Deliberation Enhancing Bot

Recent years have seen a proliferation of dialogue systems both in the industry and in academia. Advances in conversational systems lead to consumer-grade technologies such as Apple's Siri and Amazon's Alexa. While dialogue systems are a hot topic, the majority of the systems focus on interactions with a single individual, handling rather straightforward tasks.
This project proposes a novel application of dialogue systems. First, we are interested in applying dialogue agents to a small-group conversation, rather than on a single individual. Secondly, instead of solving a simple task, we suggest that a dialogue agent can support human deliberation, ultimately helping people perform better at the task at hand.
This project can be roughly split into 3 key objectives - collecting data, training single task dialogue agents, and evaluating cross-task dialogue agents.
Initially, we will assess how people engage in task-oriented group conversations, evaluating the specific cues that lead to better collaboration and task performance. As an added benefit to this step of the project, we will gather a novel corpora of task-oriented group conversations.
The lessons learned from these conversations, as well as the dataset collected, will be used to test an artificial dialogue system that will participate in a group conversation, improving the performance of the task. Once we develop the system for a single task, we are interested in applying the same premise to a variety of task setups and conversational domains. The goal is to explore to what extent such dialogue agents can be applied to different tasks with little (or ideally none) additional training and resources.