Multiple brain networks underpinning word learning from fluent speech revealed by independent component analysis

Abstract

The contribution of the different brain networks to word learning from fluent speech is still largely unknown, although neuroimaging studies using standard subtraction-based analysis have suggested that frontal and temporal regions are involved in it. However, this type of analysis cannot identify the extent of distributed networks that are engaged by a complex task such as word learning. Here we use Independent Component Analysis (ICA) – a multivariate approach – to characterize the different brain networks subserving word learning from an artificial language speech stream containing novel words. Four spatially independent networks were associated with the task: (i) a dorsal Auditory-Premotor network; (ii) a dorsal Sensory-Motor network; (iii) a dorsal Fronto-Parietal network; and (iv) a ventral Fronto-Temporal network. A further, fine-grained analysis of the network level of engagement across time showed that the engagement of these networks varied through the learning period with only the network covering auditory and motor areas being engaged across all blocks. In addition, the connectivity strength of this network in the first and second blocks correlated with the individual variability in word learning performance. These findings obtained with a data-driven approach suggest that: (i) word learning relies on segregated connectivity patterns involving dorsal and ventral networks; and (ii) specifically, the dorsal auditory-premotor network connectivity strength is directly correlated with word learning performance.