Google Squared is Google’s answer (or perhaps one of their answers) to semantic search and Linked Data.  ‘Squared gets its moniker from the matrix used for displaying results – each “square” in the matrix contains some fact derived from the content on the source site. 
Each row in the Google Squared matrix is a search result, but the interesting part is the columns.  ‘Squared relies on RDFa and microformats on the source sites to extract structure for the search category – if it’s available (I’m not entirely sure how ‘Squared derives its semantic structure in the absence of metadata, but clearly it does).  So a search on “rollercoasters” will generate columns for height, speed, construction, etc.  Essentially, ‘Squared is generating search facets on the fly using structure that is implied or explicit in the set results set. 

Because sites are inconsistent with the amount of structure they provide, ‘Squared can – and will – make errors in interpreting free text.  For example, in the video, we can see that “height” – while intended to describe the height of the rollercoaster – sometimes returns text about the minimum height requirements for riders.  Still, from the demo ‘Squared does look interesting.  Predictable categories like Restaurants (e.g. a search on “pizza”) have the dimensions you’d expect to see in columns – description, address, price range, ambiance, etc. 

Google Squared affords general-purpose faceted search, because the columns can be used to refine the search results.  The current alpha doesn’t let you sort on a column, but clearly this is where things are heading.

Google Squared will be made publicly available on the Google Labs site in the next week or so. 

Here is the video courtesy of TechCrunch (over six minutes and shaky-cam – but you’ll get the idea).

As a footnote, TechCrunch declares that ‘Squared “is how Google will crush Wolfram|Alpha”.  I’m sorry – they missed on that point by a country mile.  Alpha isn’t a search engine – it’s a user experience (and a cool one at that) built atop a “curated” database, designed to answer queries that are primarily computational in nature.  Google searches the web.  No comparison – the products aren’t in the same class, and don’t solve the same problem.  Google can’t plot the Julia Set for you – Alpha can.  That doesn’t mean Alpha “crushes” ‘Squared, either.  Geez.   Anyway, more on Alpha later.