Introducing the Recurrently Generated Football League

Several months ago I happened across an article in The Atlantic about machine learning and color. In that article, programming enthusiast Janelle Shane wondered if, given enough data, a computer could generate descriptive names for paint swatches, like what you find in hardware stores. So she fed a database of hex colors and paint names into a program called a “recurrent neural network” and ran it to see what would happen.

The results were unpredictable, and hilarious. It turns out that machines aren’t really great at matching descriptive names to actual paint colors, but it did generate some really outstanding paint names, like “Stanky Bean,” “Burble Simp,” and “Turdly.”

That article led me to Shane’s blog, aiweirdness.com, where she outlined her process and posted about similar experiments with things like My Little Pony characters, Dungeons & Dragons spell names, and even full recipes.

And then it struck me. English football clubs are often named after English cities and towns, but not always. They have names like Accrington Stanley, Quorn, and Tytherington Rocks. They are charmingly weird, and unique. So what could a recurrent neural network do with English football club names?

I’m no programmer, but I dug into the source code ( a program titled Torch-rnn which uses machine learning to generate character-by-character text from source material), compiled it on my Mac, and fed it a database of English club names going all the way down to Level 12 in the football pyramid. And then, just for fun, I added the Scottish leagues too, because what the hell, right?

I shared the results with my fellow Cartilage Free writers in Slack, and everyone loved it. We spent an entertaining evening selecting our favorite computer-generated UK football club names. We started asking ourselves: what do these clubs’ crests look like? What kind of football do they play? What is their history?

Like all good, crazy ideas that we have come up with over the years, someone said “We need to do something on the site with this!” We started organizing the best names into divisions, and then began developing backstories for them, based upon nothing except what inspired us from their completely randomly generated names. In the process, we laughed. A lot.

And that’s how the Recurrently Generated Football League was born.

It’s the international break and there’s little actual interesting football to watch, so over the next few days and weeks, we’re going to have some fun with this. Inspired by the machine learning club names, nine of us have adopted a Recurrently Generated Premier League (RGPL) club. We’ve fleshed out the clubs’ locations and histories. We somehow convinced a SB Nation graphic designer to create club crests, colors, and even kits.

We’ll be debuting these clubs in the coming days, starting tomorrow.

We’ll also be running our completely fictional computer generated clubs with completely fictional computer generated players through a custom database in Football Manager 17, to find out which club is the best recurrently generated club in the UK. For science.

Will anyone actually care about this? Who knows! Like many of the things posted on this blog over the years, this emerged as something that we did mostly just to amuse ourselves. But we’re again bringing you along for the ride, because we think this needs to happen.

Also, we’re bored.