LIUM-CVC Submissions for WMT18 Multimodal Translation Task
This work addresses the challenge of improving translation accuracy in multimodal settings for researchers and practitioners in machine translation, though it is incremental as it builds on prior architectures.
The authors tackled the problem of multimodal neural machine translation by modifying their previous attention architecture to better integrate and refine convolutional features, achieving first place for English-French and second for English-German in constrained submissions according to METEOR scores.
This paper describes the multimodal Neural Machine Translation systems developed by LIUM and CVC for WMT18 Shared Task on Multimodal Translation. This year we propose several modifications to our previous multimodal attention architecture in order to better integrate convolutional features and refine them using encoder-side information. Our final constrained submissions ranked first for English-French and second for English-German language pairs among the constrained submissions according to the automatic evaluation metric METEOR.