树转录翻译模型解码优化

Decoding Optimization in Tree Transducer based Translation Model

针对树转录翻译模型中的规则二元化和解码算法进行深入研究,通过四分化的二元化转换方法减少词汇化同步转录规则的中间项目,通过实时判断中间项目有效性的RR-CKY算法来避免冗余项目生成。实验证明,这两种方法能有效减少解码过程中的中间项目,提高机器翻译解码效率,在一定程度上提高机器翻译效果。

This paper proposes two methods to improve the efficiency of rule binarization and decoding in tree transducer based translation model. The authors convert synchronous transducer rules to four kinds of binary rules to reduce the tem- porary items, and propose RR - CKY decoding algorithm, which can avoid part of redundant items along with decoding. The experiments show that these two methods can reduce the number of temporary items and make decoding faster. They can also improve the quality of machine translation.