With that perspective, we can ask what representations of a formal system do we find easy to manipulate in ways that produce results that are significant for us (in a specific problem context).
A visual representation would certainly be a better match for a spatial problem domain (Sutherland's work, for example). Similarly, an audio representation can work well for an audio domain (ex: beardyman's realtime performance system he presents in his ted talk). Likewise, I'm quite sure a system for exploring cuisine art would match a "smell representation".
The notion of a "universal turing machine" guarantees that all of these representations can be mapped to that of a symbol stream. That is a mathematical result, but which representations appeal to humans involved in various kinds of work is not a mathematical question - it is a HCI question.