As for the the guy talking down the "bunch of words"-approach I guess one could take words from different languages and then throw in a few extra characters and numbers in a few groups here and there just to mess up if someone only use dictionaries and then it would become somewhat harder.
What you've done is make it take marginally longer to guess your password, but not impossible. By marginally, I mean minutes to hours in most cases, not days, weeks, months or years. Just try sticking a sample password of words from different languages into Google for example, and watch it cleanly cleave those words apart into a logical search.
Lexical matching + brute force is a solved problem. Password cracking doesn't just bash letters against a wall until it gets a match anymore. At least good ones don't.