Scientists Discover Neurons That Process Language at Different Time Scales

Using functional magnetic resonance imaging (fMRI), neuroscientists have identified several regions of the brain responsible for language processing.

However, uncovering the specific functions of neurons in these regions has proven difficult because fMRI, which measures changes in blood flow, does not have high enough resolution to reveal what small populations of neurons are doing.

Using a more precise technique that involves recording electrical activity directly from the brain, MIT neuroscientists have identified different groups of neurons that appear to process different amounts of linguistic context. These “time windows” range from a single word to about six words.

Time windows may reflect different functions for each population, the researchers explain. Populations with shorter time windows can analyze the meaning of individual words, while those with longer time windows can interpret more complex meanings created when words are strung together.

“This is the first time we’ve seen clear heterogeneity within the language network,” says Evelina Fedorenko, an associate professor of neuroscience at MIT. “In dozens of fMRI experiments, these brain areas all seem to be doing the same thing, but it’s a large, distributed network, so there must be some structure.”

“This is the first clear demonstration that there is a structure, but the different neuronal populations are spatially intertwined, so we can’t see these distinctions with fMRI.”

Fedorenko, who is also a member of MIT’s McGovern Institute for Brain Research, is the lead author of the study, now published in Nature Human BehaviorTamar Regev, a postdoctoral fellow at MIT, and Colton Casto, a graduate student at Harvard University, are the paper’s lead authors.

Time windows

Functional MRI, which has allowed scientists to learn a lot about the role of different parts of the brain, works by measuring changes in blood flow in the brain. These measurements act as an indicator of neural activity during a particular task.

However, each “voxel,” or three-dimensional fragment of an fMRI image, represents hundreds of thousands or even millions of neurons and summarizes activity over about two seconds. So it can’t reveal precise details about what those neurons are doing.

One way to get more detailed information about how neurons work is to record electrical activity using electrodes implanted in the brain. This data is difficult to obtain because this procedure is only performed on patients who are already undergoing surgery for a neurological disease such as severe epilepsy.

“It can take a few years to get enough data for a task because these patients are relatively rare, and within a given patient, electrodes are implanted in idiosyncratic locations based on clinical needs. So it takes some time to assemble a dataset with sufficient coverage of a target part of the cortex.

“But this data is of course the best we can get from the human brain: you know exactly where you are in space and you have very precise temporal information,” says Fedorenko.

In a 2016 study, Fedorenko reported using this approach to study the language-processing regions of six people. Electrical activity was recorded while participants read four different types of linguistic stimuli: complete sentences, lists of words, lists of nonwords, and “jabberwocky” sentences (sentences that have grammatical structure but are made up of nonsense words).

These data showed that in some neural populations in language processing regions, activity developed gradually over a period of several words when participants read sentences. However, this did not happen when they read word lists, nonword lists, or Jabberwocky sentences.

In the new study, Regev and Casto took these data and analyzed the temporal response profiles in more detail. In their original dataset, they had recordings of electrical activity from 177 speech-sensitive electrodes in all six patients.

By conservative estimates, each electrode represents an average of the activity of about 200,000 neurons. They also obtained new data from a second group of 16 patients, which included recordings from 362 other language-sensitive electrodes.

When the researchers analyzed this data, they found that in some neural populations, activity fluctuated up or down with each word. In others, however, activity accumulated over several words before falling off, and in still others, neural activity accumulated steadily over longer periods of words.

By comparing their data with predictions made by a computer model the researchers designed to process stimuli with different time windows, the researchers found that the neural populations in the language processing areas could be divided into three groups. These groups represent time windows of one, four, or six words.

“It really looks like these neural populations are integrating information across different time scales throughout the sentence,” Regev says.

Processing words and meaning

These differences in time window size would have been impossible to see using fMRI, the researchers say.

“At the resolution of fMRI, we don’t see much heterogeneity in the language-sensitive regions. If you localize in individual participants the voxels in their brain that are most sensitive to language, you see that their responses to sentences, word lists, jabber sentences, and nonword lists are very similar,” Casto says.

The researchers were also able to determine the anatomical locations where these groups were located. The neuronal populations with the shortest time window were located primarily in the posterior temporal lobe, although some were also found in the frontal or anterior temporal lobes. The neuronal populations of the other two groups, with longer time windows, were more evenly distributed across the temporal and frontal lobes.

Fedorenko’s lab now plans to study whether these time scales correspond to different functions. It’s possible that populations with shorter time scales can process the meaning of a single word, while those with longer time scales interpret the meaning represented by multiple words.

“We already know that the language network is sensitive to how words come together and what each word means,” Regev says. “So that might fit with what we found, where the longer time scale is sensitive to things like syntax or relationships between words, and maybe the shorter time scale is more sensitive to features of individual words or parts of words.”

This article is republished with kind permission from MIT News (web.mit.edu/newsoffice/), a popular site covering the latest research, innovation, and teaching at MIT.