Using an artificial language network, neuroscientists from MIT have identified the type of sentences that most effectively activate the human brain’s language processing centres. Their findings, published in Nature Human Behavior, show that the most stimulating sentences are those which are complex due to uncommon words or grammar, or unexpected meanings. Simplistic sentences or nonsensical strings of words didn’t have much effect.
The team noticed that regions of the brain involved in processing language were particularly active when participants read unusual sentences but remained relatively passive when they read straightforward ones. “The input has to be language-like enough to engage the system,” explained Evelina Fedorenko, Associate Professor of Neuroscience at MIT. Whenever sentences contained unusual constructions or unfamiliar words, the network had to work harder to decipher them.
Conducting the study involved focusing on language-processing areas found in the left hemisphere of the brain, including Broca’s area and parts of the left frontal and temporal lobes. Using an encoding model that could relate the activation patterns in the brain to those in an artificial language model, the team compiled 1,000 sentences from various sources, which five participants read whilst having their brain activity measured using fMRI.
The researchers then trained their model using these responses and subsequently used it to identify 500 new sentences designed to maximise (or minimise) activity in language processing areas of the brain. To find out what made some sentences more effective at driving activity than others, the sentences were analysed based on 11 linguistic properties, including grammar, how easy it was to visualise the sentence content, and “surprisal” (how unusual it was in comparison to other sentences).
Results revealed that sentences with higher “surprisal” and linguistic complexity generated the strongest responses. Sentences that were extremely complex or simple activated the language network the least, while sentences that made sense but required work to decipher evoked the strongest responses.
The researchers plan to extend their findings to non-English languages and explore stimuli that may activate the brain’s right hemisphere. Funding for the study was provided by various organisations, including an Amazon Fellowship from the Science Hub and the MIT-IBM Watson AI Lab.
