Learning from quantified episodic prediction errors: Individual biases in gist revision
To what extent and in which ways can episodic prediction errors (PEs) during retrieval alter episodic memories, and what properties do episodic PEs have that initiate such changes? Complementary to the first funding period, where we manipulated PE qualitatively, we will quantitatively manipulate episodic PEs in personally meaningful episodes of social interaction. We use a multi-step reconsolidation protocol that includes fMRI during both encoding and retrieval of episodes. In Exp. 1, we examine the modifiability of reactivated episodes by introducing PEs of variable strength. This will be achieved by independently manipulating the level of episodic predictedness and episodic predictability as quantified on the basis of an online rating study. We investigate whether the (dis)similarity between encoded and retrieved episodes is quantified by the PE strength at retrieval in two ways: at the behavioral level by multidimensional scaling of memory performance and at the neural level by analyzing the representative dissimilarity of BOLD responses. To identify potentially distinct contributions of hippocampus and neocortex, PEs refer either to episodic gist, i.e., the specific sequence of events that essentially constitute the episode, or to the episodic surface, i.e., the exact perceptual configuration of the episode. Exp. 2 extends the approach by accounting for idiosyncratic differences in episodic expectancy that may influence individual propensity to revise episodes after PE. The subjects' individual experiences and beliefs are captured by four moderator variables that quantify, for each episode, their subjective (a) association with an own autobiographical encounter, (b) emotionality, (c) social rule consistency, and (d) everyday typicality. Using these variables, we examine individual preconditions for the modifying effect of episodic PEs during scenario construction, with variables (a) and (b) modeling aspects of autobiographical and affective prior experience, and variables (c) and (d) rather experiential knowledge about (semantic) social rules. Exp. 3 employs episodic continuations as a specific type of PE, some of which additionally plausibly bridge previously unrelated episodes. Such continuing and bridging PEs should be easier to integrate as they leave the coherence of existing memories untouched or even increase it, and thus more readily lead to revision of encoded episodes. We examine the rating-quantified strength of PEs and their systematic effects on episode similarity (pre- vs. post-PE) in BOLD response and behavior.
Solidity Meets Surprise: Cerebral and Behavioral Effects of Learning from Episodic Prediction Errors