# Understanding Temporal Logic Consistency in Video-Language Models through Cross-Modal Attention Discriminability

arXiv: https://arxiv.org/abs/2510.08138 · HTML: https://mlsanity.com/paper/2510.08138/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 22 of its 31 references that are in the map.
- First posted on arXiv 9 Oct 2025.

## Builds on (most cited first, 22 of 22)

- [MMMU: A Massive Multi-discipline Multimodal Understanding and Reasoning Benchmark for Expert AGI](https://mlsanity.com/paper/2311.16502/) (2023)
- [Video-LLaMA: An Instruction-tuned Audio-Visual Language Model for Video Understanding](https://mlsanity.com/paper/2306.02858/) (2023)
- [Video-MME: The First-Ever Comprehensive Evaluation Benchmark of Multi-modal LLMs in Video Analysis](https://mlsanity.com/paper/2405.21075/) (2024)
- [Dense-Captioning Events in Videos](https://mlsanity.com/paper/1705.00754/) (2017)
- [Instruction Tuning for Large Language Models: A Survey](https://mlsanity.com/paper/2308.10792/) (2023)
- [TALL: Temporal Activity Localization via Language Query](https://mlsanity.com/paper/1705.02101/) (2017)
- [TimeChat: A Time-sensitive Multimodal Large Language Model for Long Video Understanding](https://mlsanity.com/paper/2312.02051/) (2023)
- [Video Understanding with Large Language Models: A Survey](https://mlsanity.com/paper/2312.17432/) (2023)
- [MMT-Bench: A Comprehensive Multimodal Benchmark for Evaluating Large Vision-Language Models Towards Multitask AGI](https://mlsanity.com/paper/2404.16006/) (2024)
- [A Practical Review of Mechanistic Interpretability for Transformer-Based Language Models](https://mlsanity.com/paper/2407.02646/) (2024)
- [HawkEye: Training Video-Text LLMs for Grounding Text in Videos](https://mlsanity.com/paper/2403.10228/) (2024)
- [VTG-LLM: Integrating Timestamp Knowledge into Video LLMs for Enhanced Video Temporal Grounding](https://mlsanity.com/paper/2405.13382/) (2024)
- [ReXTime: A Benchmark Suite for Reasoning-Across-Time in Videos](https://mlsanity.com/paper/2406.19392/) (2024)
- [MLLMs are Deeply Affected by Modality Bias](https://mlsanity.com/paper/2505.18657/) (2025)
- [Temporal Reasoning Transfer from Text to Video](https://mlsanity.com/paper/2410.06166/) (2024)
- [Debiasing Multimodal Large Language Models via Penalization of Language Priors](https://mlsanity.com/paper/2403.05262/) (2024)
- [A Survey on Mechanistic Interpretability for Multi-Modal Foundation Models](https://mlsanity.com/paper/2502.17516/) (2025)
- [How Visual Representations Map to Language Feature Space in Multimodal LLMs](https://mlsanity.com/paper/2506.11976/) (2025)
- [Chrono: A Simple Blueprint for Representing Time in MLLMs](https://mlsanity.com/paper/2406.18113/) (2024)
- [On the Consistency of Video Large Language Models in Temporal Comprehension](https://mlsanity.com/paper/2411.12951/) (2024)
- [Background-aware Moment Detection for Video Moment Retrieval](https://mlsanity.com/paper/2306.02728/) (2023)
- [Enhancing Temporal Modeling of Video LLMs via Time Gating](https://mlsanity.com/paper/2410.05714/) (2024)

## Abstract (arXiv)

Large language models (LLMs) often generate self-contradictory outputs, which severely impacts their reliability and hinders their adoption in practical applications. In video-language models (Video-LLMs), this phenomenon recently draws the attention of researchers. Specifically, these models fail to provide logically consistent responses to rephrased questions based on their grounding outputs. However, the underlying causes of this phenomenon remain underexplored. In this work, we adopt an interpretability-driven approach to analyze, statistically summarize, and intervention the potential factors of the phenomenon. We find that one of the primary reasons for the inconsistency in responses lies in the inability of cross-modal attention heads to effectively distinguish video tokens across different timestamps. To address this, we propose an attention enhancement method called Temporally Conditioned Attention Sharpening (TCAS), which constructs an enhancement objective based on attention distinctions to enhance the model's temporal resolution capability, thereby improving its temporal understanding logic consistency. Experimental results demonstrate that our method significantly enhances the temporal logic consistency of Video-LLMs. Further analyses reveal that our method indeed improves the temporal discriminability of attention heads, validating our conclusions. Additionally, our method even achieves performance improvements in general video temporal grounding tasks, suggesting that temporal logic consistency is an important factor in temporal understanding.