Papers with value alignment
Do language models practice what they preach? Examining language ideologies about gendered language reform encoded in LLMs (2025.coling-main)
Copied to clipboard
| Challenge: | Language ideologies are evaluative ideas or beliefs about language, such as ideas about what is "correct", "natural" or "articulate". |
| Approach: | They use gender-neutral variants more often when more explicit metalinguistic context is provided. |
| Outcome: | The findings show that language ideologies in LLMs can vary, which may be unexpected to users. |
ALIGN: Word Association Learning for Cultural Alignment in Large Language Models (2026.acl-long)
Copied to clipboard
| Challenge: | Large language models exhibit cultural bias from over-represented viewpoints in training data, yet cultural alignment remains a challenge due to limited cultural knowledge and a lack of exploration into effective learning approaches. |
| Approach: | They propose a cost-efficient method for fine-tuning large language models on native speakers’ word-association norms and a preference optimization method to improve cultural alignment. |
| Outcome: | The proposed model trains Llama-3.1-8B and Qwen-2.5-7B on native speakers’ word-association norms and shows that such associations capture cultural knowledge. |
Beyond Marginal Distributions: A Framework to Evaluate the Representativeness of Demographic-Aligned LLMs (2026.findings-acl)
Copied to clipboard
| Challenge: | Existing work on marginal distributions and model steering fails to account for deeper latent structures that characterise real populations. |
| Approach: | They propose a framework for evaluating the representativeness of aligned models through multivariate correlation patterns in addition to marginal distributions. |
| Outcome: | The proposed framework compares two model steering techniques against human responses from the World Values Survey. |
CONTRANS: Weak-to-Strong Alignment Engineering via Concept Transplantation (2025.coling-main)
Copied to clipboard
| Challenge: | Large language models behave consistently with human goals, values and intentions, but are computationally expensive. |
| Approach: | They propose a framework that enables weak-to-strong alignment transfer via concept transplantation. |
| Outcome: | The proposed framework surpasses instruction-tuned models in terms of truthfulness. |
Understanding How Value Neurons Shape the Generation of Specified Values in LLMs (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Current approaches to interpret value representations are limited by superficial judgments over mechanistic analysis. |
| Approach: | They propose a mechanistic interpretability framework that uses the Schwartz Values Survey to interpret value . they use a dataset that operationalizes four dimensions of universal value through behavioral contexts . |
| Outcome: | The proposed method bridges psychological value frameworks with neuron analysis in large language models. |
One fish, two fish, but not the whole sea: Alignment reduces language models’ conceptual diversity (2025.naacl-long)
Copied to clipboard
| Challenge: | Existing studies suggest large language models can capture certain behavioral patterns, but there are ongoing debates as to whether they are valid replacements for human subjects. |
| Approach: | They propose to use large language models as replacements for humans in behavioral research by relating the internal variability of simulated individuals to the population-level variability. |
| Outcome: | The proposed model can capture human-like conceptual diversity, but it is unclear whether post-training alignment affects models’ internal diversity. |
MPTA: MultiTask Personalization Assessment (2025.findings-emnlp)
Copied to clipboard
| Challenge: | MTPA tests large language models on real personas spanning demographics, beliefs, and values . aggregate metrics suggest models are truthful and safe, subgroup-specific evaluations reveal hidden pockets of degraded factuality, fairness disparities, and inconsistent value alignment. |
| Approach: | a benchmark is a tool that leverages large-scale survey data to construct real personas . they show persona conditioning exposes pluralistic misalignment . |
| Outcome: | MTPA conditions models on real personas and tests their behavior across alignment tasks. |
DEFT: Distribution-guided Efficient Fine-Tuning for Human Alignment (2024.findings-emnlp)
Copied to clipboard
| Challenge: | Experimental results show that the methods enhanced by DEFT outperform the original methods in both alignment capability and generalization ability, with significantly reduced training time. |
| Approach: | They propose a distribution-based alignment framework that integrates data filtering and distributional guidance to improve alignment efficiency and generalization ability. |
| Outcome: | The proposed framework outperforms existing methods in alignment capability and generalization ability with significantly reduced training time. |
Analyzing values about gendered language reform in LLMs’ revisions (2025.emnlp-main)
Copied to clipboard
| Challenge: | In the past years, LLMs have been used in everyday tasks, especially the formulation and revision of text. |
| Approach: | They examine LLMs' revision of gendered role nouns and their justifications using a prompt set-up to examine their alignment with feminist and trans-inclusive language reforms for English. |
| Outcome: | The proposed revision choices are based on the literature and empirical evidence. |
V-RoLoRA: RLVR-Driven MoE Routing for Steerable Pluralistic Alignment (2026.findings-acl)
Copied to clipboard
| Challenge: | Current methods for steering large language models rely on prompt engineering or reasoning-time guidance. |
| Approach: | They propose a value-controllable pluralistic alignment framework enhanced with conditioned gating that dynamically directs the flow among multiple experts based on an input value or moral vector. |
| Outcome: | The proposed method outperforms prompt-based steering and multi-task PEFT benchmarks on two 8-billion-parameter backbones. |
Unintended Harms of Value-Aligned LLMs: Psychological and Empirical Insights (2025.acl-long)
Copied to clipboard
| Challenge: | Value-aligned LLMs are more prone to harmful behavior than fine-tuned models . value-aligned models generate text according to the aligned values, which can amplify harmful outcomes. |
| Approach: | They propose to use in-context alignment methods to enhance the safety of value-aligned LLMs. |
| Outcome: | The proposed methods improve value alignment and safety, the authors say . value-aligned models are more prone to harmful behavior than fine-tuned models . |
VALUE ALIGNMENT TAX: Measuring Value Trade-offs in LLM Alignment (2026.findings-acl)
Copied to clipboard
| Challenge: | Existing work on value alignment characterizes value relations statically, ignoring how interventions reshape the value system. |
| Approach: | They propose a framework that quantifies value trade-offs by measuring how alignment-induced changes propagate across interconnected values relative to achieved on-target gain. |
| Outcome: | The proposed framework measures how value trade-offs propagate across values . it can be used to evaluate intended improvements and unintended side effects . |