Towards a deeper understanding of the complex behaviour observed in the distribution of words in written texts

Here we show that the recently reported presence of long-range correlations in the distribution of words along texts is due to the complex distribution of the keywords, while common words are not correlated. Indeed we prove that the degree of long-range correlations of a word at long scales is a good measure of its relevance to the text. Additionally, we develop a model able to reproduce the spatial distribution of a word in a text, based on the long-range correlations observed for the word. The model not only reproduces the complex behaviour characterized by the presence of correlations at long scales and the degree of relevance of the word, but also the probability distribution of the inter-occurrences distances in the whole range of scales.