Abstract: | The most flexible way to handle beliefs in a dialogue system is by using a logical representation, so that the system can consider beliefs nested to arbitrary levels (i.e. representing what A believes B believes, A believes B believes A believes, and so on) and involving arbitrary numbers of agents. Although this flexibility is desirable where the system designers cannot be sure ahead of what depth of reasoning will be required, it has one major drawback: running the associated theorem prover after every utterance in the dialogue is slower than it needs to be, since moves in a dialogue introduce predictable changes to the belief system. ..... This paper describes more fully this basic representation and different ways of handling uncertainty within it. it also shows how this simple mechanism can be connected to a more complete belief representation, allowing more complex reasoning when the need arises. ....
|