Code-switching burstiness in bilingual speech

Reliably characterizing bilingual ecologies and developing models of language activation requires metrics for code-switching (CS). This study charts the time course of CS, plotting multi-word switches by intervening distance, using Intonation Units (prosodic speech chunks) as the token level. Drawing on a community in which CS is an ingroup discourse mode, three datasets of bilingual speech are compared. The datasets were chosen to differ in the proportions of the two languages (from balanced to skewed) and the proportion of switch points (from under .05 to over .15). Despite differences in language distribution and CS frequency, CS appears bursty: clusters of activity interspersed between longer unilingual stretches. This intrinsic burstiness indicates that ‘dense CS’ is dense successions of instances separated by long gaps. If parallel activation of two languages fluctuates regardless of situational context, then rather than moving into and out of bilingual mode, bilinguals may move along a fluctuating bilingual mode.