How register and region shape the language network: evidence from Computational Construction Grammar

Abstract

While Construction Grammar has proven effective at modelling regional and register
variation separately, it has seldom been used to explore the interaction between the two. The
present paper fills this gap by applying a Computational Construction Grammar framework to a
collection of large English corpora, including two digital registers (written tweets and spoken
YouTube transcripts) and five inner-circle varieties (US, UK, Canada, Australia, and New
Zealand). We show that constructionist principles successfully capture a range of register- and
region-based distinctions across the grammar, and we report the novel finding that both sources
lead to systematic, largely independent patterns of variation. Specifically, register effects are
more pervasive and concentrated in abstract, high-level constructions, while regional effects are
relatively sparser and manifest most prominently in lower-level, surface constructions. To
account for these results, we hypothesise that register and regional associations operate along
a continuum of constructional ‘salience’: while the former require the explicit learning of variants
for communicative functions, the latter begin as products of exposure before they can acquire
indexicality. We conclude with implications of our study for a more comprehensive model of
variation in the language network, as well as for future endeavours towards intersecting
Construction Grammar and sociolinguistic theory.