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

TL;DR

This paper introduces CG-CoT, a prompting strategy that improves LLM performance on culturally-specific tasks in low-resource languages by integrating cultural context retrieval and explicit reasoning, demonstrated on Yoruba proverb interpretation.

Contribution

The paper presents a novel CG-CoT prompting method that enhances culturally-aware reasoning in LLMs for low-resource languages, addressing a key gap in current NLP approaches.

Findings

CG-CoT significantly improves culturally-aligned accuracy.

Disparities found between BLEU scores and human cultural relevance judgments.

Demonstrated effectiveness on Yoruba proverb interpretation.

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.