Explicit rules, tacit rules, and metalanguages

We can only attribute intentional states incorrigibly to an honest and competent interlocutor.  All other attributions of intention are corrigible (Kripke).

If I am theorising about a language, I am hypothesising about rule following - I am attributing intentional states to the users of the language.  If I do this in a 'meta-language' which I do not share with the users of the object language, then my attributions are always corrigible - I can never be sure they aren't quadding.

The only sense we can make of the idea of a tacit rule, is that it is a rule which can be stated in a meta-language.  Tacit rules, therefore, can only be identified provisionally (they are like hypotheses from behaviour).  If I make a tacit rule of my language explicit, however, this ambiguity disappears.  Between interlocutors, the possibility that I may not be stating the rule I appear to be stating is incoherent.

If we think of the rules of logic as being like the tacit rules of a possible language, we can see that the statement of these rules in a meta-language can only be provisional.

But if we state a tacit rule in the object language we do something to the language:  we render the provisional definite.  This must change the tacit rules of the language - it must change the way it can be described in the meta-language.

An accumulation of changes like this will produce a game very unlike the game we are presently playing.  If it can be 'translated' into the game we are presently playing, then the rule changes must, in a a sense, be trivial.  if it can't, we have no grounds for calling it a language.

But we can only show this by explicating the rules, and by changing the game ...