f[x, y, ...] is an abstract representation (AR from now on) * of an expression with head f and arguments x, y, .... (expressions, levels)
You can call 'head' a 'function' or call 'arguments' 'elements', does not really matter. (meaning of expressions) You only need to remember that both the head and each of arguments can be an expression too e.g. f[x][g[y], h[z1][z2][z3]]. For simplicity we will assume that there are no definitions associated with f, x,y,.. so it will stay inert here.
It is not parsed (in WL meaning) to anything, it is what the WL input is parsed to. The so called full form of the expression. Of course it is represented somehow in a memory but that is too deep, out of scope here (I don't know either :)). (e.g. memory management for large arrays)
It undergoes evaluation which is outlined e.g. in tutorial/Evaluation
(*) I said an AR because there are several 'stages' we can confuse.
'what it really is': ????: internal/low level representation in (?)C (as I said, out of scope here)
'how to think about it': f[x]: AR / full form of an expression (see note about context at the bottom)
'what we see'/'what we input': f[x]: piece of screen rendered by FrontEnd (terminal/Mathematica FE/CloudFE) which represents input/output of the Kernel, in a specific form or styles.
That rendered piece is really a cell expression: "f[x]" or RowBox[{"f", "[","x", "]"}] (or an html in case of cloud FE)
So f[x] as an AR of an expression is not parsed but an expression input e.g. RowBox[{"f", "[","x", "]"}] (rendered as f[x]) is parsed to our AR and then evaluated.
I need to mention that I intentionally ignored contexts to not confuse a reader.
But AR is not just f[x], every symbol has context which we should keep in mind.
That is an input is parsed to something like System`f[Global`x]. See FullForm with context for each symbol?.
For completeness, f @ x / x // f do not live 'below' input/output stage. Boxes/Strings of what you see as f @ x are parsed to System`f[Global`x].
TreeForm, but I'll warn you now that that will be unsatisfying to you if you're not comfortable with the code-as-data paradigm. – b3m2a1 Jan 25 '18 at 07:46