Culturally-Grounded Chain-of-Thought (CG-CoT):Enhancing LLM Performance on Culturally-Specific Tasks in Low-Resource Languages

2025-06-01

AI Summary

Large language models (LLMs) exhibit poor performance on culture-specific reasoning tasks in low-resource languages—e.g., Yoruba proverb comprehension—hindering their equitable global deployment. To address this, we propose Cultural Chain-of-Thought (C-CoT), a novel prompting framework that integrates dense vector retrieval for localized cultural context acquisition, explicit chain-of-thought reasoning to guide culturally grounded inference, and a dual verification mechanism combining LLM self-assessment with human-in-the-loop validation. Experiments demonstrate substantial gains in cultural alignment accuracy and reasoning depth. Crucially, we uncover a fundamental misalignment between conventional translation metrics (e.g., BLEU) and cultural relevance evaluation, thereby advocating a paradigm shift in low-resource NLP assessment. To our knowledge, this is the first work to systematically unify cultural retrieval, structured reasoning, and multi-tiered validation to enhance LLMs’ cultural intelligence.

Technology Category

• Natural Language Processing: (Large) Language Models
• Machine Learning: Large Multimodal Models (LMMs)
• Computer Vision: Language and Vision

Application Category

• Search and Retrieval-Augmented AI: Multilingual and cross-lingual Web search
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions

Abstract

Large Language Models (LLMs) struggle with culturally-specific reasoning tasks, particularly in low-resource languages, hindering their global applicability. Addressing this gap is crucial for equitable AI deployment. We introduce Culturally-Grounded Chain-of-Thought (CG-CoT), a novel prompting strategy that combines dense vector retrieval of cultural context with explicit reasoning sequences. Our extensive experiments on Yoruba proverb interpretation demonstrate that CG-CoT provides significantly higher culturally-aligned accuracy and depth than traditional prompting methods, validated through both automated metrics and LLM-based evaluations. Notably, we uncover stark disparities between token-level translation metrics like BLEU and human-judged cultural relevance, suggesting a rethinking of evaluation approaches for low-resource NLP.

Problem

• Improving LLM performance on culturally-specific tasks in low-resource languages
• Addressing disparities in cultural relevance evaluation metrics for NLP
• Enhancing reasoning accuracy with culturally-grounded prompting strategies

Innovation

• Dense vector retrieval of cultural context
• Explicit reasoning sequences in prompts
• Culturally-aligned accuracy enhancement