Scientists from MIT have used an artificial language network to investigate the types of sentences likely to stimulate the brain’s primary language processing areas. The research shows that more complicated phrases, owing to their unconventional grammatical structures or unexpected meanings, generate stronger responses in these centres. However, direct and obvious sentences prompt barely any engagement, and nonsensical sequences of words have little effect. The discovery suggests that to engage the brain’s language network, the input must resemble language and present some level of difficulty or novelty to process.
Evelina Fedorenko, Associate Professor of Neuroscience at MIT, and MIT graduate student Greta Tuckute, who led the study, focused on language-processing areas in the left hemisphere of the brain. The researchers had a compilation of 1,000 sentences from different sources read by five people, as they captured the language network activity using functional magnetic resonance imaging (fMRI). They then input these sentences into a large language model and noted this model’s activation patterns.
Using this data, the researchers trained an “encoding model” that can relate the activation patterns seen in the human brain to those observed in the artificial language model. They then used the model to select 500 new sentences that would create maximum brain activity and sentences expected to have the minimum impact on the language network. This selection was validated in three new human participants, confirming the predictions.
The researchers also analysed the sentences to understand what factors influenced activity in the processing centres. Factors included grammatical correctness, feasibility, the emotional tone of the sentence, and how easily visualisable the content was. The analysis suggested that sentences with higher unpredictability elicit higher brain responses, consistent with prior studies indicating people struggle more with such sentences. Another influencing factor was linguistic complexity, showing that sentences with any level of difficulty prompt larger responses. The researchers aim to test their findings on speakers of other languages and explore stimuli that might activate the brain’s right hemisphere.
