From Vision to Language: Investigating Causal Information Flow in Multimodal Decision-Making

• URL: http://arxiv.org/abs/2609.05149v1
• Date: Fri, 04 Sep 2026 13:52:27 GMT
• Title: From Vision to Language: Investigating Causal Information Flow in Multimodal Decision-Making
• Abstract summary: We investigate how visual information contributes to language-based decisions.<n>Our results show that visual information is mainly integrated while the model processes the candidate answer options.<n>We identify a distinct pattern in temporal reasoning, suggesting that VLMs struggle to reconstruct sequential information across video frames.
• Score: 45.85783584369188
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: Vision-Language Models are commonly evaluated through their final predictions, but understanding whether these decisions are grounded in visual evidence requires tracing how visual information contributes to language-based decisions. With this purpose in mind, we investigate cross-modal information flow in a video-based generative multiple-choice-like setting by applying a layer-wise causal intervention on video-text attention pathways. We target spatial, causal, and temporal visual reasoning. Our results show that visual information is mainly integrated while the model processes the candidate answer options, which serve as the primary textual grounding sites for the final decision. We further show that nouns play an important role as semantic anchors during multimodal enrichment, while verbs are more relevant when temporal relations are processed. Finally, we identify a distinct pattern in temporal reasoning, suggesting that VLMs struggle to reconstruct sequential information across video frames, but we remark that such fragility may also reflect linguistic biases associated with specific temporal expressions used for defining the relation between events within a scene.

Related papers

• When Discourse Pressures Conflict: Information Structure in Vision-Language Model Outputs [8.897315812066683]
  We test whether vision-language models (VLMs) distinguish discourse-old Topics from discourse-new Foci in visually grounded question answering.<n>We exploit Hungarian, a language in which Topic and Focus map onto dedicated syntactic positions, making IS choices observable in text.<n>We find that models produce IS-relevant constructions, but over-regularise this sensitivity.
  arXiv  Detail & Related papers  (2026-05-27T11:50:56Z)
• 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)
• ChronusOmni: Improving Time Awareness of Omni Large Language Models [29.685563616290352]
  Time awareness is a fundamental ability of omni large language models, especially for understanding long videos and answering complex questions.<n>We propose ChronusOmni, an omni large language model designed to enhance temporal awareness for both explicit and implicit audiovisual temporal grounding.<n>We construct ChronusAV, a temporally-accurate, modality-complete, and cross-modal-aligned dataset to support the training and evaluation on audiovisual temporal grounding task.
  arXiv  Detail & Related papers  (2025-12-10T17:22:42Z)
• Priors in Time: Missing Inductive Biases for Language Model Interpretability [58.07412640266836]
  We show that Sparse Autoencoders impose priors that assume independence of concepts across time, implying stationarity.<n>We introduce a new interpretability objective -- Temporal Feature Analysis -- which possesses a temporal inductive bias to decompose representations at a given time into two parts.<n>Our results underscore the need for inductive biases that match the data in designing robust interpretability tools.
  arXiv  Detail & Related papers  (2025-11-03T18:43:48Z)
• Deep Temporal Reasoning in Video Language Models: A Cross-Linguistic Evaluation of Action Duration and Completion through Perfect Times [0.0]
  We introduce the textbfPerfect Times dataset, a quadrilingual (English, Italian, Russian, and Japanese) multiple-choice question-answering benchmark designed to assess video-language models (VLMs) on temporal reasoning.<n> Experimental results indicate that state-of-the-art models, despite their success on text-based tasks, struggle to mirror human-like temporal and causal reasoning grounded in video.
  arXiv  Detail & Related papers  (2025-06-01T09:45:41Z)
• Multi-Modal interpretable automatic video captioning [1.9874264019909988]
  We introduce a novel video captioning method trained with multi-modal contrastive loss. Our approach is designed to capture the dependency between these modalities, resulting in more accurate, thus pertinent captions.
  arXiv  Detail & Related papers  (2024-11-11T11:12:23Z)
• Context-Aware Temporal Embedding of Objects in Video Data [0.8287206589886881]
  In video analysis, understanding the temporal context is crucial for recognizing object interactions, event patterns, and contextual changes over time. The proposed model leverages adjacency and semantic similarities between objects from neighboring video frames to construct context-aware temporal object embeddings. Empirical studies demonstrate that our context-aware temporal embeddings can be used in conjunction with conventional visual embeddings to enhance the effectiveness of downstream applications.
  arXiv  Detail & Related papers  (2024-08-23T01:44:10Z)
• MLLM as Video Narrator: Mitigating Modality Imbalance in Video Moment Retrieval [53.417646562344906]
  Video Moment Retrieval (VMR) aims to localize a specific temporal segment within an untrimmed long video given a natural language query. Existing methods often suffer from inadequate training annotations, i.e., the sentence typically matches with a fraction of the prominent video content in the foreground with limited wording diversity. This intrinsic modality imbalance leaves a considerable portion of visual information remaining unaligned with text. In this work, we take an MLLM as a video narrator to generate plausible textual descriptions of the video, thereby mitigating the modality imbalance and boosting the temporal localization.
  arXiv  Detail & Related papers  (2024-06-25T18:39:43Z)
• Structured Video-Language Modeling with Temporal Grouping and Spatial Grounding [112.3913646778859]
  We propose a simple yet effective video-language modeling framework, S-ViLM. It includes two novel designs, inter-clip spatial grounding and intra-clip temporal grouping, to promote learning region-object alignment and temporal-aware features. S-ViLM surpasses the state-of-the-art methods substantially on four representative downstream tasks.
  arXiv  Detail & Related papers  (2023-03-28T22:45:07Z)
• Multi-Modal Interaction Graph Convolutional Network for Temporal Language Localization in Videos [55.52369116870822]
  This paper focuses on tackling the problem of temporal language localization in videos. It aims to identify the start and end points of a moment described by a natural language sentence in an untrimmed video.
  arXiv  Detail & Related papers  (2021-10-12T14:59:25Z)

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.