Grammar is said to be about defining all and only the 'good' sentences of a language, implying that there are other, 'bad' sentences - but it is hard to pin those down. A century ago, grammarians did not think that way, and they were right: linguists can and should dispense with 'starred sentences'. Corpus data support a different model: individuals develop positive grammatical habits of growing refinement, but nothing is ever ruled out. The contrasting models entail contrasting pictures of human nature; our final chapter shows that grammatical theory is not value-neutral but has an ethical dimension.