Scene Graph as Pivoting: Inference-time Image-free Unsupervised Multimodal Machine Translation with Visual Scene Hallucination

TL;DR

This paper introduces an inference-time image-free unsupervised multimodal machine translation approach that uses scene graphs and visual hallucination to improve translation quality without relying on paired images.

Contribution

It proposes a novel scene graph pivoting method with visual hallucination for image-free multimodal translation, enhancing translation quality in an unsupervised setting.

Findings

Outperforms baseline BLEU scores on Multi30K.

Achieves better translation completeness, relevance, and fluency.

Enables image-free inference with improved semantic understanding.

Abstract

In this work, we investigate a more realistic unsupervised multimodal machine translation (UMMT) setup, inference-time image-free UMMT, where the model is trained with source-text image pairs, and tested with only source-text inputs. First, we represent the input images and texts with the visual and language scene graphs (SG), where such fine-grained vision-language features ensure a holistic understanding of the semantics. To enable pure-text input during inference, we devise a visual scene hallucination mechanism that dynamically generates pseudo visual SG from the given textual SG. Several SG-pivoting based learning objectives are introduced for unsupervised translation training. On the benchmark Multi30K data, our SG-based method outperforms the best-performing baseline by significant BLEU scores on the task and setup, helping yield translations with better completeness, relevance and fluency without relying on paired images. Further in-depth analyses reveal how our model advances in the task setting.