On the (non)utility of Juilland's D to measure lexical dispersion in large corpora

This paper explores the effectiveness of Juilland's D as a measure of vocabulary dispersion in large corpora. Through a series of experiments using the BNC, we explored the influence of three variables: the number of corpus-parts used for the computation of D, the frequency of the target word, and the distributions of those words. The experiments demonstrate that the effective range for D is greatly reduced when computations are based on a large number of corpus-parts: even words with highly skewed distributions have D values indicating a relatively uniform distribution. We also briefly explore an alternative measure, Gries- DP (Gries 2008), showing that it is a more reliable and effective measure of dispersion in a large corpus divided into many parts. In conclusion, we discuss the implications of these findings for quantitative methods applied to the creation of vocabulary lists as well as research questions in other areas of corpus linguistics.