An artificial intelligence can accurately translate thoughts into sentences, at least for a limited vocabulary of 250 words. The system may bring us a step closer to1speech to people who have lost the ability.
Joseph Makin at the University of California, San Francisco, and his colleagues used deep learning algorithms (算法) to study the brain2of four women as they spoke. The women, who all suffer from a certain kind of brain disorder, already had electrodes attached to their brains to monitor disease attacks.
Each woman was asked to read aloud from a set of sentences as the team 3brain activity. The largest group of sentences4250 unique words. The team fed this brain activity to a network algorithm related to nerves, training it to5regularly occurring patterns that could be linked to repeated aspects of speech. These patterns were then fed to a second network, which tried to turn them into words to6a sentence.
Each woman repeated the sentences at least twice, and the final repetition didn't form part of the training data,7the researchers to test the system. Each time a person speaks the same sentence, the brain activity associated will be similar but not exactly the sane." Memorizing the brain activity of these sentences wouldn't help,8the network instead has to learn what's similar about them so that it can generalize to this final example," says Makin. Across the four women, the AI's best performance was an average translation error rate of 3 per cent.
Makin says that using a small number of sentences made it9for the AI to learn which words tend to follow others. For example, the AI was able to10that the word "Bear" was always likely to follow the word "Teddy" in a certain set of sentences, from brain activity alone.
The team tried transforming the brain signal data into11words at a time, rather than whole sentences, but this12the error rate to 38 per cent even for the best performance." So the network clearly is learning facts about which words go together, and not just which brain activity13with which words," says Makin.
This will make it hard to scale up the system to a/an14vocabulary because each new word increases the number of possible sentences, reducing15. Sophie Scott at University College London says we are a long way from being able to translate brain signal data comprehensively.
