A study by MIT neuroscientists, utilising an artificial language network, discovered the type of sentences most likely to stimulate the brain’s key language processing centers. The study concluded that complex sentences, with unusual grammar or unexpected meaning, generate stronger responses. In contrast, simplistic sentences marginally engaged these regions, while nonsensical sequences of words had little effect.
Evelina Fedorenko, Associate Professor of Neuroscience at MIT and senior author of the study, explained that the language input needs to be challenging enough to engage the system. When language processing is easy, the response is mild, but when things are more difficult, the network must work harder to understand.
The focus of the study was on language-processing regions located in the left hemisphere of the brain, including Broca’s area and other sections of the left frontal and temporal lobes.
For this study, the scientists compiled a set of 1,000 sentences from various sources such as fiction, web text, spoken word transcriptions, and scientific articles. Five participants read these sentences while their brain activity was monitored using functional magnetic resonance imaging (fMRI). The same sentences were then processed by a large language model similar to ChatGPT, and the model’s activation patterns in response to each sentence were recorded.
An encoding model was trained to correlate the activation patterns seen in the human brain with those observed in the artificial language model. The trained model could predict how the human language network would respond to any new sentence based on its performance with the initial 1,000 sentences.
The researchers identified 500 new sentences predicted to generate maximal and minimal activity in the human brain, known as “drive” and “suppress” sentences respectively. Three new participants were enlisted, confirming that these sentences did indeed stimulate or suppress brain activity as anticipated.
The researchers found that sentences with higher “surprisal” or unpredictability triggered higher brain responses. This aligns with previous studies showing people face more difficulties in processing unexpected sentences.
Linguistic complexity, the extent to which a sentence follows English grammar rules and makes sense, also appeared to influence the brain’s responses. Sentences with extreme simplicity or incomprehensible complexity evoked little activation in the language network, whereas ambiguous sentences that require effort to understand stimulated a stronger response.
The researchers now intend to extend their investigations to speakers of other languages and to stimulate language processing regions in the brain’s right hemisphere. The research was primarily funded by the Science Hub’s Amazon Fellowship and MIT’s Department of Brain and Cognitive Sciences among others.
