Neuroscientists at MIT, assisted by an artificial language network, have discovered that complex sentences with unusual grammar or unexpected meaning, stimulate the brain’s key language processing centres more effectively. Interestingly, both straightforward sentences and nonsensical sequences of words had minimal engagement in these regions.
The findings were part of a study led by MIT graduate student Greta Tuckute and Evelina Fedorenko, Associate Professor of Neuroscience at MIT. The work focused mainly on the language-processing areas present in the brain’s left hemisphere, which includes Broca’s area and portions of the left frontal and temporal lobes. The aim was to identify what kind of linguistic input could optimally stimulate these regions.
The researchers compiled a thousand sentences from multiple sources which were read by five human participants. They simultaneously monitored language network activity using functional magnetic resonance imaging (fMRI). Those same sentences were then input into a large language model to record its activation patterns.
After collecting data, an encoding model was trained that could map the relation between activation patterns seen in the human brain and those observed in the artificial language network. Once trained, it could predict how the human language network would react to any new sentence, based on the artificial network’s response to the initial one thousand sentences.
Using this approach, the team identified 500 new sentences that would maximally stimulate human brain activity, while others would minimize it. When tested on a new group of three participants, these sentences drove and suppressed brain activity as expected.
Further analysis of these sentences based on 11 different linguistic properties (such as emotional valence, plausibility, grammaticality) showed that the linguistic complexity and “uncommonness” had more impact on brain response. Sentences that were either too simple or extremely complex hardly activated the language network, while ones that made some sense but required effort to comprehend produced the largest responses.
The MIT researchers are now planning to test their findings in non-English speakers and explore the stimuli that might activate the language processing regions in the right hemisphere of the brain. The study, which is the first to demonstrate this approach in brain areas involved in higher-level cognition, could deepen our understanding of how the brain processes language, potentially opening new avenues for studying cognition and brain health.