hide
Free keywords:
-
Abstract:
We present a new on-line annotation system that allows participants to perform manual sentiment analysis of coherent texts for emotions, as well as indicate the intensity of the emotion and the emphasis in each phrase. We have developed the following set of emotion categories: amusement, anger, contempt, despair, disgust, excitement, fear, hope, joy, neutral, pride, relief, sadness, shame, surprise. This set greatly expands the boundaries of the often used basic emotion categories and is balanced for positive and negative emotions. Using this new annotation tool and its predecessor version, we have collected two corpora of fairy tale texts manually annotated for emotions on the utterance level. One corpus encompasses 72 texts in German, each annotated by two participants. The other corpus is a work in progress and contains three fairytale texts, each annotated by seven participants. The inter-annotator agreement in both corpora is “fair”. Although annotation conflict resolution strategies can be developed for merging several annotations into one, we suggest that for manual SA, the researchers should aim at recruiting more annotators and use the consensus method for retrieving an annotation based on the opinion of the majority.