Abstract
The linguistic relativity hypothesis suggests that the way we perceive the world is shaped by the language we speak. Evidence comes from color perception, where Russian speakers, whose language distinguishes between light and dark blue ("goluboj/sinij"), show enhanced discrimination performance for these shades compared to English speakers, who typically use a single term ("blue") for both. To neuromechanistically explain this phenomenon, we built a brain-constrained neural network simulating neural activity in frontotemporal-occipital cortices. When modeling English speakers' brains, representational similarity analysis revealed similar activity for different shades of blue carrying the same verbal label. However, in virtual Russian speakers, the same colors carrying different labels induced distinct neural activations. These differences arose from microstructural neural changes, involving shifts in shared and unique neurons encoding color representations. Functionally distinct color representations before labeling were modulated by label learning, thereby facilitating or hindering discrimination. The model also reproduced neurophysiological evidence, supporting its validity. Together, these findings bridge theoretical, linguistic, cognitive, and neuroscientific accounts of how language modulates perception.