Interesting. So what I was getting at was exactly what Tarski said. Except for that he coined some cool words for his explanation, which I will steal for future use.
Except Tarski didn't really solve it, he just kinda swept it under the rug. The object language/metalanguage distinction introduces an infinite regression (since, the moment you want to talk about the metalanguage, you need a metametalanguage for those statements, which means you'll need a metametametalanguage, and so on).
