A categorical annotation scheme for emotion in the linguistic content of dialogue

Abstract

If we wish to implement dialogue systems which express emotion, dialogue corpora annotated for emotion would be a valuable resource. In order to develop such corpora we require a reliable annotation scheme. Here we describe an annotation scheme for emotion in dialogue using categorical labels to complement previous work using dimensional scales. The most difficult challenge in developing such a scheme is selecting the categories of emotions that will yield the most expressive yet reliable scheme. We apply a novel approach, using a genetic algorithm to identify the appropriate categories.