Probabilistic state machines: dialog management for inputs with uncertainty

Traditional models of input work on the assumption that inputs delivered to a system are fairly certain to have occurred as they are reported. However, a number of new input modalities, such as pen-based inputs, hand and body gesture inputs, and voice input, do not share this property. Inputs under these techniques are normally acquired by a process of recognition. As a result, each of these techniques makes mistakes and provides inputs which are approximate or uncertain. This paper considers some preliminry techniques for dialog management in the presence of this uncertainty. These techniques—including a new input model and a set of extended state machine abstractions—will explicitly model uncertainty and handle it as a normal and expected part of the input process.