Abstract

Partially Observable Markov Decision Processes (POMDPs) have been empirically demonstrated as a sound model for spoken dialogue management design. This talk will show that the model is also potentially appropriate for designing affective dialogue problems.

In the first part of my talk, I will present an affective dialogue model using factored POMD Ps and a performance evaluation for this model. Then, I will address several practical issues related to the development of POMDP -based dialogue managers for real-world dialogue systems. In accordance with researchers in the field, one of the major problems remains tractability.

To solve this problem, in the second part, I will present a hybrid DDN -POMDP method to developing a tractable affective dialogue model for probabilistic frame-based dialogue systems. The proposed method has two new features: (1) being able to deal with a large number of slots & slot values in real-time and (2) being able to take into account some aspects of the user’s affective state in deriving the adaptive dialogue strategies.

We conducted various experiments to evaluate our method and to compare it with approximate POMDP techniques and handcrafted policies. The results of these experiments will be discussed in detail.