Unveiling Intrinsic Text Bias in Multimodal Large Language Models through Attention Key-Space Analysis

• URL: http://arxiv.org/abs/2510.26721v1
• Date: Thu, 30 Oct 2025 17:22:22 GMT
• Title: Unveiling Intrinsic Text Bias in Multimodal Large Language Models through Attention Key-Space Analysis
• Authors: Xinhan Zheng, Huyu Wu, Xueting Wang, Haiyun Jiang,
• Abstract summary: Multimodal large language models (MLLMs) exhibit a pronounced preference for textual inputs when processing vision-language data.<n>We propose that the bias originates from the model's internal architecture.
• Score: 19.111897718147656
• License: http://arxiv.org/licenses/nonexclusive-distrib/1.0/
• Abstract: Multimodal large language models (MLLMs) exhibit a pronounced preference for textual inputs when processing vision-language data, limiting their ability to reason effectively from visual evidence. Unlike prior studies that attribute this text bias to external factors such as data imbalance or instruction tuning, we propose that the bias originates from the model's internal architecture. Specifically, we hypothesize that visual key vectors (Visual Keys) are out-of-distribution (OOD) relative to the text key space learned during language-only pretraining. Consequently, these visual keys receive systematically lower similarity scores during attention computation, leading to their under-utilization in the context representation. To validate this hypothesis, we extract key vectors from LLaVA and Qwen2.5-VL and analyze their distributional structures using qualitative (t-SNE) and quantitative (Jensen-Shannon divergence) methods. The results provide direct evidence that visual and textual keys occupy markedly distinct subspaces within the attention space. The inter-modal divergence is statistically significant, exceeding intra-modal variation by several orders of magnitude. These findings reveal that text bias arises from an intrinsic misalignment within the attention key space rather than solely from external data factors.

Related papers

• V-FAT: Benchmarking Visual Fidelity Against Text-bias [10.716447149075357]
  We investigate the tension between visual perception and linguistic priors.<n>We introduce V-FAT (Visual Fidelity Against Text-bias), a diagnostic benchmark comprising 4,026 VQA instances across six semantic domains.<n>Our evaluation of 12 frontier MLLMs reveals that while models excel in existing benchmarks, they experience significant visual collapse under high linguistic dominance.
  arXiv  Detail & Related papers  (2026-01-08T12:50:14Z)
• Investigating Spatial Attention Bias in Vision-Language Models [8.387055152856824]
  This work identifies and characterizes a systematic spatial attention bias in Vision-Language Models (VLMs)<n>We demonstrate that this bias persists across different architectures, with models describing left-positioned content first in approximately 97% of cases.<n>Testing on an Arabic-finetuned model reveals that the bias persists despite right-to-left language training, ruling out language reading direction as the primary cause.
  arXiv  Detail & Related papers  (2025-12-20T06:22:38Z)
• Seeing to Act, Prompting to Specify: A Bayesian Factorization of Vision Language Action Policy [59.44168425139687]
  BayesVLA is a Bayesian factorization that decomposes the policy into a visual-action prior, supporting seeing-to-act, and a language-conditioned likelihood, enabling prompt-to-specify.<n>Experiments show superior generalization to unseen instructions, objects, and environments compared to existing methods.
  arXiv  Detail & Related papers  (2025-12-12T01:59:23Z)
• Explaining multimodal LLMs via intra-modal token interactions [55.27436637894534]
  Multimodal Large Language Models (MLLMs) have achieved remarkable success across diverse vision-language tasks, yet their internal decision-making mechanisms remain insufficiently understood.<n>We propose enhancing interpretability by leveraging intra-modal interaction.
  arXiv  Detail & Related papers  (2025-09-26T14:39:13Z)
• Semantic and Structural Analysis of Implicit Biases in Large Language Models: An Interpretable Approach [1.5749416770494704]
  It proposes an interpretable bias detection method aimed at identifying hidden social biases in model outputs.<n>The method combines nested semantic representation with a contextual contrast mechanism.<n>The evaluation focuses on several key metrics, such as bias detection accuracy, semantic consistency, and contextual sensitivity.
  arXiv  Detail & Related papers  (2025-08-08T09:21:10Z)
• A Closer Look at Bias and Chain-of-Thought Faithfulness of Large (Vision) Language Models [53.18562650350898]
  Chain-of-thought (CoT) reasoning enhances performance of large language models.<n>We present the first comprehensive study of CoT faithfulness in large vision-language models.
  arXiv  Detail & Related papers  (2025-05-29T18:55:05Z)
• Interpreting Social Bias in LVLMs via Information Flow Analysis and Multi-Round Dialogue Evaluation [1.7997395646080083]
  Large Vision Language Models (LVLMs) have achieved remarkable progress in multimodal tasks, yet they also exhibit notable social biases.<n>We propose an explanatory framework that combines information flow analysis with multi-round dialogue evaluation.<n>Experiments reveal that LVLMs exhibit systematic disparities in information usage when processing images of different demographic groups.
  arXiv  Detail & Related papers  (2025-05-27T12:28:44Z)
• Beyond Early-Token Bias: Model-Specific and Language-Specific Position Effects in Multilingual LLMs [50.07451351559251]
  We present a study across five typologically distinct languages (English, Russian, German, Hindi, and Vietnamese)<n>We examine how position bias interacts with prompt strategies and affects output entropy.
  arXiv  Detail & Related papers  (2025-05-22T02:23:00Z)
• Tokens, the oft-overlooked appetizer: Large language models, the distributional hypothesis, and meaning [29.745218855471787]
  Tokenization is a necessary component within the current architecture of many language models.<n>We discuss how tokens and pretraining can act as a backdoor for bias and other unwanted content.<n>We relay evidence that the tokenization algorithm's objective function impacts the large language model's cognition.
  arXiv  Detail & Related papers  (2024-12-14T18:18:52Z)
• Context is Key: A Benchmark for Forecasting with Essential Textual Information [87.3175915185287]
  "Context is Key" (CiK) is a forecasting benchmark that pairs numerical data with diverse types of carefully crafted textual context.<n>We evaluate a range of approaches, including statistical models, time series foundation models, and LLM-based forecasters.<n>We propose a simple yet effective LLM prompting method that outperforms all other tested methods on our benchmark.
  arXiv  Detail & Related papers  (2024-10-24T17:56:08Z)
• Refining Skewed Perceptions in Vision-Language Contrastive Models through Visual Representations [0.033483662989441935]
  Large vision-language contrastive models (VLCMs) have become foundational, demonstrating remarkable success across a variety of downstream tasks.<n>Despite their advantages, these models inherit biases from the disproportionate distribution of real-world data, leading to misconceptions about the actual environment.<n>This study presents an investigation into how a simple linear probe can effectively distill task-specific core features from CLIP's embedding for downstream applications.
  arXiv  Detail & Related papers  (2024-05-22T22:03:11Z)
• Debiasing Multimodal Large Language Models via Penalization of Language Priors [38.97645845493758]
  Multimodal Large Language Models (MLLMs) have become indispensable tools in computer vision and natural language processing.<n>Despite their advancements, our investigation reveals a noteworthy bias: the generated content is often driven more by the inherent priors of the underlying Large Language Models (LLMs) than by the input image.<n>We propose two simple, training-free strategies to rectify these biases and redirect the model's focus toward visual information.
  arXiv  Detail & Related papers  (2024-03-08T12:35:07Z)
• Prototype-based Aleatoric Uncertainty Quantification for Cross-modal Retrieval [139.21955930418815]
  Cross-modal Retrieval methods build similarity relations between vision and language modalities by jointly learning a common representation space. However, the predictions are often unreliable due to the Aleatoric uncertainty, which is induced by low-quality data, e.g., corrupt images, fast-paced videos, and non-detailed texts. We propose a novel Prototype-based Aleatoric Uncertainty Quantification (PAU) framework to provide trustworthy predictions by quantifying the uncertainty arisen from the inherent data ambiguity.
  arXiv  Detail & Related papers  (2023-09-29T09:41:19Z)
• How to Handle Different Types of Out-of-Distribution Scenarios in Computational Argumentation? A Comprehensive and Fine-Grained Field Study [59.13867562744973]
  This work systematically assesses LMs' capabilities for out-of-distribution (OOD) scenarios. We find that the efficacy of such learning paradigms varies with the type of OOD. Specifically, while ICL excels for domain shifts, prompt-based fine-tuning surpasses for topic shifts.
  arXiv  Detail & Related papers  (2023-09-15T11:15:47Z)
• Behind the Scene: Revealing the Secrets of Pre-trained Vision-and-Language Models [65.19308052012858]
  Recent Transformer-based large-scale pre-trained models have revolutionized vision-and-language (V+L) research. We present VALUE, a set of meticulously designed probing tasks to decipher the inner workings of multimodal pre-training. Key observations: Pre-trained models exhibit a propensity for attending over text rather than images during inference.
  arXiv  Detail & Related papers  (2020-05-15T01:06:54Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.