In fact, there's maybe an even better way of putting it:

If I talk of a group (Group S) having a 'social norm', or of sharing a judgement (N), I am saying to the participants in my conversation (Group E) that this is the case. If the shared norm or judgement of Group S is expressed in the language of group S, then the norm must also be intelligible in the language of Group E. Otherwise we would have to doubt whether we had a correct translation (Davidson and Quine).

What I'm saying here is that if we understand the language of Group S, then we understand N.

Here's a slightly contentious step: If we understand N, then we understand how to adhere to N, at least in principle. This means that, even if we believe N to be a 'bad' norm for us, we understand how Group S has come by it. For a shared conversation to work (one that included S and E), we would need to resolve this dissonance - either S would come to reject N, or we would come to accept N, or we would both adhere to a norm N' which gave a contingent account of N that we could both accept (S would see that there were circumstances in which N didn't hold, and we would see that there were circumstances in which it did).

If, despite our best efforts, we cannot achieve a shared conversation between E and S because we cannot reach an accommodation over N, then we have, again, to be sceptical about our understanding of N, and, therefore, of our judgement that we can attribute this norm to S.

Kripke's presentation of his paradox is deceptive because 'quadding' is intelligible to us, as are a great many (maybe infinitely many) other rules that a subject might be following.

But each one of them can receive the 'N' treatment, or it doesn't count as a rule. In other words, the paradox relies on a dissonance which can always be resolved.

If we could never guess, nor check, which rule a subject was following, why would we imagine they were following a rule at all? Can it make sense to say they are definitely following a rule, but not one of the ones we have been able to check? (This is beginning to sound like Gregory Chaitin territory ...).

If we try to recast Kripke, in line with this suggestion, we find ourselves saying that we can never guarantee that our interpretation of someone's behaviour as intelligible is correct. This doesn't seem so radical.

This approach also solves a problem for 'group judgements': there is nothing especially privileged about a group over an individual with respect to the private language argument. Each may make exactly the same mistake - judging that they are following a rule when they are not. This is perfectly clear from real examples of groups amending judgements about social norms under economic or political pressure.

What is not possible is to speculate that the rules of the conversation within which we are speculating may vary in a way which fundamentally changes the nature of the activity of speculation.

Kripke and Goodman

Kripke's paradox depends on attributing a single intentional stance on the basis of a specific description of behaviour.

This is probably incoherent.

We don't describe a table as wanting to remain stationary, regardless of the "behavioural" evidence.

To attribute a single intentional stance is also to attribute the capacity to have intentional stances - and, therefore, to attribute other intentional stances. It seems unlikely that it could ever be intelligible to attribute a completely incoherent set of intentional stances to someone. In fact, for Davidsonian reasons, attributing intentional stances is probably like attributing rationality, or language: there is a normative aspect to it, and it's likely to be holistic. We're attributing a form of life.

In particular we're not depending on a restricted set of evidence, nor on a purely 'computational' argument - and most of the peculiar possibilities allowed by Kripke's account would be eliminated by holistic and 'principle of charity' types of consideration.

A similar type of argument can probably deal with Goodman's paradox. A language game needs to be playable - this is a significant constraint on 'grue' type semantics.