The Laplacian spectrum of language: Experiments with linguistic families from the Americas

To what extent would statistical mechanics approaches help to represent languages from the Americas? Is it possible to extract useful information about the relationships between these languages? This work studies a graph-based approach to extract information from text corpora of languages of the Americas. Each language is viewed as the set of eigenvalues obtained from the Laplacian matrix of co-occurrence graphs. The results suggest that our graph-based feature extraction technique is partly comparable to the knowledge contained in typological databases. We argue that our approach might propose a solution to the lack of textual resources for low-resource languages.