Hybrid Dialog State Tracker with ASR Features

TL;DR

This paper introduces a hybrid dialog state tracker that combines trainable SLU and differentiable rules, achieving state-of-the-art results on the DSTC2 dataset for belief tracking in restaurant dialog systems.

Contribution

A novel hybrid belief tracker integrating trainable SLU with differentiable rules for improved generalization and end-to-end training capabilities.

Findings

Achieved new state-of-the-art results on DSTC2 in three categories.

Demonstrated the effectiveness of combining neural networks with differentiable rules.

Enhanced generalization over purely machine-learning-based belief trackers.

Abstract

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.