Previously, On The X-Files...

February 17, 2011 11:19 AM

Previously, On The X-Files...
Markov meets Mulder! Previously, On The X-Files... is a random image generator that uses markov chains to generate random conversations between major characters from The X-Files. The text is all drawn from post-processed transcripts of the show's nine-season run; the code is perl, the images are created using the ImageMagick library.

The codebase is the same as what I used a few years ago to make Garkov, but cleaned up and refined a bit to work in this context. The generator makes a basic attempt to chain context from one panel to the next, so the interlocutors will sometimes stay on topic (and as a result sometimes repeat themselves a bit).

There's a couple of weird little undiagnosed bugs in the code, so the page will occasionally throw up a Not Found error, but a refresh should make it all better.

posted by cortex (9 comments total)

This project was posted to MetaFilter by flibbertigibbet on February 17, 2011: Previously, On The X-Files...

This is super cool. I wonder though about things like this. Is the weird text in the final panel a result of the script processing not being perfect? If so, would it be helpful to implement some sort of 'This looks weird' button so we could help cull these bad snippets? Or is it just part of the fun?
posted by lholladay at 11:26 AM on February 17, 2011

Is the weird text in the final panel a result of the script processing not being perfect?

Yeah, specifically it's a failure of my attempts to automatically post-process the source transcripts with 100% correctness. The difficulty is that the scripts for the episodes were all typed up by volunteers, and while they layout of each transcript is substantially the same there's still small differences between transcribers as well as occasional variations from conventions within any given transcript.

So to get just the dialogue by character, I've got a whole bunch of little filters and exceptions cases in a perl script that churns through the transcripts looking for dialogue; and that gets it right like 99% of the time, but there's a few cases where a stage direction or other such non-dialogue text gets mistaken for a character speaking. I cleaned the resulting files up more after the fact by hand, so that short of thing should be pretty rare, but clearly there's still a few straggling bits in there. Pointing them out if you see 'em is totally great, yeah, I can go in and clean things up accordingly, but I'm not sure I'm worried enough to build an actual flagging system into it.
posted by cortex at 11:33 AM on February 17, 2011

Too funny for work.

What went on there? What his duties included?
posted by xorry at 12:41 PM on February 17, 2011

Oh X-files. So much humor, ripe for the picking.
posted by Secretariat at 1:32 PM on February 17, 2011

Hee! I love stuff like this. This was the second one I got, and it is splendid.
posted by stavrosthewonderchicken at 5:51 PM on February 17, 2011

Oh my goodness, I love this. Thank you.
posted by flibbertigibbet at 10:45 PM on February 17, 2011

I love Markov text generators. Although sometimes they go off the rails a bit, often the result is still good even when they do.
posted by FishBike at 11:14 AM on February 18, 2011

…ahem…

Oh my. I all kinds of approve of this.
posted by yaymukund at 2:16 PM on February 18, 2011

He would have had an acetylene torch in the boxcar.
posted by expialidocious at 2:21 PM on February 18, 2011