Multilingual Prompting for Improving LLM Generation Diversity

2025-05-21

AI Summary

Large language models (LLMs) suffer from weak cultural representativeness and insufficient content diversity, leading to culture-related hallucinations. To address this, we propose a multilingual prompt engineering framework: starting from a base prompt, we construct cross-cultural variants that integrate multilingual and multicultural cues; responses are generated via high-temperature sampling, role prompting, and stepwise recall; and outputs are aggregated using a weighted fusion mechanism. Our work is the first to systematically demonstrate that LLMs’ implicit cultural knowledge can be activated via language-specific prompts, and that alignment between prompt language and cultural cues significantly mitigates cultural hallucinations. Experiments on GPT-4o, GPT-4o-mini, and LLaMA-70B/8B show consistent improvements over existing diversity-enhancement baselines. Performance gains correlate with both linguistic resource richness and model scale, highlighting the interplay between multilingual prompting and architectural capacity.

Technology Category

• Natural Language Processing: Prompt Engineering / Prompting
• Machine Learning: Large Multimodal Models (LMMs)
• Data Mining & Knowledge Management: Conversational Systems for Recommendation & Retrieval

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: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs

Abstract

Large Language Models (LLMs) are known to lack cultural representation and overall diversity in their generations, from expressing opinions to answering factual questions. To mitigate this problem, we propose multilingual prompting: a prompting method which generates several variations of a base prompt with added cultural and linguistic cues from several cultures, generates responses, and then combines the results. Building on evidence that LLMs have language-specific knowledge, multilingual prompting seeks to increase diversity by activating a broader range of cultural knowledge embedded in model training data. Through experiments across multiple models (GPT-4o, GPT-4o-mini, LLaMA 70B, and LLaMA 8B), we show that multilingual prompting consistently outperforms existing diversity-enhancing techniques such as high-temperature sampling, step-by-step recall, and personas prompting. Further analyses show that the benefits of multilingual prompting vary with language resource level and model size, and that aligning the prompting language with the cultural cues reduces hallucination about culturally-specific information.

Problem

• Enhancing LLM generation diversity using multilingual prompts
• Addressing cultural bias in LLM responses via linguistic cues
• Improving factual accuracy by aligning prompts with cultural contexts

Innovation

• Multilingual prompting enhances cultural diversity
• Generates varied prompts with cultural cues
• Combines responses to activate broader knowledge