Leveraging Visual Knowledge in Language Tasks: An Empirical Study on Intermediate Pre-training for Cross-modal Knowledge Transfer

TL;DR

This paper explores how incorporating visual knowledge into language models through intermediate pre-training can enhance performance on tasks requiring understanding of visual properties and affordances of objects.

Contribution

It provides an empirical analysis of visual knowledge transfer methods, including text and cross-modal transfer, for improving language models on visual-relevant tasks.

Findings

Visual knowledge transfer improves task performance

Cross-modal training enhances understanding of object properties

Benefits observed in both low-resource and fully supervised settings

Abstract

Pre-trained language models are still far from human performance in tasks that need understanding of properties (e.g. appearance, measurable quantity) and affordances of everyday objects in the real world since the text lacks such information due to reporting bias. In this work, we study whether integrating visual knowledge into a language model can fill the gap. We investigate two types of knowledge transfer: (1) text knowledge transfer using image captions that may contain enriched visual knowledge and (2) cross-modal knowledge transfer using both images and captions with vision-language training objectives. On 5 downstream tasks that may need visual knowledge to solve the problem, we perform extensive empirical comparisons over the presented objectives. Our experiments show that visual knowledge transfer can improve performance in both low-resource and fully supervised settings.