Website helps solve songwriter’s block

Associated Press

PITTSBURGH

The next time a stumped songwriter has a hot riff and no lyrics, a Carnegie Mellon University researcher and moonlighting musician hopes that the confused composer will turn to the Internet for inspiration.

Burr Settles’ online tool, The Muse, randomly suggests song plot lines (“Write a song in the second person in which the main character is addressing the listener ...”) and structures (verse, chorus, verse, bridge, chorus). But it can even suggest lyrics and song titles — features that grew directly out of his day job.

Settles, 32, is a postdoctoral fellow in CMU’s Machine Learning Department, which develops programs and algorithms that enable computers to learn or think. His research involves developing computers that can “read” by programming them to recognize how words relate to one another.

Settles, a pop guitarist and songwriter himself, fed the lyrics of nearly 140,000 songs into a computer and applied his research principles to them to create his musical website.

The pet project of CMU’s Machine Learning Department is NELL, or Never-Ending Language Learning, a computer system that is teaching itself how to read by trolling through Internet sites and tracking the way words are used in various contexts.

Settles turned that research on its head to create The Muse’s two key features: LyriCloud and Titular, which spit out suggestions based on patterns that his computer discerned from the lyrics and titles from songs by artists ranging from Beyonc to Van Halen.

Like all songwriters — Settles has written 150 to 200 tunes — he gets writer’s block, which is why he created LyriCloud and Titular.

Typing a word into LyriCloud produces a “cloud” of related words. “Love” produced 26 options ranging from “made,” “forever,” and “ooh” to less obvious connections. Titular is a random song-title generator.

The website isn’t meant to write the songs — it’s meant to inspire the writer.