Influence of underlying structure on human behavior in sequential categorization and object search

Human cognition can fundamentally detect, learn, and exploit underlying structure from repeated exposure to information. Everyday behavior relies on extracting temporal and spatial regularities from the environment. This dissertation investigates how humans learn and generalize underlying regularities across sequential decisions, and how such patterns in the information presented can influence search behavior when expectations are violated.

Across three chapters, this work examines (1) whether people learn and generalize temporal regularity across a sequence of decision-making tasks within and between different modalities of decision-making tasks, (2) whether learning and generalization of sequential tasks are possible in complex temporal regularity structures and whether the implementation of a formal exemplar based category learning model account for the behavior observed in sequential decision making task with temporal regularity, and (3) though based on underlying patterns this chapter makes a conceptual shift from temporal regularity to examining how prior experience of underlying pattern, with utility and effort trade-off, guides object search in everyday tasks.

Chapters 1 and Part I of Chapter 2 introduce a novel sequential decision-making task based on a real-world table-setting scenario. Rather than categorizing isolated stimuli into categorical outcomes based on feature inference, as in classical categorization tasks, participants completed sequences of categorization and estimation tasks within a single trial with the same stimulus, in which the outcome of one decision could predict the next. Across multiple studies, we manipulated temporal regularity (Type I, Type II, Type VI), motivated by the category structure proposed by Shepard, Hovland, and Jenkins (1961); temporal proximity (adjacent vs. non-adjacent) (Wilson et al., 2020); and learning modality (categorization vs. estimation). Results reveal that humans, as reported in research in other domains (Lazartigues, Mathy, & Lavigne, 2022; Shepard et al., 1961), easily learn simple rules such as Type I regularity, in which one categorical outcome predicts the other (category-category association), but fail to acquire more complex rules such as Type II. Temporal proximity also makes learning difficult when intervening tasks separate the predicted tasks. Generalization was observed for regularities in
category-category association structures, such as Type I and Type II, but was diminished for category-criterion association temporal structures.

Chapter 2, Part II, focused on evaluating whether classical exemplar-based categorization models can account for sequential decision learning under complex temporal regularity. The dissertation aimed to adapt the ALCOVE model (Kruschke, 1992) with a temporal decay mechanism (Simple Temporal Decay Mechanism: Brown, Neath, & Chater, 2007) and attention-based weighting. Simulations of the experimental setup and model fits showed that models incorporating a decay mechanism best captured the learning difficulty of temporally distant adjacent and non-adjacent sequential tasks, while maintaining the theoretical category structure and observed human behavior.

Chapter 3 examines how prior knowledge influences object search during everyday activity, for example, when a mug is missing from its usual location. The aim was to investigate how search behavior is affected, especially with respect to (1) spatial proximity: do people search closer to their current location or based on prior knowledge? (2) Does the presence of an alternative object with varying utility provide a suitable trade-off between selecting the alternative object or continuing search, and (3) does the effort of object search (virtual reality-based physical search vs. desktop environment) affect search behavior?. The chapter illustrates that, although the search environment influences overall search behavior, the critical finding is that participants tend to search in a location proximal to their current location when the search requires greater physical effort. Overall, the chapters provide an initial framework for investigating the learning of underlying patterns in an abstract temporal domain and in an everyday, task-oriented, naturalistic setting.