Annotating Errors and Emotions in Human-Chatbot Interactions in Italian

摘要: This paper describes a novel annotation scheme specifically designed for customer-service context where written interactions take place between given user and the chatbot of an Italian telecommunication company. More specifically, aims to detect highlight two aspects: presence errors in conversation on both sides (i.e. customer chatbot) “emotional load” conversation. can be inferred from emotions some kind (especially negative ones) messages, possible empathic responses provided by agent. The dataset annotated according this is currently used develop prototype rule-based Natural Language Generation system aimed at improving experience overall.