BrainGate researchers decode imagined handwriting from neural signals to enable high-speed text communication
In May 2021, Francis R. Willett and colleagues at Stanford University and Howard Hughes Medical Institute published results showing that a BrainGate2 intracortical electrode array could decode imagined handwriting movements in a paralysed person at 90 characters per minute with 94.1% raw accuracy, substantially exceeding prior neural-interface typing rates.
Background
For people who cannot move or speak, typing a message can take minutes. Most brain-computer interfaces available before 2021 worked by letting a user control a cursor on screen, moving it slowly across a grid of letters. The participant would focus attention on a letter, the system would register a neural response, and the letter would be selected. It worked, but it was slow. Typical rates hovered around 40 characters per minute under research conditions, and many systems required the user to maintain exhausting concentration throughout.
The deeper problem was one of information. A cursor gives the brain only one thing to express at a time: direction. That is a thin channel. Researchers had long suspected that the motor cortex, the part of the brain that plans and controls movement, held far richer signals than cursor control could capture. Even in people who had been paralysed for years, the cortex might still be generating detailed movement plans, letter shapes included. The challenge was reading those plans accurately enough to be useful.
What happened
Francis R. Willett, Donald T. Avansino, Leigh R. Hochberg, Jaimie M. Henderson, and Krishna V. Shenoy, working across Stanford University, the Howard Hughes Medical Institute, and the BrainGate consortium, tried a different approach entirely. Rather than asking a participant to control a cursor, they asked him to imagine writing letters by hand, one at a time, as though holding a pen on paper. The participant was a man in his late thirties who had been paralysed below the neck following a spinal cord injury.
The team used a BrainGate2 intracortical electrode array, a small grid of electrodes implanted directly into the motor cortex, to record the neural activity that accompanied each attempted letter. Each letter produced a distinct pattern of signals, a kind of neural fingerprint for that handwriting movement. A recurrent neural network, a type of model that processes sequences and carries information forward through time, was trained to match those patterns to the corresponding characters in real time. A language model then ran over the output to correct errors, in much the same way predictive text works on a phone.
The results were published in Nature in May 2021. The system decoded text at 90 characters per minute with 94.1% raw accuracy. With the language model applied, accuracy climbed further. That rate was roughly double what cursor-based interfaces had achieved, and it brought the speed closer to what a non-disabled person manages typing on a smartphone than any previous brain-computer interface had managed. The finding also confirmed something researchers had hoped but not yet shown at this scale: that the motor cortex can hold on to detailed, stable representations of fine hand movements for years after the body has lost the ability to carry them out.
Why it mattered
The study demonstrated that the motor cortex retains detailed, stable representations of intended fine-motor movements, such as forming individual letters, even years after paralysis, a finding with direct implications for restoring communication to people with conditions such as amyotrophic lateral sclerosis or spinal cord injury. By treating each attempted letter as a distinct neural trajectory rather than mapping continuous cursor movement, the approach achieved text-entry speeds closer to those of able-bodied smartphone typing than any prior brain–computer interface had managed. The combination of a recurrent neural network decoder with a language-model post-processor pointed toward a practical architecture for future high-bandwidth neural prosthetics.
People
Francis Willett, Donald T Avansino, Leigh R. Hochberg, Jaimie M. Henderson, Krishna V Shenoy
Organisations
Stanford University, Howard Hughes Medical Institute, Braingate Consortium, VA Palo Alto Health Care System
Sources
- High-performance brain-to-text communication via handwriting.Nature.Primary source
- Brain-computer interface allows paralyzed man to type by imagining writing.Stanford Medicine News Center.Official
- Brain-Computer Interface Turns Mental Handwriting into Text on Screen.Howard Hughes Medical Institute.Secondary
Cite this page
AI Achievements. (2021). BrainGate researchers decode imagined handwriting from neural signals to enable high-speed text communication. Retrieved 2026-08-22, from https://achievements.ai/milestone/brain-computer-interface-records-brain-signals-for-handwriting
@misc{achievements_brain_computer_interface_records_brain_signals_for_handwriting,
title = {BrainGate researchers decode imagined handwriting from neural signals to enable high-speed text communication},
author = {{AI Achievements}},
year = {2021},
url = {https://achievements.ai/milestone/brain-computer-interface-records-brain-signals-for-handwriting}
}