Blog

Whilst googling Stephen King’s Storm Of The Century, I stumbled upon some features of Amazon’s US site that I hadn’t noticed before. Amazon’s Inside This Book feature has, for some time, enabled customers to peek inside a book and look at a selection of scanned pages.

Recently, this feature has been extended with the ability to search the full text of a book, along with a snapshot of what to expect inside, including :

Statistically Improbable Phrases / SIPs – A set of the most distinctive phrases in a book.

Capitalised Phrases / CAPs – A set of the likely characters, people, places, topics and events.

SIPs and CAPs are actually not far from Mark’s proof-of-concept for movie dialog search. Essentially Inside This Movie, Mark’s demo indexes a DVD’s closed caption dialogue against timecodes then provides the user with a mechanism to search for particular phrases in inside a movie as well as a concordance of all spoken dialogue.

Applying Amazon’s SIP and CAP algorithms to Mark’s data would potentially yield a more useful concordance of dialogue, potentially pulling out character, people, place, topic and event names – Death Star, Tatooine, Rebel Alliance – from any encoded movie as well as any distinctive phrases- May The Force Be With You, I Have A Bad Feeling About This.

Our TV3.0 project includes a use case that allows users to attach comments to a piece of movie content, with timecode references, so that users can click straight through to individual scenes. SIP and CAP for movies would allow us to auto-comment video content with links to distinctive or significant scenes within a film, based on spoken dialogue…of course, movies from the silent era could be problematic 😉

The potential for SIP+CAP enabled movie services is compelling. Imagine, being able to search for all pop-culture references to Star Wars across episodes of Friends, The Simpsons and countless other shows, which people enjoy watching with their TVs and a rv outside tv mount in case they’re camping…assembling a playlist of TV/movie scenes as a virtual search folder syndicated to other individuals and applications.

Google yesterday added video playback to its Video Search beta (earlier covered by Mark here) using the VLC-based Google Video Viewer.

Video Search results now include a blue Play icon next to results with playable video content. Though such content is currently rare, ironically, searching on my first name will yield a playable clip!

Clicking on a playable result opens a series of thumbnails enabling the user to watch an entire video clip or start playback from highlights at at 30-second intervals. Clips playback embedded within the page, but can be played full-screen also.

The user experience is pretty good and coupled with indexing of dialogue by Google could eventually make TV content as inherently index-able and search-able as web content. Mark’s earlier experiment in indexing closed-caption data and making clips playable available though searches of dialog fundamentally achieves the same thing…albeit with more interesting content (Troy, A New Hope, The Empire Strikes Back and Return Of The Jedi!)

Curiously, the preferences dialog implies that video search is currently based on local TV content within a ZIP code defined by the user, placing an artificial geographic boundary on available results…as broadcasters, content owners and users make more content available, this should alter to reflect content categories rather then locational availability.

However, I suspect MSN and Yahoo will be better placed to exploit video search, given Yahoo’s growing status as a media network and Microsoft’s close relationships with content owners through its Windows Media business.

This much is clear, the artificial scarcity of TV distribution models is gradually unfurling as BitTorrent, PVRs, DVD ripping, RSS, Search and broadband create a confluence of technologies that are Napsteriz-ing and disintermediating the TV industry.