Fifteen modifications later, he threw it into her fantasies — The Warren Ellis Gun Machine remix

So I stripped the DRM off Warren Ellis’s Gun Machine, crunched the text and threw it into the Markov Chain engine. You can see the results here: revdancatt.github.io/CAT782-remixing-ellis

As I already have the code set up pretty well for consuming new text it’s fun to put different things in and see what comes out. I promise I’ll stick the code up on github once I’ve cleaned it up a bit more. In the meantime this is all heading somewhere, kinda.

Something I found interesting is that it seems easier to find short snappy tweetable results from Sk1n (that I’ve remixed previously) than Gun Machine. Which of course isn’t a measure of quality, more that Gun Machine’s story whips along at a frenetic pace (and is wonderful) the sentences feel like good solid strongly bonded elements. Meanwhile in Channel Sk1n and other books by Jeff Noon the sentences feel faster but looser, more easily sampled, looped, and remixed.

That’s pretty hand wavy, let’s try this instead…

It’s a bit like Gun Machine is the melody and bassline, while Channel Sk1n is the samples, synth stabs and one shots. Lord knows what would happen if I threw them together, actually I should probably try that.

The idea of using the Markov Chains is to mix things up. A remix of Gun Machine will use just the words written by Ellis and not a single new one. These tiny word atoms also remain in roughly the order Ellis put them in, but sometimes new structures similar in style and tone to the original are formed. My task is to then have a root around the output looking for these small, newly formed gems.

I’m a pretty terrible writer and so turn to code to attempt to skirt around that, by using other people’s work and algorithms to somehow generate something new. This approach works when the samplers are talented (i.e. the Beastie Boys) and not so much when they aren’t (i.e. me). It’s like I’m looking for cheats and shortcuts, you never know maybe I’ll learn something in the process.

One of these cheats or “inspiration tools” I’ve played with is the N+7 algorithm. You take each noun, find it in the dictionary, look at the 7th noun after that one and replace the first noun with the new one.

Here’s an example of a straight up Markov mix from Gun Machine…

“Fifteen minutes later, he threw it into her eyes, seemed to take out his phone and answered it with lightning pingback. As I’ve been ordered to conduct this investigation, without a clue that his leanness was turning a blind eye to calculate her alertness.”

Here it is again in N+7

“Fifteen misapprehensions later, he threw it into her eye-openers, seemed to take out his photocopy and answered it with lilt pingback. As I’ve been ordered to confederacy this invite, without a coach-and-four that his leanness was turnstile a blizzard eye-opener to calculate her alertness.”

We’ve stepped out of the Gun Machine world, new words are coming in cross contaminating and building different bonds to the original. I’m not suggesting any of this is new, but it is new for me.

N+15 with a small dictionary gives us this:

“Fifteen modifications later, he threw it into her fantasies, seemed to take out his pillar and answered it with lightning pingback. As I’ve been ordered to consequence this jail, without a collapse that his leanness was turning a blind fantasy to calculate her alertness.”

Novel remixing on the Grid.

I’ve been playing a lot with MPC-type sample/beat pads, grids of soft pads that light up and play samples as you hit/tap them… I mean obviously there’s a lot more to it than that and I’m as bad at creating music as I am writing.

Music tools are all over the place, word tools though really not so much. I know there are experiments out there — but not many. In my ideal world there’d be more for words, even better my as yet unwritten novel remixing tool would exist, more on that in a moment.

The problem with taking a whole novel and shuffling the words around is the narrative gets chopped up too. The characters are still there, Nola Blue, Alice (both of them[1]) and Tallow, but they can easily jump from one scene to the next — wormholes through the pages from front to back, and back again. Time is lost and the narrative confused.

But, say I want to make a new novel by remixing from several sources. I take five books, Gun Machine, Channel Sk1n, Automated Alice, Alice in Wonderland and The Big Sleep. Instead of taking all the text and throwing it into a bucket I divide each into roughly equal chapters and just mix those up. More than that though, to create a new Chapter 2, I take all the words from the original Chapter twos [2] and then add the words from the chapters each side of them, one and three. It looks like this…

…So I’m remixing the originals, but in a more controlled fashion. The narrative of the remix will still kind of follow the narrative of the original books, just. We won’t be throwing something in from the end of one book into the start of another. New characters will get introduced at roughly the same time, and the story will shift gradually from the start to the end.

A quick flashback to “The Kibin Magic Kettle Experiment”

But I’m still not a writer, so now over to crowd sourcing. Last year I tried an experiment, it was terrible and didn’t work, for reason I’ll get into in a moment.

I wanted to see if I could use proofreading sites that promise to fix your copy and so on (although now I think I would try paying someone on Fiverr: I will meticulously proofread and edit 1500 words of English text for $5 which is an option just to see what happens) to play Chinese Whispers on a story.

As a quirk of Kibin quickly iterating over business models I snagged a year’s subscriptions worth of proofreading, I could put one article into their system at a time which had a turn around of 24–48 hours. I figured much like photocopying an image and putting that back through the copying machine I could end up with something different after shoving a story through the machine 100 times.

I took this story: The Magic Kettle (github) and mangled it up via Google Translate a couple of times — the end result still had the main narrative but the grammar and everything else was pretty shot, but not so much that it no longer made any sense though.

I wondered if the story would get restored close to its original source, or if it’d wonder off on tangents and become something else through the process of other people’s proofreading personality and bias. I set myself a couple of quick rules, well one mainly — I’d accept all suggested corrections.

The experiment didn’t go well. Not because of the proofreading — the first couple of editors cleaned the story up fairly well, the third took all the life out of it and the fourth put some back in — but rather because the Kibin site made it hard to accept the suggestions. On top of that, I thought GitHub would be a good place to put the results so people would be able to use the tools there to see the differences (diffs) between version.

The combination of Kibin usability, GitHub not really doing large blocks of text well[3] and the story obviously just going round in circles after the 4th edit[4] all made me feel that the effort I would have to put in wasn’t going to be worth it.

I’d probably do things slightly different if I was going to do it again now.

Pulling it all together.

I like the idea of “Making Things Fast", however I seem to end up making things slowly — which is why I’ve decided just to say “to hell with it” and write up projects as I go along.

I’m going to play with the idea of pulling text into a remixing machine in the same way you can with audio tools. Until I can throw, say, 5 books into the new words-machine, chop them up into chunks and specify which chunks to remix together, much like the diagram I drew above.

I’ll then get the computer to generate version after version after version which I’ll review and somehow attempt to pick the “best” bits. See if there’s any hope for the narrative or, you know, anything usable in there at all.

If the approach yields useful prose, then maybe I can string it all together and take it to the proofreading site, Mechanical Turk and/or Fiverr. Human machines to force a bit of grammar and structure, and possibly find out if people can spot patterns and make sense of the garbled output. People are supposed to be good at that, right?

Then I’ll write about it some more.

And probably get it printed and bound, because that makes it real.

Footnotes

[1] Alice from Alice in Wonderland and Alice from Automated Alice (who is in turn kinda Alice from Wonderland)

[2] we actually double up the main chapter to statistically increase the chances of picking words from that chapter, i.e. chapter [1, 2, 2, 3] but we still blend from the previous chapter and into the next.

[3] It doesn’t do line breaks very well on plain text, which means it’s pretty much unreadable. But if I put the linebreaks in and a change made a single word roll over to the next line pretty much the whole block of text would count as a “diff”. Good for code that’s supposed to live on short lines, bad for blocks of text.

[4] By the 4th edit the original story was back into place, apart from one or two changes that came in right at the 1st edit. I could see that the grammar was going to be shuffled around but it was very unlikely that someone was going to make a change to the storyline. Maybe if I had more patience and stuck it out something weird may have happened but unless I introduced a catalyst for change it seemed unlikely.

Afterthoughts

I probably need to mention something about remixing already existing in literature, beyond Pride and Prejudice and Zombies. Taking a character and dropping them into something else is different than taking a sample. A character is some IP that the author can use again and again — you can’t just pick them up and reuse them however you like in quite the same was as a 3 second drum loop.

But, people do do this, with fanfic and slash, even to the extent that 50 Shades was a remix of sorts of the Twilight books. But then in the same way that music artist sometimes re-record vocals and samples to avoid coptright, the world of 50 Shades was rewritten away from Twilight. White label vs official release.

So while it does happen, it’s different to what I’m trying to do here which is sampling and remixing in the music spirit rather than extending or rewriting a character or creation.

Note: This post was 1st published on my own blog on 7th May 2013.