General information

This is the companion website of Frequency, dispersion, association, and keyness: revising and tupleizing corpus-linguistic measures, a book published in the Studies in Corpus Linguistics series of John Benjamins (edited by Ute Römer-Barron); I just sent off the corrections for the second proofs and the index.

For the best use of the book, you should have up-to-date versions of

• R (the book was mostly written with R 4.3.2 and Quarto)
• RStudio (the book was written with RStudio 2023.12.1 Build 402).

• download this zip file,
• unzip it (with the password provided in the book), which should return a folder <2024_STG_KLD4C> with all the files in it and an R package/source file (a tar.gz)), and then
• install the package KLD4C from the package/source file in that folder (this was tested and worked unproblematically on Linux (POP OS, an Ubuntu derivative) and Windoze 11).

Errata

• p. 198-199: There is one (infrequent) kind of situation where the way to compute upp discussed in the book is not ideal, namely when the frequency of an item is greater than the size of the smallest corpus part, which is the case here where the frequency of a (15) is greater than that of the smallest corpus part (p1, with 9 words). In such situations, one cannot practically put all 15 occurrences of a into p1 – instead, the clumpiest distribution would consist of putting 9 as into p1 and the remaining 6 as into one of the three corpus parts that has 10 words (i.e., p2, p3, or p4) and compute upp from that. (This logic applies to the discussions of computing upp for dispersion in Sections 4.4.1, 5.1, and 5.3; I am grateful to Hewei Liu for alerting me to this.) The above zip file now contains an updated version of the KLD4C package for the book that contains a new function most.uneven.distr to compute the clumpiest distribution for a word over corpus files. I have so far only tested it a bit, but it seems to work as intended and, in the vast majority of cases, the output of this function will return the result that you would get from the simpler strategy discussed in the book, but whenever a word is so frequent that it wouldn't fit in the smallest corpus part, this function will return the theoretically ideal version of upp. Also, in April 2025, I fixed an error in the C++ function CppKLDscolfromrowtotals.
• p. 199, below first bullet list: from low (2.473931) to upp (0.002899914) → from low (0.002899914) to upp (2.473931)