- Abstract: The task of visually grounded dialog involves learning goal-oriented cooperative dialog between autonomous agents who exchange information about a scene through several rounds of questions and answers. We posit that requiring agents to adhere to rules of human language while also maximizing information exchange is an ill-posed problem, and observe that humans do not stray from a common language, because they are social creatures and have to communicate with many people everyday, and it is far easier to stick to a common language even at the cost of some efficiency loss. Using this as inspiration, we propose and evaluate a multi-agent dialog framework where each agent interacts with, and learns from, multiple agents, and show that this results in more relevant and coherent dialog (as judged by human evaluators) without sacrificing task performance (as judged by quantitative metrics).
- Keywords: dialog, multi-agent reinforcement learning, emergent communication
- TL;DR: Social agents learn to talk to each other in natural language towards a goal