Hidden Markov Modelling of Viewing Behaviors Reveals Discrete “Encoding States” During Visuospatial Memory Formation

Poster Session B - Sunday, April 14, 2024, 8:00 – 10:00 am EDT, Sheraton Hall ABC

Chloe Kindell¹ (ckinde2@lsu.edu), Heather Lucas¹; ¹Louisiana State University

Visuospatial memories can include information about item details, inter-item relations, and/or relations that involve the bounds of the display space. Item and relational encoding rely on distinct neurocognitive processes, but little is known about how learners balance these encoding goals from moment to moment. We used hidden Markov models to examine whether eye movements made during intentional study can be parsed into discrete "encoding states" that emphasize different types of information. Participants (n=60) studied visual displays containing six abstract items for 16 seconds each in preparation for either a spatial reconstruction task or an item recognition task. Each gaze transition (“visit”) made to one of the display items during study was an observation in the model, and two variables—number of fixations made during the visit and mean fixation duration— served as emissions. Based on measures of model fit, we identified three states, one of which was consistent with item encoding and one with relational encoding. Task type (spatial versus item) interacted with state probabilities, with more time spent in the item state during the item versus the relational task and vice versa. In both tasks, memory errors were associated with insufficient time spent in the item state, particularly toward the beginning of a trial. Multistate modeling of eye movements is promising avenue for memory research that can be easily extended in to map gaze-defined encoding states onto concurrently obtained neural data.

Topic Area: LONG-TERM MEMORY: Episodic