Symmetries and asymmetries: their grammar, typology and parsing

Cross-linguistic variation reveals both symmetries (A+B and B+A are productive) and asymmetries (only one ordering is productive, in all languages or in certain subsets). The challenge is to better understand and predict which will be which. This paper introduces a parsing approach based on two efficiency principles. Maximize On-line Processing (MaOP) and Minimize Domains (MiD). MaOP defines a preference for parse strings that avoid misassignments (or garden paths) and unassignments of properties on-line (compared with structures that assign the relevant properties earlier). MiD defines a preference for minimal surface structure domains in the processing of all relations of combination and dependency. The predictions made by these principles for performance and grammars are defined. Symmetries are predicted across grammars when each ordering is motivated, by MaOp or Mi D or both. Asymmetries arise when one order is motivated by MaOP, and there is no or significantly less motivation for the other by either principle. Only one order is strongly preferred in performance in these cases and is conventionalized in grammars.