Back Past events: Milena Rabovsky. Modeling neural responses during human language comprehension

Milena Rabovsky. Modeling neural responses during human language comprehension

20/4/2023
- BCBL zoom room 2 only (online talk)

What: Modeling neural responses during human language comprehension

Where:  zoom room # 2 (If you would like to attend to this meeting reserve at info@bcbl.eu)

Who: Prof. Milena Rabovsky . PhD; Cognitive Neuroscience Lab, University of Potsdam, Postdam, Germany

When:  Thursday,  Apr 20th at 12:00 PM noon.

The N400 brain potential seems to reflect the brain’s default response to incoming language. Using a neural network model of predictive sentence comprehension, the Sentence Gestalt (SG) model, we simulate N400 effects across a broad range of experimental conditions as the change in the model’s hidden layer activation induced by new incoming words, corresponding to an internal temporal difference prediction error at the level of meaning. In the model, the magnitude of the simulated N400 also corresponds to the error signal driving learning. In line with this correspondence we found empirically that larger N400 amplitudes entail enhanced adaptation. In more recent work, we scaled up our cognitively motivated model to a large-scale naturalistic language corpus, which allows to quantitatively predict ERP amplitudes during naturalistic reading and to quantitatively compare different models. Our results show that the N400 as a widely observed language-related brain potential emerges naturally within a predictive comprehension system that probabilistically maps from linguistic input to events in the world.