Neural Machine Translation by Fusing Key Information of Text

When the Transformer proposed by Google in 2017, it was first used for machine translation tasks and achieved state of art at that time. Although current neural model can generate high quality results, there are still mistranslations omissions key information long sentences. On other hand, most important part traditional is information. In as translated accurately completely, even if parts results incorrect, final results’ be guaranteed. order to solve problem mistranslation missed effectively, improve accuracy completeness sentence translation, this paper proposes a fused based on Transformer. The extracts keywords source language text separately input encoder. After same encoding text, with output encoded encoder, then processed into decoder. With incorporating keyword from sentence, model’s performance task translating sentences very reliable. verify effectiveness method fusion paper, series experiments were carried out verification set. experimental show Bilingual Evaluation Understudy (BLEU) score Workshop Machine Translation (WMT) 2017 test dataset higher than BLEU WMT2017 dataset. advantages paper.