Neuroscientists at MIT, leveraging an artificial language network, have made a significant stride in understanding how our brains process language. Their research pinpoints the types of sentences that most effectively activate the brain’s crucial language processing centers, shedding light on the Learning Language Network within us.
The groundbreaking study reveals a fascinating aspect of brain function: complexity is key. Sentences that deviate from the norm, either through unusual grammatical structures or unexpected semantic content, trigger stronger responses in these language processing hubs. Conversely, straightforward sentences barely engage these regions, and nonsensical strings of words also fail to elicit significant activity.
Consider this example highlighted by the researchers: the sentence “Buy sell signals remains a particular,” sourced from the C4 language dataset, sparked considerable activity in the brain network. In stark contrast, a simple sentence like “We were sitting on the couch” resulted in minimal engagement.
Evelina Fedorenko, Associate Professor of Neuroscience at MIT and a member of MIT’s McGovern Institute for Brain Research, emphasizes this point: “The input has to be language-like enough to engage the system. And then within that space, if things are really easy to process, then you don’t have much of a response. But if things get difficult, or surprising, if there’s an unusual construction or an unusual set of words that you’re maybe not very familiar with, then the network has to work harder.” This highlights the brain’s dynamic engagement with language, working more intensely when confronted with linguistic novelty, a crucial aspect of how our learning language network operates.
Published in Nature Human Behavior, the study, with MIT graduate student Greta Tuckute as lead author and Fedorenko as senior author, delves into the intricacies of language processing.
Unpacking the Language Processing Mechanism
The research concentrated on language-processing regions predominantly located in the left hemisphere of the brain. This network includes the well-known Broca’s area, alongside other regions in the left frontal and temporal lobes. These areas are collectively understood as the core of the brain’s learning language network.
Greta Tuckute explains the motivation behind the study: “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. We wanted to discover what kinds of sentences, what kinds of linguistic input, drive the left hemisphere language network.” Understanding what stimulates this network is fundamental to grasping how we learn and process language.
To achieve this, the team meticulously curated a dataset of 1,000 sentences from diverse sources, ranging from fiction and spoken word transcripts to web content and scientific publications.
Participants underwent functional magnetic resonance imaging (fMRI) while reading these sentences, allowing researchers to monitor their language network activity in real-time. Simultaneously, the same sentences were fed into a large language model, an AI system akin to ChatGPT. By observing the activation patterns in both the human brain and the artificial model, researchers aimed to draw parallels and gain insights into the learning language network.
The wealth of data collected was then used to train an “encoding model.” This sophisticated model maps the activation patterns observed in the human brain to those in the artificial language model. Once trained, this model could predict how the human language network would respond to new sentences based on the artificial network’s reactions. This predictive capability is a testament to the model’s accuracy in capturing the essence of the learning language network.
Further validating their model, the researchers identified 500 new sentences predicted to maximize brain activity (“drive” sentences) and 500 sentences expected to minimize activity (“suppress” sentences). Testing these predictions on new participants confirmed the model’s accuracy in modulating brain activity, showcasing a “closed-loop” modulation of brain activity during language processing.
The Role of Complexity and Surprise in Language Learning
To decipher what makes certain sentences more engaging for the learning language network, the researchers analyzed the sentences based on 11 linguistic properties. These included grammaticality, plausibility, emotional valence, and ease of visualization. Crowdsourced ratings and computational techniques were used to quantify each sentence’s “surprisal,” measuring how unexpected a sentence is in comparison to typical language patterns.
The analysis revealed a strong correlation: sentences with higher “surprisal” elicited stronger brain responses. This aligns with previous research indicating that sentences with high surprisal are more challenging to process, demanding more cognitive resources from the learning language network.
Linguistic complexity also emerged as a significant factor. This encompasses both adherence to grammatical rules and plausibility of meaning. Sentences at either extreme – overly simplistic or nonsensical – triggered minimal activation. The most substantial responses came from sentences that were comprehensible but required cognitive effort to interpret, such as “Jiffy Lube of — of therapies, yes,” from the Corpus of Contemporary American English dataset.
Fedorenko concludes, “We found that the sentences that elicit the highest brain response have a weird grammatical thing and/or a weird meaning. There’s something slightly unusual about these sentences.” This “unusualness” is precisely what challenges and activates the learning language network, pushing it to adapt and refine its processing mechanisms.
Looking ahead, the researchers plan to expand their investigations to languages other than English and explore the language processing functions of the brain’s right hemisphere. This ongoing research promises to further illuminate the intricacies of the learning language network and how our brains master the complexities of language.
