CultureGuard: Towards Culturally-Aware Dataset and Guard Model for Multilingual Safety Applications

2025-08-03

AI Summary

To address the challenges of insufficient cultural adaptation for non-English languages and scarcity of high-quality annotated data in multilingual large language model (LLM) content safety, this paper proposes a four-stage culturally aligned synthetic data generation and filtering framework to extend English safety datasets to eight non-English languages. Methodologically, it integrates cultural data isolation, culturally adaptive prompting, controllable machine translation, and multi-dimensional quality filtering, followed by efficient LoRA-based fine-tuning to build a multilingual safety guardian model. Key contributions include: (1) releasing the first culture-aware multilingual safety dataset covering nine languages with 386K high-quality samples; and (2) training an 8B-parameter multilingual safety model that achieves state-of-the-art performance on cross-lingual moderation benchmarks, significantly narrowing the safety capability gap across languages.

Technology Category

• Natural Language Processing: Safety and Robustness
• Machine Learning: Large Multimodal Models (LMMs)
• Philosophy and Ethics of AI: Safety, Robustness & Trustworthiness

Application Category

• Search and Retrieval-Augmented AI: Multilingual and cross-lingual Web search
• Web Mining and Content Analysis: Mining multimedia, multimodal, multilingual, cross-lingual Web data
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation

Abstract

The increasing use of Large Language Models (LLMs) in agentic applications highlights the need for robust safety guard models. While content safety in English is well-studied, non-English languages lack similar advancements due to the high cost of collecting culturally aligned labeled datasets. We present CultureGuard, a novel solution for curating culturally aligned, high-quality safety datasets across multiple languages. Our approach introduces a four-stage synthetic data generation and filtering pipeline: cultural data segregation, cultural data adaptation, machine translation, and quality filtering. This pipeline enables the conversion and expansion of the Nemotron-Content-Safety-Dataset-V2 English safety dataset into eight distinct languages: Arabic, German, Spanish, French, Hindi, Japanese, Thai, and Chinese. The resulting dataset, Nemotron-Content-Safety-Dataset-Multilingual-v1, comprises 386,661 samples in 9 languages and facilitates the training of Llama-3.1-Nemotron-Safety-Guard-Multilingual-8B-v1 via LoRA-based fine-tuning. The final model achieves state-of-the-art performance on several multilingual content safety benchmarks. We also benchmark the latest open LLMs on multilingual safety and observe that these LLMs are more prone to give unsafe responses when prompted in non-English languages. This work represents a significant step toward closing the safety gap in multilingual LLMs by enabling the development of culturally aware safety guard models.

Problem

• Addressing lack of multilingual safety datasets for LLMs
• Developing culturally-aware safety models for non-English languages
• Improving content safety benchmarks across diverse languages

Innovation

• Four-stage synthetic data generation pipeline
• LoRA-based fine-tuning for multilingual safety
• Culturally aligned dataset expansion to 9 languages