A Method to Evaluate Systemic Constraints in Probabilistic Grammars

Systemic constraints evaluating several input-output mappings jointly have been proposed in phonology and morphology to account for contrast enhancement and homophony avoidance. However, these constraints are not directly compatible with current probabilistic models for constraint-based grammars, as these models typically assume constraints that evaluate input-output mappings individually. This squib proposes a simple method to address this issue. The method consists in reconstructing the probability of individual input-output mappings from the joint probability distribution of all input-output mappings through marginalization. The approach is illustrated with a concrete case study involving variable homophony avoidance in spoken Persian.