I have used three corpora as source text for the messages, The King James Bible, The Qur’an and the works of L Ron Hubbard. The language of the first two are broadly similar, and produce plausible text when combined. The latter, by virtue of being primarily a self-help guide, rather than a set of explicit fables and scriptures, has a very different linguistic form, and can be easily spotted in the resulting tweets.

As with @the__truth, the results are hit and miss, however there are gems of linguistic beauty every now and then.

see what it’s up to here:

http://www.twitter.com/the__truth

We live in a period of obsession with ‘celebrity’. Whether this is a function of the expansion of media sources requiring content, or some other shift in human behaviour is unclear. However, the fact remains that sites like Perez Hilton can command audiences in the millions, by providing nothing more than idle gossip. In the fast moving world of the ‘blog-o-sphere’ a tremendous quantity of information is generated, yet frequently the same ’story’ will be found at multiple sources, the only differentiating factor is the style and quality of writing. A cursory glance at these gossip sites reveals that the tone and structure of the language used is quite different to more traditional approaches to reportage. This experiment attempts to capture the tone of these texts automatically and simultaneously generate content consistent with ‘the latest celebrity gossip’.

‘The truth’ is a software application which periodically scrapes content from perezhilton, gawker and holymoly and performs a basic word-pair analysis on the resulting text. This analysis is then used to auto-generate ‘valid’ sentences from the source using a simple Markov chain procedure. In the spirit of web 2.0, these sentences are broadcast via Twitter. If you yourself are a twitterer, you can follow ‘the truth’ here: http://www.twitter.com/the__truth.

Why is it called ‘the truth’? It is, of course, a reference to our old friend Wittgenstein and his contention that there is no absolute truth beyond a functional agreement in word meaning between users of a language.

The ‘manufactured authenticity’ produced by this software is perhaps a glimpse into our near future where much of the text we read may be generated by machine. Of course in this case the text is ‘nonsense’ in the truest sense, since there is no semantic comprehension involved – yet somehow that doesn’t seem to matter when the words produced are ‘celebrity gossip’…

Below you should see the last ten generated ‘tweets’:

acrylic on canvas, 812mm x 304mm

Here’s a timelapse recording of the painting process – at the end you can see the anamorphic effect as the canvas is turned.

Tomorrow (17/10/07), sometime after 5pm (BST), I’ll be being interviewed on Reverb Radio here in Brighton. You can listen live on 97.2fm if you’re in the area, or listen to a live stream online if you’re not.

I’ll try and get a copy as an mp3 and upload it here once the interview is over.

It’s worth noting that the last time I was on the radio, it was in New York, and I managed to accidentally utter a ‘rude word’ which resulted in the program being pulled from the air. This time I shall make an effort to be more polite…