Researchers from Massachusetts Institute of Technology (MIT) have conducted a study which demonstrates that sentences with complex grammar or unexpected meaning tend to stimulate the brain’s key language processing centers significantly more than straightforward or nonsensical sentences. The study was led by Evelina Fedorenko, an Associate Professor of Neuroscience at MIT, and Greta Tuckute, a Graduate Student at MIT.
The team of researchers leveraged an artificial language network to identify which types of sentences induce heightened activity in the language-processing regions of the brain. These regions are predominantly found in the left hemisphere of the brain and include Broca’s area and parts of the left frontal and temporal lobes.
The team compiled and analyzed 1,000 sentences, taken from a diverse array of sources such as fiction texts, scientific articles, web content, and spoken word transcriptions. These sentences were read by five human participants while their brain activity was measured via functional magnetic resonance imaging (fMRI). The same sentences were then input into a large language model similar to ChatGPT, which generates and comprehends language through predicting the next word in a text. The team subsequently correlated the activation patterns within the human brain with those observed in the artificial language model.
Following this process, the researchers established an ‘encoding model’ to train the artificial language network to predict the human brain’s reaction to new sentences based on its prior reactions to the initial 1,000 sentences. Using this model, the scientists identified 500 new sentences which would likely either significantly activate (‘drive’) or nearly negate (‘suppress’) activity in the brain’s language network.
The researchers asked crowd-sourcing platform participants to rate the sentences according to emotional tone, visualization potential, plausibility, grammaticality, and other attributes. They also used a computational technique to compute the ‘surprisal’ of each sentence, or its relative novelty in comparison to common sentence structures.
The results revealed that sentences featuring high levels of ‘surprisal’ generated greater brain activity, supporting previous findings that these types of sentences can prove more challenging to process. The team has plans to see whether these findings can be replicated in speakers of languages other than English, and to explore how different stimuli can activate language processing zones in the right hemisphere of the brain. This study, published in Nature Human Behavior, might help scientists better understand brain responses in language processing.