Understanding how social interactions shape language use is a central problem in sociolinguistics, as everyday exchanges are the primary mechanism through which linguistic variants are transformed, transmitted, and reinforced. Online social media provide an unprecedented opportunity to explore this question, as they generate vast amounts of data on users’ interactions and the linguistic content they produce. In this work, we investigate how social interactions within online communities mediate both the languages employed and the prevalence of non-standard linguistic forms. Adopting a complex systems perspective, we model interactions through replies and construct a directed, weighted network comprising over one million users, each assigned to a residence cell within the
United Kingdom. Based on the analysis of more than 500 million geolocated tweets, we characterize the linguistic features of the content generated by users and detect community structures within the network. Our results reveal distinct patterns in the interplay between social interactions and language use. Communities with a marked social identity tend to reinforce the use of minority
languages. Other communities exhibit adaptive linguistic behavior, with individuals shifting from their primary language to a minority language when interacting within the social group. In the case of English, we observe that in communities where members frequently employ non-standard features, internal interactions tend to amplify such practices. This process illustrates how social reinforcement within online communities contributes to the persistence and diffusion of linguistic variation.