Proceedings of the Annual Meeting of the Cognitive Science Society

Transitive events involve an agent (source of the action) and a patient (recipient of the action), and the successful communication of such events requires conveying these roles. Languages employ various strategies for this, including word order and case marking. Differential Argument Marking (DAM), denoting systems in which case marking only occurs in certain contexts, has garnered considerable attention for its cross-linguistic prevalence. One account posits that DAM arises via efficiency: referents with properties atypical for their role (and therefore less predictable) create pressure for marking, while typical referents do not. This predicts similar numbers of languages with atypical agent-marking and atypical patient-marking. However, patient-marking is not only more common, but agent-marking systems often violate expected typicality patterns. To address this asymmetry, we develop a computational model showing that it emerges naturally from the interaction of typicality biases with production costs.