SciTech

Musical challenge bridges machine learning, creativity

Burr Settles, a postdoctoral researcher in Carnegie Mellon’s machine learning department, created an online songwriting challenge and tools to assist with the songwriting process.  (credit: Daniel Tkacik/SciTech Editor) Burr Settles, a postdoctoral researcher in Carnegie Mellon’s machine learning department, created an online songwriting challenge and tools to assist with the songwriting process. (credit: Daniel Tkacik/SciTech Editor)

This month is February Album Writing Month (FAWM), an annual online songwriting challenge for musicians to write a collection of songs in a month’s time. Though the challenge of composing 14 songs by the end of February may seem daunting, the technology embedded in the challenge’s website, FAWM.org, is designed to give participants some help.

The creator of the challenge, Burr Settles, is a Carnegie Mellon postdoctoral researcher in the department of machine learning. He thought of the project as a graduate student at University of Wisconsin-Madison in 2004. What started as a fun extracurricular activity gradually turned into a platform for musicians to share new songs globally.

Settles explained that in the last couple of years, FAWM has spawned new research used for building statistical language models to help songwriters overcome writer’s block. The interactive tools featured on FAWM.org that try to facilitate creativity include Titular, LyriCloud, Struxxure, and Plot Spline.

Musicians can use Titular to generate song titles and start the creative process. LyriCloud displays a number of semantically related words, starting from a seed word given as input. When typing in “life,” for example, some of the words that one might receive are “delish,” “freebie,” “soul,” “taxicabs,” or “unexamined.”

These are not random words, Settles explained. They are based on automated analysis of song lyrics from the internet and of frequently associated words.

Struxxure helps with generating song structures, such as verse/chorus/bridge constructions. Plot Spline is a framing device for creating a song’s plot. An example of a generated spline, according to the website, would be: “Write a song in second person (“you”) in which the main character is thinking out loud. The story takes place in the present, and the scene is a moving vehicle.”

The FAWM website has become a “testbed for interactive machine learning technology to support online communities and creativity,” Settles said. It has facilitated the study of social influences, such as whether leaving comments or collaborating on a song motivates individual success or if those behaviors are associated with the number of songs a user writes.

For example, if FAWM newcomers receive comments in the first couple of weeks after posting, they are more likely to continue to contribute. Ideally, every song will get at least one piece of feedback.

Anyone who publishes 14 songs is automatically declared a winner. So far this month, there are roughly 50 winners.

Vincent Poprocky, a senior at the University of Pittsburgh majoring in information sciences, has participated in the FAWM challenge multiple times. Although he plays bass in a band, he uses the website for his own songwriting. The one-month deadline is his motivator for writing many of the songs that he plans to use for solo shows this year.

Completing the challenge “is a good opportunity to have something that you can be proud of and show people,” Poprocky said.

In a world where artists are constantly looking for innovation, the FAWM challenge tries to give some direction, Settles said; it helps users to come up with ideas that are plausible but novel, while drawing some boundaries.

Although technology has been utilized as an art medium, Settles said, it has not been used enough to help humans in the creative process. This project tries to use computer science in support of the humanities and the creative arts.