CCL-XCoT: An Efficient Cross-Lingual Knowledge Transfer Method for Mitigating Hallucination Generation

2025-07-17

AI Summary

To address pervasive factual hallucinations in multilingual large language models (MLLMs) during domain-specific generation—primarily caused by imbalanced training data across languages, especially for low-resource languages—this paper proposes a two-stage fine-tuning framework. In Stage I, curriculum-based contrastive learning is introduced to strengthen cross-lingual semantic alignment; in Stage II, cross-lingual chain-of-thought (X-CoT) prompting is integrated to enhance factual consistency throughout reasoning. The method requires no external retrieval or model ensembling, relying solely on continual pretraining, next-token prediction, and structured prompting for joint optimization. Experiments demonstrate up to a 62% reduction in hallucination rates and significantly improved factual knowledge transfer to low-resource languages. This work establishes a lightweight, scalable paradigm for trustworthy multilingual generation.

Technology Category

• Machine Learning: Large Multimodal Models (LMMs)
• Natural Language Processing: Language Grounding & Multi-modal NLP
• Knowledge Representation and Reasoning: Knowledge Representation Languages

Application Category

• Search and Retrieval-Augmented AI: Multilingual and cross-lingual Web search
• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• User Modeling, Personalization and Recommendation: Large Language Models (LLM) for user modeling and recommendation

Abstract

Multilingual Large Language Models(MLLMs) demonstrate strong generalization across languages, yet they remain prone to hallucinations, especially in low-resource languages, due to training data imbalances. These hallucinations, which include inaccurate or fabricated outputs, are particularly problematic in domain-specific generation tasks (Chataigner et al., 2024). To address this challenge, we propose CCL-XCoT(Curriculum-based Contrastive Learning-based Cross-lingual Chain-of-Thought), a two-stage fine-tuning framework for mitigating hallucination in MLLMs. Our approach first enhances cross-lingual semantic alignment through curriculum-based contrastive learning combined with next-token prediction during continued pre-training. Building on this foundation, we then introduce a cross-lingual Chain-of-Thought (XCoT) prompting strategy during instruction fine-tuning, which guides the model to reason in a high-resource language before generating answers in the target low-resource language. Experimental results show that CCL-XCoT reduces hallucination rates by up to 62% and substantially improves factual knowledge transfer across language pairs, without relying on external retrieval or multi-model ensembles.

Problem

• Mitigates hallucinations in multilingual large language models
• Improves cross-lingual knowledge transfer for low-resource languages
• Enhances semantic alignment and reasoning via curriculum-based learning

Innovation

• Curriculum-based contrastive learning for semantic alignment
• Cross-lingual Chain-of-Thought prompting strategy
• Two-stage fine-tuning without external retrieval