Testing an agent-based model of language choice on sociolinguistic survey data

Abstract

The paper outlines an agent-based model for language choice in multilingual communities and tests its performance on samples of data drawn from a large-scale sociolinguistic survey carried out in Estonia. While previous research in the field of language competition has focused on diachronic applications, utilizing rather abstract models of uniform speakers, we aim to model synchronic language competition among more realistic, data-based agents. We hypothesized that a reasonably parametrized simulation of interactions between agents endowed with interaction principles grounded in sociolinguistic research would give rise to a network structure resembling real-world social networks, and that the distribution of languages used in the model would resemble their actual usage distribution. The simulation was reasonably successful in replicating the real-world scenarios, while further analysis revealed that the model parameters differ in importance between samples. We conclude that such variation should be considered in parametrizing future language choice and competition models.