LIUM-CVC Submissions for WMT18 Multimodal Translation Task

TL;DR

This paper presents LIUM-CVC's multimodal neural machine translation systems for WMT18, introducing architectural improvements that enhanced integration of visual features, leading to top rankings in English-French and English-German translation tasks.

Contribution

The paper introduces modifications to the multimodal attention architecture to better incorporate convolutional features and encoder-side information.

Findings

Ranked first for English-French translation

Ranked second for English-German translation

Achieved top performance with METEOR metric

Abstract

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.