From Language to Goals: Inverse Reinforcement Learning for Vision-Based Instruction Following