To produce the denotation graph, we have created an image caption corpus consisting of 158,915 crowd-sourced captions describing 31,783 images. The new images and captions (compared to an earlier, smaller Flickr dataset) focus on people involved in everyday activities and events.