You cannot just count: A statistical rethinking of semantic richness in free-generation tasks and semantic norms

Semantic richness refers to the number of distinct features people associate with a concept, a key indicator of how knowledge is represented and accessed in memory. It is typically measured through free-generation tasks-such as the property listing task-where participants list properties of everyday concepts (e.g., a BANANA might be described as "yellow," "soft," or something you "peel"). However, most studies simply count the observed features, ignoring sampling variability and leading to biased comparisons across groups. To address this limitation, we adapted the Chao2 estimator-originally developed in ecology-to infer the total number of features associated with a concept, including those not observed in a given sample. We validated this approach for psychological research through extensive Monte Carlo simulations based on empirical data from three languages. Results show that Chao2, and especially its bias-corrected version (Chao2BC), yield more accurate and interpretable estimates than simple counts. This framework reframes semantic richness as a problem of statistical inference, providing a principled basis for comparing conceptual data across languages, populations, and experimental contexts. (PsycInfo Database Record (c) 2026 APA, all rights reserved).