Delineating linguistic contexts, and the validity of context diversity as a measure of a word's contextual variability

Abstract

Abstract Research that has compared the predictive validities of word frequency (WF) and contextual diversity (CD) in lexical processing tasks has generally found that CD accounts for more unique variance than does WF. After CD's variance has been accounted for, WF accounts for little or none of the remaining variance in lexical processing times. This has led some researchers to conclude that the concept CD is thought to measure -- variety in contexts of appearance -- is more relevant to learning than is WF, which is instead an index of total exposure to a stimulus (see Adelman, Brown, & Quesada, 2006; Jones, Johns, & Recchia, 2012). We provide evidence that CD is a poor operationalization of variety in contexts of appearance. As a consequence, prior comparisons between WF and CD and the conclusions drawn from those comparisons should be re-examined. We discuss how the concept of linguistic context has been poorly treated in the CD literature and that, if we allow for linguistic contexts to be variable in size and nested, the apparent dichotomy presented by CD and WF disappears. This article also introduces a new measure of context variability that is simple to calculate and relatively uncorrelated with WF. We demonstrate that in most cases, the described measure accounts for more unique variance in lexical processing times than does CD. These results are used to motivate consideration of what an appropriate definition of context is for studying lexical processing, and ways alternative to CD that might be used to operationalize context variability.