We evaluated how language and cognitive systems associated to inhibitory control and conflict adaptation in reactive control tasks interacted using machine-learning approaches. In order to determine whether bilingual experience interacted with inhibitory control, we constructed theoretically motivated candidate models of the Simon and Number Stroop task data (N= 777 adult bilinguals, ages 18–43, living in Montréal, Canada). These models included two types of conflict adaptation: shorter term sequential congruency effects and longer term trial order effects. Models that included ongoing features of bilingual experience gave accurate predictions of novel, unmodeled data. Mixed language usage specifically indicated the change in trial order during disagreement. Numerical Stroop, which overtly incorporates linguistic or symbolic information and has substantially higher languageand response-related uncertainty, is the only task where this effect was observed.
