It'd be cool if it didn't suggest names that very obviously don't sound good in (at least common) non-english languages, eg., I got "culojoo" as a suggestion for 3 syllables, which won't sound good in Spanish at all.
Names are totally random, not checked against any dictionary, that's what I thought was fun. It would be hard to appreciate the goodness of a name, at least programmatically, how would you do that ?
tenken is right, I meant to say "not let names have vulgar connotations in other languages." In the example above, the name started with "culo", which translates into "ass" in Spanish. Here's an example of a service that does that: http://wordsafety.com/
Levenstein is an absolute metric, I think something like soreson-dice would be more useful.
Regardless, if you take the short keywords and blacklist them by approximation with curse words from several languages I think it would be really hard to get something at all.