There's obviously more than just "knowing the rules", but there is still only a fixed number of rules unlike natural language. Solving chess is algorithmic, although still complex. Natural language can imply so many different meanings based on a combination of words, wordplay, inflection, etc. For example, "bat" could mean baseball bat, cricket bat, bat (mammal), batman... and the list could continue. With millions of words and combinations the possibilities just explode.
Not to mention Watson has much more practical applications than Deep Blue. A computer that can play chess? Great, now what do you do with it? A computer that can parse and understand natural language can have many more practical uses.