Listeners can spot a lullaby from an unfamiliar culture. A love song, barely.

The team looked for one song of each kind in each of the world's 30 cultural regions: 120 recordings. It found 118. The two missing are healing songs from Scandinavia and from the British Isles: documents show both existed, but by the early 1900s, when they could first be recorded, they were rarely sung.

The team described every song in two ways: through ratings by expert musicians, who judged by ear features such as tempo, accent or rhythmic regularity, and through measurements taken automatically from the transcriptions. A statistical model trained on those descriptions, and tested on regions of the world it had not seen, classified 50.8% of songs correctly; its 95% confidence interval (40.4–61.3%) includes the listeners' 42.4%. In other words, the musical features alone carried enough information for performance comparable to people's.

Then they took the kinds in pairs and looked for the features that separate them most clearly. Below are three of them, measured on the same transcriptions you heard in the game: each dot is a song, the tick is the median.

Each dot is a field recording made between 1898 and 2012 (a few dated only approximately) and published by an archive or on an ethnographic record. Choose a dot to hear the melody and see how listeners answered.

The hollow dots are the four Aboriginal Australian songs. Some Aboriginal communities restrict who may hear certain songs, such as sacred songs or those of people who have died, and the data do not say which rules apply to these four. Their melodies are left unplayed out of caution, not because of an error; the listeners' answers are included.

The best-recognised love song is Guarani, from Paraguay: 76% got it right.

Separately from the 118 recordings, the same team pulled 4,709 descriptions of moments when someone sings from Yale's HRAF ethnographic archive, covering a different set of 60 societies: one from each of George Murdock's 60 cultural clusters, across 30 world regions, drawn pseudorandomly and historically mostly independent of one another. Song appears in all 60. For each description, readers noted who sings, on what occasion and who listens.

Song at ceremonies appears in 59 of them. In 54, ethnographers note that a particular song could not be sung by just anyone. Singing for children appears in only 31, which says more about what ethnographers watched than about children: a low count may mean only that no ethnographer described it.