Abstract: In the dialogue part of our system, we have tried to increase the user's possibilities to criticize the machine's results require explanations of them. The system must then provide a clear justification: either by furnishing the chain of reasoning or by asking a "good question" when it failed. The system must also be able to engage in a real dialogue, with more than one question/one answer. To do that, the system must build and use several kinds of representation: the reasoning, the topics and a model of the user, which is used to tailor the system's responses.
0 Replies
Loading