Language change in multidimensional space: New methods for modelling linguistic coherence

Linguistic coherence-the co-variation of language variants within speaker repertoires-has been proposed as a key process driving the divergence of language dialects. Previous studies on coherence have been often limited by dataset sizes and analyses. We analyze the use of 185 variables across 78 speakers from the Gurindji community in Australia. We use two multivariate statistical approaches to test whether clusters of variables co-vary with generation, family, household, exposure to Gurindji language speakers and education. Using Discriminant Correspondence Analysis, we find generation is the strongest grouping factor of speakers and co-varies with clusters of variants. Using the Generalized Linear Mixed Model, we find these clusters of variants not only represent a gradual loss of Gurindji language use across generations, but also contribute to distinct patterns of language usage in the different generations. Our study demonstrates the use of multivariate analyses on big datasets to identify sociolects, an important step in linking the 'micro-level' processes to the 'macro-level' outcomes.