For his native language, Spanish, it was necessary to train the system to record the activity of the neurons corresponding to each of the 200 words the patient had to try to pronounce – the man, now in his 30s, had lost the ability to use speech more than 10 years after a heart attack. The system then had to display these words on the screen and try to predict the most likely sentences that the patient was trying to say.
Previously, the same system was trained to recognize words from the same person in English. This first breakthrough led to Study in 2021, signed by a team led by neurosurgeon Edward Chang, from the University of California. The computer system was initially designed to translate English, but the researchers realized that having a bilingual patient, whose first language was Spanish, gave them the opportunity to better understand how language or languages are stored and transmitted in our brain.
Subscribe to our newsletter!
So you don't miss any scientific news and know all about our efforts to fight fake news and misinformation!
This is not the first time that such experiments have been conducted on nerve transplants, in the hope of restoring some patients’ ability to communicate: In August 2023two studies published simultaneously in the journal nature I mentioned two distinct experiences. However, this is the first time researchers have taken a second language step.
This additional step was not technically more difficult, considering that it was the center of speech Turns out it's the same in the brain, regardless of language. The difficulty for researchers and their computer model remains identifying neural “signatures,” because they have to search for one word at a time. The signature associated with the same word will be different, depending on whether it is “idea” in English or in Spanish. But the real motive, in this caseIt was for the patient, as Spanish is his first language.
By combining these 200 words with English words that have already been decoded by AI, and with the help of a dictionary, the researchers claim that their system Be able to predict complete sentences With a success rate of 75%. Their studies It was published on May 20 in the magazine Nature of biomedical engineering.
