Neural machine translation research based on the semantic vector of the tri-lingual parallel corpus

RNN Encoder-Decoder and attentional mechanism have lately been used to improve neural machine translation (NMT) on bilingual parallel corpus. In this paper, we propose tri-lingual NMT. Based on the Encoder-Decoder and attentional mechanism, we translate source language to target language, meanwhile translate another parallel source language to target language. We provides two approaches called splicing-model and similarity-model. Both of the approaches are in order to enhance the semantic representation of input sequences. Our experiments on the IWSLT 2012 Chinese to Japanese translation and English to Japanese tasks show that both of the methods provide a comparable or substantial improvement over the bilingual parallel corpus.