The "Etiquette Quotient": Evaluating Social Skills in Conversational Avatars

This work developed a “believability metric” for assessing and predicting when a given level of etiquette behavior exhibited by a computerized avatar or non-player character (NPC) in a game or simulation would be perceived as “unbelievable” by a human player or observer.

In April of 2004, SIFT and Dr. Lewis Johnson of the University of Southern California (USC)’s CARTE labs, proposed and won Phase I of DARPA SBIR SB041-009 (Contract #W31P4Q-04-C-R221, DARPA Program Manager: Dr. Ralph Chatham, 703-696-7501) titled “The Etiquette Quotient: Evaluating Social Skills in Conversational Avatars.” This work developed a “believability metric” for assessing and predicting when a given level of etiquette behavior exhibited by a computerized avatar or non-player character (NPC) in a game or simulation would be perceived as “unbelievable” by a human player or observer. SIFT completed this work in 2004 with a computational formalization of the algorithm and a demonstration of believability predictions over a set of scenarios with varying amounts of redress. The algorithm tracked and predicted the variations we had built into the test set.