The universal "shape" of human languages: spectral analysis beyond speech

Fermin Moscoso Del Prado

doi:10.1038/npre.2011.6097.1

Abstract

AbstractA common recommendation for linguists is that, when analyzing language, they should take a "man-from-mars", know-nothing, perspective to describe the structures that they observe, if possible limiting themselves to the mechanical application of a set of well-defined techniques and criteria. Unfortunately, the complex nature of linguistic structures makes it difficult to adopt such a detached perspective. A language, as represented by a corpus of text, can be described macroscopically by the symbolic periodogram of the corpus, analogous to the spectrograms commonly used for describing speech. Here, I show that the periodogram exhibits a universal "shape" from human languages, and this shape originates in known properties of the human mind. Despite the universality of the overall pattern, subtle differences also reveal particularities of individual languages. These differences demonstrate the long-held --but unproven-- hypothesis that human languages balance the amount of structure contained in different levels of description so that the total amount of linguistic structure remains fairly constant across languages. The universal pattern found in the periodograms illustrates how the biological properties of the mind constrain the structure of human languages.

Highlights

The periodogram is a fundamental tool for the acoustic analysis of speech data
Spectral analyses have revealed that English, German, and Latin texts exhibit a roughly U-shaped periodogram when plotted in double logarithmic axes[3,4,5,6]
The overall pattern is significantly different from what one would expect by chance: The flat spectrum denoting the absence of correlations plotted by the dashed line in the figure

Summary

Introduction

The periodogram is a fundamental tool for the acoustic analysis of speech data. Traditionally, research on human speech concentrates on the very high frequencies of the spectrum, so as to focus on phonemic or subphonemic information. A negative slope is indicative of anti-persistence, that is, the repetition of structures happens below chance level.

Results

Conclusion