Posts Tagged: android poetry bots

On autocorrect and algorithmic poetry

At this year’s Books in Browsers ‘opening act’ event, Creating Minds, held at UCBerkeley on Oct 23, we heard a number of speakers talking about the coming of the machine voices into our lives. That day, and at the BiB conference proper that followed, there were numerous references to machine cognition, algorithmic poetry, spambots, twitterbots, and the myriad non-humans that co-habit our social and literary spaces these days.

It does not surprise me that my phone—an Android, natch—wants to get in on the act. I’ve installed a swiping (as opposed to pecking) keyboard called SwiftKey that works pretty well; it’s much more efficient to slide one’s finger across the touchscreen than to tap away at little targets. And, of course in 2013, it learns as we go; it picks up my frequently used words—and phrases—and uses those patterns to do a supposedly better job of interpreting my smudgy finger movements as I try to achieve 40wpm. It also does the usual trick of predicting the next word, displaying the three best guesses just above the Qwerty, so I can pick the right one, if I’m so inclined.

And that allows it to generate its own prose poems, sorta. All I do is keep hitting the next guess, over and over again. On the newest beta version I installed (only a few days in to learning my stuff), it goes like this:

I am a beautiful person who is the best of luck to you by the way to get the best of luck to you by the way to get the best of…

Cheery little beast, isn’t it? The ending goes like one of those elegant irrational numbers with the repeating decimal-place patterns. Anyway, that’s pretty much the out-of-the-box functionality. Now, if I go back to the older version, where I’ve already given it six months of my typing history, things get a little more interesting:

I am a beautiful person who is the Internet and Us to the durability virtue of publishing as a manufacturing issue to publication as the author of the book of the book of the book of the book…

If you know me, you’ll recognize this fragmentary discourse as being pretty close to the blather that escapes from my mouth most days. I do like that endlessly repeating “of the book of the book of the book” mantra. It changes over time, too: when Haig Armen and I were working on our paper on index cards last spring, I would often get a big, allcaps “HINGE” in the first few words… but you’ll have to read further to get the point of that.

But they frustrate me, these autopredicting algorithms. Make no mistake: I for one welcome our spambot overlords, and I think James Bridle’s piece on how the robots are reaching out for love is one of the most poignant pieces of contemporary cultural criticism I’ve seen. But I do wish the makers of these things would take a longer view. Let me explain.

If I’m trying to type “It’s pretty interesting“—as I was just now in an email to my wife, and the thing autocorrects/autopredicts to “It’s pretty interview” or “It’s pretty girl,” (those are actual suggestions from my keyboard) well, that’s not really all that helpful. I wish it could make some broader predictions, maybe drawn from some large corpus of fine literature. Maybe then it would predict, “It’s pretty intriguing” or, “It’s pretty intense”… or how about “It’s pretty irresponsible,” perhaps I would be delighted (that word that keeps coming up in UX discussions these days) by it and it would begin to influence the way I write.

So could I please request that some DH project be set up that would glue, say, a corpus of romantic or modernist literature, with lots of good word proximity metrics worked out, into my autocorrect? Perhaps we could all choose, in the Settings dialog, which corpus we’d like to be corrected by? I’m not sure having the machine learn my typing patterns is the best way to improve my writing. But if I could have, say, Keats, or Virginia Woolf’s patterns correcting mine… then we’d have something.