The duplication there in each AST node kinda hurts the eyes :)

Also, the type-switch on interface {} is ugly.

Consider how an AST looks like in Haskell:

  data AST
    = LiteralInt Int
    | LiteralFloat Float
    | List [AST]
    ...

And then an eval looks like:

  case ast of
    LiteralInt int -> ...
    LiteralFloat float -> ...
    List nodes -> ...

and it is safe, rather than interface{}, you get the AST type and you get exhaustiveness checking that you covered all cases.

Also, if you add an annotation to each AST element (e.g: inferred types), you can very easily and safely map over them, etc.