Research And Application Of The Dependency Grammar And Valence Grammar In The Real-word Errors Correction

Nowadays, Natural Language Processing technology has been widely used in various fields. It provides important theoretical basis and implementation methods in order to realize the communication between the human and the computer. As is known to all, most of the information is expressed in the form of text. Therefore, doing processing for the text information is the key of Natural Language Processing Technology. In English, English words make up the basic unit of an article. If the word spells wrong, it will make an impact on the subsequent analysis of the article and the actual application effect of the system. At present, domestic and foreign researches on spelling correction continue. The types of studying errors can be divided into two categories roughly:non-word errors and real-word errors. Researches of non-word errors have already mature now, but real-word errors correction is relatively difficult. Other researchers have tried to use Bias classification algorithm and some rule methods to realize, but the effect is not very satisfactory. For this reason, the author introduces a new method to improve the real-word errors correction effect based on the existing Winnow statistical algorithm.Firstly, the author investigates and analyzes the related research of real-word errors correction at home and abroad and the application of dependency grammar and valence grammar in Natural Language Processing. And summarizes the advantages and disadvantage of a variety of existing methods based on it. At the same time, the author also describes the theory and technology related to this paper. After that, the author is inspired from the theory and application of dependency grammar and valence grammar which help me get the method of collocations of confusable words and prepositions. This method is mainly originated from the idea of "association" concept in dependency grammar and concept of "valence" in valence grammar. In this method, the author needs generate a preposition vector for each confusable words. At the time of testing, whether the confusable words appear correctly can be determined by the difference of the prepositions and other features. Then, the author realizes the overall architecture and every functional module of the real-word errors correction system by combining the method of collocations of confusable words and prepositions based on the Winnow algorithm to achieve checking and correcting of the English real-word errors. Finally, the author tests the whole system and contrasts with the result of the experiment which uses the Winnow algorithm completely to confirm the superiority of the new algorithm.After the system introducts the method of collocations of confusable words and prepositions, not only the errors correction performance of those confusable words whose original correct rate, recall rate, F1measures and other indicators are low is improved, but also the correct rate, recall rate and F1measures of the whole system are raised by3%,2%,3%compared to the result of the experiment which uses the Winnow algorithm completely. This shows that the author’s method is effective, and lays the foundation for further study of English real-word errors correction to other researchers.