Soon robots will be rappers
Robots can tweet about technology and do pretty much everything else (or will soon be able to), so why can’t they be lyrical wordsmiths? The answer, apparently, is that they now can.
In his paper, “DopeLearning: A Computational Approach to Rap Lyrics Generation,” (pdf) Aalto University researcher Eric Malmi and his colleagues describe how their machine-learning algorithm, nicknamed DeepBeat, can mine a database of over 10,000 rap songs from more than 100 rappers and then piece lines together to form its own.
After analyzing all the rappers in the database, the algorithm ranked them by rhyme density (average rhyme length per word). Inspectah Deck (from the Wu-Tang Clan) came in at No. 1, and Rakim placed 2nd; Notorious B.I.G. was 30th, 2Pac 33rd, Eminem 39th, Nas 40th, and Jay-Z 50th. The Lonely Island, the comedy rap group that started on Saturday Night Live, placed 94th.
The reason that Eminem, commonly considered one of the best rhyming lyricists in rap, was not near the top, is because he has a tendency to “bend” his words (video). He’ll adjust the pronunciation of certain words to make them rhyme with others—a technique that sounds awesome in songs, but can’t be picked up by the algorithm.
After ranking the rappers based on rhyme density, the robot then analyzed sequences from rap songs and, when given a multiple-choice series of options, had to identify the next line in the actual song. It did so correctly 82% of the time.
Finally, DeepBeat came up with a rap of its own, using lines from the database (including phrases from the likes of Big Daddy Kane, Lil Wayne, Snoop Dogg, Common, and Missy Elliott). You be the judge of whether or not its got chops:
For a chance at romance I would love to enhance
The algorithm can’t yet generate lyrics of its own, but that ability can’t be too far away. Jay-Z and Kanye should watch the throne.