@Argyll Unfortunately, I do not have citations for you. As for what you said, it seems you just disagree with the kind of recursive structures I proposed. Let's just say that neither of us will have much concrete computational evidence to provide, in my case because I can't be bothered to spend time to work on this. But I do claim that if you ever want robust accurate translation between multiple languages, you would find that the simplest model would be one with an internal language.

It doesn't matter how it is represented in the brain. Notice that I never claimed that it is represented in linear form or something that looks like a natural language. I just claimed that the internal language has a structure that can be generated by a certain kind of automaton (which I specified for you earlier on), and that every natural language also can.

If you don't understand pushdown automata, you won't understand why I think my proposal is correct.

As for verb valency, you are right that it is usually talked about in terms of number of direct objects. To make myself clear, I think it is obvious that native speakers understand each verb in any natural language as having a specific grammatical syntax (for each sense). More formally, that is the signature of the verb treated as a function, where the bindings of inputs are determined by the surface syntax plus the context.

I am not familiar with automaton and the related theory. So if you are interesting in talking about it. I am all ears.