Deception Detection Within and Across Cultures

In this paper, we address the task of cross-cultural deception detection. Using crowdsourcing, we collect four deception datasets, two in English (one originating from United States and one from India), one from Romanian speakers, and one in Spanish obtained from speakers from Mexico, covering three predetermined topics. We also collect two additional datasets, one for English from United States and one for Romanian, where the topic is not pre-specified. We run comparative experiments to evaluate the accuracies of deception classifiers built for each culture, and also to analyze classification differences within and across cultures. Our results show that we can leverage cross-cultural information, either through translation or equivalent semantic categories, and build deception classifiers with a performance ranging between 60-70%.