Longyue Wang - Transfer Talk - 2nd May 2017

Video Category: 
Transfer Talk
Longyue Wang

Title: Discourse Neural Machine Translation

Supervisor: Prof. Qun Liu


Due to computational complexity, machine translation (MT) systems usually translate documents by considering isolated sentences, disregarding significant discourse information beyond sentence level. As a result, machine-translated results often contain a series of problems on coherence, cohesion and consistency. It is a big challenge to integrate the cross-sentence and structural knowledge into MT frameworks. Although discourse phenomena have been investigated in conventional statistical machine translation (SMT), it is still underexplored for the state-of-the-art neural machine translation (NMT). By applying deep representation learning, it has been shown that the performance of NMT has surpassed the performance of SMT on various language pairs. The great progress on NMT encourage us to explore integrating discourse knowledge into the unified model. We expect that discourse-NMT will generate more coherent, cohesive and consistent translations.


My_Transfer_Longyue_v5.pdf3.31 MB