On explanatory theories of speech categorization

The study of speech categorization has a long history. Two phenomenological laws of speech categorization are statable based on this extensive history: monotonicity and sigmoidality of the identification/categorization functions. While prior work has heavily focussed on accounting of observed data in terms of specific theories, and computationally modelling observed patterns, far less work has been done on formally comparing theories to study their explanatory potential, particularly their explanatory value with respect to the two laws. Taking up the latter challenge, in this paper, I formally compare four theories (or frameworks) that have been popular in the literature: (a) Categorical Boundary theories, (b) Prototype theories, (c) Bayesian theories, (d) Exemplar theories. Out of the four, only Category Boundary theory was found to have a clear and simple explanatory account for the two laws. Prototype theory is shown to be consistent with the law of monotonicity but not sigmoidality. Bayesian categorization and Exemplar theories were found to not be explanatory of either law except when severe, currently ad hoc, restrictions were imposed on the category distributions.