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

Jaehyeok Lee, Xiaoyuan Yi, Jing Yao, Hyunjin Hwang, Roy Ka-Wei Lee, Xing Xie, Jinyeong Bak
Proceedings of the 43rd International Conference on Machine Learning, PMLR 306:65156-65205, 2026.

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 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.

Cite this Paper


BibTeX
@InProceedings{pmlr-v306-lee26bj, title = {Distributional Open-Ended Evaluation of {LLM} Cultural Value Alignment Based on Value Codebook}, author = {Lee, Jaehyeok and Yi, Xiaoyuan and Yao, Jing and Hwang, Hyunjin and Lee, Roy Ka-Wei and Xie, Xing and Bak, Jinyeong}, booktitle = {Proceedings of the 43rd International Conference on Machine Learning}, pages = {65156--65205}, year = {2026}, editor = {Zhang, Tong and Dudik, Miroslav and Jaggi, Martin and Agarwal, Alekh and Li, Sharon and Schuurmans, Dale and Zhu, Jerry and Berkenkamp, Felix and Dong, Hanze and Bietti, Alberto}, volume = {306}, series = {Proceedings of Machine Learning Research}, month = {06--11 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v306/main/assets/lee26bj/lee26bj.pdf}, url = {https://proceedings.mlr.press/v306/lee26bj.html}, 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 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.} }
Endnote
%0 Conference Paper %T Distributional Open-Ended Evaluation of LLM Cultural Value Alignment Based on Value Codebook %A Jaehyeok Lee %A Xiaoyuan Yi %A Jing Yao %A Hyunjin Hwang %A Roy Ka-Wei Lee %A Xing Xie %A Jinyeong Bak %B Proceedings of the 43rd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2026 %E Tong Zhang %E Miroslav Dudik %E Martin Jaggi %E Alekh Agarwal %E Sharon Li %E Dale Schuurmans %E Jerry Zhu %E Felix Berkenkamp %E Hanze Dong %E Alberto Bietti %F pmlr-v306-lee26bj %I PMLR %P 65156--65205 %U https://proceedings.mlr.press/v306/lee26bj.html %V 306 %X 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.
APA
Lee, J., Yi, X., Yao, J., Hwang, H., Lee, R.K., Xie, X. & Bak, J.. (2026). Distributional Open-Ended Evaluation of LLM Cultural Value Alignment Based on Value Codebook. Proceedings of the 43rd International Conference on Machine Learning, in Proceedings of Machine Learning Research 306:65156-65205 Available from https://proceedings.mlr.press/v306/lee26bj.html.

Related Material