Learning to Play Guess Who? and Inventing a Grounded Language as a Consequence

Acquiring your first language is an incredible feat and not easily. Learning to communicate using nothing but a few pictureless books, corpus, would likely be impossible even for humans. Nevertheless, this is the approach in most natural language processing today. As an, we propose the use of situated interactions between agents as a force for communication, and the framework of Deep Recurrent Q-Networks evolving a shared language grounded in the provided environment. We task agents with interactive image search in the form of the game Guess Who? . images from the game provide a non trivial environment for the agents to and a natural grounding for the concepts they decide to encode in their. Our experiments show that the agents learn not only to encode concepts in their words, i. e. grounding, but also that the agents to hold a multi-step dialogue remembering the state of the dialogue from to step.