What information is sufficient to learn the full richness of human scene understanding? The distributional hypothesis holds that the statistical co-occurrence of language and images captures the conceptual knowledge underlying visual cognition. Vision-language models (VLMs) are trained on massive paired text-image corpora but lack embodied experience, making them an ideal test of the distributional hypothesis. We report two experiments comparing descriptions generated by 18 VLMs to those of over 2000 human observers across 15 high-level scene understanding tasks, spanning general knowledge, affordances, sensory experiences, affective responses, and future prediction. Because many tasks lack ground truth answers, we developed a Human-Calibrated Cosine Distance (HCD) metric that measures VLM output similarity to the distribution of human responses, scaled by within-human variability. In Experiment 1, VLMs approached human-level performance on general knowledge tasks, but showed a robust deficit for affordance tasks that resisted prompt engineering and did not improve with newer model releases. In Experiment 2, we tested six mechanistic hypotheses for explaining this affordance gap, finding that the deficit was structural rather than stylistic and was not resolved by providing explicit spatial information. Corpus analyses revealed that image captioning datasets contain sparse agent-addressed affordance language, consistent with Gricean accounts of why embodied knowledge may be systematically underrepresented in language. Together, these findings suggest that distributional learning from images and text is insufficient for affordance-based scene understanding, implying that some dimensions of human visual cognition may require the kind of agent-centered, three-dimensional experience that no photograph or caption can encode.
Vision-language models (VLMs) learn strong statistical regularities during training, which can make them fail to properly respond to visual evidence when input images violate those regularities. It is challenging to disambiguate these prior-induced errors from errors caused by the model missing the underlying visual capabilities. In this work, we use controlled visual in-context learning to encourage the model to focus on the visual evidence, and demonstrate that this alone is frequently sufficient to correct the errors -- suggesting that the models in fact do possess the relevant visual capabilities. Specifically, we show that across five vision-centric tasks, demonstrations yield large gains on prior-conflicting counterfactual examples (+27.2% for Qwen3-VL 235B and +15.1% for Qwen3-VL 32B) while leaving accuracy on prior-aligned examples nearly unchanged. These findings generalize across five VLM families with gains from +9.7% to +30.3%. We analyze visual representations and find that demonstrations indeed redirect attention toward grounding cues while reducing reliance on cues that support the learned prior. Finally, we find that in-context learning is limited when the underlying visual capability is weak or absent. Together, these results show that current VLMs possess visual grounding capabilities that standard evaluations may overlook, underscoring the need for controlled evaluation design in diagnosing and improving model capability.
How much scene context a single object carries is a well-studied question in human scene perception, yet how this capacity is organized in vision-language models (VLMs) remains poorly understood, with direct implications for the robustness of these models. We investigate this question through a systematic behavioral and mechanistic analysis of contextual inference from single objects. Presenting VLMs with single objects on masked backgrounds, we probe their ability to infer both fine-grained scene category and coarse superordinate context (indoor vs. outdoor). We found that single objects support above-chance inference at both levels, with performance modulated by the same object properties that predict human scene categorization. Object identity, scene, and superordinate predictions are partially dissociable: accurate inference at one level neither requires nor guarantees accurate inference at the others, and the degree of coupling differs markedly across models. Mechanistically, object representations that remain stable when background context is removed are more predictive of successful contextual inference. Scene and superordinate schemas are grounded in fundamentally different ways: scene identity is encoded in image tokens throughout the network, while superordinate information emerges only late or not at all. Together, these results reveal that the organization of contextual inference in VLMs is more complex than accuracy alone suggests, with behavioral and mechanistic signatures
Martina G. Vilas, Timothy Schaumlöffel, Gemma Roig
Goethe University Frankfurt · The Hessian Center for AI
Visual inputs are often assumed to improve language understanding in multimodal models. We examine this assumption by asking whether vision-language models (VLMs) can distinguish useful visual evidence from incidental image context in lexical judgments. We use human concreteness and imagery ratings because they span words with varying expected visual relevance, from abstract and low-imagery words to concrete and high-imagery words. We find that real-image contexts do not yield consistent gains and often hurt alignment with human ratings, most sharply when visual evidence is least relevant. Through probing and canonical correlation analysis, complemented by an attribution case study, we find that real-image contexts are associated with representational shifts and greater sensitivity to spurious visual cues, coinciding with weaker recoverability of the targeted lexical properties. We further show that instructing models to focus solely on textual content at inference time can reduce this degradation, with the clearest gains on these vulnerable subsets. Our findings suggest that current instruction-tuned VLMs need better calibration of when visual context should inform lexical judgments.