Hi and sorry for only getting to this now! What you’re trying to do definitely makes sense, but we don’t currently have a combined interface for that.
One solution would be to do the task in two steps and start with the answers and then stream in the examples again to select the paragraphs. But I understand that this isn’t ideal. Alternatively, you could also add some JavaScript to the ner_manual interface that mimicks the choice options (which is probably easier than doing it the other way around). When the user selects an option, it can call window.prodigy.update and add the selected option to the task. And selecting the spans would just work out-of-the-box, just like in the regular manual annotation workflow.