don't understand the verb environment leading to semantic errors and confusion.However,it is so difficult to define the semantic errors of verbs directly that there is no research related to verb semantic misuse.Compared with the grammatical errors that are relatively fixed in the wrong form,there is no existing rules to dealt with the semantic misuse of the verb directly.Therefore,in order to solve the semantic misuse of verbs,this paper mainly does the following work.Based on these sentences obtained from the Lang-8 website,this paper extracts a collection of verbs which are easy to be misused,and then we design several methods to correct the semantic errors of English verbs for ESL learners.First,we use the traditional machine learning method which introduces a strong classifier SVM,and we train a multi-classifier for those verb(because each verb may have a variety of misuse form)through the feature selection.Then,we propose a sequence to sequence Attention GRU(SSAG)model.After the experimental verification,we find that the effect of the model is not stable for the verbs which have multiple misuse forms.One of the main reasons is that appropriate classification surface cannot be chosen since the feature space is too sparse.The SSAG algorithm with word embedding and Attention model is stable and the model accuracy is 87.84%.Finally,we combine SVM model with SSAG model.Through the evaluation on the Giga dataset,the above method shows satisfactory results,and the final accuracy of the model is 92.52%.