If you try taking some well-written code and converting it to natural language with utterly no ambiguity, you'll quickly find that:
- the natural language version is far, far longer and harder to read - it takes a lot of work to remove all ambiguities, and you will find yourself wanting to put a dictionary with strict rules and extensions -- totally unlike a REAL dictionary.
=> Word "allow" will always mean generally the 1st (verb) definition in Mirriam-Webster online, and will further imply that if the subsequent test (see below for definition) succeeds (definition below), then execution (see below) will continue; if the test fails (see below) then standard error (see below) handling procedure (see below) will be followed, in which...
It's ridiculous, isn't it?
You'd be inventing a new (but probably horrible) coding language, not using natural language at all any more.
The thing about natural language code -- it does exist, sort of. Think about the checklists that airline pilots run through before takeoff, for a good example.
That's sort of natural language process execution... but it's the same every time, the context is known (it's being executed by an airline pilot sitting in the cockpit of a specific airplane), the pilot is trained such that s/he has practiced running the checklist many, many times before doing it "for real". And: it's a friggin' checklist. Is logic in there whatsoever, beyond "if this is true, continue; if not, abort"?
The limitations on it are incredible, and we still have human beings executing that "code"; they are not automated. The answer to one of the questions might unexpectedly be "sort of", or "yes, value in the safe range, but why is it steadily dropping?"
(the second sentence is a joke.)
I think this has been on HN already: http://www.hpl.hp.com/personal/Hans_Boehm/c++mm/why_undef.ht...
Essentially it has to be that way, or on some architectures every access to a potentially shared variable would have to be wrapped in fences by the compiler.