[ilaksh]

I think that this is the obvious path to more robust models -- grounding language on video.