Shaping Memory from the Start: Initial Prediction Errors during First Encoding

The brain constantly makes predictions about upcoming input, and prediction errors (PEs) have been shown to promote encoding of the unexpected information. So far, previous experimental designs have left it unclear if PEs that may be evoked by the first exposure to a coherent novel stimulus, based on individual knowledge, experiences, and beliefs, can affect subsequent memory processes. In the current study, we aimed to test the neural and mnemonic consequences of these initial PEs and how they influence such outcomes together with later induced, experimental PEs. To this end, participants (N = 42) listened to naturalistic dialogues, which induced an initial PE, while undergoing fMRI scanning. Later, the dialogues were modified to induce a second, experimental PE, and memory for the original and modified versions was assessed using a recognition test. The results showed that initial PEs, like experimentally induced PEs, shifted the balance from top-down predictions to bottom-up processing, as reflected in reduced predictive reinstatement and stronger activation in the auditory cortex upon re-exposure. Moreover, semantic components of both initial and experimental PEs enhanced learning, while IFG activation biased memory towards the currently activated representation rather than the novel input. Taken together, these findings provide first evidence for the existence and relevance of initial PEs that are evoked during the encoding of coherent episodes not obviously violating world knowledge based on individual experiences and beliefs, indicating that they should be taken into consideration in paradigms investigating episodic PEs.