Z-LaVI: Zero-Shot Language Solver Fueled by Visual Imagination

TL;DR

Z-LaVI enhances large language models by integrating visual imagination through image retrieval and synthesis, enabling better zero-shot performance on language understanding tasks.

Contribution

The paper introduces Z-LaVI, a novel approach that combines visual retrieval and synthesis to endow language models with visual imagination capabilities for improved zero-shot learning.

Findings

Significant performance improvements across diverse language tasks.

Effective use of visual knowledge enhances language understanding.

Visual imagination bridges the gap caused by reporting bias.

Abstract

Large-scale pretrained language models have made significant advances in solving downstream language understanding tasks. However, they generally suffer from reporting bias, the phenomenon describing the lack of explicit commonsense knowledge in written text, e.g., ''an orange is orange''. To overcome this limitation, we develop a novel approach, Z-LaVI, to endow language models with visual imagination capabilities. Specifically, we leverage two complementary types of ''imaginations'': (i) recalling existing images through retrieval and (ii) synthesizing nonexistent images via text-to-image generation. Jointly exploiting the language inputs and the imagination, a pretrained vision-language model (e.g., CLIP) eventually composes a zero-shot solution to the original language tasks. Notably, fueling language models with imagination can effectively leverage visual knowledge to solve plain language tasks. In consequence, Z-LaVI consistently improves the zero-shot performance of existing language models across a diverse set of language tasks.