This narrative doesn't really ring true to me. In any case, if you're saying that this result indicates that the system is no better than a bag of words model,

This narrative doesn't really ring true to me. In any case, if you're saying that this result indicates that the system is no better than a bag of words model, doesn't that indicate that humans are equally no better, considering the errors made by the two commenters in this thread? To me it just seems like a query that is likely to trip any imperfect entity up, since a "when" question usually implies that the event in question is known to have happened.