- From: John F. Sowa <sowa@bestweb.net>
- Date: Fri, 26 Jun 2009 06:16:57 -0400
- To: Danny Ayers <danny.ayers@gmail.com>
- CC: SW-forum <semantic-web@w3.org>, "Manola, Frank" <fmanola@acm.org>, Azamat <abdoul@cytanet.com.cy>
Danny, I never raised any objections to URIs. In fact, the ISO standard for Common Logic supports them as names. > ... for FOL this may be trivial, but in the context of the Web > it's hugely powerful, the possibility of using a simple protocol > to retrieve more information about the topic at hand. I am always in favor of supporting simple but powerful things. What I am against is making simple things difficult. My recommendation for the next version of the Semantic Web is very simple: 1. Keep the URIs. 2. Replace RDF with JSON (which is as readable as any of the recommended syntaxes for triples, but it also supports n-tuples). (And JSON, by the way, is the notation that Google uses instead of RDF.) 3. Replace OWL with a DL that has equivalent logical power, but a much cleaner syntax and the ability to use JSON. 4. Adopt ISO 24707 for Common Logic as the semantic foundation for multiple dialects. For example, a Horn-clause subset or a DL subset would be two different subsets of full CL. 5. Use a tag such as <script> ... </script> for embedding such notations in a web page. (But Common Logic also supports an XML-ified dialect called XCL, which is more compact than RDF for triples -- and it also supports full Common Logic.) I realize that some people claim that triple stores are useful, but there are far more efficient internal representations that give the programmer (or the logician) a view as either tables or as graphs. For just one example, see the following paper: http://citeseerx.ist.psu.edu/viewdoc/download;jsessionid=911FFAC5BC8B7B7A60B5E9197850E6AD?doi=10.1.1.52.3727&rep=rep1&type=url&i=0 The GMAP: a versatile tool for physical data independence Tsatalos, the first author, did the work for his PhD dissertation, in which he demonstrated that Gmaps (Generalized Combinatorial Maps) provide a physical representation that is more efficient for SQL than conventional tables and more efficient for object-oriented access than conventional graphs. He was hired by IBM Research, but as might be expected, he was not able to budge the DB2 behemoth. So he left IBM to start his own company. For our company, VivoMind Intelligence, we use Gmaps to represent graphs, and they support very efficient operations with very compact code. Gmaps are also widely used in architectural systems to represent huge graphs with billions of nodes. They enable graphs that represent a building or a complex of buildings to be mapped to any perspective for virtual reality -- and the mappings are extremely fast, even on huge graphs. They can run circles around anything that could be done with SPARQL. And as Tsatalos showed, they can support SQL-like queries against arbitrarily large graphs. That is just one example of Knuth's dictum: "Premature optimization is the root of all evil." The choice of triples to support the implementation of triple stores was a premature optimization by people who did not understand the state of the art for processing graphs. John
Received on Friday, 26 June 2009 10:17:33 UTC