How Do Vision-Language Models Process Conflicting Information Across Modalities?

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Hua, Tianze, Yun, Tian, Pavlick, Ellie
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866915369585737728
author Hua, Tianze
Yun, Tian
Pavlick, Ellie
author_facet Hua, Tianze
Yun, Tian
Pavlick, Ellie
contents AI models are increasingly required to be multimodal, integrating disparate input streams into a coherent state representation on which subsequent behaviors and actions can be based. This paper seeks to understand how such models behave when input streams present conflicting information. Focusing specifically on vision-language models, we provide inconsistent inputs (e.g., an image of a dog paired with the caption "A photo of a cat") and ask the model to report the information present in one of the specific modalities (e.g., "What does the caption say / What is in the image?"). We find that models often favor one modality over the other, e.g., reporting the image regardless of what the caption says, but that different models differ in which modality they favor. We find evidence that the behaviorally preferred modality is evident in the internal representational structure of the model, and that specific attention heads can restructure the representations to favor one modality over the other. Moreover, we find modality-agnostic "router heads" which appear to promote answers about the modality requested in the instruction, and which can be manipulated or transferred in order to improve performance across datasets and modalities. Together, the work provides essential steps towards identifying and controlling if and how models detect and resolve conflicting signals within complex multimodal environments.
format Preprint
id arxiv_https___arxiv_org_abs_2507_01790
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle How Do Vision-Language Models Process Conflicting Information Across Modalities?
Hua, Tianze
Yun, Tian
Pavlick, Ellie
Computation and Language
Artificial Intelligence
Computer Vision and Pattern Recognition
Machine Learning
AI models are increasingly required to be multimodal, integrating disparate input streams into a coherent state representation on which subsequent behaviors and actions can be based. This paper seeks to understand how such models behave when input streams present conflicting information. Focusing specifically on vision-language models, we provide inconsistent inputs (e.g., an image of a dog paired with the caption "A photo of a cat") and ask the model to report the information present in one of the specific modalities (e.g., "What does the caption say / What is in the image?"). We find that models often favor one modality over the other, e.g., reporting the image regardless of what the caption says, but that different models differ in which modality they favor. We find evidence that the behaviorally preferred modality is evident in the internal representational structure of the model, and that specific attention heads can restructure the representations to favor one modality over the other. Moreover, we find modality-agnostic "router heads" which appear to promote answers about the modality requested in the instruction, and which can be manipulated or transferred in order to improve performance across datasets and modalities. Together, the work provides essential steps towards identifying and controlling if and how models detect and resolve conflicting signals within complex multimodal environments.
title How Do Vision-Language Models Process Conflicting Information Across Modalities?
topic Computation and Language
Artificial Intelligence
Computer Vision and Pattern Recognition
Machine Learning
url https://arxiv.org/abs/2507.01790