The traditional world model of parser games involves objects in rooms, connected to one another by directions. The textual conventions that we have arose in order to communicate that model: often a room description in an Inform 7 game looks fairly similar in its outlines to the code for that room. So the conventions are traditional, but not arbitrary.
Of course there are exceptions already: there are one-room pieces or heavily plot-driven things where the player doesn’t do a lot of room navigation but is just moved to new locations as the story demands. Those designs cut down on the need to communicate space, because spatial navigation is in fact not a thing you’re asking the player to do.
In any case, I’d be the last to discourage experimenting if you have some ideas you want to experiment with, but as a design principle I tend to hesitate when I see a suggestion that boils down to “let’s work harder to obfuscate the underlying model of this game.” In my experience, that much more often leads to player confusion and pain than to a good and coherent experience; I typically find that players are willing to look past a fair amount of scaffolding if the result is a coherent story in which they have a strong sense of agency.
So for that reason I personally would be more interested in exploring not “can we narrate this same world model in a different fashion?” but “can we do parser IF that does not make use of that type of world model? what types of story would we tell if we did?” And in fact I think we are seeing a little experimentation in that direction, and that said experimentation does perhaps reflect the influence of Twine. I’d point to Castle of the Red Prince and Toby’s Nose and Enigma here.