A man on the radio is talking about what it was like to come out to his family as gay.
The sound of his voice streams into the ear of the listener and vibrates in the snail shell-shaped cavity of her cochlea. The sounds are translated into electric impulses, which shoot along her nerves into her auditory cortex. Language processing centers start parsing the story for syllables, words, rhythm and syntax. And somehow, they're able to figure out what it all means.
Inside the MRI machine, the listener's brain is aglow with activity.
For the first time, researchers at the University of California at Berkeley have mapped that activity, determining where in the brain certain concepts — family, numbers, texture and touch — are processed and understood. The result of their work, which was published this week in the journal Nature, is an entirely new kind of tool for neuroscientists. Researcher Jack Gallant, a psychology professor at Berkeley, calls it a "semantic atlas" — an atlas of ideas.
He hopes neuroscientists will use it the way navigators use a globe. It can't tell them anything about brain function on its own, but it can guide their exploration. "It's a tool that you can use to answer other questions," he said.
Other researchers, or anyone else who is interested, will soon be able to look at the atlas online. (A rough version is up now, but it only shows the results of one brain scan and requires a pretty fast computer). It's based on scans of the brains of seven Berkeley graduate students and post-docs as they listened to two hours worth of stories from the public radio program "The Moth Radio Hour" — stories about love, faith, abuse, regret, gender identity, exotic dancing and Yankees baseball, among other things.
Gallant and his colleagues matched spikes in activity in each brain to the words being uttered, and found that words associated with related ideas tended to elicit similar responses. For example, an area that lit up in response to "pregnant" was adjacent to the one stimulated by "house," suggesting that a broader notion — family — was focused there. Natural language processing software let the researchers translate the stories into groups concepts, then map those concepts onto each of the seven brains. Intriguingly, all seven of the maps were strikingly similar.
Then Gallant's team used a statistical tool of their own invention to identify functional areas the students all had in common (essentially, a more sophisticated version of taking an average) and create a more general model. When they tested that model on a story none of the subjects had heard yet, it turned out to be a fairly good predictor of how they would respond.
With just seven subjects, the study is a lot smaller than is considered reliable in neuroscience (or pretty much any scientific field, for that matter). Traditionally, a bigger sample size indicates a more accurate result; a poll of 100,000 people is generally more reliable than one that questions just 10, because there's less chance of random variations and mistakes skewing the average result. A larger sample size is also more likely to accurately represent humanity as a whole.
But it's harder to collect a lot of information from your subjects as your sample size gets larger — running his experiment on 700 or even 70 subjects instead of seven would have taken an extraordinary amount of time and limited how many stories and concepts Gallant could examine. Instead, he opted to use just a few subjects to develop his model map of the brain, and then tested that model on a new story to see if it held up. He argues that this process of proving the model's accuracy is just as good a test of the validity of his results. And he'll continue bumping up the sample size as well, adding results of future brain scans to refine his atlas further.
For now, all it takes is a quick glance at the atlas to see that the semantic system of the brain is still an uncharted jungle. Meaning isn't processed in specific centers so much as within vast, intricate networks. Social concepts, for example, which are coded in red on the atlas, are splashed across both hemispheres of the brain.
That's something of a surprise for neuroscientists, who have traditionally believed that language was the purview of the left brain – the side that deals with logic, computation and facts. Then again, most neuroscience studies look at responses to specific words and sounds, not broad concepts.
It may be true that word processing and production happen in the left brain, Gallant said, but the search for meaning seems to require the entire organ.
"This doesn't mean that localization is false," he said. "It's just that the brain is really, really complicated."
Gallant believes that various parts of the brain are marshaled into action to analyze ideas. For example, a mention of "family" might stimulate memories of the listener's own family from one sector and an abstract ideal of family from another.
"The brain is an efficient organ, so presumably we have so many different representations because they are necessary," Gallant said. The question is: Why?
That's just one of the unsolved puzzles provoked by Gallant's study. Another is why the seven brain maps generated from the seven subjects looked so much like one another. This may be a function of the fact that the subjects were so similar — all were successful, English-speaking students at the same school. Perhaps, if Gallant mapped concepts in the brain of an artist in Brazil or a toddler in Japan, the results would look different. Or perhaps they would look like the initial seven, indicating that the semantic systems of diverse human brains share a fundamental architectural plan — the way apartments in the same building can have the same layout but be filled with wildly different furniture.
These questions are studies waiting to happen, Gallant said. And now he has a road map for running them.