On September 25, CBS broadcast an interview with Republican Vice-Presidential candidate Sarah Palin. The next day, interviewer Katie Couric discussed an additional, particularly newsworthy segment of the interview, stating that Palin "is not always responsive." The interview garnered intense negative attention from commentators, editorialists, comedians, and even conservative pundits.

The incoherency reminded of us of talking to rudimetary artificial agents like ELIZA. A web app just had to be made.

This is a parody. We don't think Sarah Palin is "stupid." Some older interviews on local Alaska issues demonstrate a more coherent Palin. But these new interviews are simply ludicrous.

PalinSpeak strings together sentences and phrases from several of Sarah Palin's own interviews and speeches. (1 2 3 4 5)  It tries to come up with an answer to your question; but mostly, it randomly transitions between phrases and topics. The basic algorithm is to take thousands of three-word chunks from the texts and connect chunks that share words. (That is, traverse the n-gram Markov model.) This ensures that all small phrases were actually said, but the generator makes wild tangents and has little overall coherency.

Is a crude algorithm distinguishable from Palin's Couric interview? You be the judge!

Please leave us feedback — and send your funniest answers — to palinspeak@gmail.com.

PalinSpeak was created by Doug Wilson and Brendan O'Connor.

Here's the code, if you're interested.

1/09 update: More thoughts from Doug.

p.s.: Shoutout to interviewpalin.com — great minds thinking alike! They also use a Markov model, which seems well suited to Sarah Palin statistical language modeling.