Contextual clarity during training hinders learning transfer but spares task-switching

An emerging hypothesis proposes that overlap between task representations aids learning transfer butinvokes greater control demands during task-switching. A parallel line of research shows that we flexiblyinfer what is relevant to a task, and thus what forms the task representations, by observing what eventsco-occur in time and space. Thus, varying how much two tasks are interleaved, which affects whether wecan rely on temporal context to differentiate them, could alter their representations and, in turn, how wegeneralise or switch between them. Over two experiments (N=100 each), we tested whether contextualclarity during training impacts both learning transfer (Experiment One) and task-switching (ExperimentTwo). Participants searched through grids of squares to find hidden targets. Grids were bordered by one oftwo coloured cues; each colour signalled the relevance of a different set of target squares. Thus, participantslearned two task-sets, each signalled by a different task cue (colour) and associated with a unique set ofcontingencies. Participants then trained on the task without explicit colour cues. The task setpseudorandomly switched over trials at one of two rates, to modify contextual clarity across groups. TheStable group experienced rare switches (5%), whereas the Variable group experienced more frequentswitches (30%). At test, explicit colour cues were reintroduced. Rare switching impaired transfer learningon a novel task that combined the trained tasks, but not on a novel task that replicated a single trainedtask (Experiment One). However, the training regimen did not impact task-switching performance at test,where the switch rate increased to 50% (Experiment Two). These findings show that contextual clarity candissociate task representations at the expense of learning transfer, and constrains theories regarding thetransfer/task-switching trade-off.