README.md

---
license: cc-by-nc-4.0
language:
- en
pretty_name: SubgroupValuesSG model outputs
tags:
- cultural-alignment
- persona
- llm-as-a-judge
---

# SubgroupValuesSG model outputs

Model outputs for **"Can Persona-Prompted LLMs Emulate Subgroup Values? An Empirical Analysis of Generalisability and Fairness in Cultural Alignment"** (ACL 2026).

[Paper (ACL Anthology)](https://aclanthology.org/2026.acl-long.1127/) · [Code](https://github.com/Social-AI-Studio/SubgroupValuesSG)

Answers of the 18 evaluated models (seven open-source models before and after SFT, and four GPT models) to the persona-prompted survey questions, and the LLM judge's verdicts on their open-ended answers. Together with the code repository, which recreates the dataset from the official WVS file, they reproduce every table in the paper on a CPU. The judge's reasoning is omitted; only each verdict's final JSON is kept.

| Path | Contents |
|---|---|
| `results/raw_inference/<model>/<model>_<split>_{numerical,open-ended}_raw.json` | Answers: `[{"custom_id": <sample_id>, "response": <text>}]` |
| `results/raw_inference/<judge>__judge/<judge>_judge_<model>_vs_gpt-4.1-2025-04-14_<split>_pos{1,2}_raw.json` | Judge verdicts (persona, value, overall) in both answer orders |

Download into a clone of the code repository:

```bash
hf download Incomple/SubgroupValuesSG-outputs --repo-type dataset --local-dir . --include "results/*"
```

## Licence

CC BY-NC 4.0. The terms of the evaluated models' providers and of the judge ([RedHatAI/Mistral-Small-3.1-24B-Instruct-2503-quantized.w8a8](https://huggingface.co/RedHatAI/Mistral-Small-3.1-24B-Instruct-2503-quantized.w8a8)) also apply.

## Citation

```bibtex
@inproceedings{tanCanPersonaPromptedLLMs2026,
  title = {Can {{Persona-Prompted LLMs Emulate Subgroup Values}}? {{An Empirical Analysis}} of {{Generalisability}} and {{Fairness}} in {{Cultural Alignment}}},
  shorttitle = {Can {{Persona-Prompted LLMs Emulate Subgroup Values}}?},
  booktitle = {Proceedings of the 64th {{Annual Meeting}} of the {{Association}} for {{Computational Linguistics}} ({{Volume}} 1: {{Long Papers}})},
  author = {Tan, Bryan Chen Zhengyu and Liu, Zhengyuan and Yi, Xiaoyuan and Yao, Jing and Xie, Xing and Chen, Nancy F. and Lee, Roy Ka-Wei},
  editor = {Liakata, Maria and Moreira, Viviane P. and Zhang, Jiajun and Jurgens, David},
  year = 2026,
  month = jul,
  pages = {24571--24590},
  publisher = {Association for Computational Linguistics},
  address = {San Diego, California, United States},
  doi = {10.18653/v1/2026.acl-long.1127},
  urldate = {2026-09-14},
  abstract = {Despite their global prevalence, many Large Language Models (LLMs) are aligned to a monolithic, often Western-centric set of values. This paper investigates the more challenging task of fine-grained value alignment: examining whether LLMs can emulate the distinct cultural values of demographic subgroups. Using Singapore as a case study and the World Values Survey (WVS), we examine the value landscape and show that even state-of-the-art models like GPT-4.1 achieve only 57.4\% accuracy in predicting subgroup modal preferences. We construct a dataset of over 20,000 samples to train and evaluate a range of models. We demonstrate that simple fine-tuning on structured numerical preferences yields substantial gains, improving accuracy on unseen, out-of-distribution subgroups by an average of 17.4\%. These gains partially transfer to open-ended generation. However, we find significant pre-existing performance biases, where models better emulate young, male, Chinese, and Christian personas. Furthermore, while fine-tuning improves average performance, it widens the disparity between subgroups when measured by distance-aware metrics. Our work offers insights into the limits and fairness implications of subgroup-level cultural alignment.},
  isbn = {979-8-89176-390-6},
}
```