Hybrid Dialog State Tracker with ASR Features
This work addresses dialog state tracking for spoken dialog systems, offering incremental improvements in a specific domain.
The paper tackles the problem of dialog state tracking for slot-filling systems by introducing a hybrid tracker with trainable SLU and differentiable rules, achieving new state-of-the-art results in three out of four categories on the DSTC2 dataset.
This paper presents a hybrid dialog state tracker enhanced by trainable Spoken Language Understanding (SLU) for slot-filling dialog systems. Our architecture is inspired by previously proposed neural-network-based belief-tracking systems. In addition, we extended some parts of our modular architecture with differentiable rules to allow end-to-end training. We hypothesize that these rules allow our tracker to generalize better than pure machine-learning based systems. For evaluation, we used the Dialog State Tracking Challenge (DSTC) 2 dataset - a popular belief tracking testbed with dialogs from restaurant information system. To our knowledge, our hybrid tracker sets a new state-of-the-art result in three out of four categories within the DSTC2.