A Persian spoken dialogue system using POMDPs

Abstract

This paper represents a statistically framework for a Persian spoken dialogue system. The framework is based on the Partially Observable Markov Decision Process (POMDP). A Bayesian network is used to represent the states of the POMDP model. It is shown that Bayesian approaches can improve the spoken dialogue system performance by handling uncertainties. Also Natural Actor Critic (NAC) algorithm is used for learning in spoken dialogue system and finally a framework for collecting training data is proposed. We compare the system with a handcrafted spoken dialogue system to show the efficiency of the proposed framework.