An interdisciplinary research team from the Friedrich-Alexander University Erlangen-Nuremberg (FAU) and the University of Heidelberg has gained new insights into how the human brain processes language. The study shows that when listening, the brain continuously predicts which words will follow next – similar to modern AI language models.

For the investigation, the researchers combined natural listening situations with high-resolution EEG and MEG measurements of brain activity. The results were then compared with the predictions of large language models.

Brain predicts words before they occur

It was found that the more likely a word was in a given context, the lower the neural response was during its processing. At the same time, the researchers were able to detect measurable activations in the brain even before the actual word began. Unexpected words, on the other hand, triggered significantly stronger reactions.

The findings suggest that the brain and AI language models organize and process language according to comparable principles. Although both systems are fundamentally different in structure, they seem to use similar mechanisms for predicting linguistic information.

The study strengthens central assumptions of cognitive neuroscience and could also explain why modern language models are so effective at many tasks.

In the long term, the researchers hope for new insights into the development of brain-computer interfaces, personalized therapies, and more transparent and explainable AI systems.