The Application Of Machine Learning To Chinese Grammatical Error Detection

Chinese has become an important commuicstion took and gradually recognized by people all over the world.In the process of learning Chinese,writing is considered as one of the most effective way to evaluate the learning results of the learners.However,manual correction of Chinese composition requires a lot of human support and usually takes a long time,so it can not give timely and effective feedback and analysis to language learners.Moreover,the complexity of Chinese itself increases the difficulty of manual correction.The purpose of this paper is to build a machine learning algorithm model that can detect Chinese grammartical errors,and apply the model to the Chinese composition grammar detection system,whciah can save a lot of time and labor cost of reviewing the compositions manuuly.The system can free teachers from heavy and repeated evaluation activities,and give them more time to focus more on teaching.Meanwhile,the system can give learners timely feedback and help them find out the common mistakes they make.This study combines theoretical research and practical research methods to find Chinese grammar detection methods based on machine learning.The specific work includes:(1)revealing the frequency and trend of different machine learning models in the education field these years through literature research(2)classifying and summarizing the common strategies of grammar detection through the current research(3)Building three different machine learning algorithm models,include conditional random field model,LSTM-CRF model and Multi-task learning model,to detect Chinese grammar,evaluating the models and choosing one which has best performance(4)Designing and developing The Chinese Composition Grammartical Error Detection System,and applying the algorithm model to the system.In recent years,the proportion of the use of neural network algorithm model is significantly higher than other methods in the application of artificial intelligence in education.In the task of grammar detection,researchers have gradually shifted their attention from the original statistical based method to the neural network method.Among the three algorithm models built in this paper,the performance of Multi-task learning model is better than Conditional Random Field model and LSTM-CRF model,because Multitask learning architecture which contain a supplementary objective is able to learn the bias in the label distribution without obtaining much additional information from the majority labels.It allows the model to make full use of the training data and get better results than other sequence labeling task in Grammatical error detection task.