I'm well aware that English isn't a formal language, that's why I added the qualifier "mostly". The great majority of expressions in natural languages can, in

I'm well aware that English isn't a formal language, that's why I added the qualifier "mostly". The great majority of expressions in natural languages can, in fact, be accounted for with CFGs, and purely CFG-based Phrase Structure Grammars have been proposed (see the work of Gazdar and Pullum on Generalized Phrase Structure Grammar, from the early 80's, if you're interested). Many of Chomsky's original claims about the weak generative capacity of CFGs with respect to natural lanaguage that gave rise to transformational syntactic frameworks have since been disproven. Whether or not there is an absolutely snug fit between CFGs formally and natural language "in the wild", so to speak, is another topic, and rather beside the point of the analogy. Context Sensitive Grammars are overly expressive, Regular Grammars much too weak, for much the same reason why they are too weak for HTML. Were there a perfect English language parser, you would not need it in order to match regular subsets of English, just as you do not need a full HTML parser in order to match regular subsets of HTML.