The problem with this angle is that while sure, you can always make a compiler more intelligent in ambiguous situations, you must also now train all the humans interfacing with it to be equally as intelligent (or at least, able to sufficiently disambiguate).

We don't see many context-sensitive grammars for programming languages, not because very difficult to implement, but I suspect because they're hard for humans to understand.

There is something to be said for the simplicity of naming things to describe what they are (e.g. Hungarian notation), but I really think it's a mistake for languages (configuration, programming, etc.) to even allow identifiers to be reused for differing types.