# Slot Self-Attentive Dialogue State Tracking

**Authors:** Fanghua Ye, Jarana Manotumruksa, Qiang Zhang, Shenghui Li, Emine Yilmaz  
**arXiv:** [2101.09374](https://arxiv.org/abs/2101.09374)  
**Date:** 2021-01-22  
**Summary page:** https://www.summarizepaper.com/en/paper/2101.09374v1/

## Key points

- Dialogue state tracking is important in task-oriented dialogue systems
- The dialogue state tracker helps understand users' intentions during conversations
- Existing methods often predict slots independently and overlook slot correlations
- The paper proposes a slot self-attention mechanism to learn slot correlations automatically
- Slot-token attention is used to extract slot-specific features from the dialogue context
- Stacked slot self-attention captures correlations among different slots
- Experiments on MultiWOZ 2.0 and MultiWOZ 2.1 datasets show state-of-the-art performance
- Considering slot correlations improves accuracy in dialogue state tracking

## Summary

The paper titled "Slot Self-Attentive Dialogue State Tracking" addresses the importance of dialogue state tracking in task-oriented dialogue systems. The dialogue state tracker plays a crucial role in understanding users' intentions during conversations. Traditionally, this involves filling in pre-defined slots that are necessary to complete the task. However, existing methods often predict the value of each slot independently and overlook the correlations among slots. To address this limitation, the authors propose a slot self-attention mechanism that can automatically learn slot correlations. The proposed approach first utilizes slot-token attention to extract slot-specific features from the dialogue context. Then, a stacked slot self-attention is applied to these features to capture the correlations among different slots. The authors conduct comprehensive experiments on two multi-domain task-oriented dialogue datasets, namely MultiWOZ 2.0 and MultiWOZ 2.1. The experimental results demonstrate that their approach achieves state-of-the-art performance on both datasets, highlighting the necessity and effectiveness of considering slot correlations for accurate dialogue state tracking. In summary, this paper introduces a novel method for dialogue state tracking that improves upon existing approaches by incorporating slot correlations through a slot self-attention mechanism which shows promising results in accurately capturing users' intentions during task-oriented conversations.