Research On Dependency-to-String Model For Chinese To English Example-Based Machine Translation

As one of the most popular topic in natural language processing, machine translation has huge research value and wide commercial application prospect. In recent years, example-based machine translation has developed rapidly and obtained an outstanding achievement. The performance of example-based machine translation depends on the quality of example base. And the quality of example base reflected as the probability of correct retrieve of examples. Consequently, improving the quality of example base is the orientation of the common efforts of researchers and also one of the hotspots of example-based machine translation. In recent years, example-based machine translation has widely adopted structured example base. More syntactic structure and semantic information are involved to promote the reliability of examples. The acquisition of structured examples needs to preprocess raw corpora by word segmentation, POS tagging and dependency parsing. Traditional methods process the three basic tasks in turn. However, the transmission of the iterative error may affect the veracity and reliability of structured examples. Motivated by this, this paper proposes a new Chinese to English example-based machine translation and mainly studies the construction method of example base.The main contribution are identified as follows:(1) Proposition of the construction method of example base. On the one hand, fusing a joint model to integrate word segmentation, POS tagging and dependency parsing. The motivation of this study is targeting to relieve the affections of error propagation and failure of feature extraction at different levels with the joint model. The reliability has been approved under a large scale corpus. On the other hand, based on the traditional methods and combining the practical situation, this paper proposes an example generalization method.(2) Based on the Chinese to English dependency tree to string example base constructed, improving the traditional methods of similar example retrieval and translation generation.(3) Organic integration of methods of each module. The system performance has been approved under a large scale corpus.This paper integrates joint model of word segmentation, POS tagging and dependency parsing for the first time in this area. Experimental results of translation system performance prove that our method performs well and can obtain high-quality translation results. Contrast experimental results of integrating joint model prove that the integration of joint model can effectively improve the quality of translation and system performance.