Designing a platform and algorithm for event-cognitive unit prediction experiments

Abstract

Event cognition is a core cognitive process that enables humans to segment a continuous stream of experiences into discrete, coherent units (i.e., events) for perception, memory, prediction, and decision-making at a high level. Despite its importance, the role of event cognition in structuring prediction processes has not been extensively explored. In particular, a framework for measuring predictions within the context of event cognition remains underdeveloped. This study addresses this gap by leveraging computational concepts from machine learning to define an event as a segment consisting of a sub-goal state and its corresponding trajectory during prediction. Based on this event unit, the study proposes two key contributions: (1) a human experiment platform designed to collect event unit prediction data and (2) a deep learning-based computational model inspired by event cognition principles. Results from the human experiments reveal that human event unit predictions reflect key aspects of event cognition in terms of (1) segmentation, (2) fine-grained updates of trajectories, and (3) updates of sub-goals. The deep learning model shows similar behavior, segmenting its predictions at boundary states and updating intermediate states between boundary states. Overall, this study offers a pioneering framework for understanding the role of event cognition in predictive processes, offering a benchmark for interactive research between human cognition and machine learning.