Enriching a Model's Notion of Belief using a Persistent Memory