Eliza meets the wizard-of-oz: evaluating social acceptability

What authoring possibilities arise by blending machine and human control of live embodied character experiences? This paper explores two different "behind-the-scenes" roles for human operators during a three-month gallery installation of an embodied character experience. In the Transcription role, human operators type players' spoken utterances; then, algorithms interpret the player's intention, choose from pre-authored dialogue based on local and global narrative contexts, and procedurally animate two embodied characters. In the Discourse role, human operators select from semantic categories to interpret player intention; algorithms use this "discourse act" to automate character dialogue and animation. We compare these two methods of blending control using game logs and interviews, and document how the amateur operators initially resisted having to learn the Discourse version, but eventually preferred having the authorial control it afforded. This paper also outlines a design space for blending machine and human control in live character experiences.