Multifractal correlations in natural language written texts: Effects of language family and long word statistics

During the last years, several methods from the statistical physics of complex systems have been applied to the study of natural language written texts. They have mostly been focused on the detection of long-range correlations, multifractal analysis and the statistics of the content word positions. In the present paper, we show that these statistical aspects of language series are not independent but may exhibit strong interrelations. This is done by means of a two-step investigation. First, we calculate the multifractal spectra using the word-length representation of huge parallel corpora from ten European languages and compare with the shuffled data to assess the contribution of long-range correlations to multifractality. In the second step, the detected multifractal correlations are shown to be related to the scale-dependent clustering of the long, highly informative content words. Furthermore, exploiting the language sensitivity of the used word-length representation, we demonstrate the consistent impact of the classification of languages into families on the multifractal correlations and long-word clustering patterns.