meaning the syntax is not the semantic, I can try imperative algol things in lisp clothing, or prolog things or both .. but I won't waste time on syntax, more on semantics.

I think that's valid, even desirable. Others have said lexing/parsing are a fraction of the language implementation problem. So, when using s-expression like syntax (xml as some have done) you are basically writing code in the AST. Then you can concentrate on interpretating the AST, or translating it to bytecode for some vm (maybe of your own design) or performing passes on the AST to do things like type checking, up casting values in expressions, optimizations, etc. There is really a lot more to language implementation than lexing parsing.

That's the point made by the course instructors, focus on semantics/interpretation/compilation rather. Another point I'd like to see is that with a small syntactic surface it may be easier to compare linguistic traits. Instead of having 219 languages we'd have one core and libs (with their macro sugar) giving the desired features. But maybe that would only move the problem further.

I think exist some gap in how build languages. I wanna make one, but need to re-create a lot of basic stuff just to bring my own thing. I know .net/java/llvm is suposed to solve this, but exist something that give just a AST and I plug onto it? So I don't need to invent how declare a var or how make a function, but just control the syntax, the sugar, etc.. (ie: Like have a API to build languages).

Of course, where this dream get killed is when I'm interested in do it in C, not in .NET, or need a VM (or not). But when I see julia/GO/lua/nimrond and think "Hey, I could build on that" but is too highly coupled and then the option is do everything from the start...