Translating physics concept inventories using Large Language Models

This study explores the use of artificial intelligence in translating and validating physics concept inventories across multiple languages, with a specific focus on the Force Concept Inventory (FCI). Although the FCI has been translated by expert physicists into many different languages, many physics concept assessments remain unavailable in languages other than English, creating barriers to global physics education research. We discuss the challenges that arise when machine-transcribing and translating physics concept inventories. Subsequently, we analyze how an AI large language model performs for translated versions in fifty different languages. While acknowledging that human physics experts with appropriate language skills remain essential for full validation, we demonstrate that AI can provide preliminary insights into translation validity. We analyze how formatting issues, special characters, equations, figures, and connected question sequences affect machine translations. Our findings offer practical guidance for physics educators and researchers seeking to use AI for translating educational content and highlight both the limitations and opportunities of machine translation in physics education. This research contributes to broader conversations about AI's role in making STEM education more globally accessible while maintaining conceptual accuracy across linguistic boundaries.