Enhancing Empathic Accuracy: Penalized Functional Alignment Method to Correct Misalignment in Emotional Perception

Abstract

Empathic accuracy (EA) is the ability of one person to accurately understand thoughts and feelings of another person, which is crucial for social and psychological interactions. Traditionally, EA is measured by comparing perceivers` real-time ratings of emotional states with the target`s self--evaluation. However, these analyses often ignore or simplify misalignments between ratings (such as assuming a fixed delay), leading to biased EA measures. We introduce a novel alignment method that accommodates diverse misalignment patterns, using the square--oot velocity representation to decompose ratings into amplitude and phase components. Additionally, we incorporate a regularization term to prevent excessive alignment by constraining temporal shifts within plausible human perception bounds. The overall alignment method is implemented effectively through a constrained dynamic programming algorithm. We demonstrate the superior performance of our method through simulations and real-world applications to video and music datasets.