Statistical Machine Translation Approach Based on Dependency-to-String Model
Xinlian Hu · 2016
This article establishes a translation model from source language dependency treelet to target language string. This model performs syntactic analysis on source language terminal so it belongs to tree-to-string model. In phrase of training, source language sentence is performed dependency relationship analysis to obtain source language dependency tree. Then, according to bilingual alignment information, generalization method is adopted to summarize the learnt vocabulary template, to extract source language dependency treelet to target language string and to compute correlation In decoding stage, for the input source language dependency analysis tree, dependency treelet is matched from bottom-up block by block for integration and pasting, to generate the final translation. The experiment results show that the improved approach can effectively integrate grammar knowledge and dependency tree information to Chinese- English statistical machine translation. It has effective translation performance and smaller rule-table scale with fast decoding speed and strong scalability.