The Sapir Whorf hypothesis (language defines what we can perceive and think) is mostly wrong for natural language, but true for programming. Computer languages don't differ in what they can do but in how they let us think.
-
Yep. I do wonder how much it has to do with Miller-style "chunks". That, roughly speaking, most "natural" languages require roughly the same number of chunks to represent a given concept. But specialized representations can greatly reduce that number.
A speculative, almost tongue-in-cheek just-so story in this vein is that g is pretty well correlated with working memory; learning far more efficient chunks in some domain is like effectively expanding your working memory, and thus increasing g in that domain.
