Christopher A. Miller, Peggy Wu, Harry B. Funk
A significant and prevalent aspect of human behavior (as opposed to attitudes and beliefs) which is affected by cultural differences is communication style. We have been developing computational models of an important culturallyvarying aspect of communication: ”politeness” and “etiquette” in social interactions and its role in establishing and managing power and familiarity relationships, urgency, indebtedness, etc. A valid computational model of such interactions would enable the creation of better simulations and games for language and culture training, as well as aid in the design of materials and machines to better serve members of a given culture. We have developed such a model based on a rich, universal theory of human-human politeness and “etiquette”. This model links observable and inferred aspects of social context (power and familiarity relationships, imposition, knowledge about character), which have culture- specific values, to produce expectations about politeness behaviors (also culturally defined). By using observations of politeness behaviors (or their lack), the same model permits inferences and updates about those attributes. We briefly describe the algorithm we have developed and describe results from two validation experiments involving first trained, and later untrained raters. We have used this model in other work to guide simulated game agents in interpreting etiquette directed at them and in generating politeness behaviors in response. While other methods of interactive behavior generation are available (e.g., behavior scripting) our modular, computational approach shows promise for reducing software development costs and/or increasing the breadth of an agent’s social interaction behavior through the creation of modular, cross-cultural etiquette libraries.
Subjects: 7.1 Multi-Agent Systems; 12.1 Reinforcement Learning
Submitted: Jun 20, 2008