Abstract
Frequency spectra for words of coinciding length are observed. To this end a method of representation of empirical word frequency spectra is proposed. The exponent of the Zipfian power law shows a strong dependency on word length. For short words this exponent assumes values which cannot be accounted for by Mandelbrot's deduction of the Zipfian law. A theoretical model based on a stochastic process is proposed to explain the unusual values of the exponent.
Notes
Address correspondence to: Edda Leopold, FB II, Linguistische Datenverarbeitung, Universität Trier, Universitätsring 15, D ‐ 54286 Trier, Germany. E‐mail: [email protected]‐Trier.de.