People can decide what sorts of additional assumptions they want to add to flesh out the problem. For instance:

Similar Sensorium: Observer at the same positions in the same VR would record same or similar vector of base-level perception/sensation. (is base-level completely pre-linguistic or…?)

Similar Elementary Training – Some set of comparable elementary VR & language training are shared. Drawn from a common distribution?? Perhaps identical to start with & then add complexity later.