Mathematical modeling of Spoken Human - Machine dialogues including erroneous confirmations

摘要

In this paper, we present a method that enables the designer to investigate a spoken dialogue system's performance by employing diagnostic evaluation during the initial phases of the system's development. Results from glass box assessment (e.g.. recognition success rate), combined with the dialogue strategy in the proposed mathematical model, can be used to predict the system's performance. The model incorporates erroneous confirmations by the user, which affect the overall performance.