Love it or hate it, improv — though it may appear random — is often more purposeful and patterned than it may seem. And, improbable as it may seem, supercomputing is at play here, too: a team of Penn State-led researchers applied the university’s Roar supercomputer to map and analyze the patterns in the improvisational music of all-time jazz greats. The goal: to understand the methods and motifs that underpin creativity.
The research began with the Weimar Jazz Database, which contains compositions and improvisations from renowned musicians like John Coltrane and Miles Davis. The research team approached this data from a syntactic, linguistic perspective, breaking down the improvisations into more manageable chunks.
“We’ve largely been thinking about improvisation as something called a hierarchical behavior,” explained Hannah Merseal, a graduate student at Penn State and member of the Cognitive Neuroscience of Creativity Lab, in an interview with Penn State’s Matt Swayne. “This is something that occurs in real time and is organized so that smaller subunits fall under larger units, which is also kind of the way that language works. We have a sentence and then smaller chunks of the sentence underneath that.”
These chunks, she explained, are core to the improvisational process.
“When we think about improvising, we’re largely thinking about players taking smaller chunks of music — such as musical notes that they have stored in memory,” she said. “They’re chunking up that information in ways that are easy to store and rearrange.”
The researchers noticed that the sequences found in the improvisations in the Weimar Jazz Database increased in complexity over the course of the phrases, exemplifying an “easy-first” principle. They parsed the solos into five-note sequences, then recruited participants to quantify how related those sequences were to one another.
Modeling the resulting data was a computationally daunting task — but luckily, the team had access to Penn State’s Roar supercomputer, which contains over 1,000 servers and more than 23,000 cores, alongside 18PB of storage. Roar’s power allowed the researchers to model and analyze the network, such as by measuring semantic distance — a common metric for creativity that relies on the conceptual distance between words and phrases.
“Our memories might organize categories of things by clustering them together and we can model this as a network,” explained Roger Beaty, an assistant professor of psychology at Penn State and principal investigator of the Cognitive Neuroscience of Creativity Lab. “Earlier, there have been investigations into modeling of networks and semantic memory. More recently, we, and some of our colleagues, have been trying to test this out using the computational approaches of network science.”
The researchers found that the jazz greats tended to begin with simple, familiar sequences, slowly building into more complex, creative choices later in the solos.
To learn more about this research, read the reporting from Penn State’s Matt Swayne.