Zero-weighted constraints in Noisy Harmonic Grammar
Bruce P. Hayes, Aaron Kaplan
January 2023

In Noisy Harmonic Grammar, a stochastic version of Optimality Theory, constraints are weighted and the output of the grammar is a probability distribution over GEN, computed by adding a noise factor to the constraint weights at each evaluation time. Intuitively, one might expect that under this system constraints bearing zero weights would have no empirical effect, but this is not so: we explore cases in NHG in which a zero-weighted constraint affects the probability of candidates that violate it. Sometimes this effect is a straightforward lowering of the probability of the violator. But in other cases, the effect is actually reversed: the zero-weighted constraint helps candidates that violate it. Finally, the effect is not just a probability shift, but can be absolute: when a zero-weighted constraint is responsible for harmonic bounding, it rules out violating candidates categorically. We explore what sort of empirical data could serve to test these predictions and point out alternative frameworks we might appeal to if they emerge as false.
Format: [ pdf ]
Reference: lingbuzz/007077
(please use that when you cite this article)
Published in: To appear in Linguistic Inquiry
keywords: noisy harmonic grammar, harmonic bounding, clipping, zero weight, phonology
Downloaded:77 times


[ edit this article | back to article list ]