Abstract
The availability of social media language data, such as from Reddit, offer us a unique opportunity to analyse a linguistic system in its entirety. More importantly, this enables us to observe variation and change within a complex adaptive system (CAS). Dialectologists frequently rely on the distribution of discrete variables. As an example, 'in the weekend' in New Zealand English is preferred to 'on the weekend' in American English in contrast to 'at the weekend' in older varieties of British English. However, this top-down approach does not account for the interactions that exist within a linguistic system. Computational Construction Grammar (C2xG) takes a bottom-up approach to ‘learn’ a grammar based on lexical, syntactic, and semantic constraints. Unlike large language models (LLMs), C2xG extracts interpretable grammatical constructions within a linguistic system. We therefore ask, can we predict grammatical similarity based on user behaviour within a digital network.
Oral presentation.