Inferred and Residual Parameter Interactions in Syntax Dynamics

Abstract

What makes natural languages diversify, in their syntactic structure, even when sharing a largely common origin? We refine our simple model, which focuses on the glassy dynamics induced by a network of interactions between syntactic parameters. Distinguishing between symmetric interaction weights that can be inferred from the observed statistics of syntax across many languages, and additional asymmetric weights assigned at random, we show that both components are needed, with similar overall strength, to account for plausible long-term syntactic dynamics. An approximate balance can combine the glassy metastability provided by the inferred weights with the chaotic diversification induced by the residual asymmetric weights.