Collaborative Research: Interactive Deception and its Detection through Multi-modal Analysis of Interviewer-Interviewee Dynamics

欺骗是社会生活中普遍存在的特征，但往往不被发现，因为骗子利用了人际交往过程的特点。他们在互动过程中调整自己的语言和非语言行为，以逃避检测。为了揭示使成功欺骗或检测成为可能的通信过程的复杂性和动态，将由来自亚利桑那大学、芝加哥大学、密歇根州立大学和罗格斯大学的通信、语言学、心理学、计算机科学和管理信息系统研究人员组成的多学科团队进行合作研究。来自科隆大学、伦敦帝国理工学院和冰岛大学的国际专家将加入他们的行列。该项目将建立一个人际欺骗的理论模型，将重点从稳定的个人行为转移到动态的互动模式。它将建立5个试验台，通过测量视频录制的人际交流实验中的语言和非语言特征。这五个实验分别是作弊实验、模拟盗窃实验、欺骗访谈、小组协作任务和叙述任务。测量将包括广泛的自动化和人工注释的视觉、声音和语言特征测量。作为注释工作的一部分，该团队将改进和验证用于自动测量非语言和语言特征的软件。利用时间序列、基于贝叶斯的主题分析和人工智能数据挖掘技术，对测量的传播行为进行理论驱动的假设检验和探索性检验，以识别动态适应模式。该项目将通过组建最大的国际和多学科欺骗专家研究团队，整合多学科的知识和方法，完善人类交流的计算机辅助分析，在自动化欺骗检测方面取得进展，并通过实验室交流培养新的研究人员，来推进欺骗研究的科学基础设施。社会将受益于更有效的欺骗图像，可以指导公共、商业、政府和安全环境中的培训和检测工作。

Deception is a pervasive feature of social life yet often goes undetected because deceivers capitalize on features of the interpersonal communication process. They adjust their verbal and nonverbal behavior over the course of an interaction in ways that evade detection. To uncover the complexities and dynamics of the communication processes that make successful deception or detection possible, collaborative research will be conducted by a multidisciplinary team of communication, linguistics, psychology, computer science, and management information systems researchers from the University of Arizona, University of Chicago, Michigan State University, and Rutgers University. They will be joined by international experts from the University of Cologne, Imperial College London, and University of Iceland. The project will develop a theoretical model of interpersonal deception that shifts emphasis from stable individual behaviors to dynamic interaction patterns. It will create five test beds by measuring the verbal and nonverbal features from video-recorded interpersonal communication experiments. The five experiments are a cheating experiment, a mock theft experiment, deceptive interviews, a group collaboration task, and a narration task. Measurement will consist of extensive automated and human-annotated measurement of visual, vocal, and verbal features. As part of the annotation work, the team will refine and validate software for automated measurement of nonverbal and verbal features. Theory-driven hypothesis tests and exploratory tests will be conducted on the measured communication behaviors to identify dynamic adaptation patterns using time-series, Bayesian-based Theme analyses, and artificial intelligence data mining techniques. The project will advance the scientific infrastructure for studying deception by forming the largest international and multidisciplinary research team of deception experts of its kind, integrating the knowledge and methods of multiple disciplines, refining computer-aided analysis of human communication, making progress in automating deception detection, and educating new investigators through laboratory exchanges. Society will benefit from a more valid picture of deception that can guide training and detection efforts in public, business, government, and security settings.