Search papers, labs, and topics across Lattice.
The paper introduces AlignCultura, a two-stage pipeline for creating and benchmarking culturally aligned LLMs based on UNESCO's cultural taxonomy. The first stage constructs CULTURAX, a new HHH dataset with 1,500 samples across 30 cultural subdomains, using query construction and two-stage rejection sampling to ensure cultural grounding and minimize data leakage. Benchmarking on CULTURAX reveals that culturally fine-tuned models achieve a 4-6% improvement in joint HHH, an 18% reduction in cultural failures, and 10-12% efficiency gains compared to general-purpose models.
Fine-tuning on a new UNESCO-aligned cultural dataset boosts LLM helpfulness, harmlessness, and honesty by up to 6% while slashing cultural faux pas by nearly a fifth.
Cultural alignment in Large Language Models (LLMs) is essential for producing contextually aware, respectful, and trustworthy outputs. Without it, models risk generating stereotyped, insensitive, or misleading responses that fail to reflect cultural diversity w.r.t Helpful, Harmless, and Honest (HHH) paradigm. Existing benchmarks represent early steps toward cultural alignment; yet, no benchmarks currently enables systematic evaluation of cultural alignment in line with UNESCO's principles of cultural diversity w.r.t HHH paradigm. Therefore, to address this gap, we built Align-Cultura, two-stage pipeline for cultural alignment. Stage I constructs CULTURAX, the HHH-English dataset grounded in the UNESCO cultural taxonomy, through Query Construction, which reclassifies prompts, expands underrepresented domains (or labels), and prevents data leakage with SimHash. Then, Response Generation pairs prompts with culturally grounded responses via two-stage rejection sampling. The final dataset contains 1,500 samples spanning 30 subdomains of tangible and intangible cultural forms. Stage II benchmarks CULTURAX on general-purpose models, culturally fine-tuned models, and open-weight LLMs (Qwen3-8B and DeepSeek-R1-Distill-Qwen-7B). Empirically, culturally fine-tuned models improve joint HHH by 4%-6%, reduce cultural failures by 18%, achieve 10%-12% efficiency gains, and limit leakage to 0.3%.
