Songwriting by numbers

Can a crowd write a song? That’s what an online experiment by computer programmer Brendon Ferris in the Dominican Republic is hoping to determine. Users are invited to vote on the notes of a melody, one note at a time, and the most popular choice secures the next note. The melody is selected to fit an anodyne chord sequence, and as far as I can make out the choices of notes are restricted to those in the major scale of C, the key signature of the composition. I’m not sure if the notes are allowed to stray out of the single octave range beginning on middle C (the New Scientist article provides very few details), but so far they haven’t. In other words, the rules are set up to ensure that this will be a pretty crappy song come what may, with all the melodic invention and vocal range of Morrissey (oh ouch, that’s not going to be popular!).

Even putting that aside, the experiment bears no relation to how music is composed. No one decides on a melody note by note, or at least not outside of the extremes of, say, total serialism, where everything is determined algorithmically. Neither do we hear a melody that way. We group or “chunk” the notes into phrases, and one of the most salient aspects of a melodic line that we attend to – it’s what infants first discern, irrespective of the exact relationships between successive pitches – is the overall contour. Does it go up or down? Is the pitch jump a big or small one? The melodic phrase is, in general, a single meaningful unit, and its musicality disappears once it is atomized into notes. The very basis of our musicality lies in our propensity to arrange sound stimuli into groups: to bind notes together.

But this doesn’t mean that the experiment is worthless (even if it’s worthless as music). It potentially raises some interesting questions (though as I say below, the answers in this case are highly compromised by the constraints). Will this democratic approach to making melody result in a tune that shares the characteristics of other conventional tonal melodies? In other words, can the crowd as a whole intuit the “rules” that seem empirically to guide melodic composition? It seems that to a certain extent they can. For example:

- the crowdsourced melody (to the extent that can be judged so far) exhibits the same kind of arch contour as many common tunes (think of “Ode to Joy” or “The Grand Old Duke of York”, say), rising at the start and then falling at the end of the phrase.

- the contours tend to be fairly smooth: an ascent, once started, persists for several notes in the same direction, before eventually reversing.

- the statistics of pitch jumps between one note and the next exhibit the same general pattern, within the limited statistics so far, as is seen for music pretty universally: that’s to say, there are more small pitch steps than large ones, with most being just zero, one or two semitones (especially two, since this corresponds to the distance between most successive note pairs in the diatonic scale). Here’s the comparison: the statistics for a sample of Western classical music are shown in grey, the thick black line is for this song:


But there are some anomalies, like those weird downward jumps of a seventh, which I suspect are a consequence of a silly restriction on the span of the allowed note to exclude the upper note of the tonic octave: you have to go back down to C because you can’t go up. So perhaps all we really learn in this case is totally unsurprising: people have assimilated enough from nursery rhymes not to be picking notes at random or putting rests in weird places, they have intuited some basic principles of harmony (so that we’re not getting B naturals against an F chord), and that if you permit only the blandest of note choices against the blandest of chord sequences, you’ll get a tune that is of no real interest to anyone.

That’s the opposite of what Ferris was hoping for. “My way of thinking was, if the crowd decides what the next note is, then there must be something there that appeals to the most people,” he has said. “The song should sound good to everybody.” But even if the rules weren’t so badly chosen, this totally misunderstands what music is about. What snags our attention is not the obvious, the consensual, the average, but the unusual, the unexpected. But that can’t be arbitrary: there are also rules of a sort that help to make the unexpected work and prevent it from seeming unmotivated. Whether the crowd could, if given the right options, find its way to that sort of inventiveness remains to be seen; I’d be astonished if it could do so note by note.

Something of this same nature was tried before, with more insight, by the avant-garde jazz musician David Soldier, who is the pseudonym of the neuroscientist David Sulzer at Columbia University. Sulzer wrote a song based on surveys of hundreds of people to discover what elements, such as instrumentation, tempo and lyrics they liked best. He called the result "Most Wanted Song". I haven’t heard it myself, but some people have described it as a sickly abomination, while others have said that it sounds a bit like Celine Dion. Which I suppose is the same thing.

Sulzer’s whole point is that trying to define the perfect song according to some kind of measure of popularity is liable to end badly. I think Ferris is discovering that too.