Reducing Cultural Hallucination in Non-English Languages Via Prompt Engineering for Large Language Models
Kanato SATO, Haruto Kaneko, Mei Fujimura
Abstract
Advancements in prompt engineering offer significant potential for mitigating cultural hallucinations in large language models (LLMs). The strategic formulation of prompts, when combined with deep cultural and linguistic insights, enhances the accuracy and cultural sensitivity of LLMs, particularly in non-English contexts. This paper explores the application of prompt engineering across three major LLMs—OpenAI ChatGPT, Google Gemini, and Anthropic Claude—demonstrating how tailored prompts can effectively reduce cultural biases and improve user interaction. Through case studies and comparative analysis, the research identifies best practices and provides strategic recommendations for further development. The findings emphasize the importance of continuous innovation and ethical considerations in AI to ensure inclusivity and respect for cultural diversity in global technology applications.