The web speaks

[darius]

Here's the 'best' of the first half-dozen 'poems' from my new program: it babbles at random according to a Markov model using the most common bigrams from Google's whole-web corpus, constrained to a Shakespearean sonnet.

Comments

[dariusk]

Oh, that is amazing. Is the source up anywhere yet? (I've been working on a different sort of sonnet-generator myself.)

[dariusk]

Err, I just realized you're probably accessing that 6-DVD set of data, though?

[darius]

Not directly, though I might go and buy it. The data's from someone at Google who wrote a great article about some of the cool things you can do with that dataset -- but the article hasn't been published yet so I don't know if I can say more or give out this condensation of the data.

[darius]

Thanks! I guess I'll put the code up on github as it is now (an ugly mess). Different how?

[dariusk]

I helped a poet I know write something that takes existing sonnets and does some simple regex to kind of rearrange them in interesting ways.

[darius]

Ah, cool. If it weren't simple I wouldn't have the patience.

[darius]

Code now at http://github.com/darius/languagetoys/tree/master