Speaker
Description
Evolutionary models of genetic drift have been successfully applied in the last decade to the analysis of diachronic change in corpus data. Amongst all of them, the Wright-Fisher model stands out as a simple but powerful paradigm that is mathematically equivalent to models of the cultural transmission of language, like Iterated Bayesian Learning and the Utterance Selection Model. Wright Fisher characterises language change as a process where different expressions (variants) realising the same function (meaning) compete against each other for usage in the speech community. Corpus data analyses using Wright-Fisher aim at detecting and quantifying the evolutionary forces (e.g. drift, selection, mutation) shaping this competition process, which can shed light on the underlying diachronic phenomena driving language change.
This approach is limited in that it assumes isolation of the competition process for each function. In this work, we present an Iterated Bayesian Learning model of grammar change involving the co-evolution of interrelated functions and expressions that better reflects the complex interdependencies often present in language change. We show that this model is equivalent to a modified Wright-Fisher paradigm, and maps effects including learning biases, analogy and social preferences to evolutionary forces. This enables its application to hypothesis testing and model selection in the analysis of corpus data, which we illustrate through applications to the study of the evolution of relativisers in Middle and Modern English, and the emergence of periphrastic do in Early Modern English. Our results show that evolutionary models incorporating co-evolving functions are relevant towards our empirical and quantitative understanding of language change. The model we introduce is a promising first step towards this.
