Hi PetaMem!

Thanks for that additional information! Chatbots are a really interesting topic!

Looking at your specific problem (and not the general generalicity of regexes) makes me wonder if the empirical approach (do a check on how many possible strings are matched by the regex) really doesn't work. However with some modifications:

• Your problem doesn't seem to be real-time - and the order of your rules seem to be static:
  therefore you could do the generalicity-rating in some offline preprocessing phase, e.g. by letting your computer work on it all night/weekend/holliday
• Instead of looking at all possible input strings, just narrow it to the expected data. You will probably have tons of chat-logs which you can use
  So your rating could be: how many matches are found by a regex in a given set of logs

Regarding the high number of possible characters (as also mentioned by JavaFan): Here you could do some preprocessing, e.g. by replacing the german A-Umlaut by ae. This will also help with some other "languages" like leetspeak ... your chatbot seems to get confused when greeted by a friendly "h3110" ;-)

However I fear that any automatic ordering would just be one criteria for determining the order of the rules. You will probably add additionally some rating done by a human. And the idea of giving additional context to the rule (like you wrote: try this before ruleX ...) sounds great to me. Have you also experimented with randomness? (Using a random order in case of several rules have a similar rating.) That way the answers might not be so predictable...

HTH, Rata