An expectation-driven approach to Q-A processing

The authors explore three aspects of an expectation-driven approach to understanding a human´s response to a system´s ad hoc question: (i) where expectations come from; (ii) how expectations can be expressed; and (iii) how they can be processed to understand subsequent user input. They describe various sources from which expectation can be derived `on-the-fly´: intention, world knowledge, conversational moves, previous context, and specific wording. They present a frame-based expectation representation scheme, and they identify eight expectation resolution modes: fulfilled, mismatched, tested, challenged, requisitioned, denied, annulled,and overruled. They present an ATN-based method for processing expectations and illustrate the method with an example