Field notes

How Birds Learn to Sing

field-notes.example

A young white-crowned sparrow does not hatch knowing its song. It hatches with an appetite for one. In the first weeks out of the egg it listens far more than it sings, and what it hears in that window it will spend the rest of its first year trying to reproduce.

Biologists call that window the sensory phase. In white-crowned sparrows it opens around ten days after hatching and closes near fifty. A bird that hears an adult of its own species inside those forty days keeps the memory. A bird kept in silence through them grows up singing something thin and disordered, and no amount of later tutoring fully repairs it.

What comes next sounds like nothing at all. The fledgling produces subsong: soft, rambling, unstructured, closer to a hum than a melody. It is quiet enough that a person standing three metres away can miss it. Nobody is being impressed. The bird is finding out which muscles make which sounds.

Then the babble tightens. Recognisable syllables surface, drop out, come back in a different order. This stage is called plastic song, and it is the part that looks most like practice: the bird is comparing what it produces against the recording it made months earlier and trimming the difference.

Eventually the song stops changing. Crystallisation is the word for it, and it is oddly literal — one version hardens and the alternatives fall away. From then on a male sparrow will sing that song, with small seasonal drift, for the rest of his life.

Underneath all of this sits a small circuit of brain nuclei. HVC feeds the robust nucleus of the arcopallium, which drives the muscles of the syrinx. A second loop through Area X handles the learning itself. Damage the second loop in a young bird and it never learns properly; damage it in an adult whose song has already hardened and the song mostly survives.

The most direct evidence that hearing matters comes from birds that lose it. A young bird deafened before it starts to practise never converges on its tutor's song, even though the memory is intact. It cannot hear its own attempts, so it has nothing to compare them against.

Because the song is learned rather than issued at birth, it drifts geographically. Sparrows in one valley sing a version that differs audibly from the population two valleys over, and the boundaries can be sharp enough to map. These are dialects in the ordinary sense of the word: local, inherited by ear, and stable across generations.

Zebra finches learn differently again. A young male attaches to a particular adult — usually his father — and copies him closely. Take the father away and he will copy whichever adult male is available, but he copies a bird, not a sound source.

That distinction turns out to matter. Finches tutored by a loudspeaker learn poorly compared with finches tutored by a live adult, even when the audio is identical. Something about the presence of another bird, the back-and-forth of it, sharpens the copying.

Not every species shuts the window. Canaries reopen theirs each year and rebuild parts of the song each breeding season, adding and discarding syllables. Their song nuclei physically grow and shrink on an annual cycle.

Why go to the trouble? A long, accurate, well-executed song is expensive. It takes months of practice, brain tissue that must be fed, and time spent conspicuous to anything hunting. Females of many species prefer larger repertoires and cleaner delivery, which is a plausible reason the cost is worth paying.

There is a territorial argument as well. Song carries further than a display, and it works in the dark and through foliage. A bird holding ground can advertise continuously from one perch instead of patrolling a boundary it cannot see.

The comparison with human infants is close enough to be useful and loose enough to be misleading. Babbling in babies and subsong in birds both involve a long noisy period of self-generated practice guided by hearing. Both have a window in which input matters disproportionately. The similarity is functional, not ancestral — the two lineages arrived at it separately.

Recording matters to all of this. Sound spectrograms, which turn a song into a picture of frequency over time, are what allowed anyone to prove that a young bird's output converges on a specific tutor rather than approximating a species average.

Long-running archives now hold decades of recordings from the same populations. They show dialect boundaries moving, syllables disappearing from a valley over twenty years, and populations near cities shifting to higher frequencies where traffic noise masks the low end.

None of it is instinct alone, and none of it is learning alone. A sparrow will only ever learn a sparrow's song; the appetite is specific. But what it ends up singing depends entirely on which adult happened to be within earshot during a few weeks it will not remember.

Which is a strange thing to sit with. The song a bird sings for the whole of its life is a copy of something it heard before it could produce a single clean note, held in memory for months, then practised in the dark until the difference closed.