This computer model, developed by a team led by Tom Mitchell of Carnegie Mellon University in Pittsburgh, Pennsylvania, may reveal the mechanism by which the brain processes words and language and could ultimately come up with a technique that can deduce people's thoughts.
First, the computers were 'trained' in order to identify brain patterns linked with 60 images, which represented a different noun, such as 'celery' or 'aeroplane,' reports Nature.
The researchers proceeded further with the assumption that the brain processes words in terms of how they relate to movement and sensory information. For example, words like 'hammer', triggers the movement-related areas of the brain. However, the word 'castle' initiates activity in those regions of the brain that process spatial information.
In addition, the researchers were aware of the fact that different nouns are usually associated with some verbs more than that with others. For example, the verb 'eat' seems to be related with 'celery' than with 'aeroplane'. Thus, they designed the model apparently for using these semantic links to work out how the brain would react to particular nouns and thus they had put in 25 such verbs into the model.
Later, they scanned the brains of 9 volunteers for determining the images of the nouns by using functional magnetic resonance imaging (fMRI). Then they trained the model by feeding the model with 58 of the 60 nouns. And for every noun, this model scanned through a trillion-word body of text to know if it was related to the 25 verbs, and how that related to the activation pattern.
Training was then followed by testing the models, where they had to predict the pattern of activity for the two missing words from the group of 60, and then to figure out which word was which. The results indicated that the models were right almost over seventy five percent of the time on an average.
The next step for the team was to train the models on 59 of the 60 test words, and then showing them a new brain activity pattern where they were provided with a choice of 1,001 words to match it. It was found that the models performed extraordinarily well even in this case. The results of the study are reported in the journal Science.