PREDICTING CLUSTERING FROM SEMANTIC STRUCTURE

This study presents a process model for predicting the strength of semantic clustering within homogeneous semantic domains. The key element of the model is the assumption that clustering between adjacent items in recall is a function of their semantic similarity defined by proximity in a multidimensional space. Data from 17 word lists drawn from various homogeneous semantic domains were collected by a method that simultaneously provides interitem proximity data for similarity scaling and creates a memory list for later recall. Wide variation in the strength of semantic clustering was observed among the 17 word lists. Subjects' clustering performance was in close correspondence to predictions derived from the process model. Moreover, the degree of observed and simulated clustering across lists was strongly associated with distributional features of the semantic structure of the word lists.