The Abstraction Gap in Vision-Language Causal Reasoning
On the map: LLM reasoning → Multimodal reasoning
Abstract (arXiv)
Vision-language models (VLMs) generate fluent causal explanations, but current evaluations cannot distinguish linguistic plausibility from faithful causal reasoning. We introduce a dual-probe methodology that isolates these properties. The Text-Only Probe measures linguistic quality. The Chain-Text Probe requires models to first generate explicit causal chains. The Abstraction Gap (AG) metric quantifies the normalized performance difference. Evaluating eight VLMs on CAGE (Causal Abstraction Gap Evaluation), a benchmark of 49,500 questions across 5,500 images spanning Pearl's causal hierarchy, we find seven models exhibit AG exceeding 0.50 with text scores of 6--8 but chain scores below 2.5. Fine-tuning on 45,000 chain-annotated examples fails to close the gap. However, one model achieves near-zero AG. The capability exists within current VLM architectures and depends on pretraining and architectural choices. CAGE provides a diagnostic tool for assessing faithful causal reasoning in VLMs.
Key facts. No citing papers in the map yet (data through 14 Oct 2026). Placed in Multimodal reasoning, one of 17 directions in LLM reasoning. Builds on 15 of its 80 references that are in the map. First posted on arXiv 27 May 2026.
Builds on
15 of 80 references.
Closest ancestors · 5
Same direction, newest first
- 2025MM-CoT:A Benchmark for Probing Visual Chain-of-Thought Reasoning in Multimodal Models9 Dec
- 2025InfoCausalQA:Can Models Perform Non-explicit Causal Reasoning Based on Infographic?8 Aug
- 2025CausalVQA: A Physically Grounded Causal Reasoning Benchmark for Video Models11 Jun
- 2024LLaVA-CoT: Let Vision Language Models Reason Step-by-Step15 Nov
- 2023CausalVLR: A Toolbox and Benchmark for Visual-Linguistic Causal Reasoning30 Jun
Other references · 10
- 2025Gemini 2.5: Pushing the Frontier with Advanced Reasoning, Multimodality, Long Context, and Next Generation Agentic Capabilitiescited by 3,796
- 2025Qwen3-VL Technical Reportcited by 2,793
- 2023Qwen-VL: A Versatile Vision-Language Model for Understanding, Localization, Text Reading, and Beyondcited by 2,224
- 2023AMBER: An LLM-free Multi-dimensional Benchmark for MLLMs Hallucination Evaluationcited by 300
- 2024CogVLM2: Visual Language Models for Image and Video Understandingcited by 217
- 2019On the Measure of Intelligencecited by 144
- 2022VL-CheckList: Evaluating Pre-trained Vision-Language Models with Objects, Attributes and Relationscited by 114
- 2024Faithfulness vs. Plausibility: On the (Un)Reliability of Explanations from Large Language Modelscited by 105
- 2025MM-IQ: Benchmarking Human-Like Abstraction and Reasoning in Multimodal Modelscited by 29
- 2025TimeCausality: Evaluating the Causal Ability in Time Dimension for Vision Language Modelscited by 9
Built on by
Not cited yet.
Similar, not cited
Close in wording, no citation either way
- 2025What's Missing in Vision-Language Models? Probing Their Struggles with Causal Order Reasoning
- 2025ISO-Bench: Benchmarking Multimodal Causal Reasoning in Visual-Language Models through Procedural Plans
- 2025CausalVLBench: Benchmarking Visual Causal Reasoning in Large Vision-Language Models
- 2026Diagnosing Causal Reasoning in Vision-Language Models via Structured Relevance Graphs
- 2026Evaluating Explanation-Driven Vision-Language Reasoning via Generation Order Interventions
- 2024Multimodal Causal Reasoning Benchmark: Challenging Vision Large Language Models to Discern Causal Links Across Modalities
- 2026Causal Scaffolding for Physical Reasoning: A Benchmark for Causally-Informed Physical World Understanding in VLMs
- 2026From Prompts to Tokens: Internalizing Causal Supervision in Vision-Language Model for Multi-Image Causal Reasoning