Text & Lyric Generation with Markov Chains

Find the text generator here, and the lyric generator here.

It fascinates me how simple weighted probabilities can produce semi-comprehensible text. (Playing with text generation, and then writing fiction, I became aware of how my mind also operates like the Markov chain.)

Based off an old college assignment, this project uses basic natural language processing to generate semi-plausible text.

Britney Spears text generation, my original inspiration.

Text generation for Nick Bostrom’s Superintelligence .

Markov Chains

The algorithm divides the input text into k-grams, and for each k-gram, forms a frequency table for the following character.

When generating text, the algorithm looks at the current k-gram and selects a random character based off the weighted distribution of the frequency table.

The algorithm incorporates the character into the next k-gram and repeats the process.

This algorithm is called a Markov chain and produces long-strings of semi-believable text.

Try it out for yourself with text or lyrics.

Project Design

I coded the Markov chain in Python and hosted it on Heroku with the Flask micro web framework. I wrote a CSS/HTML/JavaScript GUI to mimic the effect of the Terminal when originally testing the algorithm. For the animation I used the Typed.js library.

Because the output was especially amusing for song lyrics, I connected a version to the RapGenius API.

For Lyric Generator, you enter the name of an artist. The algorithm scrapes the artist’s 10 most popular songs from RapGenius and produces a pseudo-random song.

The GitHub repos for the text and lyric generators are here and here.