Over at The Stone and the Shell there’s an interesting post by Ted Underwood, Hoyt Long, Richard Jean So, and Yuancheng Zhu, You say you found a revolution. It’s a critique of Mauch et al. “The Evolution of Popular Music: USA 1960-2010”, from 2015. Mauch et al. 17,000 recordings that topped the Billboard charts during that period, assessed their similarity on harmonic and timbral properties, and argued for three ‘revolutions’ during that interval, at roughly 1964, 1983, and 1991. Underwood et al. argue that the claim is overstated and that they’ve mis-analyzed their data. As Mauch at al. have made their data public, Underwood et al. were able to reanalyze it, to more modest conclusions.
In the course of explaining their work, Underwood et al. made some assertions I found to be problematic. So I wrote to Underwood about it, he replied, and has asked me to post my observations to my blog. That’s what’s in the rest of this post.
* * * * *
Why Assume Linear Direction in Time?
I’ve read my way through this and I’m not quite sure what I think. I have no strong attachment to the revolution argument – seems to me too many “revolutions” for that stretch of time ¬– but I think you have a hidden assumption in your argument. Here’s two passages where that assumption shows up:
History doesn’t repeat itself in the same way. It’s extremely likely (almost certain) that music from 1992 will resemble music from 1991 more than it resembles music from 1965. That’s why the historical distance matrix has a single broad yellow path running from lower left to upper right.
As a result, historical sequences are always going to produce very high measurements of Foote novelty. Comparisons across a boundary will always tend to create higher distances than the comparisons within the half-spans on either side, because differences across longer spans of time always tend to be bigger.
In short, the tests in Mauch et al. don’t prove that there were significant moments of acceleration in the history of music. They just prove that we’re looking at historical evidence! The authors have interpreted this as a sign of “revolution,” because all change looks revolutionary when compared to temporal chaos.
The assumption you’re making is that history has a default direction and that it is linear. That is, linear change of the kind we see in that data set requires no explanation, though acceleration and deceleration do. But I think that the direction itself requires explanation, though just how to go about that is not clear to me.
Of course, proposals of various kinds have been made that this or that historical process, or even all of history, is cyclical. And there are processes that happen in history that are cyclical. Think of the motions of the planets and other bodies in our solar system and their effects on earth (e.g. seasonal changes in weather). Now, on the one hand we (humanists) don’t normally think about such processes when we think of history and, on the other hand, we (scientists) know a great deal about how those processes work.
I suppose one can say that that’s all well and good, but it really isn’t relevant to this case because however it is that musical culture changes over time, it doesn’t change like that. I agree that musical culture doesn’t change like that, but I think that ultimately, if not here and now, the nature of musical change is something that does require an explanation. Imagine, for a moment, that pop music is cyclic on a 20-year period. If we take 1980 as our reference point then as we move to 1981, 1982, etc. we see the music getting more and more different from that of 1980. But then we hit 1991 and the degree of difference starts getting smaller until we reach 2000, which brings us back to the point (in the cycle) where we were in 1980. That’s not what we observe, of course, but why not?
Of course, it’s not hard to begin telling some story about that. My point is simply that a story IS required. We can’t assume linear change.
We need to distinguish between direction in time and direction in the feature space we’re using to evaluate, in this case, popular music. And at this point I’ll change examples, to Jockers’ 19th century novels (see links below). As you know at the end of his book he investigated influence by measuring similarity between novels. Each text was assessed on some 600 features and assigned a point in 600D feature space. When he then did this and that and visualized the resulting graph he was surprised to see that it laid the books out in rough temporal order and yet there was no temporal information in his data, no dates of any kind.
So there we’ve got linear order in a 600D feature space. Just what this order is, how we’d characterize it, that’s an interesting question, but we don’t need to deal with it now. But when it comes time to EXPLAIN why that order coincides with time, then we’re going to have to figure out what’s going on. What’re the mechanisms that produced those books over the course of a century? Why is it that they unfolded roughly over a certain direction in 600D feature-space?
The same question exists for popular music. Sure, people’s interests don’t change that much from one year to the next, nor does the output of the music business. That’s one thing. But why do small local differences seem to accumulate over time so that music doesn’t go back the way it was?
* * * * *
As my note indicates, I’ve been through this before, with Matt Jockers’ work in Macroanalysis. Here’s a downloadable working paper on that:
- On the Direction of Cultural Evolution: Lessons from the 19th Century Anglophone Novel, https://www.academia.edu/12112568/On_the_Direction_of_Cultural_Evolution_Lessons_from_the_19th_Century_Anglophone_Novel
Here’s two of the blog posts I put into that working paper: