The component support understanding of the emotional state of a speaker from â€œwhatâ€ and â€œhowâ€ a speaker is linguistically expressing itself. It is based on an annotated corpus consisting of sentences that present typical expressions used in a conversation. Natural language is generated without disregarding the affective aspects of a conversation, considering a reference corpus which has to be annotated with categories and topics related to what sentences are about and which is the emotional state that they denote.
Availability of the component: Reference contact is Michael Wissner of University of Augsburg
Notes: EmoNLG will be made available upon request
The component is used in the Interactive Storytelling CALLAS Proof-of-Concept application to produce verbal utterances that convey an ECAâ€™s emotional state.