Abstract
Recent advances in machine learning have allowed for the use of natural language responses to predict outcomes of interest to memory researchers such as the confidence with which recognition decisions are made. The present experiments were designed to leverage this novel methodological approach by soliciting free-response justifications of judgments of learning (JOLs) whereby people not only assess the probability with which they will later recognize individual items but also (for some items) justify the reasoning behind their judgment. Across all experiments and conditions, regression models trained on justification language showed above-chance prediction of subsequent memory success and outperformed models trained on numeric JOLs alone. Conditions that improved predictive accuracy of scale JOLs also improved the accuracy of language models. Further, the predictors (word choices) retained by the regularized models provide insight into the mechanisms underlying differences in metamemory performance.
Committee Chair
Kathleen McDermott
Committee Members
Ian Dobbins, David Balota, Mark McDaniel, Justin Kantner,
Degree
Doctor of Philosophy (PhD)
Author's Department
Psychology
Document Type
Dissertation
Date of Award
Summer 8-15-2021
Language
English (en)
DOI
https://doi.org/10.7936/gm9p-9c27
Author's ORCID
http://orcid.org/0000-0002-5781-0042
Recommended Citation
Anderson, Nathan Lloyd, "The Use of Introspective Reports to Predict Subsequent Memory: Implementing Machine Learning for Judgment-of-Learning Paradigms" (2021). Arts & Sciences Theses and Dissertations. 2479.
The definitive version is available at https://doi.org/10.7936/gm9p-9c27