I recently posted <a href="http://www.livejournal.com/users/jerith/3708.html">this</a> in my LiveJournal and it generated some discussion on #Schlock_Mercenary. If you don't feel like reading the link, it describes a basic random name generator I wrote to play with the idea of context-free grammars.
Some of the relevent comments (edited for readability) are below:
Personally, what I think a name generator needs most is a rating system. Basically determine how "hard" a name is, or whether the "feel" is a certain way (like masculine or feminine, or elvish or dwarvish), and then let you filter and maybe sort based upon it. So instead of trying to generate good names to begin with you instead take a bunch of bad names and figure out which ones of the bunch aren't so bad
Perhaps have a rating system that adjusts the generating grammar instead of a filter. But how would you do the filter?
Well I'd do stuff like counting the number of hard sounds in a name
We could always use a CS or two-level grammar to handle that kind of rule.
Any other comments? Suggestions? Criticisms? Shiny things?