I. Introduction
When science fiction writers imagined a future for smarter computer systems, they often envisioned all-knowing, immensely capable, and purposeful machines. In February 2011, IBM's Watson computer won the Jeopardy! contest, displaying how close machines are to being ‘all- knowing’, thanks to their ability to store and sift through enormous amounts of factual information. While we may philosophically argue that ‘knowing’ is not applicable here in the strict sense, we cannot deny that Watson surpasses humans in its knowledge of trivia. However, when it comes to social intelligence, computers have a lot of room for improvement. We all use computers, and we know that they are never responsive to our emotions, moods, or to any kind of social context. Even an ‘all-knowing’ Watson computer would be a very poor dinner companion, as it lacks the skills to decide when to speak, and what to say. The newly emerging field of social signal processing aims at providing computers with the means of analysing and adequately representing human social signals, which will allow them to adapt and properly function in social settings.