Each case that we discover can be seen as demonstrating the inadequacy of the existing rules. This is not something to be upset about. Rather, it helps us work towards a fuller account of the language and of what native speakers implicitly know about it.
Linguists are not content with vague hand-waving in the direction of ‘sprachgefühl’ every time we are faced with something we cannot explain. We seek a fuller, more principled account.
I’ve been thinking about the phrase I don’t know. Apart from its obvious use to say that one doesn’t know some fact or the answer to some question, we also sometimes use it to demur — that is, to show that we disagree slightly with what has just been said.
How do people typically demur? Politely. Politeness being what it is, we also use this phrase, in BrE at least, to show that we disagree strongly (rather than slightly) but don’t wish to enter into an immediate fierce argument.
Here’s LDOCE’s example of this usage (s.v. know 21 c).
‘I couldn’t live there.’What intonation would be appropriate for I don’t know in this sense?
‘Oh, I don’t know. It might not be so bad.’
I think it has to be a rise.
Oh, I ˌdon’t /know.
\Oh, | I ˌdon’t /know.
Ah, say followers of Brazil’s theories, that’s because we’re referring (to our lack of knowledge) rather than proclaiming it. To which I reply, but how do you know? Why should demurring be referential, whereas straightforward disagreement is proclamatory? It sounds nice as a post-hoc explanation, but if you were a NNS and needed to use this phrase — say, as an actor in a play — how would you know that a falling tone is inappropriate and that you should use a rise?