Announcement. The protocol is GData, which is mostly Atom Publishing Protocol. The Google Base data model is a kind of RDF lite - lots of names types, simple strings in a lot of places you'd expect namespaces. But it's part of Google's own subweb rather than the web at large, so I guess this was predictable.
I've not dug deep, but I imagine a useful mapping to RDF would be possible, in fact I bet Google Base could hold arbitrary RDF, although probably needing a clunky URI/string map somewhere.
The queries are similarly string-oriented, e.g.
[employer: Apple] [job type: full time] [job function: Marketing]
I'm sure a SPARQL endpoint could be made to map to that. Dunno, presumably something very similar must be inside e.g. SquirrelRDF's LDAP mapping. If all this stuff is considered in Google's Own Namespace, keywords mapped to URIs etc, it could make a pretty useful Semantic Web system.
A couple of months ago I was invited onto a Google Beta, the hint was that it was data-oriented, I assume it was this one. Thing is I never got around to faxing off the NDA...Â@en