Evaluating Object Hallucination in Large Vision-Language Models