Structured Discriminative Model For Dialog State Tracking

Many dialog state tracking algorithms have been limited to generative modeling due to the influence of the Partially Observable Markov Decision Process framework. Recent analyses, however, raised fundamental questions on the effectiveness of the generative formulation. In this paper, we present a structured discriminative model for dialog state tracking as an alternative. Unlike generative models, the proposed method affords the incorporation of features without having to consider dependencies between observations. It also provides a flexible mechanism for imposing
relational constraints. To verify the effectiveness of the proposed method, we applied it to the Let’s Go domain (Raux et al., 2005). The results show that the proposed model is superior to the baseline and generative model-based systems in accuracy, discrimination, and robustness to mismatches between training and test datasets.