Analogical Reasoning for Visually Grounded Language Acquisition