Intermittency and scale-free networks: a dynamical model for human language complexity

TL;DR

This paper models human language complexity using a dynamical system on scale-free networks, linking linguistic features with statistical mechanics and explaining Zipf's law through long-range correlations.

Contribution

It introduces a novel dynamical model based on scale-free networks to explain features of language complexity and connects it with empirical data and linguistic paradigms.

Findings

The diffusion entropy method reveals anomalous scaling in language data.

A random walk on a scale-free network reproduces observed complexity.

The model explains Zipf's law via the generalized central limit theorem.

Abstract

In this paper we try to model certain features of human language complexity by means of advanced concepts borrowed from statistical mechanics. We use a time series approach, the diffusion entropy method (DE), to compute the complexity of an italian corpus of newspapers and magazines. We find that the anomalous scaling index is compatible with a simple dynamical model, a random walk on a complex scale-free network, which is linguistically related to Saussurre's paradigms. The network complexity is independently measured on the same corpus, looking at the co-occurrence of nouns and verbs. This connection of cognitive complexity with long-range time correlations also provides an explanation for the famous Zipf's law in terms of the generalized central limit theorem.