CALM introduces a technique that dramatically improves cross-lingual knowledge consistency in large language models without additional training or human supervision.

• Uses self-alignment to synchronize knowledge across different languages
• Achieves up to 23.5% improvement in cross-lingual performance
• Eliminates the need for language-specific fine-tuning
• Creates more equitable AI systems that perform consistently regardless of language

For medical applications, CALM ensures critical healthcare information remains accurate across languages, reducing potential for harmful misinformation and creating more globally accessible medical AI systems.

CALM: Unleashing the Cross-Lingual Self-Aligning Ability of Language Model Question Answering