Masked IRL: LLM-Guided Reward Disambiguation from Demonstrations and Language | ResearchPod