Davis, M. H.
Medical Research Council, Cognition and Brain Sciences Unit, Cambridge, UK.
This presentation provides neuroimaging evidence in support of two neural processes in initial learning and later consolidation of novel spoken words. This proposal builds on neuro-computational accounts of lexical processing and complementary learning systems (CLS) models of memory. As in other domains, the CLS account suggests a division of labour between medial temporal systems responsible for rapid encoding of novel items and the contexts in which they occur, followed by slower, offline integration of novel and existing items in cortical representations. A review and meta-analysis of recent fMRI studies of spoken word learning shows that: (1) successful initial acquisition is associated with the magnitude of hippocampal activity, (2) rapid changes in cortical responses to pseudowords following familiarization are best explained as task-specific repetition priming, or consequences of hippocampal encoding, rather than new, word-like cortical representations, (3) cortical responses to pseudowords (e.g. in the superior temporal gyrus) only become word-like in test sessions that follow a period of overnight consolidation. A recent MEG study of spoken word learning and consolidation (with Pierre Gagnepain and Rik Henson) provides additional evidence concerning the specific cortical computations that are consolidated overnight. We assessed the time-course and location of evoked responses to trained novel words (cathedruke), real word (cathedral) and non-word neighbours (cathedron) at different stages during learning and consolidation. Consolidation-induced differences in evoked responses to learned items are: (1) time-locked to the phonetic deviation point between these triples, (2) localized to the same cortical regions that show consolidation effects in fMRI, and (3) simulated by changes in segment prediction error rather than lexical competition in neural network inspired models. These findings suggest an important role for consolidated cortical representations in supporting efficient recognition of newly learned words and provide initial ideas concerning the integration of behavioural, computational and neural evidence in spoken word learning.