MIT neuroscientists have used an artificial language network to determine the type of sentences that most stimulate the brain’s key language processing regions. Their study reveals that the brain reacts more to complex sentences with unusual grammar or unexpected meaning. Straightforward sentences or nonsensical sequences show little engagement. The researchers focused on language processing regions in the left hemisphere of the brain, including Broca’s area.
They compiled 1,000 sentences from various sources and had five human participants read them while their brain activity was recorded using fMRI. The scientists then fed the sentences into a large language model similar to ChatGPT to measure its activation patterns in response to each sentence. They trained an encoding model to relate the activation patterns from the human and artificial language network, enabling predictions of human responses based on the model’s responses to the sentences.
To identify key drivers of brain activity, they then used this model to select 500 sentences they predicted to maximize or minimize activity in the human brain. When presented to a new group of participants, these sentences did indeed increase or decrease brain activity as predicted.
‘Closed-loop’ modulation of brain activity during language processing is a new discovery and the first instance of this approach was used in brain areas related to higher-level cognition such as the language network.
The researchers analyzed the sentences based on linguistic properties like plausibility, grammaticality, emotional valence and ease of visualization. Using crowd-sourced ratings and a computational technique, they assessed how uncommon each sentence was. Their findings revealed that sentences with higher ‘surprisal’ generated stronger responses in the brain. This aligned with past studies showing people struggle more with processing less common sentences.
Linguistic complexity, which refers to how much a sentence adheres to English grammar rules and how plausible its content is, also influenced brain responses. Sentences that are extremely simple or too complex garnered little reaction, while ones that made sense but required effort to understand provoked the most response.
The team plans to explore if these findings apply to other languages. They also hope to investigate stimuli that could activate language processing regions in the brain’s right hemisphere. Funding for the research came from various sources including an Amazon Fellowship from the Science Hub and an International Doctoral Fellowship from the American Association of University Women.
