The Communicative Face
One of the most fundamental human activities consists of communication through human language. The most important aspect of human language is face-to-face interaction, suggesting human language is a multimodal phenomenon. There is an enormous variation in the face's articulators and the potential signals they can produce. This research will investigate whether any regularities regarding those facial signals occur when comparing questions to responses. Data from dyadic conversations were used where participants talked freely for 60 minutes, in an attempt to take a more naturalistic approach compared to most existing literature that uses conversational data from highly controlled environments. Facial signals are recognized with the help of OpenFace, a tool for the automatic detection of facial signals from video data. General (co-)occurrence counts of facial signals, as well as sequences of facial signals, and their timing were analyzed while comparing questions to responses. Signi cant di erences between questions and responses were found both agreeing, as well contradicting existing literature. Therefore this research could provide more insight to what facial signals occur systematically during questions and responses and possibly help to addressee to predict the content or ending of the incoming turn. Furthermore, SPeeding Up the Detection of Non-iconic and Iconic Gestures (SPUDNIG): a toolkit for the automatic detection of hand movements and gestures in video data was presented. This toolkit was developed since there was no existing toolkit for the automatic detection of hand movements, in contrast to toolkits for the automatic detection of facial signals such as OpenFace. It was demonstrated that SPUDNIG accelerates the process of annotating hand gestures. Therefore SPUDNIG could be used in order to facilitate the timeconsuming and labour-intensive task of manually annotating hand gestures.
Faculteit der Sociale Wetenschappen