Starting with what we can say, rather than with what we can think, or experience, may seem perverse. It's certainly at odds with the whole tradition of rational empiricism from Hume to the present.

This tradition has, however, broadly failed to produce solutions to some fundamental problems - particularly those to do with the reliability of scientific theorising, and the relationship between language and 'the world'.

Also, and contrary to common conception, we cannot just 'say what we like'. At some point, playing fast and loose with the rules of the language game undermines it's playability. And it is not only the rules of logic which constrain us - we are also constrained by the capacities of our interlocutors and, in some way, by 'the way the world is'. Part of my position is that we cannot give a full account of these constraints, because to do so would be to articulate a complete theory of how our language works - effectively a theory of truth for it. A consequence of this is that there are constraints we can only explore experimentally - by trying out new ways of talking, to see if they work.