In 1933 a French neurologist named Théophile Alajouanine began examining the composer Maurice Ravel. The neurologist was musical himself, and he added a piano to his tests.
By the neurologist's account, Ravel's ear seemed sharp. Parts of one of his own pieces were played to him with small mistakes, and he heard them and wanted the music played properly. He tried two notes on the neurologist's piano and could tell it was out of tune.
Then they asked him to play. He got through the opening bars of one of his own pieces almost perfectly, and couldn't finish. He managed only two or three notes of a piece he didn't know. He could barely read music, and writing it down was slow and full of errors. His speech and his writing were failing too. The neurologist's description of the case begins, "Maurice Ravel was struck down by an aphasia." Aphasia is a loss of language after brain damage. Ravel composed nothing more, and he died in December 1937.
Ravel lost words and music together. Was his music just another kind of language, failing along with the rest? Put bluntly, is there a music center in the brain?
Ravel's own case can't settle it. Losing two abilities at once proves nothing about whether they share machinery, because a single injury can damage two separate systems. And nobody knows for sure what Ravel's illness was. An operation on his brain days before he died found no tumor, and his family didn't allow an autopsy.
What his case does show is a split inside music. He lost the making of it: playing it, reading it, writing it down. He kept the hearing of it: recognizing it, judging it, remembering it. These tests were made before modern tests for musical disorders existed. Still, Alajouanine called Ravel's musical thinking "comparatively well preserved." For music against language, we need other evidence.
It comes from two directions, and at first they disagree. People with damaged brains show music and language coming apart. Brain scans of healthy people seem to show them in the same places. We'll take the patients first, then the scans. Next we'll look closer than an ordinary scan can, at populations of neurons that answer almost only to music, and one that answers to song. Last come the researchers who say most of it is shared, even up close.
Two words need keeping apart. Language, here, means words and grammar, which is what aphasia takes away. Speech means the sound of someone talking.
Showing that two abilities run on separate machinery takes a pair of cases. In one person, the first ability is lost and the second is kept. In another person, it's the other way round. That two-way split is called a double dissociation. For music and language, the pair exists.
Isabelle Peretz is a psychologist in Montreal who studies music and the brain. A review by Peretz gathers the cases. In one, an opera singer had brain damage and could no longer sing intervals, the steps between notes. The same singer still spoke with normal expression, and with normal intonation, the rise and fall of the voice. The reverse is common. Many people with aphasia can no longer talk, and yet they can sing a familiar tune, and some can learn a new one. Peretz reads the evidence as showing that the tune survives while the words fail, sung or spoken. Peretz also notes that patients who keep some words when they sing are reported more often among English speakers than French ones. A later lecture comes back to singing and speech.
There's an obvious objection. Most of us are amateurs at music and experts at talking, so damage might just wreck the weaker skill first. Peretz answers with the patients who sing. The objection predicts that talking, the expert skill, should be the one to survive. Yet the common case runs the other way, with talking lost and singing kept.
A team in Finland tested the split in 97 stroke patients. Within three weeks of each stroke, the patients heard pairs of short piano melodies and said whether the two in each pair were the same. In half the pairs, a note or a rhythm had changed. Their language was tested too.
About two-thirds of the patients had amusia, a failure to make sense of music. Forty-five of those had no aphasia. Fifteen other patients showed the reverse. They had aphasia, and they made sense of music normally. A stroke can take music and spare language, or take language and spare music.
Some people show the music side of the split from birth. Congenital amusia is amusia from birth, with no brain injury behind it. It's often called tone deafness. People with it miss a wrong note in a tune they haven't heard before. They can't tell when someone sings out of tune, themselves included. Many can't recognize a familiar song without its words. Yet they follow everyday speech normally, and they recognize people by their voices.
One person born with amusia described it to researchers in London: "When the music finished, the sound was always gone—as though it had never happened." How common is it? Peretz's lab in Montreal tested more than sixteen thousand adults online. By a strict measure, one and a half percent were born with amusia.
You might call yourself tone-deaf because you can't sing in tune. You probably aren't, though. The London group notes that people who say that rarely have amusia. Their trouble is with singing, and their hearing of music tends to be normal.
Damaged brains and brains that were different from the start agree, then. Music and language can come apart, in either direction.
Healthy brains in a scanner seemed to say the opposite. These scanners track blood flow as a stand-in for the activity of neurons. For years, scans showed music lighting up areas known for language, and speech lighting up areas that preferred music.
By 2014 the idea of shared machinery had caught on, and it carried a hope. Shared neurons would mean that training one skill might help the other, and music lessons might help children learn to read. A later lecture weighs that hope. Peretz's team counted the papers on music and language overlapping, year by year, and found the number climbing for a decade. They wrote that much of that work had no direct comparison behind it.
Their example was a brain-scan study of professional musicians. It found a stronger response to rhythm in musicians than in non-musicians, in a region usually tied to grammar. Its authors concluded that musicians seem to handle rhythm with the brain's resources for grammar. The study never tested grammar.
Peretz's answer is that overlap doesn't mean sharing. A brain scan is built of tiny dots. Each dot is the smallest piece of brain the scanner can measure, and it holds hundreds of thousands of neurons. What the scanner reads from a dot is their combined activity.
Say you hang one microphone over a section of a stadium, where home and away fans sit mixed together. The home team scores, and the microphone picks up a roar. The away team scores, and it picks up a roar again. By the microphone's account, the section cheers for both teams. It has no way to tell you that the section holds two crowds, each cheering for its own side.
A dot that lights up for music and for language is that microphone. It might hold one set of neurons doing both jobs. It might hold two sets side by side, each with a job of its own. From the dot alone, the scanner can't tell which.
The old scans had a second problem, and the lab of Evelina Fedorenko at MIT went after it. Most scans averaged many people's brains together. But each person's language areas sit in a slightly different place. Averaging can smear two neighboring areas into one blur, even when they never overlap in any single person. So Fedorenko's group found each person's language areas first, from how they responded to reading sentences. Then they played that person music.
In 2011 the language areas barely responded to music. In four more experiments in 2023, their response to music was generally low, often lower than when nothing was playing at all. Songs drove them no more than the same words spoken. And three people with severe aphasia, who couldn't tell whether a sentence was grammatical, judged melodies well.
Looked at person by person, then, the language areas leave music alone. That fits the patients. But in the brain's hearing areas, in and around the hearing cortex, music and speech still seem to share dots. Both pictures could still be true there, if each dot holds more than one crowd. The way to find out is to look inside the dots.
In 2015 another team at MIT found a way to do it. The study was led by Sam Norman-Haignere, then a postdoctoral researcher there. The team put ten non-musicians in a scanner and played them 165 everyday sounds, two seconds each. Among them were a ringing telephone, footsteps and a car engine starting up, with many kinds of speech and music mixed in. Nobody asked the listeners about music. Their only task was to press a button when a repeated sound came out a little quieter.
Norman-Haignere's team didn't ask each dot whether it preferred music. They treated every dot as a blend of a few underlying populations, and let the mathematics find them. A population, in this sense, is a group of neurons that all answer to the same kinds of sound.
The stadium shows how that can work. One microphone can't tell you there are two crowds. Thousands of microphones can, if each one hangs over a different mix of fans. The home team scores, and some microphones get much louder than others. The away team scores, and a different set gets louder. From those differences, you can work out that there are two crowds, and what each one cheers for.
The team had thousands of dots, each a different blend. The analysis was never told which sound was music. Nobody saw these neurons directly, either. The populations are what the mathematics says must be in the dots.
Six populations accounted for most of what the dots did. Four answered to plain features of sound. Two followed the last lecture's place map, the ear's layout from low sounds to high: one took low sounds and one high. That made a useful check: working blind, the method had found something everyone knew was there. A third took sounds with a clear pitch, and a fourth took sharp, fast changes in a sound.
The last two were new. Speech drove one of them, even in languages the listeners didn't know. Music drove the other. Of the thirty sounds that drove the music population hardest, all but two were music. The two exceptions were wind chimes and a ringtone, which at least have tunes. Measures of the raw sound couldn't explain its taste. Music chopped into tiny pieces and shuffled drove it less.
Why had nobody found a music population before? The stadium suggests a guess. In the dots most tuned to speech, the speech population supplied about seventy percent of the response. Speech had patches almost to itself, and ordinary scans had found them. In the dots most tuned to music, the music population supplied under half. Much of the rest came from the population tuned to pitch. The music crowd never had a section of the stands to itself, so no single dot looked very interested in music.
Had they found the music center, then? A warning came from Josh McDermott, one of the team's senior scientists: "It's inappropriate at this point to conclude that this is the seat of music in the brain." The team had looked mostly at the hearing areas.
Seven years later, Norman-Haignere's team looked closer still. Some people with epilepsy have electrodes placed on the surface of the brain before surgery, to show doctors where their seizures start. They wait in hospital for several days. Fifteen such patients agreed to hear the same 165 sounds. Electrodes on the brain give a much sharper picture than a scanner.
A song makes a neat test, because a song is speech and music at once. Suppose the brain handled a song with a speech population plus a music population. Then a song could draw no more than speech and music would draw added together. The analysis turned up a population that answered almost only to music with a singing voice in it. Every song drew a strong response from it. Speech and instrumental music drew little or none. And a song drew more from it than speech and instrumental music added together. The team thinks this population answers to the combination itself, a voice singing with music. Exactly what it does is still unknown. Norman-Haignere said, "This was a finding we really didn't expect."
The song population came from all fifteen patients' electrodes taken together. Seven single electrodes, in three of the patients, showed the same taste on their own, and the analysis still found it with those seven left out. One of the three patients had no musical training, so this isn't something only musicians have. Norman-Haignere said that to a scanner, the song and music populations are "so close that you can't disentangle them."
Not everyone reads these results as music having machinery of its own. Aniruddh Patel, who studies music and language, offers a different picture. In his view, language and music keep separate stores of knowledge, which is why a stroke can take one and spare the other. But they share a limited pool of resources for fitting each new word or chord into a structure as it arrives. That idea was built to square the patients with the scans.
In 2009 a team including Patel tested it. Ninety-six students in San Diego read sentences on a screen, one chunk at a time, while a chord played under each chunk. Some sentences set a trap, as in an example from Patel's paper: "The attorney advised the defendant was guilty." Most readers first take the defendant as the one being advised. The word "was" forces a rethink, and readers slow down there. They slowed down more when the chord under "was" came from a distant key. A word that was merely odd in meaning, like angry pigs where you'd expect angry dogs, didn't interact with the chords. Nor did a chord played on an unexpected instrument.
Fedorenko's group answers that if such a shared pool exists, it lies outside the language areas, in the brain's general machinery for hard tasks. Some studies have also found subtle trouble with music in people with aphasia, though that evidence is mixed.
The widest look came in 2024, from a team in Marseille. Eighteen people with epilepsy had thin electrodes running deep into the brain, again to find their seizures. Most of them were recorded in their hospital bedrooms. Each heard ten minutes of a spoken story and ten minutes of instrumental music. Between them, the electrodes reached most regions of the brain, well beyond the hearing areas.
The Marseille team called a response shared when a spot answered to the story and the music about equally. About seven in ten of the spots that responded at all were shared. No region of the brain was given over to speech or to music. The team concluded that speech and music mostly run on the same resources.
Put the two studies side by side, and ask each the same two questions. First, the sounds. The Marseille team argue that two-second clips are too short for music, whose phrases run far longer than speech's. Marseille, for its part, used one story and one piece of music. Second, the tool. A reviewer of the Marseille study pointed out that each of its electrodes takes in thousands of neurons. A small population mixed among others is what such a tool can miss. Even Norman-Haignere's single electrodes showed the song taste only weakly. I think the tool matters more than the clips. Short clips were enough to reveal a music population, and one story and one piece of music can't rule it out.
There's one more piece of evidence for sharing, and it comes from people born with amusia. Early tests found they could tell a statement from a question as well as anyone. "He speaks French," with the voice falling at the end, is a statement. The same words with the voice rising are a question. The London group made that rise smaller. With small rises, people with amusia struggled, in speech as well as in plain tones. Only two of the sixteen they tested had noticed any trouble in daily life, probably because faces and context fill the gap. So the music deficit reaches into the pitch of speech. Some of the machinery for pitch, at least, seems to be shared.
The stroke patients point the same way: more than a quarter of the Finnish patients with amusia had aphasia as well.
Here is how I'd put the two sides together. It's my reading. Much of what the brain does with a song, it also does with speech, from taking in the raw sound to tracking its pitch. Patel thinks some of the work of building structure is shared as well. Inside that shared machinery sit small populations that answer almost only to music, and one to singing. The Marseille team allow for exactly that. They write that their results and Norman-Haignere's can be reconciled if separate populations sit inside the same region. In 2015, Peretz and colleagues called the overlap between music and speech "an open question for the field." Nine years later, it still is.
Is there a music center in the brain? Not one you could circle on a scan. I'd say the question gets two answers, because it can be asked from two distances. From a scanner's distance, music shares a neighborhood with speech. Up close, some populations answer almost only to music, and one to song. How close you look decides whether music has a place of its own.
So the patients and the scans only seemed to clash. Strokes and amusia from birth pull music and language apart. Blurry averages and crowded dots push them back together. Look person by person, or inside the dot, and the music population appears, tangled among neurons tuned to pitch. Pitch stays common ground, and perhaps the building of structure does too.
One skill looks most like music's own: knowing which notes belong in a tune. Peretz points to one candidate for a music-only system, the one that fits pitches into a key. My guess is that Ravel kept that sense. He could still hear small mistakes in his own music while he struggled to play it. People born with amusia miss a wrong note in a tune they've never heard. Most people catch one without a single music lesson. That's where the next lecture begins. How does a brain that has never studied music know which note is wrong?