“Batman isn’t paid to destroy crime corners. He is actually attracted to the crimes and also The Penguin.”
This is a line from the internet story “Batman Loves Him a Criminal,” which may at first glance seem like a poorly written piece of fanfiction, but in actuality was created by a predictive text generator. It was posted on Tumblr last week and it’s the stuff the internet is made of.
The emulator was created by Jamie Brew, the head writer for Clickhole, who told Gizmodo that he wanted to play with the idea of smartphone text prediction. The emulator offers words up in a way that looks similar to what text prediction looks like on a smartphone. You have a word, and the program will offer suggestions on what to choose next.
You can do this on your phone. Type a word, any word. Then just keep inputting suggestions from the autocorrect and see what you come up with. “Last summer a friend showed me that you can just keep taking the phone’s suggestions and write things like ‘I have a great time in my head and neck and shoulders and the rest.’ I couldn’t get enough of that,” Brew said.
But that wasn’t enough. It could be funnier. “I wrote a lot of stuff and found that some of the best stuff was when it hit a particular niche, like words related to cooking,” he said. “So the thinking was something like ‘I wish it could all be that specific.’”
So Brew started inputting source text from various websites into his program, his first and only project. He would take text from things such as Yelp to make fake Yelp reviews, so his generator would have context to work with in the same way that your smartphone’s autocorrect grabs examples from your emails and messages to see how you talk.
The internet makes this easy. Transcripts are easy to come by, especially when Brew wants to create a play starring people from Anthony Bourdain: Parts Unknown, for example.
It uses a Markov chain-like algorithm, which is a random mathematical system that offers a next step only based on the current state a system is in, and not the one that came before it. Basically, the predictive text generator will only offer suggestions based on the last two words entered. Unlike most Markov chains, it’s not automatic, so it’s not exactly a proper chain. The user has to then choose a word, and the chain continues.
“[It’s] like a choose your own adventure book that’s running on autopilot,” Brew said. “With my program, it still has the same idea that you’re looking at what just happened but rather than making the decision randomly, you choose the next one.”
Brew has used it to mock a number of different websites, from Batman wikis to IMDB parental advisories, and to create fake Carl Sagan speeches. It really draws a person’s attention to the language used online, and can highlight the utter weirdness of internet slang. Do you know anywhere else where being called “trash” is considered a positive? It is on Tumblr.
While all of the text in his posts are generated on the fly, some do read eerily close to what you might find online. It’s sometimes like bad internet poetry. We’ve seen worse writing than this on Craigslist.