Good–Turing frequency estimation is a smoothing estimator for estimating a multinomial distribution. It seems very convoluted.
- From mathematical statistics point of view, what is the rationale behind the construction of Good–Turing frequency estimation?
- Is it a shrinkage estimator?
- Is it based on the posterior distribution with respect to some prior distribution on the parameters, in Bayesian inference view?
- Is it not additive smoothing, is it? What type of smoothing is it?