I grabbed all of Drake's lyrics from lyrics.com and put them into a text file. Only the verses and chorusses he performs, lyrics by guests aren't included.
It uses RiTa.js to create Markov Chains that generate the lines. Basically, the program looks at each word the text file and then the word after it to determine the probability the word A is followed by word B. "You" is sometimes followed by "used," which is sometimes followed by "to" ... "call," "me" ... you get the idea. But sometimes "you" is followed by "know" (as in "You know they all sentimental now").
RiTa.js allows you to determine the probability - or likeness - of the sentences it generates. At one end, it will create lines verbatim to the original text file - so the output would be "You used to call me on my cell phone." At the other end, it creates random gibberish. I set the bot to be in the middle - I wanted it to have a bit of nonsensical playfulness, but it also needs to be recognizable as something Drake might say.
I made it as part of NaNoGenMo as a way to become more familiar with text generation, elit, and that whole world. It was fun, I definitely want to do more. I need to learn more about poetic structures and the building blocks of creative literature, as well as Python, NLP, and the like.