Neural Dialogue State Tracking with Temporally Expressive Networks

TL;DR

This paper introduces Temporally Expressive Networks (TEN), a novel model combining recurrent networks and probabilistic graphical models to improve dialogue state tracking by explicitly modeling temporal dependencies across dialogue turns.

Contribution

The paper proposes TEN, a new model that jointly captures temporal feature and state dependencies in dialogue state tracking, enhancing accuracy.

Findings

TEN improves turn-level state prediction accuracy.

TEN enhances state aggregation performance.

Model effectively captures temporal dependencies.

Abstract

Dialogue state tracking (DST) is an important part of a spoken dialogue system. Existing DST models either ignore temporal feature dependencies across dialogue turns or fail to explicitly model temporal state dependencies in a dialogue. In this work, we propose Temporally Expressive Networks (TEN) to jointly model the two types of temporal dependencies in DST. The TEN model utilizes the power of recurrent networks and probabilistic graphical models. Evaluating on standard datasets, TEN is demonstrated to be effective in improving the accuracy of turn-level-state prediction and the state aggregation.