自动后处理和语法错误校正驱动的译文质量提高方法

如何提高机器翻译系统输出译文的质量、增强译文的可读性和直接可用性，是机器翻译研究中的一个重要问题。针对目前统计机器翻译解码过程中难以引入需要进行深层次分析的模块，且额外资源不易融入等问题，本项目将以译文自动后处理和语法错误校正为切入点，在以下方面进行深入的研究和探索：（1）研究最新神经网络机器翻译和传统统计机器翻译互相提高的方法，利用源语言句子信息结合交叉熵和余弦距离方法进行译文自动后编辑，提高译文的忠实度；（2）探索译文中语法错误的校正方法，运用马尔可夫逻辑网络和递归神经网络语言模型分别校正译文中常见语法错误和其它错误，提高译文的流利度；（3）我们不仅研究英文译文的校正方法，还通过标注一定数量的汉语译文语料并进行深入错误分析，利用基于汉字的汉语译文自动后编译和语法错误校正方法来提高汉语译文的质量。最终达到减少译文人工修改和润色工作量的目的，促进机器翻译的研究和应用。

How to improve the quality of the outputs produced by machine translation system, and enhance machine translation readability and usability is an important issue in the study of machine translation. In view of the module needed to perform deeper text analysis and the additional resources are both difficult to introduce in currency machine translation decoding process, we will take the automatic post-editing and grammatical error correction as the breakthrough point, investigate and delve deeply in the following areas: (1) Study of how the new neural machine translation and the traditional statistical machine translation can interact with and benefit from each other, utilize the source language sentence combining with cross entropy and cosine distance approach for automatic post-editing to enhance the translation adequacy; (2) investigate the grammatical error correction approach for machine translation, exploit the Markov logic network and the recurrent neural network language model, respectively, to correct the common grammatical errors and others errors to improve the translation fluency; (3) Not only investigate the correction approach for English translation, but also for Chinese translation. The character-based automatic post-editing and grammatical error correction approaches are applied to Chinese translation to improve translation quality by annotating a certain number of Chinese translation corpus and analyzing grammatical error deeply. The goal of this project is to reduce the workload of manual revision and polishing, to promote the research and application of machine translation.