Proceedings of the Annual Meeting of the Cognitive Science Society

A graph-based semantic map is a visual representation of presumptively universal conceptual structure underlying seman-tic variation across languages. In such maps, vertices (nodes) represent semantic functions (e.g., the spatial relation ofsupport) and edges connect conceptually similar functions. Using an algorithm that selects edges based on the frequencywith which pairs of semantic functions co-occur across words (or other linguistic forms), Regier and colleagues inferredparsimoniousbut not maximally informativesemantic maps from cross-language data on indefinite pronouns and spatial re-lations. Here, using the same data, we present several alternative map inference methods that prioritize informativeness byaccounting for typological prevalencethe frequency with which pairs of semantic functions co-occur across languagesviathe selection and/or weighting of edges. We suggest that these methods may provide a more complete picture of theuniversal conceptual bases of cross-language semantic variation.