Saved in:
Bibliographic Details
Main Authors: Xiao, Cihan, Shao, Yiwen, Li, Chenxing, He, Xiang, Liang, Zhenwen, Yves, Steve, Khudanpur, Sanjeev, Bo, Liefeng
Format: Preprint
Published: 2026
Subjects:
Online Access:https://arxiv.org/abs/2605.27741
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866917538141569024
author Xiao, Cihan
Shao, Yiwen
Li, Chenxing
He, Xiang
Liang, Zhenwen
Yves, Steve
Khudanpur, Sanjeev
Bo, Liefeng
author_facet Xiao, Cihan
Shao, Yiwen
Li, Chenxing
He, Xiang
Liang, Zhenwen
Yves, Steve
Khudanpur, Sanjeev
Bo, Liefeng
contents Audio and omni-modal large language models exhibit impressive cross-modal reasoning capabilities. However, applying standard reinforcement learning post-training algorithms to these models exposes a critical structural vulnerability: methods like GRPO apply uniform policy gradients across all tokens, ignoring their unequal dependence on the non-text source modality. This exacerbates late-stage modality collapse during extended chain-of-thought generation, where models progressively abandon the primary source signal in favor of compressed textual priors, leading to confident but ungrounded hallucinations. To address this, we introduce Modality-Aware Policy Optimization (MAPO), a novel dual-branch reinforcement learning framework. First, MAPO dynamically concentrates the policy gradient on modality-critical tokens using a modality relevance mask, which is derived from the cross-modal differential entropy between an audio-ablated reference and the multimodal policy. Second, it integrates an auxiliary attention loss branch that applies a targeted, temporally scaled penalty to the model's internal attention distributions. This ensures the model actively sustains cross-modal grounding deep into the reasoning trace. Evaluations on complex audio reasoning benchmarks demonstrate that MAPO substantially improves long-horizon reasoning fidelity and multimodal instruction following, achieving highly competitive performance and setting new state-of-the-art results on several key benchmarks among open-weight models. By relying strictly on native statistical signals rather than domain-specific inductive biases, MAPO offers a promising foundation for mitigating epistemic collapse across diverse multimodal systems.
format Preprint
id arxiv_https___arxiv_org_abs_2605_27741
institution arXiv
publishDate 2026
record_format arxiv
spellingShingle Escape the Language Prior: Mitigating Late-Stage Modality Collapse in Audio Reasoning via Modality-Aware Policy Optimization
Xiao, Cihan
Shao, Yiwen
Li, Chenxing
He, Xiang
Liang, Zhenwen
Yves, Steve
Khudanpur, Sanjeev
Bo, Liefeng
Computation and Language
Audio and omni-modal large language models exhibit impressive cross-modal reasoning capabilities. However, applying standard reinforcement learning post-training algorithms to these models exposes a critical structural vulnerability: methods like GRPO apply uniform policy gradients across all tokens, ignoring their unequal dependence on the non-text source modality. This exacerbates late-stage modality collapse during extended chain-of-thought generation, where models progressively abandon the primary source signal in favor of compressed textual priors, leading to confident but ungrounded hallucinations. To address this, we introduce Modality-Aware Policy Optimization (MAPO), a novel dual-branch reinforcement learning framework. First, MAPO dynamically concentrates the policy gradient on modality-critical tokens using a modality relevance mask, which is derived from the cross-modal differential entropy between an audio-ablated reference and the multimodal policy. Second, it integrates an auxiliary attention loss branch that applies a targeted, temporally scaled penalty to the model's internal attention distributions. This ensures the model actively sustains cross-modal grounding deep into the reasoning trace. Evaluations on complex audio reasoning benchmarks demonstrate that MAPO substantially improves long-horizon reasoning fidelity and multimodal instruction following, achieving highly competitive performance and setting new state-of-the-art results on several key benchmarks among open-weight models. By relying strictly on native statistical signals rather than domain-specific inductive biases, MAPO offers a promising foundation for mitigating epistemic collapse across diverse multimodal systems.
title Escape the Language Prior: Mitigating Late-Stage Modality Collapse in Audio Reasoning via Modality-Aware Policy Optimization
topic Computation and Language
url https://arxiv.org/abs/2605.27741