Suppose (hypothesis 1) A sees the same colours that B sees, only systematically transposed, red for green and blue for yellow, for instance. Or suppose (hypothesis 2) A sees a range of visual presentations which qualitatively resemble nothing seen by B — are not colours at all, in fact — but which are related to one another in exactly the ways in which the colours B sees are related to one another. Suppose, too, that A and B are native speakers of the same natural Language L. Would such discrepancies in their visual experience necessarily make any difference to what either says in L? Can't we, on the contrary, easily imagine versions of both hypothesis 1 and hypothesis 2 on which the envisaged discrepancies couldn't make any difference to anything either A or B might say in L?