LIUM-CVC Submissions for WMT18 Multimodal Translation Task

Ozan Caglayan (1) , Adrien Bardet (1) , Fethi Bougares (1) , Loïc Barrault (1) , Kai Wang (2) , Marc Masana (3) , Luis Herranz (4) , Joost van de Weijer (5)

1 LIUM - Laboratoire d'Informatique de l'Université du Mans
2 M2DisCo - Geometry Processing and Constrained Optimization
3 CVC - Computer Vision Center (Centre de visio per computador)
4 Institute of Computing Tech.
5 UAB - Universitat Autònoma de Barcelona = Autonomous University of Barcelona = Universidad Autónoma de Barcelona

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.

Domaines: Informatique et langage [cs.CL]; Vision par ordinateur et reconnaissance de formes [cs.CV]

Ozan Caglayan, Adrien Bardet, Fethi Bougares, Loïc Barrault, Kai Wang, et al.. LIUM-CVC Submissions for WMT18 Multimodal Translation Task. Third Conference on Machine Translation, Oct 2018, Brussels, Belgium. pp.597 - 602, ⟨10.18653/v1/W18-64065⟩. ⟨hal-02498133⟩