Probability and randomness in phonology: Deep vs. shallow stochasticity
Sam Tilsen (Cornell University)
Abstract
This paper argues that there are two different ways that probability and randomness
can be used in models of phonological cognition. One of these is termed shallow
stochasticity and refers to a situation in which probabilities are cognitively
represented and provide a causal basis for variation in behavior. The other is termed
deep stochasticity and refers to a situation in which probabilities are not cognitively
represented but merely useful due to our ignorance of the detailed dynamics of a
system. The argument is made specifically in relation to the framework of Maximum
Entropy Harmonic Grammar (MaxEnt) and its theoretical analysis of the
phonological pattern of nasal substitution in Tagalog. A number of critiques of the
MaxEnt approach are presented.
Keywords
Maximum Entropy Harmonic Grammar, MaxEnt, probability, randomness,
stochasticity, Tagalog