There’s an interesting parallel between “circular analyses of concepts” and the way people figure out whether they understand each other, in conversation.
If I repeat back ~the exact same words, there’s no guarantee I’m picking up your intended sense, because I’m simply repeating the same surrogates back (the letter that represents the spirit of the message).
If I repeat back very different sorts of words or descriptions, and you read them and say, “yes this is another valid way of expressing the spirit of what I was saying in different letters,” then you know we’re aligned on the “spirit” of meaning.
This seems roughly analogous to the multilateration/hyperbolic positioning thing (“triangulation,” loosely/generally) that Beiser recently mentioned.
Another way of conceptualizing it is: there is some “thing” being referenced, and a given description specifies some of that thing’s properties explicitly, and then other of properties are only “implicitly referenced”: they’re aspects of the thing which are not mentioned in the map but are instead part of the territory. If you are able to describe parts of the referenced territory that are not mentioned in the original referencing map, you are able to ~confirm that you understand which sector your interlocutor’s referring to.