Early language dissociation in bilingual minds: Non-invasive magnetoencephalography evidence through a machine learning approach

Abstract

Is language selection in balanced bilinguals decodable from neural activity? Previous research employing various neuroimaging methods has not yielded a conclusive answer to this issue. However, direct brain stimulation studies in bilinguals have detected different brain regions related to language production in separate languages. In the present MEG study, we addressed this question in a group of proficient Spanish-Basque bilinguals (N=45), who performed two tasks (picture naming and word reading). They were asked to name the line drawing or read the word out loud, either in Basque or Spanish, if the ink turned from black to green after one second (randomly, in 10% of trials). Sensor-level evoked activity was similar and could not be differentiated for the two languages in either task. Crucially however, decoding analyses classified the language used in both tasks, starting [~]100 ms after stimulus onset. Searchlight analyses revealed that activity detected in the right occipital-temporal sensors contributed the most to language decoding in the picture naming task, while the left occipital-temporal sensors contributed the most to decoding in word reading. The cross-task decoding analysis highlighted robust generalization effects from the picture naming to the word reading task in a later time interval. The present findings bridge the gap between non-invasive and invasive experimental evidence on bilingual lexical processing and provide novel evidence about the role of the two hemispheres in activating each language for picture naming and word reading.