Linking Strategies to Think Aloud in A Stochastic Learning Task
Abstract
Understanding human thoughts is a key goal of cognitive science. Behavioral observations alone limit insight into cognition. The think-aloud protocol, where participants verbalize thoughts, offers a direct probe into reasoning but is underutilized due to challenges in subjectivity and scalability. Advancements in natural language processing (NLP) enable computational analysis of think-aloud data, yet little work explores its role in strategy learning. We test whether think-aloud reports reveal strategy use in a stochastic learning task where participants verbalized their strategies. Our results show diverse strategy usage, with a preference for persistent choices. Think-aloud analysis suggests participants rely on distinct meta-strategies to guide learning. Clustering and predictive modeling reveal strong alignment between choices and verbalized strategies. These findings highlight think-aloud as a scalable tool with NLP techniques for studying high-level cognition, shedding light on a promising paradigm for cognitive sciences.