Williams, Sandra; Power, Richard and Piwek, Paul
Simulating emotional reactions in medical dramas.
In: Proceedings of the Symposium on Affective Language in Human and Machine, Volume 2, The Society for the Study of Artificial Intelligence and the Simulation of Behaviour (AISB) 2008 Convention: Communication, Interaction and Social Intelligence, 1-2 April 2008, Aberdeen, UK.
Full text available as:
Presenting information on emotionally charged topics is a delicate task: if bare facts alone are conveyed, there is a risk of boring the audience, or coming across as cold and unfeeling; on the other hand, emotional presentation can be appropriate when carefully handled, but when overdone or mishandled risks being perceived as patronising or in poor taste. When Natural Language Generation (NLG) systems present emotionally charged information linguistically, by generating scripts for embodied agents, emotional/affective aspects cannot be ignored. It is important to ensure that viewers consider the presentation appropriate and sympathetic.
We are investigating the role of affect in communicating medical information in the context of an NLG system that generates short medical dramas enacted by embodied agents. The dramas have both an informational and an educational purpose in that they help patients review their medical histories whilst receiving explanations of less familiar medical terms and demonstrations of their usage. The dramas are also personalised since they are generated from the patients' own medical records. We view generation of natural/appropriate emotional language as a way to engage and maintain the viewers' attention. For our medical setting, we hypothesize that viewers will consider dialogues more natural when they have an enthusiastic and sympathetic emotional tone. Our second hypothesis proposes that such dialogues are also better for engaging the viewers' attention.
As well as describing our NLG system for generating natural emotional language in medical dialogue, we present a pilot study with which we investigate our two hypotheses. Our results were not quite as unequivocal as we had hoped. Firstly, our participants did notice whether a character sympathised with the patient and was enthusiastic. This did not, however, lead them to judge such a character as behaving more naturally or the dialogue as being more engaging. However, when pooling data from our two conditions, dialogues with versus dialogues without emotionally appropriate language use, we discovered, somewhat surprisingly, that participants did consider a dialogue more engaging if they believed that the characters showed sympathy towards the patient, were not cold and unfeeling, and were natural (true for the female agent only).
Actions (login may be required)