Adding Chit-Chat to Enhance Task-Oriented Dialogues

Abstract

Existing dialogue corpora and models are typically designed under two disjoint motives: while task-oriented systems focus on achieving functional goals (e.g., booking hotels), open-domain chatbots aim at making socially engaging conversations. In this work, we propose to integrate both types of systems by Adding Chit-Chat to ENhance Task-ORiented dialogues (ACCENTOR), with the goal of making virtual assistant conversations more engaging and interactive. Specifically, we propose a Human ↔ AI collaborative data collection approach for generating diverse chitchat responses to augment task-oriented dialogues with minimal annotation effort. We then present our new chit-chat-based annotations to 23.8K dialogues from two popular task-oriented datasets (Schema-Guided Dialogue and MultiWOZ 2.1) and demonstrate their advantage over the originals via human evaluation. Lastly, we propose three new models for adding chit-chat to task-oriented dialogues, explicitly trained to predict user goals and to generate contextually relevant chit-chat responses. Automatic and human evaluations show that, compared with the state-of-the-art task-oriented baseline, our models can codeswitch between task and chit-chat to be more engaging, interesting, knowledgeable, and humanlike, while maintaining competitive task performance.

SalesBot: Transitioning from Chit-Chat to Task-Oriented Dialogues

“…Considering that both skills are essential for a human-like dialogue system, the recent work (Sun et al, 2021) merged those two capabilities by inserting chit-chat sentences into the existing taskoriented dialogue data. The idea is to allow the agent gains more social, personalized communication skills when focusing on task-oriented dialogue generation.…”

“…In more details, all task-oriented dialogues in the SGD dataset are grouped by intents, and one TOD dialogue is sampled based on the detected task-oriented intent to append to the transition turn and form a new dialogue containing both chit-chat and TOD. Note that the delexicalized version of SGD (Sun et al, 2021) is used to avoid severe inconsistency between open-domain and task-oriented parts. Task-Oriented Simulation Different from opendomain social chatting, the roles in task-oriented dialogues are important.…”

Where to Go for the Holidays: Towards Mixed-Type Dialogs for Clarification of User Goals

“…Recently, there are multiple efforts on developing dialog systems that can multi-task on multiple types of dialogs (Kim et al, 2020;Smith et al, 2020;Mosig et al, 2020;Saha et al, 2018;Sun et al, 2021;Young et al, 2021). Specifically, Kim et al (Kim et al, 2020) propose to handle out-of-API requests, by accessing unstructured domain knowledge in task-oriented dialogs.…”

“…Specifically, Kim et al (Kim et al, 2020) propose to handle out-of-API requests, by accessing unstructured domain knowledge in task-oriented dialogs. Sun et al (Sun et al, 2021) and Yong et al (Young et al, 2021) propose to fuse task-oriented and opendomain dialogs in conversational agents, in order to generate more engaging and interactive dialogs.…”

Style Control for Schema-Guided Natural Language Generation

“…Instead, we focus on the task-oriented dialogue framework where specific values must be communicated, and conduct a rigorous evaluation of different methods and their efficacy on different forms of style generation. Other recent work has explored adding additional information such as chit-chat data to task-oriented dialogue (Sun et al, 2021; and could potentially provide new opportunities for stylistic control.…”