Ev Fedorenko, a neuroscientist at the Massachusetts Institute of Technology (MIT), published a 15-year longitudinal study in Nature Neuroscience, which for the first time mapped a millimeter-level probabilistic map of the human brain's language network — a volume smaller than a strawberry, yet responsible for all functions of "word-meaning mapping + sentence assembly," completely decoupled from thinking and emotional modules.
1,400 fMRI scans pinpointed the "language chip"
- The experiment had participants complete tasks such as reading, copying, and listening to stories, comparing them with mathematical calculations and visual stimuli to isolate "language activation."
- Results: the average volume of the language network was only 4.2 cm³, located at the intersection of the left inferior frontal gyrus and the temporal lobe, with highly stable signal patterns, with more than 92% overlap across subjects.
Clinical evidence: the strawberry is spoiled, but thinking remains
- The study included 212 aphasic patients: after damage to the language network, patients could still perform complex reasoning and spatial planning, but could not produce complete sentences.
- Fedorenko pointed out, "This proves that language is merely a 'output interface' of thought, not thought itself."
The probabilistic map is open source, benefiting AI and brain-computer interfaces
- The team released a 1mm resolution probabilistic map, which has already been cited by Meta and Google DeepMind, used to guide the architecture of large language models and electrode placement on brain-computer chips.
- MIT plans to release a "language-thought" dual-area stimulation protocol in Q2 2025, which may help aphasic patients regain the ability to generate sentences through external stimulation.
Implications for AI: language ≠ cognition
