Put Chatbot into Its Interlocutor’s Shoes: New Framework to Learn Chatbot Responding with Intention

Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 1559–1569 June 6–11, 2021. ©2021 Association for Computational Linguistics 1559 Put Chatbot into Its Interlocutor’s Shoes: New Framework to Learn Chatbot Responding with Intention Hsuan Su1*, Jiun-Hao Jhan2*, Fan-yun Sun2†, Saurav Sahay23 and Hung-yi Lee1* *National Taiwan University, Taipei, Taiwan †Stanford University, Stanford, CA, USA 3Intel Labs, Santa Clara, CA, USA {b04203058, r06942141, hungyilee}@ntu.edu.tw fanyun@stanford.edu saurav.sahay@intel.com Abstract Most chatbot literature that focuses on improv- ing the fluency and coherence of a chatbot, is dedicated to making chatbots more human- like. However, very little work delves into what really separates humans from chatbots – humans intrinsically understand the effect their responses have on the interlocutor and of- ten respond with an intention such as propos- ing an optimistic view to make the interlocutor feel better. This paper proposes an innovative framework to train chatbots to possess human- like intentions. Our framework includes a guiding chatbot and an interlocutor model that plays the role of humans. The guiding chat- bot is assigned an intention and learns to in- duce the interlocutor to reply with responses matching the intention, for example, long re- sponses, joyful responses, responses with spe- cific words, etc. We examined our framework using three experimental setups and evaluated the guiding chatbot with four different metrics to demonstrate flexibility and performance ad- vantages. Additionally, we performed trials with human interlocutors to substantiate the guiding chatbot’s effectiveness in influencing the responses of humans to a certain extent. Code will be made available to the public. 1 Introduction Humans have evolved to become sensitive to their social interactions. The more they interact, the more they generally learn what to say and what not to say to light up people’s mood or to avoid upsetting others. In this paper, we aimed to train a chatbot to emulate these human-like qualities by making it learn from interactive conversation. A chatbot that understands the effect its utterances have on the interlocutor could be a significant step towards achieving human-level chatbots. A chatbot that understands the effect of its ut- terances on the interlocutor is also critical in real- world applications. For instance, as shown in Fig- ure 1, given a context from the interlocutor, both Figure 1: An example of the dialogue to show that how the chatbot interacts with the interlocutor, and how our chatbot affects the interlocutor’s response when as- signed the intention, making people respond joyful. responses "I did. They were really nice and fun and smart people." and "I did. I was so bummed out. I was so lonely." were relevant and reasonable responses, and were equally suitable for a typical chatbot. However, we could give an intention to the proposed chatbot (guiding chatbot), such as making the interlocutor feel joyful. In this way, the chatbot would respond in a positive way to induce joy in the interlocutor. Much literature combine Reinforcement Learn- ing(RL) (Kaelbling et al., 1996) with transformer- based …

Content truncated to 500 words