The Effects of Sleep on Category Structure Knowledge

Jayme Banks




Project Summary

What is the role of sleep in preserving our memories of recently learned categories? Sleep is believed to facilitate memory consolidation, or the gradual process of changes in memory storage after initial encoding that are vital to preserving the memory. Research suggests that sleep is vital for abstraction and creative thinking, though time delays are also hypothesized to facilitate this process of generalization. How does one night of sleep impact our memory beyond what we would expect within the same timespan? Using a two-session design, this thesis addresses the effect of one night of sleep versus the same amount of time (12 hours) awake on memory performance on a feature association task.

We recruited 80 workers on Amazon Mechanical Turk to complete two sessions of a statistical learning task involving features of a new species of beetle. Counter to my hypothesis, sleep was not a statistically significant factor in overall accuracy scores for session two on either of the two category structures tested (ring structure with low modularity, and more clustered mod structure with high modularity). Thus, sleep did not preserve knowledge of category structures in a way that was significantly different from the same amount of time spent awake. However, in the ring sleep condition, knowledge of core structure was preserved, while memory performance decreased for peripheral features. This suggests that sleep may prioritize better encoded information (core knowledge) over modular structure when the category structure itself is harder to learn.

Throughout this process, I have learned to engage with my curiosity while maintaining focus on what is important. Even more critically, I learned that conducting reliable and valid research is hard. It involves dedication and time, and good researchers question themselves constantly. I am proud to say that I conducted a study that found no statistical significance because the sample size is small. Finding robust significance between conditions with this sample size would have been a red flag given the implicit learning design of this study.