While I haven't tested those built-in commands with text outside the current scope, I'm still under the impression of what my German Dragon used to do when trying to use select-and-say in a not supported window:

Saying "Markier <xyz>" always yielded "Marquis xyz".

I interprete this as Dragon creating a selection grammar with the current text content (empty or just the previous utterance in case of an unsupported control) and if the utterance does not match this selection grammar, you get the result from the dictation grammar.

Regards

mav