Learning, Consolidating, and Generalising Novel Morphology

Despite a central role for morphological knowledge in supporting linguistic generalisation, the neural representations supporting its learning remain largely unexplored. This thesis addressed this gap by exploring the role of memory consolidation in morphological learning and generalisation. In three experiments, adult participants learned an artificial language in which stems (e.g. gleet, shiln) combined with plural affixes (e.g. –aff, -opp; gleetaff, shilnopp) to refer to the occupation of multiple male and female characters. Mimicking properties of morphological systems in natural languages, the plurals varied in their phonological consistency/ambiguity and type/token frequency. Two sets of plurals, distinguished by gender, were trained on two successive days. Experiment 1 revealed that generalisation to novel phonologically ambiguous forms measured on the second day showed a greater influence of token frequency for plurals trained on the previous day, suggesting overnight changes in their underlying representations. Experiment 2 examined this effect further by using fMRI to compare the neural representations underlying plurals learned on the day of scanning or on the previous day. Representational Similarity Analysis revealed increased similarity structure among high type frequency plurals and reduced similarity structure among high token frequency plurals following overnight consolidation in the left superior temporal gyrus (STG). These results are consistent with a Complementary Learning Systems (CLS) model in which overnight consolidation supports the development of overlapping representations among several items sharing the same feature (here, an affix; type frequency) and strengthens item-specific representations for frequently occurring items (token frequency). Additionally, connectivity analyses showed that the functional coupling between the left STG and the left dorsolateral prefrontal cortex was weaker for high type frequency plurals and stronger for high token frequency plurals following overnight consolidation. These results suggest that the engagement of prefrontal control processes in retrieving the newly-learned plurals is subject to overnight consolidation and sensitive to the similarity structure underlying the plurals to be retrieved. However, the overnight changes in similarity structure and functional networks observed in Experiment 2 were not mirrored by changes in generalisation to novel forms as were observed in Experiment 1. Experiment 3 aimed to address the discrepancy in consolidation-related changes in generalisation behaviour between the first two experiments. Type/token frequencies were manipulated to bias learning, consolidation, and generalisation towards high token frequency plurals. Despite this manipulation, no consolidation-related changes in generalisation were observed. Findings from all three experiments are interpreted in the context of the CLS model and a role for overnight consolidation in morphological learning and generalisation is discussed.Economic and Social Research Counci