A young indigo bunting in its first summer doesn't know which way is north. By autumn, when the time comes to fly south, it will know. What happens in between is one of the stranger things I've come across: the bird watches the sky rotate.
The stars rise in the east and set in the west. They pivot around a fixed point — Polaris, more or less — and that fixed point doesn't move. An indigo bunting spends its first summer nights watching this rotation, and from it extracts a direction. Not by learning that Polaris is north, but by finding what doesn't move. North, for the bird, is the still center of a turning system.
In the 1970s, Stephen Emlen tested this by raising young indigo buntings in a planetarium. He manipulated the sky. For one group, the stars rotated normally, around their usual axis. For another, he shifted the rotation to a different point — an arbitrary location in the sky that had nothing to do with north. The birds in this group, come autumn, oriented toward their newly learned "north." They had calibrated to the fake axis. Their south was not our south.
A third group was raised without any sky at all. They couldn't orient at all. The information wasn't there to begin with, and nothing was inherited to replace it.
What I find strange about this is that the bird is not memorizing a position. It's detecting a relation. The star compass isn't built from a map of where specific stars are — it's built from a pattern observed over time: this region of the sky barely moves, that region moves a lot. The still point is what matters, and the still point is found by watching the whole thing turn.
You can think of this as the bird solving a problem without knowing it's solving a problem. It has no concept of north yet. It has no label to attach to what it's learning. It's just watching the sky, night after night, and something in the visual processing is tracking which stars maintain their relative position over time, and building from that a sense of where the rotation axis is. After roughly fourteen nights of exposure, the compass is set. After that, the bird knows which way is north — but "knows" in a specific, narrow sense: it has calibrated its directional sense to a structural feature of the sky it watched.
There's a version of this that sounds like learning. A young bird watches, extracts, updates. But what it extracts isn't a fact — it's not "Polaris is north." It's more like: the space around that region is the low-motion region of the sky. The navigation system runs off that invariant. Polaris could move (it does move, over thousands of years, as Earth wobbles), and a bird exposed to a sky where Polaris had drifted would calibrate to wherever the new still point was. The specific star doesn't matter. The axis matters.
I'm not sure what this means for what the bird "knows." It knows north in some functional sense — show it a planetarium sky and it will orient correctly. But its north is contingent on which sky it happened to observe in its first summer. If Emlen had rotated the sky differently, the bird's north would be different. The orientation is genuine, the calibration is learned, but the calibration is parasitic on a specific visual history. Change the history, change the north.
There's also something I don't fully understand here: how does the critical period end? The bird watches for fourteen nights and then — what? Closes the window? Stops updating? The pied flycatcher work suggests that some birds can still learn the star compass in spring, during their first return migration, which makes the timing more flexible than it first appeared. But there has to be a point where the compass is fixed and stops being overwritten by new observations. I don't know what closes that window. The bird doesn't decide to stop learning. Something does it for the bird, by some mechanism I can't find clearly described anywhere.
What I keep returning to is the simpler version of the question: the bird is looking for what doesn't move. Everything else turns. It finds the exception. And that exception becomes its north, its anchor, the reference against which everything else is measured. The whole compass is built on one invariant, discovered from motion.