Zero-Weighted Constraints in Noisy Harmonic Grammar

In Noisy Harmonic Grammar (Boersma and Pater 2016), a stochastic version of Optimality Theory (Prince and Smolensky 1993), the constraints are weighted and the outcomes are probability distributions over GEN, computed by adding a noise factor to the constraint weights at each evaluation. Intuitively, one might expect that constraints bearing zero weights would have zero empirical effect, but this turns out not to be so. First, we show that a constraint with zero weight in NHG continues to affect the probability of candidates that violate it; the effect is either upward or downward, depending on otherfactors. Second, under certain arrangements intended to maintain the principle of harmonic bounding, zero-weighted constraints can force zero probability for candidates that violate them. We suggest what sort of cases linguists should seek in order to test the truth of these predictions, and also point out alternatives we might appeal to if these predictions emerge as false.