$500,000 grant for music research at MIT
Michael Scott Cuthbert uses computational methods to study Western music
Michael Scott Cuthbert
Michael Scott Cuthbert, associate professor of music, was recently awarded a $500,000 grant from the Digging into Data consortium. This grant will support his work in using computational techniques to study changes in Western musical style. He has received $175,000 specifically for his music21 project . On Thursday, Cuthbert sat with The Tech to discuss his work with music21 and his passion for combining computational techniques with music.
The Tech: Many of us with a musical background must be interested in your computational work and how it applies to music. What is the motivation behind your project?
Cuthbert: One of the main ways artists analyze art work or music is examining a piece very carefully, from all possible dimensions. But it’s really hard to put the work into the context of the time. How is the piece representative of its time period, or how does it break the mold? It takes us a very long time to look at one piece. In contrast, computers are good at getting an overview of a particular problem. For example, what patterns exist in how chords progress from one to another? Is the piece being looked at representative of the music grammar for the period?
People have been working on these questions for a long time in computational musicology, but the programming necessary to answer these questions has been difficult. My goal is to create basic building blocks to build libraries and analyze models of behavior that are easy enough for someone with basic programming knowledge and strong musical background to be able to handle. I have been working to make the algorithm easier to use so that it can be put together with other algorithms.
TT: What does the music21 project specifically address with its computational techniques?
MSC: Music21 is a suite of computer tools that deals with musicology — understanding chord behavior, in particular. How do people perceive sounds over the course of time, say from the 1300s to the present time? Certain sounds may be considered stable, conclusive sounds now — in other words, they can easily conclude a piece or movement. However, such sounds may not have been considered appropriate for concluding a piece several centuries ago. If we look at large number of pieces — such as 100,000 pieces — we are probably looking at hundreds of millions, even billions of notes. For each chord, we can then make comparisons between what other instruments are playing at the same time. This is a huge amount of data, and computers may be able to deal with such data more quickly than humans can.
I really strive to bring statistical and computational relevance to my work, and the grant has been extremely helpful in that regard.
TT: How did your interest in computational musicology grow?
MSC: I was always as much of a geek as a musician. I remember sitting in my college music theory classes and thinking about writing computer programs to do the homework for me. I was really interested in this area as an undergraduate. Around 2000, in grad school, I was in music graduate school. I wanted to take skills I had and worked as programmer for the National Bureau of Economic Research — while at the job, I would be writing code and thinking about how I could apply it in some form to music. When I came to MIT and started working with the UROP program, I could further develop my code.
TT: What kinds of material have you been involving your students with?
MSC: The combination of technical and musical knowledge at MIT is outstanding. At MIT, I could do work with undergraduates that I may not be able to do with postdocs elsewhere.
In one of my classes, 21M.269, Studies in Western Music History, we can address issues of statistical significance in music. Did one computer use phrases of music from another composer? There are only so many combinations of notes — do the similarities in notes come by chance? I try to bring in computational and quantitative approaches to my classwork. For example, I may apply limits to dotted notes as a way of understanding ideas.
TT: What kinds of research projects do your students work on in UROPs?
MSC: My students often work on computer programs. I am a huge proponent of good code — you want someone to be able to use the code to solve their own problems. My students have worked on programs that can do tasks such as solve figured bass assignments from theory classes, add an accompaniment to Baroque pieces, and translate pieces to Braille. The research opportunities at MIT have opened the door for a lot of music opportunities and discoveries.
TT: You have focused a lot on medieval music, is that right?
MSC: Yes, definitely. In fact, I have worked with some of my students to try to determine how much medieval music is still preserved today. In biology, animals are randomly tagged and put back into the wild, and samples of the species are taken in order to determine the population of the species in the wild. We are using a similar tagging technique with music samples to determine how much music from this time era has been lost. While many people seem to think most medieval music is already gone, our findings have shown that we may even have ⅓–½ of medieval music left today. This may be more intuitive for a computer scientist, but contradicts general notions of how much music is still preserved in the present time.
Medieval music has been a big area of interest for me. If I don’t examine a piece from, say, 13th-century Poland, who will? I find that getting acquainted with more obscure music allows me to be more discerning of the music that I hear.