Carl Zimmer in The New York Times:
In 2011, a 66-year-old retired math teacher walked into a London neurological clinic hoping to get some answers. A few years earlier, she explained to the doctors, she had heard someone playing a piano outside her house. But then she realized there was no piano. The phantom piano played longer and longer melodies, like passages from Rachmaninov’s Piano Concerto number 2 in C minor, her doctors recount in a recent study in the journal Cortex. By the time the woman — to whom the doctors refer only by her first name, Sylvia — came to the clinic, the music had become her nearly constant companion. Sylvia hoped the doctors could explain to her what was going on.
…For their experiment, Sylvia put on earphones and sat with her head in a scanner that detects the magnetic field produced by the brain. On the day of the study, she was hearing selections from Gilbert and Sullivan’s “H.M.S. Pinafore.” Every few minutes the scientists would switch to Bach for 30 seconds, to tamp down the hallucination. When the real music stopped, Sylvia pressed numbers on a keyboard to rate the strength of her hallucinations while the scanner recorded her brain activity. Dr. Kumar and his colleagues later pored over the data. They compared Sylvia’s brain activity when the hallucinations were strongest with when they were at their weakest. They found that a few regions consistently produced stronger brain waves when the hallucinations were louder. It turned out that they are regions that we all use when we listen to music. One region becomes active when we perceive pitch, for example. Another region becomes active when we recall a piece of music. Dr. Kumar argues that these results support a theory developed by Karl Friston of the Wellcome Trust Center for Neuroimaging. (Dr. Friston is a co-author of the new study.) Dr. Friston has proposed that our brains are prediction-generating machines. Our brains, Dr. Friston argues, generate predictions about what is going to happen next, using past experiences as a guide. When we hear a sound, for example — particularly music — our brains guess at what it is and predict what it will sound like in the next instant. If the prediction is wrong — if we mistook a teakettle for an opera singer — our brains quickly recognize that we are hearing something else and make a new prediction to minimize the error.
More here. (Note: Could this explain deja vu?)