Text this: Alleviating Hallucination in Large Vision-Language Models with Active Retrieval Augmentation.