Evaluating theories of bilingual language control using computational models

Abstract

Bilingual language control refers to a bilingual’s ability to speak exclusively in one language without the unintended language intruding. It has been debated in the literature whether bilinguals need an inhibitory mechanism to control language output or whether a non-inhibitory mechanism can be used. This paper presents mathematical models instantiating the two accounts. The models explain how participants’ reaction times in language production (naming) are impacted by across-trial semantic relatedness and consistency of language (same or different language across trials). The models’ predictions were compared to data from an experiment in which participants named semantically-related and -unrelated pictures in their first and second language. Results indicate that within-language facilitation effects are abolished after a language switch, supporting the predictions of the Inhibitory Model. However, within-language facilitation was observed over the course of ‘stay’ trials in which no language switch was required, contrary to the predictions of both models. A second experiment was conducted to determine the origin of this unexpected facilitation, by separating spreading activation effects from incremental learning effects. The results suggest the facilitation observed in Experiment 1 was due to spreading activation. Together, the modeling and data suggest that language switching abolishes spreading activation effects, but cumulative semantic interference (created by incremental learning) is unaffected by language switching. This suggests that (1) within-language control is non-competitive, (2) between-language language control is competitive and (3) incremental learning plays a role in bilingual language speech production.