Annotating and analyzing emotions in a corpus of first encounters

The paper is about the annotation and analysis of emotions and attitudes in a Danish corpus of first encounters. Emotions are classified using an open-ended list of values following the MUMIN annotation scheme and bipolar values in three emotional dimensions: Pleasure (P), Arousal (A) and Dominance (D), PAD) as proposed by Kipp and Martin [1]. The strategy of combining a coarse grained dimensional annotation with a fine grained list of emotions helped the coders to reach a common understanding of the semantics of the emotion labels. It also contributed to a significant improvement of the inter-coder agreement. A first analysis of the emotions annotated in six first encounters indicates that there is a tight relation between type of emotions and the social activity. Furthermore, some emotion types are related to specific communicative functions, such as feedback and turn management in these data. These findings can be useful for modeling plausible emotional devices.