On the Effects of Intra-episodic Language Signals in the Minigrid Environment

Master Thesis at University of Potsdam (WIP)

Research Questions

1. Are intra-episodic language signals (provided by a simple heuristic) beneficial for follower success in the environment? And in what situations especially?
2. Can we learn intra-episodic language production in the environment? And does such a learnt generator (neural model) perform better than a simple heuristic?
3. Can joint learning of adaptable (neural) models succeed simple heuristics? How do the resulting policies behave and why are they better or worse?