Sleeping birds might be proofing songs.

Are birds learning--or at least fine-tuning--music in their sleep?

That's a suggestion proposed by Daniel Margoliash and his colleagues at the University of Chicago to explain an odd finding in their new study of the song machinery in a bird's brain.

They focused on a structure called the robustus archistriatalis (RA), the control center for the nerves that drive singing movements. Researchers have known that this brain structure can respond to sounds, though just why has been perplexing. In the Dec. 18 SCIENCE, Margoliash's team reports an even more puzzling development: RA is less sensitive to sound, particularly to the bird's own voice, when the creature is awake than when it's asleep or anesthetized.

The birds don't sing in their sleep, and no one is suggesting that the brain has evolved some pathway to detect midnight serenades. Instead, Margoliash speculates that sleeping brain activity might play a role in learning and maintaining a song. Sleeping brain activity in rats seems necessary for them to learn how to navigate new spaces, according to experiments by Bruce L. McNaughton of the University of Arizona in Tucson and his colleagues.

Margoliash studies male zebra finches, which learn to sing as youngsters. He says their chattery strings of syllables sound like "a cross between Bugs Bunny and a squeaky' door." To keep singing well, a bird seems to need auditory feedback. If an adult is deafened, his song gradually deteriorates, developing uncharacteristic variety, as well as pops and clicks.

In laboratory experiments, Margoliash and his colleagues implanted electrodes in the birds' brains. When researchers played a recording of a bird's own song, a finch showed 5 to 20 times the RA activity when asleep as when awake.

When no recording was playing, anesthetized birds showed bursts of RA activity synchronized with impulses in a song structure called HVc. Pathways from the cells that detect sound feed into HVc, so Margoliash speculates that it might be downloading the day's sounds to RA. This activity could provide the feedback that keeps a bird singing properly.

The idea that RA plays a role in learning does not seem out of the question to Mark Konishi of the California Institute of Technology in Pasadena, Calif.

He also predicts that the finding will not settle the debate over the motor theory of sound perception, which holds that nerves controlling speech movement also respond to sound and help the brain interpret speech. Fernando Nottebohm of Rockefeller University in New York and his colleagues proposed in 1985 that the theory applies to bird song, as well. Margoliash argues that the theory can't be correct if the motor control center hardly responds until the bird sleeps.

Heather Williams of Williams College in Williamstown, Mass., who co-authored Nottebohm's proposal, agrees that RA's delay makes "far-fetched" the idea that RA serves as an aid in song perception. However, other song-related brain structures, like HVc, might still play that role.

What's more significant about the new report is the novel research angles it opens, she says. "Basically, you 'dream' of your song, and that's how you make adjustments in it. It's an interesting idea."