A Stochastic Model of Language Change through Prediction-driven Instability

Language change is paradoxical: Children acquire their native language accurately, yet over time, the language can change, apparently even without external influence. A possible resolution to this paradox comes from the facts that language is flexible and children do not learn from the entire population equally. Individuals vary in how they use their native language, and chance local fluctuations in such variation might be enough to trigger a language change. As we will see in Section 2, a mean-field model in which children learn from the entire population equally does not lead to spontaneous change, even in the presence of random variation. However, in Section 3 we will discuss a modification of the model, in which children can detect age-correlated patterns in variation. This does exhibit spontaneous language change: Children can detect accidental correlations between age and speech, predict that the population is about to undergo a language change, and accelerate the process, a process we will call prediction-driven instability.