MIT neuroscientists, using an artificial language network, have learned that more complex sentences, due to either odd grammar or unexpected meanings, trigger stronger responses in the brain’s key language processing centers. On the other hand, plain sentences barely stimulate these regions, and nonsense word sequences have little effect on them.
Evelina Fedorenko, an Associate Professor of Neuroscience at MIT, explains that for the brain network to engage, the input must resemble language. If the input is very easy to process, there won’t be much of a response. However, if the input is challenging to process, maybe because of unusual word choices or sentence structure, the network has to work harder.
The study, appearing in Nature Human Behavior, centered on the language-processing regions located in the brain’s left hemisphere. These include Broca’s area and other areas in the left frontal and temporal lobes. The scientists aimed to determine what kind of linguistic input would stimulate these regions the most.
Five human participants read 1000 sentences while researchers measured their language network activity using FMRI scans. The same sentences were then input into an artificial language model, and its activation patterns were measured. Based on these data, an ‘encoding model’ was created to predict how the human language network would respond based on how the artificial network responded to these sentences.
The researchers used the encoding model to identify 500 new sentences thought to trigger the highest and lowest amounts of activity in the human brain. On testing with three new human participants, the researchers confirmed their predictions.
The researchers also examined what made some sentences trigger more activity. By analyzing the sentences on 11 different linguistic properties and asking participants to rate these properties, the researchers found sentences with higher ‘surprisal’, a quantity indicating how rare they are, elicited higher brain responses. This echo findings from previous studies showing that people struggle more to process sentences with higher surprisal.
Linguistic complexity, meaning how much a sentence adheres to English grammar rules and how plausible its content is, also affected the language network’s responses. Sentences that were either overly simple, or so intricate they didn’t make sense at all, hardly activated the language network. Sentences that made partial sense, but required in-depth comprehension of their content, triggered the most significant responses.
The researchers now aim to investigate if these findings apply to speakers of other languages. They also plan to look into which stimuli activate language processing regions in the right brain hemisphere.
