Irish Lit AutoTweets
This page provides background information about the IrishLitTweets Twitter account and the Automated Irish Literature Discourses Tumblr account, both of which contain automatically generated text based on course-related writing that I did when I was a TA for a course in Anglo-Irish literature during the fall quarter of 2013 at UC Santa Barbara. The automatically generated text is sometimes gibberish and sometimes funny. Occasionally, it's even rather surprising or insightful. The project itself was thrown together quickly because I needed a break from more substantial work.
The tweets at @IrishLitTweets are generated by my Markov chain-based text generator. Like Jamie Zawinski's DadaDodo, which was previously used to generate these tweets, this Markov chain-based generator thinks of itself in terms of William S. Burroughs's cut-ups
, hypertext, musical sampling, programmatic computational dissociation, and other things. The text generator generates these tweets on its analysis of all electronically captured writing that I did over the course of fall quarter 2013. This analysis transformed the aggregate text that I'd captured into a model based on Markov chains, which indicate the frequency with which given chunks of text follow each other in the reference corpus. The corpus analyzed, after some massaging (removing student names and grades, for instance, and making an automated minimalist effort to remove duplicated text that was cut and pasted), consisted of 117,018 words. Before massaging, the corpus consisted of text from the following sources:
Despite this, there are of course many words that went into teaching this course that were not captured or included in the corpus: handwritten margin comments on student papers, for instance, and everything that I said verbally.
The script on my laptop that produces the tweets simply uses the text generator to produce a tweet, makes sure the generated text meets specific criteria (being an acceptable length, for instance), and sends the first tweet that it generates that meets those criteria over to Twitter. More information about this script is available in the technical write-up; you can also take a look at the script itself (it's written in Python 3.X) on GitHub.
In summary, then, the Twitter stream consists of more or less randomly generated text that's based on a statistical analysis of text that I typed. It bears some difficult-to-determine relationship to the hours and hours and hours and hours of writing that I actually did in the course of teaching during fall quarter of 2013. About all that can really be said about the tweets is that, for each pair of consecutive words in a tweet, that pair of words actually occurred, sequentially, in that order, next to each other, at some point in the writing that I performed that was incorporated into the corpus.
The script that writes the tweets at @IrishLitTweets actually generates a lot of text that doesn't get used, mostly because it often coughs up chunks of text that are too long to fit in a tweet. This text is collected and automatically posted at irregular intervals to the Automated Irish Literature Discourses. There's more information about that in the technical write-up, too.
The content of the Twitter stream is actually generated by a script running on a spare laptop, so this content is only updated when that computer is on and connected to the Internet. It's on most of the time but my crappy monopolistic ISP is not great at keeping me on line, and if it's not on line, no updates happen. If you're not seeing updates, this is probably why, although if you don't see updates for a very long time, I'd appreciate hearing about it. (You can tweet at my personal Twitter account to get my attention; I don't really monitor tweets at the @IrishLitTweets Twitter account closely, though I'd probably notice eventually if you sent something.)
You can read more about the technical aspects of the text generation, if you'd like. You're welcome to download and adapt the script that's used there, subject to certain conditions. If you find the script helpful or the Twitter account amusing, you're welcome to Flattr me, if you'd like, and I'd appreciate that. (I'm a grad student, and money is always appreciated.) If you do anything useful with it or have suggestions for improvement, why, I'd love to hear about that, too. (Again, you can tweet at my personal Twitter account, or reach me via the GitHub project page; or there are a number of other ways to reach me.)
You can find me on various other places on the Internet.