Tag
This paper introduces REGARD, a study using Valence-Arousal-Dominance profiling to measure affective framing differences across LLMs on post-Soviet entities, revealing that models cluster by emotional intensity and generic-answer rate rather than origin or size.
Introduces PLURAL, a large-scale preference dataset grounded in the Integrated Values Survey across 92 countries, with ~500,000 preference triplets from 20 diverse countries, aimed at improving cultural value alignment in LLMs.
Analysis of 2.6 billion sketches from 236 countries reveals hidden cultural variation in how common concepts are visually represented, showing that visual imagery preserves rich semantic and cultural structure that language models compress.
A study comparing the worldviews of various AI models to those of 88 different countries, analyzing cultural and ideological alignments.
This paper proposes a framework for probing and steering latent cultural values in LLMs using scenario-based behavioral dilemmas and activation steering, applied across three models and four cultures, finding steerability variation and latent entanglement between cultural dimensions.
A new study reveals that Italian and Dutch adults instinctively adapt their hand gestures in similar ways when teaching children, suggesting a shared communicative strategy across cultures.
This paper explores methods for adapting large language models to cultural contexts in political discourse, aiming to improve cross-cultural understanding and reduce bias.
XL-SafetyBench is a benchmark of 5,500 test cases across 10 country-language pairs to evaluate LLM safety and cultural sensitivity, distinguishing jailbreak robustness from cultural awareness.
Researchers from Tianjin University and Alibaba Group propose EA-RLVR, a reinforcement learning framework with verifiable rewards that improves cross-cultural entity translation in LLMs by activating parametric knowledge already encoded during pre-training, without relying on external knowledge bases. Training on 7k samples boosts Qwen3-14B's entity translation accuracy from 23.66% to 31.87% on unseen entities.
Research paper examining how large language models express social emotions compared to human cultural norms, finding systematic misalignment where LLMs show inconsistent patterns of engaging vs. disengaging emotion expressivity across cultural personas (European American and Latin American) compared to human responses.