Distributional Open-Ended Evaluation of LLM Cultural Value Alignment Based on Value Codebook

As LLMs are globally deployed, aligning their cultural value orientations is critical for safety and user engagement. However, existing benchmarks face the Construct-Composition-Context (C$^3$) challenge: relying on discriminative, multiple-choice formats that probe value knowledge rather than true orientations, overlook subcultural heterogeneity, and mismatch with real-world open-ended generation. We introduce DOVE, a distributional evaluation framework that directly compares human-written text distributions with LLM-generated outputs. DOVE utilizes a rate-distortion variational optimization objective to construct a compact value codebook from 10K documents, mapping text into a structured value space to filter semantic noise. Alignment is measured using unbalanced optimal transport, capturing intra-cultural distributional structures and subgroup diversity. Experiments across 12 LLMs show that DOVE achieves superior predictive validity, attaining a 31.56% correlation with downstream tasks, while maintaining high reliability with as few as 500 samples per culture.

“Should I prioritize my family or my personal goals?” People from different cultures may answer this question differently, revealing the values that shape how they think and communicate. As LLMs are used worldwide, it is important to know whether they appropriately reflect such cultural nuances. Yet most evaluations still rely on rigid surveys, asking models to rate statements like “How important is family to you on a scale of 1 to 5?” These surveys miss how people usually interact with AI: through open-ended writing, where values emerge indirectly. We introduce DOVE, a framework for evaluating cultural alignment in open-ended writing rather than fixed-choice answers. DOVE identifies culturally meaningful value codes, such as independence, duty, or personal ambition, from real-world texts and compares how often those value codes appear in human-written texts and AI-generated responses. Across 12 LLMs and four cultures, DOVE predicted culturally related model behavior more accurately than existing methods and remained reliable with relatively small datasets.