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

• URL: http://arxiv.org/abs/2605.12515v1
• Date: Thu, 02 Apr 2026 14:04:06 GMT
• Title: Mitigating Cross-Lingual Cultural Inconsistencies in LLMs via Consensus-Driven Preference Optimisation
• Abstract summary: We show that multilingual large language models (MLLMs) frequently exhibit inconsistent behaviour when the prompt's language changes.<n>We propose Cross-lingual Cultural Consistent Preference optimisation (C-3PO), a consensus-driven alignment framework.<n>C-3PO achieves up to a 0.10-point absolute increase in $_S$ over unaligned models, outperforming strong prompting and representation steering baselines.
• Score: 58.01855677487771
• License: http://creativecommons.org/licenses/by/4.0/
• Abstract: 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 $κ_S$, 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 0.10-point absolute increase in $κ_S$ over unaligned models, outperforming strong prompting and representation steering baselines. Empirical evaluations show this inconsistency disproportionately affects lower-resource languages like Indonesian and Persian. A layer-wise interpretability analysis reveals the underlying mechanism: by early-decoding intermediate layer representations, we find that MLLMs implicitly personalise outputs towards the prompt language's stereotypical culture as forward-pass representations stabilise.

Related papers

• A Systematic Evaluation of Cross-Lingual Consistency Enhancement Methods in Multilingual Language Models [50.27606276000142]
  We present a unified evaluation of representative CLC-enhancement methods for question answering, spanning inference-time interventions and post-training approaches.<n>Results show that post-training methods are generally more reliable, with direct distribution alignment consistently improving CLC across all model-dataset combinations.<n>We investigate whether CLC enhancement hurts models' ability to respond differently *when needed*, that is, when asked culture-dependent questions.
  arXiv  Detail & Related papers  (2026-09-03T19:18:44Z)
• Inference-Time Steering for Cross-Lingual Factual Consistency in LLMs [51.56484100374058]
  Large Language Models (LLMs) demonstrate remarkable multilingual fluency, but their internal knowledge representations remain disproportionately biased toward high-resource languages.<n>We investigate whether these biases can be mitigated at inference time, forcing an English-prompted model to answer as if it were queried in target languages.<n>Experiments on Gemma 3 12B Instruct reveal persona prompting to be the strongest overall intervention, balancing efficacy, safety, and out-of-domain generalization.
  arXiv  Detail & Related papers  (2026-07-21T16:15:05Z)
• MSQA: A Natively Sourced Multilingual and Multicultural SimpleQA Benchmark [31.67992649991894]
  A model that can speak a user's language must also understand the culture encoded by that language.<n>We introduce MSQA, a benchmark of 1,064 natively sourced questions across 11 language groups, five cultural dimensions, and three difficulty tiers.<n>We find substantial cultural degradation and a pronounced Locality Effect.
  arXiv  Detail & Related papers  (2026-07-01T10:12:03Z)
• Cross-Lingual Consensus: Aligning Multilingual Cultural Knowledge via Multilingual Self-Consistency [1.338174941551702]
  Large Language Models (LLMs) exhibit significant performance discrepancies across languages.<n>We propose a novel self-supervised framework to bridge this knowledge gap.
  arXiv  Detail & Related papers  (2026-05-21T08:11:01Z)
• Dialect vs Demographics: Quantifying LLM Bias from Implicit Linguistic Signals vs. Explicit User Profiles [0.42970700836450476]
  Explicit identity prompts activate aggressive safety filters, increasing refusal rates and reducing semantic similarity compared to reference text for Black users.<n>In contrast, implicit dialect cues trigger a powerful dialect jailbreak,'' reducing refusal probability to near zero.<n>We find that current safety alignment techniques are brittle and over-indexed on explicit keywords, creating a bifurcated user experience.
  arXiv  Detail & Related papers  (2026-04-22T23:33:18Z)
• When Meanings Meet: Investigating the Emergence and Quality of Shared Concept Spaces during Multilingual Language Model Training [57.230355403478995]
  We investigate the development of language-agnostic concept spaces during pretraining of EuroLLM.<n>We find that shared concept spaces emerge early and continue to refine, but that alignment with them is language-dependent.<n>In contrast to prior work, our fine-grained manual analysis reveals that some apparent gains in translation quality reflect shifts in behavior.
  arXiv  Detail & Related papers  (2026-01-30T11:23:01Z)
• CLM-Bench: Benchmarking and Analyzing Cross-lingual Misalignment of LLMs in Knowledge Editing [5.137059606366328]
  We propose CLM-Bench, a culture-aware benchmark constructed using a native Chinese-first methodology.<n>We conduct extensive experiments on representative LLMs and reveal a significant Cross-lingual Misalignment.<n>Our findings challenge the effectiveness of current methods in cross-lingual transfer and underscore the importance of culturally native benchmarks.
  arXiv  Detail & Related papers  (2026-01-24T09:55:34Z)
• Beyond Instrumental and Substitutive Paradigms: Introducing Machine Culture as an Emergent Phenomenon in Large Language Models [9.785535924216765]
  This study proposes textbfMachine Culture as an emergent, distinct phenomenon.<n>We employed a 2 (Model Origin: US vs. China) $times$ 2 (Prompt Language: English vs. Chinese) factorial design across eight multimodal tasks.<n>We conclude that LLMs do not simulate human culture but exhibit an emergent Machine Culture.
  arXiv  Detail & Related papers  (2026-01-23T13:11:28Z)
• Tears or Cheers? Benchmarking LLMs via Culturally Elicited Distinct Affective Responses [28.3173238194554]
  We introduce CEDAR, a benchmark constructed entirely from scenarios capturing culturally underlinetextscElicited underlinetextscDistinct underlinetextscAffective underlinetextscResponses.<n>The resulting benchmark comprises 10,962 instances across seven languages and 14 fine-grained emotion categories, with each language including 400 multimodal and 1,166 text-only samples.
  arXiv  Detail & Related papers  (2026-01-19T13:04:26Z)
• Code-Switching In-Context Learning for Cross-Lingual Transfer of Large Language Models [64.54005959758733]
  We introduce code-switching in-context learning (CSICL) as a principled and robust approach for overcoming the translation barrier during inference.<n>We conduct extensive experiments across 4 LLMs, 6 datasets, and 10 languages, spanning both knowledge-intensive and reasoning-oriented domains.<n>Our results demonstrate CSICL consistently outperforms X-ICL baselines, achieving gains of 3.1%p and 1.9%p in both target and unseen languages.
  arXiv  Detail & Related papers  (2025-10-07T08:35:42Z)
• Which Cultural Lens Do Models Adopt? On Cultural Positioning Bias and Agentic Mitigation in LLMs [53.07843733899881]
  Large language models (LLMs) have unlocked a wide range of downstream generative applications.<n>We find that they also risk perpetuating subtle fairness issues tied to culture, positioning their generations from the perspectives of the mainstream US culture.<n>We propose 2 inference-time mitigation methods to resolve these biases.
  arXiv  Detail & Related papers  (2025-09-25T12:28:25Z)
• Can you map it to English? The Role of Cross-Lingual Alignment in Multilingual Performance of LLMs [12.334510055293535]
  Large language models (LLMs) pre-trained predominantly on English text exhibit surprising multilingual capabilities.<n>We introduce cross-lingual alignment metrics to quantify the alignment at an instance level for discriminative tasks.<n>We find that while cross-lingual alignment metrics strongly correlate with task accuracy at the language level, the sample-level alignment often fails to distinguish correct from incorrect predictions.
  arXiv  Detail & Related papers  (2025-04-13T00:01:22Z)
• Lost in Multilinguality: Dissecting Cross-lingual Factual Inconsistency in Transformer Language Models [49.16690802656554]
  We find that Multilingual factual models struggle to provide consistent responses to semantically equivalent prompts in different languages.<n>We propose a linear shortcut method that bypasses computations in the final layers, enhancing both prediction accuracy and cross-lingual consistency.
  arXiv  Detail & Related papers  (2025-04-05T19:43:10Z)
• Simple Linguistic Inferences of Large Language Models (LLMs): Blind Spots and Blinds [59.71218039095155]
  We evaluate language understanding capacities on simple inference tasks that most humans find trivial. We target (i) grammatically-specified entailments, (ii) premises with evidential adverbs of uncertainty, and (iii) monotonicity entailments. The models exhibit moderate to low performance on these evaluation sets.
  arXiv  Detail & Related papers  (2023-05-24T06:41:09Z)

This list is automatically generated from the titles and abstracts of the papers in this site.

This site does not guarantee the quality of this site (including all information) and is not responsible for any consequences.