Modeling Coverage for Neural Machine Translation

Моделирование покрытия для нейронного машинного перевода
Hang Li, Zhengdong Lu, Yang Liu, Zhaopeng Tu, Xiaohua Liu
2016-01-01

alignment qualityattention mechanismcoverage vectorneural machine translationover-translation and under-translation
Attention mechanism has enhanced stateof-the-art Neural Machine Translation (NMT) by jointly learning to align and translate.It tends to ignore past alignment information, however, which often leads to over-translation and under-translation.To address this problem, we propose coverage-based NMT in this paper.We maintain a coverage vector to keep track of the attention history.The coverage vector is fed to the attention model to help adjust future attention, which lets NMT system to consider more about untranslated source words.Experiments show that the proposed approach significantly improves both translation quality and alignment quality over standard attention-based NMT. 1
1
Feeding the coverage vector into the attention model adjusts future attention to reduce translation errors.
2
Introducing a coverage vector that tracks attention history helps the model consider untranslated source words.
3
Standard attention-based NMT ignores past alignment information, causing over-translation and under-translation.
4
The coverage-based NMT significantly improves both translation quality and alignment quality compared to standard attention-based NMT.

Attention-based Neural Machine Translation system (with coverage vector)

Modeling and using a coverage vector to track attention history for reducing over-translation and under-translation, improving translation and alignment quality

Publication Details
Publication Date
2016-01-01
Journal
Publisher
ISSN
Cited by
715
Access Type
Author Information
Authors
Hang Li
Zhengdong Lu
Yang Liu
Zhaopeng Tu
Xiaohua Liu
Explore further
Open the scid.ai AI chat with a ready-made request: it will find papers on a similar topic and help build a literature review.
Find similar papers in the chat →
Make a presentation
100%