When I started generating these poems, I expected what in hindsight is a fairly high level of…well…comprehensibility and not-suckiness. I expected them to be something like Emery Allen’s poems – pretty, prose-looking writing that appeals to lost young adults. (Well, actually I’m not sure why people are liking my intentionally crappy generated poems, but it’s nice to know that someone appreciates my crimes against the literary arts.)
They’re gibberish. And they’re short. They’re short because if I keep them going any longer, they turn into an unrecognisable parody of the English language. The reason my poems are such crap is because even though I’m aiming for cut-up prose, the tools I’m using aren’t generating coherent prose. I think this is by design.
I’m going to have to change my methods. More on this later.
I’m honestly quite surprised. Even though I’ve had to reject a lot of my prospective poems for turning into nonsense within the first two lines, what’s left has turned up some really sad, touching, occasionally actually not terrible things I wasn’t expecting. That said, it makes sense; I do a lot of communication through my phone, so I’ve definitely bared my psyche to its algorithms. I actually giggled a little when it started telling me I should put my pet name for my boyfriend in my poems!
What’s more, I forgot how fun writing could be. These days I write technical reports (not fun), popular science articles (very fun) and personal blog posts (also very fun), but I gave up writing poetry at about thirteen, when I realised I actually sucked really, really badly at it. Not having to worry about being good is liberating.
That said, it’s trite and poorly-generated. It’s enjoyable, yes, but I need to do something cathartic. Generating bad poetry is pulling work out of an algorithm. I need to pull something out of myself.
A little more coherence, please?
Generative poetry does not begin or end with predictive text algorithms. There’s some really fascinating stuff out there, spanning the funniest Twitter bots all the way up to AI projects.
I want to look into things like Markov text generators. They produce superficially real-looking text and there’s a lot of poetry out there for me to play with. Writing some code to do it is an option – I found a tutorial to make your own Twitter bot using Markov chains. If I can get the libraries to play nicely, I should have a fun little poetbot ready to generate doggerel!
That’s the next step I want to take. This goes all the way up to research on neural networks – which is fascinating, but also usually funded by staggeringly large grants and done by multiple people, i.e. not the logical next step for an undergraduate working alone and doing this in her spare time.
Still, this is actually producing some surprising new insights. Just as I hoped it would!