Using Linguistic Metrics and Task Characteristics to Investigate and Manage Group Deception

Abstract

Interactive deception among groups of humans regularly occurs. Barriers to studying this type of interaction include practical constraints as well as experimental design challenges that balance realism of deception with ground truth knowledge that deception is occurring. As our ability to capture and process more aspects of this type of interaction grows with the use of computers, so do opportunities for understanding and managing deception. This dissertation begins with a review of linguistic characteristics of group speech that have been found to have some relationship with deception. Then, this dissertation characterizes a deceptive group interaction, using a lens model to consider individual and interactive components that relate to cue generation, deception judgements, and group performance. This dissertation then describes an experimental method and technology tool created to facilitate the study of this type of interaction. The dataset used for this dissertation was collected using this tool and involved participants completing a hidden identity game and providing periodic ratings of other participants with whom they interacted. Lastly, this dissertation considers performance of both automated methods and performance of actual individuals in the hidden identity game and explores reasons why certain cues and individuals performed better or worse than expected using simulation and exploratory analyses.