Back

Hierarchical Signatures of Language in the Human Brain

Eichert, N.; Favre, C.; Cabalo, D. G.; Ngo, A.; DeKraker, J.; Hwang, Y.; Smith, M. E.; Crooks, V.; Bautin, P.; Rodriguez-Cruces, R.; Watkins, K. E.; Jones, O. P.; Paquola, C.; Jbabdi, S.; Bernhardt, B.

2026-02-16 neuroscience
10.64898/2026.02.15.705992 bioRxiv
Show abstract

Language relies on a hierarchy of sensory and cognitive processes, yet how different levels of this hierarchy are supported by distinct neural architectures remains unclear. Here we show that semantic processing, compared with phonological processing, is associated with higher-level brain networks, as characterized by resting-state fMRI connectivity, in vivo measures of cortical myelin, and cortical types derived from a cytoarchitecture-defined reference atlas. These relationships were established using individualized ultra-high field fMRI in both English and French speakers leveraging a multi-session, multi-modal 7T MRI protocol including a language localizer. For comparison, we developed an artificial neural network, in which a representational hierarchy spontaneously emerged where phonological information was captured in earlier layers and semantic information in later layers. By integrating individualized functional mapping, neuroanatomical characterization and artificial intelligence, this study advances understanding of the neural basis of language and provides a framework for linking biological and artificial systems of communication. Significance StatementLanguage is widely described as hierarchical, yet how this functional organization is implemented in the brains biological architecture remains unclear. By combining ultra-high field, individualized neuroimaging with in vivo measures of cortical microstructure and large-scale connectivity, this study establishes a framework for linking distinct levels of linguistic computation to the brains structural and functional organization. Integrating these findings with artificial neural network modeling further reveals shared principles between biological and machine systems. Together, this work advances a biologically grounded account of language, bridges cognitive neuroscience and artificial intelligence, and provides a roadmap for understanding how complex cognition emerges from structured brain architecture.

Matching journals

The top 5 journals account for 50% of the predicted probability mass.

50% of probability mass above

"Similar papers" are the closest papers from that journal in the model's embedding space. They show what the match is built on, but the ranking comes mostly from a classifier over the whole training set, not from these examples alone.