ValueGround: Evaluating Culture-Conditioned Visual Value Grounding in MLLMs
This addresses the gap in evaluating cultural values in models beyond text-only settings, providing a controlled testbed for cross-modal transfer, though it is incremental as it builds on existing World Values Survey data.
The paper tackles the problem of evaluating whether multimodal large language models (MLLMs) can ground culture-conditioned judgments when response options are visualized, introducing the ValueGround benchmark, and finds that average accuracy drops from 72.8% in text-only settings to 65.8% when options are visualized across six MLLMs and 13 countries.
Cultural values are expressed not only through language but also through visual scenes and everyday social practices. Yet existing evaluations of cultural values in language models are almost entirely text-only, making it unclear whether models can ground culture-conditioned judgments when response options are visualized. We introduce ValueGround, a benchmark for evaluating culture-conditioned visual value grounding in multimodal large language models (MLLMs). Built from World Values Survey (WVS) questions, ValueGround uses minimally contrastive image pairs to represent opposing response options while controlling irrelevant variation. Given a country, a question, and an image pair, a model must choose the image that best matches the country's value tendency without access to the original response-option texts. Across six MLLMs and 13 countries, average accuracy drops from 72.8% in the text-only setting to 65.8% when options are visualized, despite 92.8% accuracy on option-image alignment. Stronger models are more robust, but all remain prone to prediction reversals. Our benchmark provides a controlled testbed for studying cross-modal transfer of culture-conditioned value judgments.