A hybrid strategy method for patent machine translation

Patent documents involve many professional fields and include multiple paragraphs, which have brought great difficulties for patent machine translation. This paper presents a hybrid strategy for patent machine translation, which combines the template-based machine translation (TBMT), the semantic rule-based machine translation (RBMT) and the statistics-based machine translation (SMT). The TBMT and RBMT are combined to translate the claim texts in patent documents, and the RBMT and SMT are combined to translate the other texts. In order to find the optimal hybrid strategy, three larger patent fields are selected to train specific models, and then compared with the general model. The experiment results show that the hybrid strategy method outperforms each individual system across a varied set of automatic translation evaluation metrics, verifying the effectiveness of the proposed hybrid strategy method in patent machine translation.