If there are exceptions, I would think the problem isn’t solved. Perhaps the article is being loose with the wording. Close only counts in horseshoes and hand grenades.
But almost means one can easily list the remaining special cases.
If that is not good enough, I suspect that no possible answer would prove to be satisfactory.
If you want to construct the entire field of linguistics mathematically that would certainly be interesting to watch. Since a function that is described by a simple equation is mathematically equivalent to a function that is described by a table assigning the appropriate values to each element of the function domain, you could try to come up with a definition for what constitutes the "optimal" way to describe a function. Do the same for any kind of relation.
Until you've done so linguists will sadly be stuck applying nothing more than common sense.
Moose -> moose Goose -> geese House -> houses Mouse -> mice Etc.