mlsanityData to 14 Oct 2026

Mitigating Cross-Lingual Cultural Inconsistencies in LLMs via Consensus-Driven Preference Optimisation

arXiv 2605.12515 ↗First posted 2 Apr 2026cs.CLcited by 1

On the map: LLM alignment / RLHF / preference optimisation → Human values and AI feedback alignment

Abstract (arXiv)

Despite their impressive capabilities, multilingual large language models (MLLMs) frequently exhibit inconsistent behaviour when the prompt's language changes. While such adaptation is generally desirable, it becomes a critical failure when a user's identity is explicitly defined. For instance, given a fixed British persona and an ambiguous everyday knowledge query about literature, the prompt's language frequently overwrites the system persona -- yielding Shakespeare in English but Cervantes in Spanish. To robustly quantify this Cross-lingual Cultural Inconsistency, we introduce Singleton Fleiss's kappaS, a metric mathematically resilient to hallucinations. For mitigation, we propose Cross-lingual Cultural Consistent Preference Optimisation (C-3PO), a consensus-driven alignment framework. C-3PO achieves up to a 0.13-point absolute increase in kappaS over unaligned models, consistently outperforming strong prompting and representation steering baselines whilst preserving explicit user identities, cultural neutrality and intrinsic cultural knowledge. Empirical evaluations demonstrate this inconsistency disproportionately affects lower-resource languages like Indonesian and Persian. Finally, early decoding of intermediate layers reveals that MLLMs implicitly personalise outputs towards the prompt language's stereotypical culture as forward-pass representations stabilise.

Key facts. Cited by 1 paper in the map (data through 14 Oct 2026). Placed in Human values and AI feedback alignment, one of 10 directions in LLM alignment / RLHF / preference optimisation. Builds on 21 of its 32 references that are in the map. First posted on arXiv 2 Apr 2026.

Builds on

21 of 32 references.

References · 21

  1. 2023Direct Preference Optimization: Your Language Model is Secretly a Reward Modelcited by 8,065
  2. 2024Gemma 2: Improving Open Language Models at a Practical Sizecited by 1,998
  3. 2023Okapi: Instruction-tuned Large Language Models in Multiple Languages with Reinforcement Learning from Human Feedbackcited by 190
  4. 2024BLEnD: A Benchmark for LLMs on Everyday Knowledge in Diverse Cultures and Languagescited by 173
  5. 2023Few-shot Fine-tuning vs. In-context Learning: A Fair Comparison and Evaluationcited by 136
  6. 2023Cross-Lingual Consistency of Factual Knowledge in Multilingual Language Modelscited by 118
  7. 2024Survey of Cultural Awareness in Language Models: Text and Beyondcited by 103
  8. 2025Lost in Multilinguality: Dissecting Cross-lingual Factual Inconsistency in Transformer Language Modelscited by 44
  9. 2024Who's asking? User personas and the mechanics of latent misalignmentcited by 40
  10. 2025Multilingual Machine Translation with Open Large Language Models at Practical Scale: An Empirical Studycited by 38
  11. 2024Does Mapo Tofu Contain Coffee? Probing LLMs for Food-related Cultural Knowledgecited by 29
  12. 2025Reading Between the Prompts: How Stereotypes Shape LLM's Implicit Personalizationcited by 24
9 more
  1. 2024Beneath the Surface of Consistency: Exploring Cross-lingual Knowledge Representation Sharing in LLMscited by 21
  2. 2025Presumed Cultural Identity: How Names Shape LLM Responsescited by 21
  3. 2025Localized Cultural Knowledge is Conserved and Controllable in Large Language Modelscited by 20
  4. 2025LLMs and Cultural Values: the Impact of Prompt Language and Explicit Cultural Framingcited by 17
  5. 2025Disentangling Language and Culture for Evaluating Multilingual Large Language Modelscited by 16
  6. 2025Aligning LLMs for Multilingual Consistency in Enterprise Applicationscited by 11
  7. 2025AlignX: Advancing Multilingual Large Language Models with Multilingual Representation Alignmentcited by 6
  8. 2025Enhancing LLM Language Adaption through Cross-lingual In-Context Pre-trainingcited by 4
  9. 2024Beyond Dataset Creation: Critical View of Annotation Variation and Bias Probing of a Dataset for Online Radical Content Detectioncited by 3

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2026: 1.

1 paper.

  1. 2026Lower-Resource, Higher Scores: Language Bias in LLM Evaluatorscited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2026Inference-Time Steering for Cross-Lingual Factual Consistency in LLMs
  2. 2026DiSCo: A Distribution-First Steering and Cultural Prior Evaluation Framework for Measuring Cultural Preference Bias in LLMs
  3. 2026Prompt Programming for Cultural Bias and Alignment of Large Language Models
  4. 2025The American Ghost in the Machine: How language models align culturally and the effects of cultural prompting
  5. 2026Cross-Lingual Consensus: Aligning Multilingual Cultural Knowledge via Multilingual Self-Consistency
  6. 2025CM-Align: Consistency-based Multilingual Alignment for Large Language Models
  7. 2025Whispers of Many Shores: Cultural Alignment through Collaborative Cultural Expertise
  8. 2026Mitigating Cultural Bias in LLMs via Multi-Agent Cultural Debate