The new brain implant works with a neural network to map brain patterns and movements of the vocal tract.

After 10 years of research, scientists from the University of California, San Francisco have demonstrated a unique brain implant for the first time. It turns neural activity into full words. The first test participant, a paralyzed man in his 30s, can now speak using a vocabulary of 50 words just by thinking about pronouncing them.

The innovative new technology differs from previous brain-computer interfaces. Instead of forcing a person to hover over the screen to spell words, the new device monitors brain activity in areas that control the vocal systems. Thus, while paralyzed subjects may lose the ability to literally move their mouths and speak words, their brains may still attempt to send these unique signals to organs in the vocal tract, such as the jaw and larynx.

A new study published in the New England Journal of Medicine describes the first person to test an experimental implant. Subject suffered a stroke 15 years ago and can only communicate by typing words on the screen using a pointer attached to a baseball cap.

High density electrodes were surgically implanted over the subject’s verbal motor cortex. The brain activity was then recorded over several months, correlating certain signals with a vocabulary of 50 words. The researchers then taught custom neural network models to recognize brain activity and identify words in real time as they thought about them.

Early tests showed that the man responded to researchers’ requests with full sentences. When asked, “Would you like some water?” Subject was able to reply, “No, I don’t want to drink.”

The implant now decodes about 18 words per minute. And the average accuracy is only 75%, so there is plenty of room for developers to improve. As the authors of the development note, improving the algorithms will increase the accuracy and speed of the device.

It is planned that the trial will be expanded to include more participants. Researchers are also looking to expand the system’s vocabulary and speed up speech decoding.