A team of neuroscientists from the Massachusetts Institute of Technology (MIT) have used an artificial language network to identify the type of sentences that activate the brain’s critical language processing centres. The team learned that more complex sentences, which feature unusual grammar or unexpected meanings, generate strong responses from these centres, while straightforward sentences barely engage them. Nonsensical word sequences also have little effect.
The brain network was most active when participants read unusual sentences like “Buy sell signals remains a particular.” However, it was largely quiet when the subjects read straightforward sentences such as “We were sitting on the couch.”
The study, which is published in Nature Human Behavior, focused on language-processing regions found in the left hemisphere of the brain. The neuroscientists compiled a set of 1,000 sentences from various sources and administered them to five human participants while using functional magnetic resonance imaging to measure their language network activity. The same sentences were input into a large language model, and the team observed the activation patterns in response to each sentence.
The researchers then used these data to train an encoding model, which relates the activation patterns seen in both human and artificial language models. They discovered this could predict how the human language network would respond to a new sentence, based on the artificial language network’s responses. Consequently, the scientists used the encoding model to identify 500 new sentences that would optimise activity in the human brain. These new sentences were then confirmed to have the predicted effect on brain activity in a revised group of participants.
The scientists assessed the sentences based on various linguistic properties, such as grammar, emotional valence, and “surprisal” – a measure of how statistically unlikely the sentence is. They found that sentences with higher surprisal and language complexity elicited higher responses from the brain. The linguistic complexity was measured by the sentence’s compliance with English grammar rules and plausibility.
Future studies will test if these findings are extensible to speakers of languages other than English, and assess what stimuli may trigger the brain’s right hemisphere language processing regions.