What do Models Learn From Training on More Than Text? Measuring Visual Commonsense Knowledge
Paper in proceeding, 2022

There are limitations in learning language from text alone. Therefore, recent focus has been on developing multimodal models. However, few benchmarks exist that can measure what language models learn about language from multimodal training. We hypothesize that training on a visual modality should improve on the visual commonsense knowledge in language models. Therefore, we introduce two evaluation tasks for measuring visual commonsense knowledge in language models(1) and use them to evaluate different multimodal models and unimodal baselines. Primarily, we find that the visual commonsense knowledge is not significantly different between the multimodal models and unimodal baseline models trained on visual text data.

Author

Lovisa Hagström

Chalmers, Computer Science and Engineering (Chalmers), Data Science and AI

Richard Johansson

University of Gothenburg

PROCEEDINGS OF THE 60TH ANNUAL MEETING OF THE ASSOCIATION FOR COMPUTATIONAL LINGUISTICS (ACL 2022): STUDENT RESEARCH WORKSHOP

252-261
978-1-955917-23-0 (ISBN)

60th Annual Meeting of the Association-for-Computational-Linguistics (ACL)
Dublin, Ireland,

Subject Categories

Other Computer and Information Science

Language Technology (Computational Linguistics)

General Language Studies and Linguistics

Computer Science

DOI

10.18653/v1/2022.acl-srw.19

More information

Latest update

11/6/2024