On Vision Features in Multimodal Machine Translation

TL;DR

This paper investigates how the quality and type of vision models influence multimodal machine translation, highlighting that stronger vision models improve translation and emphasizing the importance of careful evaluation on biased, small-scale benchmarks.

Contribution

It systematically examines the impact of various advanced vision models on MMT and introduces a selective attention approach to analyze image contributions at the patch level.

Findings

Stronger vision models enhance translation quality in MMT.

Visual features from advanced models contribute significantly to translation.

Current benchmarks may be biased and insufficient for comprehensive evaluation.

Abstract

Previous work on multimodal machine translation (MMT) has focused on the way of incorporating vision features into translation but little attention is on the quality of vision models. In this work, we investigate the impact of vision models on MMT. Given the fact that Transformer is becoming popular in computer vision, we experiment with various strong models (such as Vision Transformer) and enhanced features (such as object-detection and image captioning). We develop a selective attention model to study the patch-level contribution of an image in MMT. On detailed probing tasks, we find that stronger vision models are helpful for learning translation from the visual modality. Our results also suggest the need of carefully examining MMT models, especially when current benchmarks are small-scale and biased. Our code could be found at \url{https://github.com/libeineu/fairseq_mmt}.