Improved template-driven neural machine translation

At present, neural networks are widely used in machine translation and have achieved good translation results. However, neural machine translation is more sensitive to sentence length and has poor translation effect on long and complicated sentences. In order to solve this problem, based on template-based neural machine translation, this paper adds a sub-sentence level template named sub-block template to the sentence level template and proposed a method using compressed sentences to match the sentence level template, and extract the noun phrase and verb phrase with the compressed sentence word as the center word and match the sub-block template. The experimental results show that the proposed method can effectively improve the quality of machine translation.