That's true, but the article mentions an interesting twist that might not be so clear. The key statement (which, AFAIK, is a direct quote from the FSF) is: "Subclassing is creating a derivative work." In certain programming languages (Java is the example given in the article), you can't write code without subclassing one of the classes built into the language (in Java, you have to, at a minimum, subclass Object), so if subclassing is creating a derivative work, you can't write a program in such a language that isn't a derivative work of the language. Even in languages that don't force you to subclass to write a program at all, there can still be a large set of programs that can't be written without subclassing (for example, in Python 3 any time you use the class statement you are subclassing, at a minimum, the built-in object class).
I don't think the FSF's intent was to make any Java program a derivative work of Java, or any Python 3 program that uses the class statement a derivative work of Python. But if so, they need to qualify their bald statement about subclassing being a derivative work.