Paper ID: 2404.18624

Do Vision & Language Decoders use Images and Text equally? How Self-consistent are their Explanations?

Letitia Parcalabescu, Anette Frank

Vision and language model (VLM) decoders are currently the best-performing architectures on multimodal tasks. Next to predictions, they can also produce explanations, either in post-hoc or CoT settings. However, it is not clear how much they use the vision and text modalities when generating predictions or explanations. In this work, we investigate if VLMs rely on modalities differently when they produce explanations as opposed to providing answers. We also evaluate the self-consistency of VLM decoders in both post-hoc and CoT explanation settings, by extending existing unimodal tests and measures to VLM decoders. We find that VLMs are less self-consistent than LLMs. Text contributions in VL decoders are more important than image contributions in all examined tasks. Moreover, the contributions of images are significantly stronger for explanation generation compared to answer generation. This difference is even larger in CoT compared to post-hoc explanations. Lastly, we provide an up-to-date benchmarking of state-of-the-art VL decoders on the VALSE benchmark, which before only covered VL encoders. We find that VL decoders still struggle with most phenomena tested by VALSE.

Submitted: Apr 29, 2024