Mapping and decoding natural language representations from the human cortex

Deep neural network language models enable us to model how the human brain responds to natural language with unprecedented accuracy. Applied to very large within-subject datasets, these models enable us to both map language representations with high fidelity and decode language that a person is hearing or thinking from non-invasive functional measurements. I will discuss recent advances in mapping and decoding language representations using BOLD fMRI in combination with modern AI tools and the integration of HD-DOT for these methods.