☰

Research

Sharing our thoughts using language

How do we transmit our thoughts—complex patterns of neural activity—from one brain to another? How do our brains become aligned through interactive conversations? Historically, research on the neurobiology of language focused on isolated words and sentences, and was unable to capture the richness of everyday communication. The development of large language models (LLMs), however, has allowed us to quantify the unique meaning of words in context in a way that was previously not possible. These models use simple statistical learning algorithms to encode the patterns of language in a neural population code. Much of our recent work uses LLMs to better understand the neural machinery supporting real-world language and commnication.

Key publications

    Zada, Z., Goldstein, A. Y., Michelmann, S., Simony, E., Price, A., Hasenfratz, L., Barham, E., Zadbood, A., Doyle, W., Friedman, D., Dugan, P., Melloni, L., Devore, S., Flinker, A., Devinsky, O., Hasson, U.*, & Nastase, S. A.* (2024). A shared model-based linguistic space for transmitting our thoughts from brain to brain in natural conversations. Neuron, 112(18), 3211–3222. DOI PDF

    Kumar, S.*, Sumers, T. R.*, Yamakoshi, T., Goldstein, A., Hasson, U., Norman, K. A., Griffiths, T. L., Hawkins, R. D., & Nastase, S. A. (2024). Shared functional specialization in transformer-based language models and the human brain. Nature Communications, 15, 5523. DOI PDF

Ecological human neuroscience

How can we build more holistic models of human behavior? How can we ensure that our neuroscientific theories generalize outside the laboratory to everyday life? Cognitive neuroscientists typically confront the complexity of human cognition with a divide-and-conquer approach: they use tightly-controlled experimental manipulations to isolate neural signatures of particular cognitive processes. In our lab, we use a complementary approach: first, we develop naturalistic experimental paradigms to measure brain activity in more ecological contexts; second, we formalize our hypotheses as computational models and test these models against naturalistic brain activity. Our work increasingly relies on deep neural networks to accommodate the complexity and contextual richness of real-world cognition.

Key publications

    Nastase, S. A., Goldstein, A., & Hasson, U. (2020). Keep it real: rethinking the primacy of experimental control in cognitive neuroscience. NeuroImage, 222, 117254. DOI PDF

    Hasson, U., Nastase, S. A., & Goldstein, A. (2020). Direct fit to nature: an evolutionary perspective on biological and artificial neural networks. Neuron, 105(3), 416-434. DOI PDF

Coordinating our actions with others

How do we coordinate our own behavior with the actions of others? How do complex social dynamics emerge from multiple interacting agents? Our lab studies the neural systems that allow us to understand the actions and intentions of others. The dynamic, interactive structures of our social life are notoriously absent or underdeveloped in computational models of vision and language. We are also interested in using multi-agent deep learning to model social interaction more holistically. Simulated multi-agent ecosystems can help us better understand how individual neural systems learn to interact and cooperate with one another.

Key publications

    Jung, H., Han, X., Feilong, M., Han, J., Shokeen, D., Van Uden, C., Hanson, I., Connolly, A. C., Haxby, J. V., & Nastase, S. A. (2025). Action features dominate cortical representation during natural vision. bioRxiv. DOI PDF

    Han, J., Chauhan, V., Philip, R., Taylor, M. K., Jung, H., Halchenko, Y. O., Gobbini, M. I., Haxby, J. V.*, & Nastase, S. A.* (2024). Behaviorally-relevant features of observed actions dominate cortical representational geometry in natural vision. bioRxiv. DOI PDF