Being Bayesian about Categorical Probability