Visually-augmented pretrained language models for NLP tasks without images

Code

• github.com/rucaibox/vawi
  OfficialIn papernone★ 1

Abstract

Although pre-trained language models~(PLMs) have shown impressive performance by text-only self-supervised training, they are found lack of visual semantics or commonsense. Existing solutions often rely on explicit images for visual knowledge augmentation (requiring time-consuming retrieval or generation), and they also conduct the augmentation for the whole input text, without considering whether it is actually needed in specific inputs or tasks. To address these issues, we propose a novel Visually-Augmented fine-tuning approach that can be generally applied to various PLMs or NLP tasks, Without using any retrieved or generated Images, namely VAWI. Experimental results show that our approach can consistently improve the performance of BERT, RoBERTa, BART, and T5 at different scales, and outperform several competitive baselines on ten tasks. Our codes and data are publicly available at~https://github.com/RUCAIBox/VAWI.

Tasks

Reproductions