Scholar-robots sound out musical mystery
Acertain nostalgia passed over me when I thumbed through a new scientific paper titled “Universality and diversity in human song” this week. The paper, which appears in Science and has 19 named authors from hither and yon, is an interesting example of a new research frontier sometimes called the “digital humanities.” It represents a large collective effort to arrive at an answer to the question whether music has existed in every human culture. I’ll go ahead and spoil the ending: it definitely looks like a yes.
Poets and critics have always talked of music as a universal, even defining characteristic of the human species. This is a belief most people would think of as natural, even axiomatic. But for precisely this reason, it has always been doubted by anthropologists, ethnomusicologists, sociologists, and others working in the modern scholarly tradition.
They are trained to be skeptical, or to deny, whether anything can be “universal” among humans or cultures. And even if everybody has some kind of music, there is no reason to suppose that any feature of it is innate, biologically programmed, or cross-cultural.
A question like this is smack- dab in the centre of the humanities, so how would we go about settling it? If you were a traditionalist, a monk or an Oxford don perhaps, you could dedicate a few decades to ransacking the world’s ethnomusical literature and seeing what conclusions might result from squinting at it. Or maybe you could attack the problem by means of experiment. As the Science paper explains, there have been stabs at taking music into remote places and seeing if Isolated Ethnic Group X attaches the same descriptors to a collection of tunes as does Ragbag of Broke American College Students Y.
In the 21st century, we have the possibility of a third approach to the question: to wit, the digital humanities. The DH way starts by noticing that all the difficult legwork has already been done. We have detailed ethnographic notes gathered from hundreds of cultures that were still in a relative state of innocence vis- a- vis global influences. And we have a vast treasury of recordings of their music. No one can read all of the notes, or listen to all of the music, but we can now get at it with — ta- da! Computers!
We are speaking here of “artificial intelligence.” In this setting it means that the researchers sampled a cross- section of world cultures from a vast database of ethnographic observations, as well as a corpus of sound recordings, and used statistics and text- mining to look for trends and common features.
It turns out that all cultures seem to have music ( as well as dancing), and all appear to possess songs with actual words in the local language. The researchers identified four known types of songs that all cultures appear to possess — dances, lullabies, healing songs, and love songs — and found that when they converted the written ethnomusical descriptions of performances to variables in a three- dimensional space, the points clustered well, falling fairly neatly into the four groups.
Different societies have different distributions of musical features — some seem to sing in a more severely formal way than others, for example — but they all overlap somewhat. Study of the sound recordings — made by means of both human musicologists and automated algorithms — suggests that the tendency for songs to adhere to a tonal centre may be universal as well.
The researchers could not, however, find any sign that singing evolved from any “single, prototypical adaptive function”: hopes of deriving some grand unified theory of the evolution of music from the exercise were dashed. ( The horde of authors includes Steven Pinker, the Canadian- born linguist who is a renowned popularizer of “evo-psych” ideas.)
The paper offers a lot to think about, much of it speculative or exploratory. The point of the thing is surely as much to demonstrate the power and
As I say, it made me nostalgic.
promise of the digital- humanities idea — the idea of turning computers loose on the documentary record of humankind, compiled at large scale — as it is to make a particular argument about the nature of music.
The approach taken by the team has a certain virtue of convenience: at a bare minimum we now have a cost- effective “third thing” to try that might be no worse than the don’s approach and the experimenter’s. But their work is also practically checkable in a way that the obiter dicta of an ethnomusicologist or the findings in one experiment are not. The databases behind the Science paper are open to scrutiny, and their various calculations are in principle reproducible.
As I say, it made me nostalgic. I went to graduate school in a history department, at a time and place at which the don’s view of the universe still prevailed. If any of us had been told there was a 45-million-word corpus of historians’ notes newly available for scrutiny, the expected reply would surely have been “I suppose I’d better get started.” Even in a faculty that had some first- class economic historians, we scholar- flunkies somehow absorbed a defiant attitude toward our own ignorance of the quantitative or the algorithmic: “if it’s graphable,” went one proverb, “it’s laughable.” This is, in retrospect, recognizable as a fearful precognition — a historians’ instinctive anticipation of the future in which we now reside.