Supervisor: Prof Massimo Poesio
Humanity has long wished for methods for telling whether somebody is lying have, particularly in criminal investigation, and extensive research has been carried out leading to tools such as the polygraph and interrogation methods thought to be effective at capturing deception, but with not entirely satisfactory results (Vrij, 2008). The need for such methods has however become even more acute with the explosion of social media and the increasing importance of online reviews of products. Almost everybody has ran across a fake review on a site such as Amazon, and/or is aware of claims about the effect of fake news on recent elections and referenda. In recent years, it has been shown that NLP methods can be extremely effective at identifying deception in language--e.g., at identifying deceptive witness statements, or fake reviews-- and a lot of research is now being carried out also on identifying fake news. The project could focus on either fake news or fake reviews. A particular focus of interest of our lab is the development of methods for creating genuine datasets for the study of deception. We are also interested in methods for combining evidence from language with multimodal evidence, e.g., to identify deception in video recordings.