Edit: I've forked this to implement the classification based on bigrams, and to test the results against the current unigram-based classification.
Original README:
This is an implementation of a simple double entendre classifier in Python.
This currently uses a Naive Bayes classifier (the NLTK implementation), a simple SocketServer server implementation, so you can fire her up and build your own client to interact with her using basic sockets; and the bvandenbos data corpus. We might build a better API in the future. There is a sample PHP client included.
A live version of this can be found at http://kt4.in/twss
Ranjeet and I wrote this in 3 hours on the eve of Barcamp Mumbai 8 and presented it there. This was fairly well received.
Suggestions welcome. Do file bugs. Fork away. Send us pull requests.