Some Modalities are More Equal Than Others: Decoding and Architecting Multimodal Integration in MLLMs

TL;DR

This paper evaluates the robustness of Multimodal Large Language Models (MLLMs) to conflicting modalities, revealing their vulnerabilities and proposing an alignment tuning method to improve multimodal reasoning and reliability.

Contribution

The paper introduces MMA-Bench for testing modality reliance, analyzes MLLMs' brittleness, and proposes a modality alignment tuning strategy to enhance cross-modal reasoning.

Findings

MLLMs are sensitive to misaligned audio-visual inputs.

Current MLLMs struggle with simple misleading text.

Alignment tuning improves multimodal grounding.

Abstract

Despite remarkable advancements in Multimodal Large Language Models (MLLMs), a fundamental question remains: are MLLMs robust to contradicting modalities? To rigorously study this, we introduce MMA-Bench comprising videos and tasks that probe a model's reliance on specific modalities. Using black-box and white-box interpretability techniques, we provide a critical analysis of the brittleness of both open- and closed-sourced MLLMs. We show that current MLLMs struggle under misaligned audio-visual pairs and simple misleading text, thereby lacking robust multi-modal reasoning. Building on these findings, we propose a modality alignment tuning strategy to teach the model when to prioritize, leverage, or ignore specific modality cues. Through extensive experiments and analysis, we show that our alignment tuning yields demonstrably stronger multimodal grounding. This work provides both interpretability tools and a clear path toward developing MLLMs with intrinsically reliable cross-modal reasoning. Code and dataset will be publicly available.