Grammatically Derived Factual Relation Augmented Neural Machine Translation

Transformer-based neural machine translation (NMT) has achieved state-of-the-art performance in the NMT paradigm. This method assumes that model can automatically learn linguistic knowledge (e.g., grammar and syntax) from parallel corpus via an attention network. However, network cannot capture deep internal structure of a sentence. Therefore, it is natural to introduce some prior guide model. In this paper, factual relation information introduced into as knowledge, novel approach named Factual Relation Augmented (FRA) proposed decoder NMT. encoding procedure, mask matrix constructed generate representation for source sentence, while decoding procedure effective incorporate original sentence decoder. Positive results obtained several different tasks indicate effectiveness approach.