Viñals, L. 1, 2 , Mirkovic, J. 3, 4 , Gaskell, G. 3 & Davis, M. 1
1 Cognition and Brain Sciences Unit, Cambridge, UK
2 Department of Theoretical and Applied Linguistics, University of Cambridge, UK
3 University of York, York, UK
4 York St John University, York, UK
Language learning and generalisation are tuned to input statistics. In two experiments, we explored the role of overnight memory consolidation in learning and generalising novel inflectional affixes trained with different type and token frequencies. We used an artificial language to train participants on two sets of plural affixes, distinguished by grammatical gender, on two successive days. Within each set, a subset of words contained an ambiguous phonological cue (e.g. arb) which was associated both with a high type frequency regular affix (e.g. farbaff[fem,plur], tarbopp[masc,plur] but also gleetaff[fem,plur], shilnopp[masc,plur], etc.) and a high token frequency irregular affix (e.g. varbesh[fem,plur], yarbull[masc,plur]). In Experiment 1, productive generalisations to untrained phonologically ambiguous singulars (e.g. zarbi[fem,sing], zarbu[masc,sing]) showed greater influence of token frequency for affixes trained on the previous day than for affixes trained on the same day. In Experiment 2, we observed overnight changes in hippocampal and neocortical responses to high type and high token frequency affixes trained in the context of an ambiguous phonological cue. These results suggest a role for overnight memory consolidation in the extraction of frequency statistics underlying inflectional morphology. We discuss these findings with reference to a Complementary Learning Systems account of learning and memory.