Researchers at the Korea Advanced Institute of Science and Technology (KAIST) have developed BehaVERT, an artificial intelligence framework that interprets rodent movement sequences as language-like tokens. Led by Professor Dae-Soo Kim and first author Dr. Seungjae Shin, the interdisciplinary team recently published their findings in the International Journal of Computer Vision. By converting skeletal coordinates of mice into discrete behavioral tokens, the researchers trained a transformer-based architecture to analyze motor patterns without manual annotations or explicit biological guidance. The system successfully identified core social behavioral abnormalities in Shank3B knockout autism-model mice, automatically isolating reduced oral-oral contact as a defining deficit. This outcome aligns with established neuroscience literature but was derived entirely from data-driven pattern recognition. Beyond simple classification, BehaVERT generates interpretability maps that highlight which specific movements influence its decisions, confirming that animal locomotion possesses an underlying semantic structure comparable to linguistic syntax. The model achieved state-of-the-art performance across five international benchmarks covering social interaction, multi-agent tracking, and three-dimensional behavioral analysis. A defining technical advantage of the framework is its self-supervised learning pipeline, which extracts meaningful representations directly from raw motion data. The architecture demonstrated strong cross-species adaptability, with a model trained on rat behaviors transferring effectively to mouse analysis. This capability positions BehaVERT as a scalable foundation model for broader life science applications. The development also underscores a significant methodological crossover, as biologists on the team independently mastered deep learning optimization and transformer design to tailor the system for neurobehavioral research. Researchers anticipate that BehaVERT will serve as a critical instrument for next-generation psychiatric diagnostics, behavioral genetics, and preclinical drug development. By treating complex motor patterns as interpretable semantic sequences, the AI framework offers a systematic pathway to decode neurological and behavioral disorders. The project builds upon the laboratory’s previous work in virtual behavioral reconstruction, further establishing KAIST as a central node at the intersection of artificial intelligence and computational neuroscience.