A population of male humpback whales, during the winter breeding season, sings a long, repeated song that can last more than twenty minutes and be heard many kilometres away through the water. Within any one population, the individual singers sing almost identical versions of the same song; across populations in different ocean basins, the songs differ considerably. The most remarkable feature of humpback song, however, is that it changes continuously over time. New phrases appear; old phrases disappear; the whole song gradually shifts. Over a period of several years, a population's song may become quite different from what it was at the start, while remaining recognisably related from one year to the next.

Systematic recording of humpback song began in the 1960s, when improvements in underwater microphones allowed long recordings to be made at whales' normal singing depths. The pioneering biologist Roger Payne produced the first widely distributed recordings, which contributed to public support for the international moratorium on commercial whaling that came into effect in 1986. Since then, a succession of research programmes has produced continuous recordings from populations across most of the world's oceans. The accumulation of this data has made humpback song one of the best-documented examples of animal culture.

The individual song is hierarchical. The smallest unit is the note, a sound lasting a few seconds. Notes are organised into phrases, which are repeated several times to form a theme. Themes are strung together in a specific order to form the full song. A typical humpback song contains between five and ten themes, and the whole song lasts fifteen to thirty minutes before repeating. Singers within a population conform closely to the current version of the song; variation between individual singers is modest, and the main singers all begin and end with the same themes in the same order.

The dynamics of change are distinctive. Changes typically begin with a small number of innovators in a population and then spread rapidly, as other singers adopt the innovation. Researchers tracking a population from year to year can often identify a specific innovation - a new note, a modified phrase, an altered theme - and watch it spread through the population over a few months. Not every innovation succeeds; many are adopted by a few singers and then abandoned. Successful innovations, however, can spread to nearly the entire population within a single breeding season.

Most remarkable, perhaps, is that the songs also spread between populations. The Australian bioacoustician Nathan Fraser and his colleagues have tracked the eastward spread of song innovations from the population of east Australian humpbacks to the population of New Zealand humpbacks, a distance of over two thousand kilometres. A new song that appears in east Australia in one year is typically adopted by a portion of the New Zealand singers by the following year, and by the year after that, the New Zealand song has shifted substantially toward the Australian version. The direction of spread, east from Australia, corresponds to the geographic arrangement of breeding areas and migration routes. Fraser has described the eastward flow as the most thoroughly documented example of large-scale cultural transmission in any non-human animal.

The mechanism of spread depends on males hearing one another sing. Males from different populations occasionally overlap during migration, particularly at the high-latitude feeding grounds where populations that winter in separate breeding areas sometimes feed in the same Antarctic waters. During these overlaps, a male from one population can hear the song of another, and apparently, in some cases, can adopt elements of it. The elements are then carried back to the original breeding area, where they may be adopted by the other males. The mechanism is, in effect, cultural transmission via migration, and it produces a pattern of long-range cultural exchange that no other non-human species is known to display on such a scale.

What the song is for remains debated. The most widely held view is that it is part of reproductive display, though in humpbacks, unlike many other songbirds and mammals, the song does not seem to serve as a simple territorial or courtship signal. The singing takes place at breeding areas where receptive females are present, but many of the singers are not obviously approaching or being approached by females during their singing. Other hypotheses include the possibility that the song serves to coordinate male social relationships or to advertise the singer's location to other males for loose social purposes. Fraser has argued that humpback song probably has multiple overlapping functions, and that its continued dramatic cultural evolution may reflect, among other things, ordinary aesthetic preference among the whales themselves.

The cultural perspective on humpback song raises broader questions. If an entire population of whales shares a detailed, continuously evolving song that can be transmitted across ocean basins, the distinction between culture as a human possession and genetically coded behaviour as an animal possession looks less sharp than it once did. Some researchers have proposed that humpback song should be considered a non-human cultural tradition in the full sense of the term, with its own history, its own regional variation, and its own evolution. Fraser has described the accumulated catalogue of humpback recordings - now covering more than fifty years of continuous observation - as one of the most detailed records of cultural change for any species, human or otherwise, and he has argued that the implications of such a record have not yet been fully absorbed by either biology or cultural history.

The conservation implications are, perhaps unexpectedly, substantial. A population of whales that has preserved a distinctive song tradition across generations may be doing so through a web of social relationships that is itself the subject of conservation concern. Disruptions to the social structure - through shipping noise, selective harvesting of particular individuals, or habitat loss at migration staging points - could, in principle, damage the transmission of the song tradition even while leaving the population's numbers intact. Fraser has argued that acoustic monitoring of humpback song should be treated as an indicator of population health more generally, because a healthy song tradition implies a functioning web of social interactions among the singers that other monitoring methods may miss. The song, on this view, is more than a curious phenomenon; it is a signal of the internal state of the population that sings it.