Text this: Modular Multi-Objective Deep Reinforcement Learning with Decision Values