Small Worlds and Big Data: Examining the Simplification Assumption in Cognitive Modeling

Approaches to modeling semantic memory fall into two main classes: Those

that construct a model from a small and well-controlled artificial dataset (e.g.

Collins and Quillian, 1969; Elman, 1990; Rogers & McClelland, 2004) and

those that acquire semantic structure from a large text corpus of natural language

(e.g. Landauer & Dumais, 1997; Griffiths, Steyvers, & Tenenbaum, 2007; Jones

& Mewhort, 2007). We refer to the first class as the small-world approach1 and

the latter as the Big Data approach. The two approaches differ on the method by

which to study semantic memory and exemplify a fundamental point of theoretical

divergence in subfields across the cognitive sciences.