Keynotes | K&Ts | GACs | Talks | Posters | Search

Poster D in Poster Session D: Wednesday, August 5, 2:00 – 3:45 pm, Kimmel Center, Shorin & Rosenthal Rooms

Human medial-frontal neurons integrate language and vision to structure decision-making

Tomas Aquino¹, Reidar Riveland², Robert Kim³, Adam Mamelak³, Chrystal Reed³, Alexandre Pouget², Ueli Rutishauser³, Nuttida Rungratsameetaweemana¹; ¹Columbia University, ²University of Geneva, ³Cedars-Sinai Medical Center

Presenter: Tomas Aquino

A central question in cognitive neuroscience is how linguistically informed goals interact with sensory evidence to guide decisions. We recorded single neurons from 11 intracranially implanted epilepsy patients performing a verbal-instruction task and found a functional dissociation across cortical regions. Ventrotemporal cortex (VTC) neurons exhibited stable encoding of visual categories and features, whereas neurons in medial frontal cortex (MFC: dorsal anterior cingulate (dACC) and pre-supplementary motor area (preSMA)) encoded task demands and target choices following instruction-vision integration. A vision-language model (VLM) coupled to a sensorimotor recurrent neural network (RNN) trained on the same task provided a mechanistic account: vision-transformer features best predicted VTC activity, and recurrent decision signals best predicted MFC activity. These results delineate how task goals and stimulus features are decomposed across cortical systems to support language-guided choice. We test two competing hypotheses: instructions either propagate to early sensory representations or primarily reconfigure downstream decision circuitry. The evidence supports the latter: VTC neurons are largely insensitive to instruction contents, whereas MFC populations rapidly reorganize to represent rules, framing, and target selection.

Topic Area: Decision-Making, Cognitive Control & Event Cognition