Lune

ACL Findings2026

Distributional Alignment for Large Language Models under Domain Shift

Viet Thanh Pham, Lizhen Qu, Zhuang Li, Gholamreza Haffari

Abstract

Distributional alignment enables large language models (LLMs) to predict how a target population distributes its responses across answer options, rather than collapsing disagreement into a single consensus answer. However, existing LLM-based distribution prediction is often unstable and degrades under cultural and domain shift. Token score-based estimates can change with minor option wording or formatting, response sampling-based estimates are expensive and sensitive to prompts and decoding settings, and directly generated distributions are frequently miscalibrated.

We propose Evi-DA 1 , an evidence-based alignment technique that improves the fidelity and robustness of LLM-based distribution estimation under domain and cultural shift. Given a target country and a multiple-choice question, Evi-DA retrieves related World Values Survey items and their answer distributions, predicts a coarse Welzel value signature for each option, and infers the country-conditioned answer distribution in a structured format. We train the LLMs using a two-stage pipeline, where reinforcement learning optimizes survey-derived rewards that encourage accurate intermediate value predictions, faithful final distributions, well-formed structured outputs, and reduced cultural bias. Across in-domain and out-ofdomain benchmarks and multiple open-source backbones, Evi-DA reduces Jensen-Shannon divergence between predicted and gold distributions relative to strong baselines, with average relative improvements of up to 44%.

Ask Lune

Know this paper in minutes, not an afternoon.

Lune can read every page of it and the top-tier papers that cite it, and gives you its claims, its evidence and its limits, each one cited to the paper it comes from. No skimming, no AI slop.

Questions to start from

Lune can draw on

the full paper · related work

Get the answer

Free to start. No credit card required.

Lineage

  • Modular Pluralism: Pluralistic Alignment via Multi-LLM Collaboration cites Whose Opinions Do Language Models Reflect?
Builds onCited byLeft to right by year; the more cited, the higher.

Builds on8

  1. Whose Opinions Do Language Models Reflect?Shibani Santurkar, Esin Durmus, Faisal Ladhak, Cinoo Lee et al.ICML 2023 · 764 citations
  2. Verbalized Sampling: How to Mitigate Mode Collapse and Unlock LLM DiversityJiayi Zhang, Simon Yu, Derek Chong, Anthony Sicilia et al.ICML 2026 · 102 citations
  3. Modular Pluralism: Pluralistic Alignment via Multi-LLM CollaborationShangbin Feng, Taylor Sorensen, Yuhan Liu, Jillian Fisher et al.EMNLP 2024 · 12 citations
  4. SurveyPilot: an Agentic Framework for Automated Human Opinion Collection from Social MediaViet Thanh Pham, Lizhen Qu, Zhuang Li, Suraj Sharma et al.ACL 2025 · 4 citations
  5. Proverbs Run in Pairs: Evaluating Proverb Translation Capability of Large Language ModelMinghan Wang, Viet-Thanh Pham, Farhad Moghimifar, Thuy-Trang VuACL Findings 2025 · 4 citations

Related papers

  1. Distribution Shift Alignment Helps LLMs Simulate Survey Response DistributionsJi Huang, Mengfei Li, Shuai ShaoACL Findings 2026
  2. Beyond Marginal Distributions: A Framework to Evaluate the Representativeness of Demographic-Aligned LLMsTristan Williams, Franziska Weeber, Sebastian Padó, Alan AkbikACL Findings 2026
  3. Leveraging robust optimization for llm alignment under distribution shiftsMingye Zhu, Yi Liu, Zheren Fu, Yongdong Zhang et al.NeurIPS 2025 · 2 citations
  4. Robust LLM Alignment via Distributionally Robust Direct Preference OptimizationZaiyan Xu, Sushil Vemuri, Kishan Panaganti, Dileep Kalathil et al.NeurIPS 2025 · 18 citations
  5. ALIGN: Word Association Learning for Cultural Alignment in Large Language ModelsChunhua Liu, Kabir Manandhar Shrestha, Sukai HuangACL 2026
Dusk over the sea between two cliffs drawn in fine vertical lines