Learning from quantified episodic prediction errors: On the neural basis of gist revision

1070 

Abstract Submission 

Nina Liedtke^1,2, Marius Boeltzig^1,2, Ricarda Schubotz^1,2

¹University of Muenster, Muenster, NRW, ²Otto Creutzfeldt Center for Cognitive and Behavioral Neuroscience, University of Muenster, Muenster, Germany

Nina Liedtke  
University of Muenster|Otto Creutzfeldt Center for Cognitive and Behavioral Neuroscience, University of Muenster
Muenster, NRW|Muenster, Germany

Marius Boeltzig  
University of Muenster|Otto Creutzfeldt Center for Cognitive and Behavioral Neuroscience, University of Muenster
Muenster, NRW|Muenster, Germany

Ricarda Schubotz  
University of Muenster|Otto Creutzfeldt Center for Cognitive and Behavioral Neuroscience, University of Muenster
Muenster, NRW|Muenster, Germany

Episodic memories do not always accurately reflect our past experiences, but can be changed through new input. It is hypothesized that such changes are driven by mnemonic prediction errors which arise when an expectation based on memory representations does not match the actual experience (Sinclair & Barense, 2019). Further research suggests that there might be a qualitative difference in brain responses and memory performance after prediction errors that affect the gist of an episode (gist modifications) compared to ones that leave the gist intact (surface modifications; Siestrup & Schubotz, 2023). However, it remains unknown whether the magnitude of prediction errors is related to the likelihood of memory updating. It has been suggested, that especially prediction errors of medium strength might lead to changes in memory (Ritvo, 2019). Therefore, we elicited surface and gist prediction errors of different strengths to investigate their distinct effects on brain activity and memory.

The sample consisted of forty-three healthy, right-handed participants (36 women, age 18-32 years, M= 22.66 years). The study used short dialogues as highly naturalistic and meaningful stimuli and it was designed as a five day paradigm. Day one to three were functional Magnetic Resonance Imaging (fMRI) sessions, day four and five took place on the computer. On the first day, 30 dialogues were presented for the first time in the MR scanner and a second time outside of the scanner. On day two, 24 of the presented dialogues were partly modified, while six remained unchanged. The modification was either on the surface level (i.e., synonyms or different phrasings) or on the gist level (i.e., changing the meaning of the utterance) and designed to be either low or high in degree. For each dialogue, each participant listened to only one modification. On day three, the fMRI session from day one was repeated. Imaging on the first three days was conducted on a 3T Siemens Prisma MR tomograph (TR/TE= 2000/30ms, FOV= 192x192mm2, 33 slices, slice thickness= 3mm). Recognition tests and cued recall were conducted on day four to assess memory for both the original versions and the heard modifications. On day five, participants rated to what extent the original and modified versions differed. FMRI data were preprocessed (slice time correction, realignment, co-registration of functional to structural scans, normalization, and smoothing) and analyzed with SPM12 using a general linear model.

For surface modifications compared to non-modified dialogues we found activation in the inferior frontal gyrus, left dorsal premotor cortex, left (pre-) supplementary motor area and in the right anterior as well as left middle superior temporal sulcus (p<.005, FDR corrected). Gist modifications also caused activations in these regions, but compared to surface modifications we found additional brain responses in medial frontal areas, the ventral precuneus, the posterior cingular cortex and the temporo-parietal junction (p<.005, FDR corrected). In the recognition test, participants recognized low gist modifications less confidently than other modification types (F(1,1050) = 5.02, p = .025). In accordance with that, modeling the BOLD response with participant-rated degree of difference revealed brain activation in both parahippocampal gyri for lower compared to stronger gist changes (p<.001, uncorrected).

We demonstrated that gist modifications differ from surface modifications not only quantitatively but also qualitatively. While there were no effects for the strength of surface modifications, we found a behavioral effect of low gist modifications on new learning and the original memory. This could possibly reflect the postulated particular effect of medium-strength prediction errors, which was also associated with specific parahippocampal activity.

Social Cognition

Long-Term Memory (Episodic and Semantic) ¹

Learning and Memory Other ²

BOLD fMRI

FUNCTIONAL MRI

Learning

Memory