‘Jiffy Lube of — of therapies, yes’: Complex Sentences Can Generate a Higher Brain Response Than Simple Sentences

Complex sentences activate certain brain networks more than simple sentences, according to study authors at Massachusetts Institute of Technology (MIT) who published their findings in Nature Human Behavior. These complex sentences induced more “surprisal” (uncommon elements) and had greater linguistic complexity that was not as easy to decipher.

For instance, “Jiffy Lube of — of therapies, yes,” is an example of a complex sentence, according study authors. It is a sentence from the Corpus of Contemporary American English dataset that is brain-activating because elements of it are understandable, but it requires deeper thinking to understand it.

Image credit: Caphira Lescante | stock.adobe.com

“There’s something slightly unusual about these [complex] sentences,” said senior author Evelina Fedorenko, PhD, associate professor of neuroscience at MIT and a member of MIT’s McGovern Institute for Brain Research, in a recent press release. “The sentences that elicit the highest brain response have a weird grammatical thing and/or a weird meaning.”

However, overly complex sentences did not engage the brain’s language processing centers. Simple sentences did not either.

In the study, the investigators aimed to evaluate the types of sentences and linguistic input that activates the brain’s language network (found in the left hemisphere of the brain), according tolead study author Greta Tuckute, a PhD candidate at MIT.

“This language network is highly selective to language, but it’s been harder to actually figure out what is going on in these language regions,” Tuckute said in the press release.

Investigators first took 1000 sentences from different literary sources and had 5 participants read the sentences. They used a functional magnetic resonance imaging (fMRI) to see the participant’s language network activity.

The team also put the 1000 sentences into a large language model (LLM)—which first used predictive capabilities to understand the language—then the team evaluated the model’s activation patterns while engaging with the text. Then, using a mapping (encoding) model, they compared how the sentences activated the LLM compared to how they activated the human brain, explains Tuckute.

With all this data, the mapping model then created 500 new “drive” sentences, which can generate peak human brain activity; the model also identified 500 “suppress” sentences, which generate some of the least activity in the brain’s language network.

“This ‘closed-loop’ modulation of brain activity during language processing is novel,” Tuckute said in the press release. “Our study shows that the model we’re using, that maps between language-model activations and brain responses, is accurate enough to do this.”

Finally, investigators evaluated why certain sentences increased brain activity, looking at grammaticality, plausibility, emotional valence (positive or negative), ease of visualizing sentence content, and other linguistic properties.

The computer then analyzed each sentence’s surprisal based on these properties and noticed that complex sentences with more surprisal elements, like unusual grammar and/or an unexpected meaning, activated the brain more.

Possible studies on the horizon include evaluating these findings in non-native English speakers or focusing on stimuli that activate language learning processes in the right hemisphere of the brain.

REFERENCE

Complex, unfamiliar sentences make the brain’s language network work harder. MIT. News Release. January 3, 2024. Accessed on January 12, 2024. https://eurekalert.org/news-releases/1030107