Are there research trends that try to research code generation directly from the (operational or some other kind of) semantics? One can imagine cognitive architecture that discovers and reasons about the tasks that the software should do and constructs the operational semantics of some program. And later there should be only one additional small system that generates software code from the semantics for the execution on well established platforms.
In such a way one can avoid the uncertain and rigorous path that goes through discovery and generation of formal specification that requires rigorous reasoning (sometimes undecidable, sometimes exponentially hard). Use of cognitive architecture and generation of the operational semantics (or some other semantics) allow one to do approximate, uncertain reasoning which cancels undecidability or exponentiality by discovery of additional axioms or shortcuts.
This is reference request question. Of course, I can use Google, but my experience is that computer science often names the research themes and terms differently than I have them imagined firstly and there is no reliable semantic search. So - reference request is necessary.