A comparison of several Zipf‐type distributions in their goodness of fit to language data
Beth I. Krevitt, Belver C. Griffith · Journal of the American Society for Information Science · 1972
Abstract Raw language data dealing with frequency of letters, phonemes, words and categories were transformed to determine the goodness of fit of such obtained data to several Zipf‐type distributions: log percentage‐log rank, percentage‐Whitworth, log percentage‐linear rank, and linear percentage‐log rank. For comparison, 9 sets of 50 three‐digit random numbers and the 9 sets of intervals created by assuming those numbers to be randomly dissecting a line, were ordered by size, as in a Zipl distribution, and tested as if they were raw language data. Two distinct types of frequency distributions seem to emerge in language data with the lesser studied type applying to phonemes, letters, and conceptual data being fit by the Whitworth or loglinear (Linear percentage vs. log rank) distributions and words by the hyperbolic (Log percentage vs. log rank) distribution.