Not all birdsong is learned. In certain species a bird reared in isolation, hearing nothing, will produce the adult song characteristic of its kind at the appropriate season and reproduce it accurately. In others, isolation generates something recognisable as song and defective in every particular: the phrases are present in outline, but their sequencing is arbitrary and the fine structure disorganised. The distinction does not correspond to taxonomy in any orderly fashion, and for a considerable period the phenomenon was documented as a curiosity rather than explained.
The explanation that has emerged turns on what the song is required to accomplish. A song that functions principally to announce a species -- I am one of these, and not one of those -- can be incorporated without cost, because the target it must hit does not move. Every generation requires the identical signal, and a bird that inherits it entire is spared the labour of acquisition and the risk of defective acquisition. Where song is employed in this fashion it is dependably innate. Furthermore it is, characteristically, brief and stereotyped, for nothing is gained by elaborating a badge.
Learning appears where the target does move. In some species the song must correspond to a local population, because females discriminate against outsiders, or because a male must answer his neighbours in kind to maintain a territory among them. In such species the correct song is whatever the birds within earshot are singing, and no genome can specify that beforehand. A young bird in such a species listens throughout a sensitive period, stores what it hears, and devotes the subsequent months to bringing its own production into conformity with the accumulated store, often by a long stretch of quiet approximate singing that resembles nothing so much as practice.
The account predicts an intermediate case, and the intermediate case exists. Several species inherit a rudimentary template -- sufficient to predispose the young bird towards the songs of its own kind and away from everything else it hears -- and subsequently learn the particulars within it. A bird of this description reared among the songs of another species will ordinarily disregard them, and reared in silence will produce the same defective outline as previously. What is innate is not the song but a filter on what counts as one. That filter is precisely what a system would require if the target moved a little and not a great deal.
Consequently the traditional question, whether birdsong is learned or innate, was improperly formulated. The useful question is how much of a moving target the song has to hit, and the answer differs according to species because their circumstances differ. Learning is expensive and is deployed where a fixed answer would be erroneous. Nothing in the account requires a species to occupy a single pole, and several appear to have transitioned between them as their circumstances altered.
