30 Oct, 2020 (Fri)Time:
5:00 pm - 6:00 pmWebinar Link:
Meeting ID: 981 7466 9133
SpeakerMr. Guanhua CHEN
Department of Electrical and Electronic Engineering
The University of Hong Kong
Lexically constrained neural machine translation (NMT), which leverages pre-specified translation to constrain NMT, has practical significance in interactive translation and NMT domain adaption. Previous work either modify the decoding algorithm or train the model on augmented dataset. These methods suffer from either high computational overheads or low copying success rates. In this seminar, the speaker introduces the background of lexically constrained neural machine translation and proposes ATT-INPUT and ATT-OUTPUT, two alignment-based constrained decoding methods. These two methods revise the target tokens during decoding based on word alignments derived from encoder-decoder attention weights. The speaker further presents EAM-OUTPUT by introducing an explicit alignment module (EAM) to a pretrained Transformer. It decodes similarly as EAM-OUTPUT, except using alignments derived from the EAM. Experiments on WMT16 De-En and WMT16 Ro-En show the effectiveness of the approaches on constrained NMT task.
OrganizerProfessor Victor O.K. Li
Most seminars are open to the general public, free of charge, unless otherwise stated. Registration is not required. Arrangement for car parking facilities on campus please contact us for details.
For enquiries, please contact:
Department of Electrical and Electronic Engineering,
Room 601, Chow Yei Ching Building,
Pokfulam Road, Hong Kong
Tel: (852) 3917 7093