From specific examples to general knowledge in language learning

The extraction of general knowledge from individual episodes is critical if we are to learn new knowledge or abilities. Here we uncover some of the key cognitive mechanisms that characterise this process in the domain of language learning. In five experiments adult participants learned new morphological units embedded in fictitious words created by attaching new affixes (e.g., -afe) to familiar word stems (e.g., “sleepafe is a participant in a study about the effects of sleep”). Participants’ ability to generalise semantic knowledge about the affixes was tested using tasks requiring the comprehension and production of novel words containing a trained affix (e.g., sailafe). We manipulated the delay between training and test (Experiment 1), the number of unique exemplars provided for each affix during training (Experiment 2), and the consistency of the form-to-meaning mapping of the affixes (Experiments 3–5). In a task where speeded online language processing is required (semantic priming), generalisation was achieved only after a memory consolidation opportunity following training, and only if the training included a sufficient number of unique exemplars. Semantic inconsistency disrupted speeded generalisation unless consolidation was allowed to operate on one of the two affix-meanings before introducing inconsistencies. In contrast, in tasks that required slow, deliberate reasoning, generalisation could be achieved largely irrespective of the above constraints. These findings point to two different mechanisms of generalisation that have different cognitive demands and rely on different types of memory representations.