Earlier quoted context omitted.
How can you automatically detect that from a video? And if instead of automatically detecting it, you just let the users tag the video, why does text search not work? Finally, even if they manage to get a useful video search working, why do they need Youtube for it, when they can crawl the web for video?
One thing you can search is closed captioning, like SnapStream does: http://www.snapstream.com/tvtrends/
YouTube has also added a closed captioning feature, but I'm not sure how it is done. I assume this is also working towards video search.