IBM’s Belief Tracker: Results On Dialog State Tracking Challenge Datasets

Accurate dialog state tracking is crucial for the design of an efficient spoken dialog system. Until recently, quantitative comparison of different state tracking methods was difficult. However the 2013 Dialog State Tracking Challenge (DSTC) introduced a common dataset and metrics that allow to evaluate the performance of trackers on a standardized task. In this paper we present our belief tracker based on the Hidden Information State (HIS) model with an adjusted user model component. Further, we report the results of our tracker on test3 dataset from DSTC. Our tracker is competitive with trackers submitted to DSTC, even without training it achieves the best results in L2 metrics and it performs between second and third place in accuracy. After adjusting the tracker using the provided data it outperformed the other submissions also in accuracy and yet improved in L2. Additionally we present preliminary results on another two datasets, test1 and test2, used in the DSTC. Strong performance in L2 metric means that our tracker produces well calibrated hypotheses probabilities.