Affect corpus 2.0: an extension of a corpus for actor level emotion magnitude detection

Improvement in human computer interaction requires effective and rapid development of multimedia systems that can understand and interact with humans. These systems need resources to train and learn how to interpret human emotions. Currently, there is a relative small number of existing resources such as annotated corpora that can be used for affect and multimodal content detection. In this paper, an extension of an existing corpus is presented. The corpus includes new annotations for affect magnitude detection and anaphora resolution. The format of the collected data is presented, along with the annotation methodology, basic statistics, suggestions for possible uses, and future work. This corpus is an extension of the UIUC Affect corpus of children's stories. The corpus includes new automatic annotations using Natural Language Processing toolkits as well as new manual annotations for affect magnitude detection and anaphora resolution. Results of inter-annotator agreement analysis on a subset of the corpus are also presented.