Research On Text Generation Technology Oriented To Template Based Machine Translation

For all these years, machine translation has always been one of the most important and challenging task in the area of nature language processing. Based on many years of research between Northeastern University and Fuji Xerox company, this paper discuss and implement the template-based machine translation method. The paper focus on the key technologies needs to be solved during text generation process in template-based machine translation.Best template selection and phrase translation are two key factors which influence the performance of machine translation. By applying sentence-template direct matching and best template selection methods, the paper improves the optimum template selection. At the same time, the paper improves the performance of phrase translation through the introduction of word sense disambiguation method and articles selection method, and this improves the translation performance on the whole.The main work of the paper:(1)For the selection of candidate templates, this paper proposes direct matching method between text and template. Usually the traditional method extract template from sentence to be translated with the aid of parser firstly, then calculate similarity between the extracted template and the templates in the library. But so far the performance of Parser can’t meet the requirements of actual demand, the result of extraction is unsatisfactory, and this will influence the following calculation of similarity. This paper transform the sentence directly according to the candidate template, then calculates the similarity between them with edit-distance method, finally the system get the mapping relations through backtracking algorithm.(2) In template-based translation, the quality of phrase translation will affect the final translation directly. This paper set up a separate phrase translation module, and make the phrase template based method as one of the sub-module of phrase translation mudule. WSD(Word sense disambiguation) problem is an urgent problem needs to be solved in phrase translation, this paper tries a variety of WSD method and attempt to merge these method to phrase translation engine. Also the paper discusses the problem of article choices during the English text generation.The experiment results show that the direct matching method based can achieve satisfactory performance, and the introduced WSD method and the article choices method also improve the quality of translation.