But are there really enough people out there writing BlueBBerries to be statistically significant in the dataset? Is it viewing letters as shapes as well? I don't understand how no one has taught the predictive text generator to count, even if actual maths is too challenging?