Abstract

In many applications, spoken dialogue is a compelling method for interacting with computers. With the popularity of mobile devices, voice interfaces are becoming increasingly important, but the technology for building these interfaces is often very poor. This talk will discuss how statistical methods can aid in the decision making processes of these spoken dialogue systems. In particular, we will discuss how Expectation Propagation (EP) can be used to build models of user behaviour in spoken dialogues and how reinforcement learning can be used to optimise the decision making. EP provides an efficient way to train the parameters and update the beliefs of a spoken dialogue systems based on the partially observable Markov decision process. These parameters can even be learned using noisy observations, and do not require any annotations besides semantic representations of the speech recognition output of a dialogue. The resulting systems are shown to be more robust to errors than standard approaches, largely because the models are able to handle the uncertainty in the dialogue in a principled way.