Cornell NLVR is a natural language grounding dataset with 92K examples. Each example shows an image and a sentence describing it, and is annotated with the truth-value of the sentence. - View it on GitHub