# When to Talk: Chatbot Controls the Timing of Talking during Multi-turn Open-domain Dialogue Generation

**Authors:** Tian Lan, Xianling Mao, Heyan Huang, Wei Wei  
**arXiv:** [1912.09879](https://arxiv.org/abs/1912.09879)  
**Date:** 2019-12-20  
**Summary page:** https://www.summarizepaper.com/en/paper/1912.09879v1/

## Key points

- Existing dialogue systems in multi-turn open-domain conversations are considered boring
- Most existing systems only respond when the user's utterance is accepted, without considering context
- The authors propose a dialogue system that autonomously controls the timing of talking based on conversation context
- A decision module is incorporated into existing dialogue models to control response timing
- Graph neural networks are used to effectively model conversation context using a natural graph structure
- The proposed model improves the quality of dialogue generation by controlling the timing of talking
- Extensive experiments on two benchmarks demonstrate the effectiveness of the proposed model
- The methods presented in this paper enhance the accuracy of timing control
- The authors have made their model's codes publicly available

## Summary

The existing dialogue systems in multi-turn open-domain conversations have made significant progress but are still considered boring. Most of these systems only respond when the user's utterance is accepted, without considering the context to decide whether to continue the conversation. To address this limitation, the authors propose a dialogue system that autonomously controls the timing of talking based on the conversation context, enabling more natural and engaging interactions with humans. The paper focuses on incorporating a decision module into existing dialogue models to control the timing of responses. Additionally, effectively modeling the conversation context is crucial for accurate timing control. To achieve this, the authors adopt graph neural networks, which can process contextual information using a natural graph structure. The proposed model is evaluated through extensive experiments on two benchmarks, demonstrating that controlling the timing of talking significantly improves the quality of dialogue generation. The methods presented in this paper also enhance the accuracy of timing control. Furthermore, the authors have made their model's codes publicly available. In summary, this paper introduces a dialogue system that autonomously controls when to respond during multi-turn open-domain conversations. By incorporating a decision module and utilizing graph neural networks for context processing, the proposed model improves dialogue generation quality and enhances timing control accuracy.