The Role of Linguistic Priors in Measuring Compositional Generalization of Vision-language Models

ICML 2023 Workshop SCIS Submission13 Authors

Keywords: multi-modal learning, vision-language models, compositional generalization, linguistic prior

TL;DR: We show that current improvements to vision-language compositional generalization mainly rely on linguistic priors.

Abstract: Compositionality is a common property in many modalities including natural languages and images, but the compositional generalization of multi-modal models is not well-understood. In this paper, we identify two sources of visual-linguistic compositionality: linguistic priors and the interplay between images and texts. We show that current attempts to improve compositional generalization rely on linguistic priors rather than on information in the image. We also propose a new metric for compositionality without such linguistic priors.

Submission Number: 13