The idea of putting a 'bot on IRC that will interact with other chatters is not, of course, novel. Many other IRC bots and chatter programs have been made over the years, with varying levels of success. This is what I've managed to find out about them:
Not the granddaddy of all chatbots, but probably one of the few publically available bots to feature in the Loebner Contest. The MegaHAL site has a few cute things, including an IRC chatbot and source-code -- however, the source is some of the ugliest C I've seen. Nevertheless, the How It Works page says MegaHAL is a 4th-order bidirectional Markov chain. I could do that.
I've never found this on the web, but it inspired gNiall and Nicole.
This is the website for gNiall, a chatbot thingy for GNOME. The How It Works page describes a first-order Markov chain engine, although it never mentions the word "Markov". It *does* appear to be more space-efficient than MB-01, because it stores each word once, and other entries are just links to "word numbers".
This is the website for Nicole. No implementation details are given on the site, but the first version ran on an IBM XT (!) and took 20 minutes to generate a sentence (!!).
Bizzarely, it requires MySQL.
This is the website for Ummon. Source-code available, but no description of algorithm. From the fact that they stress "must learn language given no prior knowledge", I assume it's Markov chaining again. Has a good Links page, thought.
No docs, no source, no details. Looking at the output, I'd say it's a first order markov chainer. For some reason, it seems to be a fair bit less random than Moby - maybe because it has 10% of the words? :) Anyway, I'd like to steal the statistics module. More information here.
- Nemo sez: another reason why it might be less random - it's only listening to one person speak, not many.