Music and Evolutionary Biology? You might be wondering what those two topics have to do with the technology stories you are reading on O'Reilly ? Well, aside from being a welcome, interesting break from stories about our continued global financial demise, this particular story involves an interesting confluence of Science, Art, and Computing.
Armand Leroi is an Evolutionary Biologist with the Imperial College in London. Leroi is leveraging the ability of computers to analyze sound to create a Cantometric description of traditional music from various cultures. Cantometrics is a form of musical analysis developed by Victor Grauer and an Ethnomusicologist named Alan Lomax in the 1950s. Lomax and Grauer developed a series of objective measures which could be used to characterize music - a codification of musical structure and qualities. By applying these measurements to traditional music, Lomax uncovered similarities between different culture's musical artifacts. In the 50s, Lomax, Grauer, and anyone else interested in conducting a Cantometric survey would listen to and analyze music with the aid of a handbook and a series of rules.
Fast forward nearly fifty years from the dawn of Cantometrics and scientists like Leroi are able to write programs to analyze music and create a digital Cantographic survey. Being an evolutionary biologist, one of Leroi interests is in exploring how music "evolves" and how one could measure similarities and differences between geographically distributed populations. Much in the same way the indigenous population of Patagonia in Chile shares genetic history with the indigenous population of the Kamchatka Penninsula in Russia, Leroi has shown that there are similarities in the Cantometric signatures of the population's traditional music.
In this interview, Leroi refers to Ani Patel, a Senior Fellow in Theoretical Neurobiology at the Neurosciences Institute. Patel is focused . Patel's research is groundbreaking, entertaining, and involves birds dancing to the Backstreet Boys. For more about Patel's research watch one of his lectures "Music and the Mind".
Note: The interview was filmed outside of Google Headquarters in Mountain View during this year's Scifoo. Unbeknowst to me, Google's Headquarters are adjacent to some sort of stadium, about half way through the interview you'll hear some persistent cheering, the crowd isn't cheering for Leroi, but if it makes the video interesting to pretend that they are, go for it. It you are paying attention, you'll see the wings of space-age, car-sized jet being deployed and retracted several times.
The Evolution of Music
Armand Leroi: I work at Imperial College in London, where I am a Professor of Evolutionary Development in Biology.
Tim O'Brien: That's good, that's good. What brings you here?
AL: SciFoo! Goodness knows why. I was one of the lucky, the gifted, to be invited by O'Reilly and Google and Nature, to come to this rather strange and wonderful place to speak to all these people.
TO: Tell me about the session you gave earlier this afternoon, it had something to do with music. How does music relate to evolutionary biology?
AL: Music is an evolutionary problem. Most people don't appreciate that, but there's a few of us who increasingly think that it is. And in two ways: First of all, the question of why do humans make music and most animals don't or is that indeed true? And tied to that, the question of where does the human music-making instinct come from and how is it related to language and what is the cognitive basis for the both.
In the session you saw this morning, Patel, very bright, spoke about that and finished up the session with a clip of the dancing cockatoo, which was delightful. And the cockatoo was dancing to Back Street Boys and it was all good. My presentation, which came after that, was naturally rather subdued and quite frankly a bit lame by comparison. Because what I was talking about was the other kind of evolution that music has namely cultural evolution.
TO: You talked about something called Cantometrics.
AL: There was an American, Allen Lomax, he died a few years ago, who was one of the great treasures of American cultural life. He and his father, before him, from around the 1920-30's through the 1960's, went around south and recovered and recorded much of what is important of the American indigenous music, by which I mean the blues and so forth in the south and immortalized that music and revived it.
What makes Jr. (Allen), was a little bit unusual for a folk music collector. He had a fine, analytical mind. And he had a few theories about what shapes folk music which he chose to test using a system of his own devising called Cantometrics
Cantometrics was, in effect, a way of measuring the properties of songs, how they are sung more than melodies; what are the performance qualities, whether people were singing with an open voice, nasally, whether they were singing in rhythm, and if so, are they singing collectively, and so on and so forth. He collected about 5000 songs from around the world, traditional songs, and stored them in a system and attempted to track the major patterns in them.
I heard about this a few years ago, and to me it was obvious what should be done. One should use it to try and cover the history of human song. And so that is what we've been doing.
TO: Cantometrics is the quantitative study of music. Tell me some of the data points that you get when you analyze the song catemetrically.
AL: When you analyze a song, catemetrically, some of the things you study is not so much the melody of the song, but how it is sung -- What is technically known in the trade as performance style. What you are interested in are the songs, what is the kind of vocal attack, are the songs being sung with an open mouth, sung with a closed mouth, in a compressed tone, sung nasally, is there a lot of glissando, or does the song progress in a "ha, ha, ha, ha" step wise fashion and so on. These are the kinds of things that can be easily measured in a song.
It turns out they are the things that can discriminate songs from different parts of the world very easily and that could possibly tell us about their evolution.
TO: Could you say that the cantometric data from a culture's music is similar to a population's DNA? Is that your take?
AL: Yeah. The thing about a song, it's most abstract, or it can be thought of as a string of information, rather like a sentence, a word or like DNA. What it all has in common with words, or a sentence or DNA is that music is passed from one individual to another.
Music can be transmitted laterally. I can give it to you, even though you're not my child. But DNA can be transmitted vertically from parent to child. But that's just a detail. The point is that it's transmitted from one person to another. And like all things transmitted from one person to another, it's transmitted perfectly, and the result of that is that you get what Darwin calls a modification by descent process, in which other times, music changes. If it changes like DNA, like languages, you can see the results in the songs that we have today; and like DNA and like languages, you can, in principle, reconstruct the history of song. But, if you're prepared to make assumptions about what that evolution looks like.
What we aimed to do, initially, which proved to be a little bit difficult, is to reconstruct what the earlier songs that humans actually sung, what they actually sounded like.
TO: So given songs from two cultures, could you do the math just like we do in genetics and say these two cultures have songs which are derived from songs from another culture or do we not have the historical records?
AL: So the whole principle, you should be able to take any two songs, from a given culture, do the math and reconstruct the ancestral condition -- a practice that turns out to be very difficult and it's for a very straightforward reason.
The math that evolutionary biologists depend upon to re-construct the evolution of trees, assumes, that in fact you have an evolutionary tree, that evolution happens... you have one entity, divided into two and so on and so on.
Linguists assume the same thing, languages do that too. It's an assumption they're prepared to make. It's pretty clear, however, when you look at music that that assumption does not hold. The reason is music gets mixed up. Songs recombine, and mutate at a relatively high rate -- at least they suddenly recombine that they obscure that pattern of descent. So you can't just apply the math that evolutionary biologists have developed to reconstruct song history. You have to develop some math of your own.
We've done that and we've been recovering what we think are some deep historical world signals out of the song.
TO: So tell me how you do that. What media do you use, do you use software, is it stored in a database, tell us about the research that's been going on.
AL: I can tell you something about the research. We have a database of songs, which is large by the standards that Lomax is working on. Its 5000 songs, belonging to 800 cultures, 44 variables each, which has about a dozen variable states which they can adopt.
Small beer these days, big in 1965 when it was originally developed. What we've developed is just a particular form of the data. It's a combination of ordinal and numerical data. It's a particular kind of cluster, a kind of means, not rocket science by any means, but it happens to be well suited to this data.
What we've been doing is asking if you give the data set to the clustering algorithm, what is the best way of dividing up that world of songs. How many clusters is best, secondly what are the clusters; and thirdly once you've got those clusters, what do they need.
At the moment, we're working with the Lomax data collection, culled together the hard way, Splicing together and taking records, all done in the 1960's which was overwhelmingly from what we call musicological records. Some of the musicologists who have gone out since the invention of the gramophone into different parts of the world and brought back the songs and stuck them into their own databases or in libraries of music collection and published them.
Smithsonian, for example, has a very big collection.
Lomax did a lot of collection and a lot of the best and wonderful parts of the database are songs he collected in various parts of the world. That's the raw data which forms the basis of the database that he developed.
Lomax worked hard for many years even up to his death in the 1990's. Adding songs to his collection and . . . the fact of the matter is even after he managed to get 5000 traditional songs, which he thought was traditional, it's now very small beer compared to the number of songs that are available these days. Every undergraduate has 5000 songs in his iPod.
We should be able to do better especially as out there in the great musicological libraries, there are songs sitting on tapes, or which the libraries are now beginning to digitize.
What we want to do is analyze much larger set of songs. In fact we're trying to get 100,000 songs which is just an arbitrary number, from all the cultures around the world and begin to analyze them using new digital tools to bring about a new kind of Cantometrics and in effect and revive a new Cantometrics that doesn't depend on actually listening to the songs the hard way as Alan Lomax did.
TO: I've heard you talk about musicology, yet you are an evolutionary biologist, by training, defining the new field of "neo-cantology". Have you received any push back from people who do that in musicology, from people who do that in anthropology, to people who do linguistics? Has anyone ever said this is not the purview of an evolutionary biologist?
AL: well, nobody has said to me this is not the purview of an evolutionary biologist. People have been quite polite as far as this is concerned. It must be said, however, ethnomusicologists, although generous and polite, have, for the most part, been skeptical. They were skeptical to Lomax when he presented his work in the 1960's, which is indeed fallen into oblivion or at least to neglect until we approached Lomax Foundation and we began the work to revive, figure out the data, and clean it up and begin to reanalyze it.
The reason it had fallen/had become neglected was very much a cultural one. Ethnomusicologists are in the humanities. They don't do numbers, they don't do statistics and Lomax did. He invented the statistics on the fly back in the 1960's and put it into a computer which was as big as a house and they could not deal with the results.
We as evolutionary biologists found it entirely natural way to deal with the data sets and analyze them historically. I think they are starting to recognize that music is, their belief that music be studied entirely in humanistic terms of its meaning, in terms of identity and politics and individual performance. But those days must be numbered or at least a revival of the science of music is necessary. They're suspicious but fascinated too.
TO: Music evolves so that we could . . .
AL: What is music? It's a very good question and one that Patel could answer better than I, not that he has an answer any better than I. The first distinction you made is "why did music evolve?" The first thing you have to ask is does it serve a purpose, or is it Pinker charmingly called it "auditory cheesecake", a by-product of the language so to speak.
In my mind, something that is ubiquitous, and important and very difficult to deny, at least on an intuitive level, is that it has evolved repeatedly throughout the animal kingdom. There are some qualitative differences. Whales and birds do indeed have some sort of music-making ability.
So it's perfectly reasonable to suppose it does something for us. What? Guesses, that it's a form of communication, social bonding that hits neurons, parts of our nervous systems that straightforward verbal communication does not. Why it should do so, why it should be so powerful is less clear. My guess is that it in fact has something to do with sexual selection. Not an original suggestion of mine, but that's actually driving a lot of it. It has that same slightly ridiculous and extravagant quality about it that the peacock's tail does.
TO: So you're saying that great singers just reproduce more?
AL: That certainly been the suggestion. There are some statistics on the philoprogenitiveness of rock singers and jazz singers to back it up.
TO: You talked about your system as being able cluster music. Can you talk about what clusters have you identified, are you identifying genre, are you identifying types of music?
AL: I did identify clusters that depend entirely on the properties of songs themselves. That did not contain any additional information, no cluster information or geographic information. They could be anything, happy songs versus sad songs, working songs versus wedding songs, for example.
The obvious guess, however, after all we are dealing with 5000 songs from around the world, is that the songs/clusters are geographic in origin. People in different parts of the world sing differently. So we've mapped the statistics; and perhaps surprisingly or unsurprisingly, as you look at it, we found that clusters do pop up dominantly in one part of the world or another. Not entirely. There is a lot of variation in human song. The individual cluster may pop up in different parts of the world. You can look overwhelmingly that Africa is one color; America is another, and northeast Europe is another and so on and so on.
The question is why is this? If you look at the distribution of the clusters, the most parsimonious explanation of why they are the way they are or why they are where they are. They can be explained by very ancient migration patterns.
For example, one of our strongest clusters identifies people whose songs sung by Selk'nam, right at the tip of South America; and they are really similar to the songs sung by the Inuit of the Arctic, right at the top of North America; and those are in turn very similar to the songs sung by the Ainu of Japan; and a lot of other people in the Siberian Peninsula. The songs aren't much the listen to, basically they are chaps sitting on rocks screaming to themselves for the most part.
TO: Could you give me some sort . . . could you sing a song? What does it sound like?
AL: A Selk'nam man sitting on a rock, he's not happy, and he rocks back and forth and goes [groaning] and does that about an hour or so. I must say that one expects music to grow on one as time goes by. But I must say that Cluster F (Americas) has proven to be a tough one. And I say that after having listened to a lot of those songs.
TO: You didn't personally have to listen to them to get the data. It was mostly the software that was subjected to the rhythmic moaning.
AL: That's right. You're quite right. We did a lot of listening ourselves to ensure the software is doing what it is supposed to be doing. There's nothing quite like having a human ear sit through it all.
TO: Essentially the software is a "sentry". It listens the music finds a similarity and then a human listens to it and says, "Yep, that's some rhythmic moaning."
AL: You want to take a look at it yourself and make sure it's doing what it's supposed to be doing.
TO: Patagonia and Siberia, talk about the pattern.
AL: This is one pattern of many. But what is so striking about it is that it is so similar. The best explanation is it's not the only one, but it is the most parsimonious explanation. What you're looking at is the cultural after effects of the cultural legacy of the original peopling of the Americas, by the people from Siberia some 13,000 years ago. If you buy that, I can't go into all the evidence suggesting why that is true, it's just that when we listen to songs from people around the world, we're listening to song styles that are incredibly ancient, that may be tens of thousands of years old or at least thousands of years old.
Another example, if you go to the Congo, you listen to the Arche Pygmies . Then you go to San Bushmen of the Kalahari -- People thousands of miles away from each other, very different environments, and yet sing strikingly similar songs. Why? They could have I supposed evolved from songs sung south independently and convergently upon each other. But actually a much better explanation is that they are the relic of song style, once widely spread in Africa prior to the invasion of the Bantu speaking peoples who swept down Africa some 2000 years ago, and obliterated everybody in their party, except for a hand full of people at the Congo and in the Kalahari.
And all that is entirely consistent with the archeological and genetic evidence. So one of the exciting things we're discovering is it looks as though the music and the way people sing is very conserved. We tend to think of music as high speed, if you used to living in a high speed musical environment, in which the top ten are changing all the time, styles are changing all the time.
In fact, music is really conservative. In fact, it looks as if it has a historical signal that may be older than languages. You can see the historical depths in music that you can't see in languages but you can see in genes.
TO: It's difficult to imagine music remaining static over 13,000 years.
AL: It is.
TO: Music is in the hyper evolutionary sort of state. Could you talk about is there anything similar in evolutionary biology?
AL: I think so. We know very little about how music is evolving. Or should I put it this way. we know enormous amounts about it because we all have iPods and we listen to music all the time and we see how it changes. We know remotely little about it in a quantitative way. We should be able to quantify the properties of music and should be able to do it better than Lomax. Just use digital analysis machines in order to see exactly how music is changing right now in front of our eyes.
I think it is the case that music is evolving much faster than in the past. And I think the reasons for that are relatively straight forward from an evolutionary theory.
If you ask where do you find conservation in populations versus rapid evolution? The answer must be that rapid evolution tends to occur in big populations there is a high probability of change and which new innovations could spread rapidly in the population.
Historically, the reason why songs have not changed that much is most certainly because people have lived in isolated, small populations. I'm speaking of it in a cultural sense and so the rate of evolution has been relatively slow and we've actually gotten to see just how conservative that is.
TO: You're saying in a world without flight and interstate highways, we might actually see 13,000 year old rhythmic moaning?
AL: Yeah, i think that's absolutely right. We see the rate of musical evolution speeding up now and a lot of what we see is in effect a combination to use population genetics jargon . . . and that is to say, mixing up.
A lot of musical evolution is taking the musical element from one culture and mixing it up with the cultural element of another and making it a de novo thing. That is the consequence of big networks, big populations, rapid movement -- all to the speed of evolution. And that, historically, hasn't been the case.
There's been a huge shift. Of course when I say all this, it is assertion because we don't really have all that much data. The study of music as a cultural artifact has really just begun to be a science. It was Lomax' brilliance, I think, to have conceived that it could be a science and we have been very slow in following up on that insight and pursuing his legacy.
TO: Are you worried about licensing issues when it comes to studying music? It seems like one of the flash points for copyright disputes. Who has the right to copy what? If the idea that all scientific data should be open, if your scientific data is covered by a copyright from 1958, how do you manage that?
AL: This is a problem we've been . . .one of the problems of studying music scientifically, is indeed, it is surrounded by all sorts of copyright and ownership issues. I must say, I have had difficulty, even for myself as a geneticist, as a human geneticist, even though scientific data is freely available to anybody who wants it. Clearly modern music is, of course, owned by people, companies, record companies and artists.
TO: It's a cultural artifact, but it's not necessarily owned by the culture in the way rhythmic moaning in Patagonia is.
AL: Well that's fort problem itself because every recording made of a particular tribal group, the recording in some sense is initially owned by the person who made the recording. Increasingly as people from different cultures are becoming aware of their own cultural legacies, they are laying claim to their rights they have over their own cultural productions.
All this means, I don't mean to dispute the rights/wrongs of it, it does mean it's hard to see, it's a problem, a potential impediment to the study of music. I would like to build up big database, which we are beginning to do that, that would then become freely available for anybody to study, to run their own digital analysis programs on and do analysis on to replicate our results, to test for results we haven't found. It's the result of free inquiry.
I must say never before have I come across so many impediments, and "tut tutting", and "we just can't do that", etc., as I have in dealing with the distribution of music. And I say that regardless of whom I'm talking to about it.
TO: The project you're about to embark upon, or that you've already started, sounds like it involves a lot of interesting technology. How would other scientists interested in collaborating with you or technologists interested in helping out, is there a way for them to participate?
AL: Yes. I will tell you what I think.
The vision I have for this project is quite a grand one in a way. The thing I love about it is once you get a notion that music has a history which can be recovered and studied, innumerable questions, a world of questions about processes, dynamics, form and history, and patterns, all of these question that evolutionary biologists ask all the time about the natural world, all this can be immediately asked about music in specific. More than that you can ask them, I believe, about any cultural artifact.
Bear in mind that music is just one kind of cultural artifact that has a modification by descent type process. Language is of course another. Linguists are pretty sophisticated (although they're not as good as they think they are). But they are better than the musicologists.
Here's another thought, what about images. Everywhere you go in the world, people make sculptures, patterns. And those images contain information and those images have come down to those cultures by a descent type process.
Imagine reconstructing in a proper scientific way the history of the representation of the human face for example. Practically every culture represents the human face. It's utterly distinctive. You just have to look at 50 African masks for example.
There is a world of historical information that remains to be discovered and can be reconstructed. It's not just one project, I think if you understand it; it's a whole new way looking at the diversity of the cultural world around us.
TO: Do you have a website?
AL: Yes, armandleroi.com