Landmark-Based Location Belief Tracking in a Spoken Dialog System

摘要: Many modern spoken dialog systems use probabilistic graphical models to update their belief over the concepts under discussion, increasing robustness in face of noisy input. However, such are ill-suited reasoning about spatial relationships between entities. In particular, a car navigation system that infers users' intended destination using nearby landmarks as descriptions must be able distance measures factor inference. this paper, we describe tracking for location identification task combines semantic tracker categorical based on DPOT framework (Raux and Ma, 2011) with kernel density estimator incorporates landmark evidence from multiple turns hypotheses, into posterior probability candidate locations. We evaluate our approach corpus setting dialogs show it significantly outperforms deterministic baseline.