In a lab setting, animals need multiple trials to learn new tasks through positive and negative reinforcement since they lack verbal instructions. Humans, on the other hand, can quickly learn new tasks with verbal or written instructions and describe them using natural language. The brain’s ability to integrate linguistic information to reorganize sensorimotor mappings for novel tasks remains unclear. Recent research shows that language models can help RNNs perform novel tasks with 83% accuracy. By training models to interpret instructions, they can produce linguistic descriptions of tasks based on motor feedback. This study aims to understand how the human brain processes language to generalize across sensorimotor tasks.
https://www.nature.com/articles/s41593-024-01607-5