Learning Phonotactic Distributions