Bilinguals routinely shift between their languages, changing languages between communicative settings. To test the consequences of such changes in language use, 48 Arabic–Hebrew bilinguals named pictures in Arabic (L1) before and after a brief exposure manipulation, including either reading a list of Hebrew (L2) words aloud or performing a nonlinguistic task. Half of the items post-exposure were new and half were translation equivalents of the words presented during the L2 exposure task. Further, half of the items were very low-frequency L1 words, typically replaced by borrowed L2 words. Results show that across word types bilinguals were less accurate and produced more L2 cross-language errors in their dominant L1 following brief L2 exposure. Error rates were comparable for translation equivalents and new items, but more cross-language errors were observed post-exposure on translation equivalents. These findings demonstrate the engagement of both global whole-language control mechanisms and item-based competitive processes, and highlight the importance of language context and the dynamic nature of bilingual performance.