To find language in the brain, map each brain on its own
The language network is the same across people, but it sits in a slightly different spot, so group averages smear away the detail.
Wu Tsai Neurosciences Institute, Stanford · A new precision neuroscience of language (Big Ideas in Neuroscience) | Cory ShainIf you want to find where language lives in the brain, averaging 1,200 brains together can blur into something no one has.
That is the argument of Cory Shain, a neurolinguist who studies how the brain handles language. The usual method in brain imaging goes like this. Researchers give each person a language task, often a harder versus an easier condition, such as more or less complicated sentences. They make a map of which areas respond more during the hard condition in each individual. Then they stack all those maps and average them, spot by spot, on the assumption that the same location does the same job in everybody.
The problem is the assumption. The idea traces to a 2009 paper by Ev Fedorenko, then a postdoctoral researcher at MIT, and her adviser. They argued that brain research had been using the wrong unit of analysis, in a line of work that goes back to the 1800s, when the French physician Paul Broca studied people who could understand language but had a hard time producing speech, and found damage recurring in systematic parts of the brain. If the patch of brain that parses word order and sentence structure sits a few millimeters higher in you than in me, averaging smears the two patches together. The average may look nothing like the individuals who went into it, merging separate functions into one or missing a real function because it never lines up in the same place twice.
Shain puts it as structure, not location. In a study of almost 1,200 people, with nearly 2,000 separate scanning sessions because some people were scanned more than once, his group mapped the language network in each brain first. Only then did they compare people, making what he calls apples to apples comparisons that respect each person's layout.
The network still looks like the same network. In unpublished follow-up work, Shain says, if two people listen to the same story, the activity pattern from a language area in one person can pick out the matching language areas in the other person. Location differs, and the strength of response differs, but the signature matches.
The practical lesson is to map first, then compare. Find language where it actually sits in that head, then ask how that version compares to the next head.
Questions people ask
Why don't brain scans line up on their own?
A brain scan aligns heads by anatomy, by folding patterns and landmarks. But function does not follow anatomy exactly. Two people can do the same job with tissue that sits a few millimeters apart, so lining up the anatomy does not line up the function.
Does this matter for medicine, or only for research?
For surgery and for brain computer interfaces that aim to restore speech, it matters a great deal. Surgeons already map language in individual patients before operating near it, and implant teams need to know where to listen for speech signals in that particular person. A group average is too blurry to guide a probe.
Do speakers of very different languages use the same language network?
Yes, in a broad sense. Work across about three dozen languages finds the same kind of network, with differences between languages no larger than differences between individuals who speak the same language. The implementation shifts, but the system looks shared.