A probabilistic model for integration of strong dependent cues in category identification

Semantic features are relationally connected to one another and they may show different levels of importance in indexing a familiar basic-level category. The article proposes a new Bayesian model to efficiently capture the relative importance of possibly dependent semantic features in the process of category identification. Unlike the Bayesian model of category identification (naive Bayes identifier), the new model does not require independence between cues and uses a simple property, called strong stochastic dependence (SSD), to efficiently represent dependence-type patterns in semantic features. We applied the model to empirical data from a category identification task administered to a group of 30 participants. Results show that the new Bayesian model is consistently superior to the naive Bayes identifier in predicting the distribution of the participants’ responses in the empirical semantic task.