Towards An Algorithmic Account Of Phonological Rules And Representations