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

TL;DR

This paper introduces DOVE, a novel open-ended evaluation framework for assessing LLMs' cultural value alignment by comparing distributions of human and machine-generated texts within a structured value space.

Contribution

DOVE is the first distributional evaluation method that directly compares text distributions using a value codebook and optimal transport, addressing limitations of existing benchmarks.

Findings

DOVE achieves 31.56% correlation with downstream tasks.

It maintains high reliability with only 500 samples per culture.

DOVE outperforms existing benchmarks in predictive validity.

Abstract

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 sub-group 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.