Speaker or Listener? The Role of a Dialog Agent

Findings of the Association for Computational Linguistics: EMNLP 2020, pages 4861–4869 November 16 - 20, 2020. c⃝2020 Association for Computational Linguistics 4861 Speaker or Listener? The Role of a Dialogue Agent Yafei Liu♦♥†, Hongjin Qian△†, Hengpeng Xu♥, Jinmao Wei♥‡ ♦ Jarvis Lab Tencent △Renmin University of China ♥Nankai University davenliu@tencent.com ian@ruc.edu.cn xuhengpeng@mail.nankai.edu.cn weijm@nankai.edu.cn Abstract For decades, chitchat bots are designed as a listener to passively answer what people ask. This passive and relatively simple dialogue mechanism gains less attention from humans and consumes the interests of human beings rapidly. Therefore some recent researches attempt to endow the bots with proactivity through external knowledge to transform the role from a listener to a speaker with a hypoth- esis that the speaker expresses more just like a knowledge disseminator. However, along with the proactive manner introduced into a dialogue agent, an issue arises that, with too many knowledge facts to express, the agent starts to talks endlessly, and even completely ignores what the other expresses in dialogue sometimes, which greatly harms the interest of the other chatter to continue the conversa- tion. To the end, we propose a novel model named Initiative-Imitate to interact with adap- tive initiative throughout a dialogue. It forces the agent to express in parallel with the ap- propriate role during the whole conversation. The corresponding experiments show the pro- posed Initiative-Imitate obtains competitive re- sults both on the automatic and manual met- rics. And the fluency and engagement of the chatbot have also been improved significantly. Besides, the case study indicates the Initiative- Imitate can constantly transfer to appropriate role timely and response more properly during the whole continuous conversation. 1 Introduction Automatic human-machine conversation lies in the core of artificial intelligence (AI) and natural lan- guage processing (NLP). Many researchers have developed lots of dialogue systems, such as rule- based (Weizenbaum et al., 1966; Webb, 2000), †Equal contribution ‡Corresponding author Leader Follower human2 human1 1 2 3 4 5 7 6 8 9 0 turns Topic Role Figure 1: The role transformation of the two humans in a multi-turn human-human conversation started by human1. The dashed line means one’s silent period, the solid line for the speaking period. retrieval-based (Wu et al., 2017) and generation- based like neural networks (Vinyals and Le, 2015; Vougiouklis et al., 2016; Yavuz et al., 2019). For decades, machine agent is generally pre- sumed to be a passive role with the ability to answer what humans ask. However, this passive and rel- atively simple response mechanism consumes the interest of the other dialogue participant rapidly (Li et al., 2016b). Actually in a continuous human- human conversation, both participants need to be a speaker to lead current topics. This phenomenon is statistically summarized from the analysis on a real human-human dataset named DailyDialog (Li et al., 2017) as depicted in Figure 1. Two humans take the topic leading role like a speaker to introduce something new in turns. Thus in human-machine conversation, the dialogue agent side needs to act as a speaker …

Content truncated to 500 words