OpenCyc and UMBEL are now linked to Open GUIDs.  This was accomplished based on the latest OWL extract found at http://sw.opencyc.org/ (dated Sept 13, 2008).

These are manual links to the WordNet 2.0 sysnsets so have high trust values.  Ones that didn’t match were not loaded.  There were also DBPedia links included for some.  The numbers were as follows:

• 10835 OpenCyc & Cyc links
• 6221 UMBEL links
• 7288 DBPedia links

Here is an example of the resulting linkage: Integer

Thought I’d mention a few of the improvements recently deployed:

• There is now a shortcut to view the RDF+XML version of an Open GUID without messing with Accept-Encoding headers.  Just append “?encoding=xml” to the URI.  This trick will also be supported for future encodings like N3 and Turtle.
• The Abstract and Background sections of the Open GUID Specification we re-written for brevity and clarity.  A brief contrast between oguid:identical and owl:sameAs was also added.
• Open GUID Search was improved with performance enhancements and better grouping of keywords by GUID.
• A BCS WordNet is now being linked via the serbian dictionary.

More to come…

Open GUID was orginally seeded with WordNet 3.0 noun and verb synsets.  It is now correctly mapped to the WordNet 2.0 synsets via the RDF/OWL Representation of WordNet.

Using the Legacy 2.0 - 3.0 sense maps provided by WordNet, I was able to obtain the corresponding version 2.0 synsetid.  These are directly mapped to the RDF representation in a download available at the W3C.

Several details:

• 41 synsets in 3.0 combined synsets from 2.0.  These are represented as oguid:identical statements to each 2.0 URI.
• 138 concepts in 3.0 were split from muddled 2.0 synsets.  These were a problem because two Open GUIDs ended up with relations to the same 2.0 URI.  Because of the transitive nature of oguid:identical, this declared the Open GUIDs to be identical.  For most of these, I unlinked the Open GUID that least matched the 2.0 gloss.  Eight of them were merged back together, because they only differed by being a slang or local term for the same concept.
• 2896 concepts were brand new.  Since there is not an RDF representation of 3.0, these are maintained with a fictional URL to the W3C, approximately what it would be if they were to publish a new version.
• 37 links had a low map quality score and are not the same concept.  These were unlinked.

A good example of the result of this mapping is a station wagon.  It is the combination of two WordNet 2.0 synsets, and additionally has a merged relation.  The term ’shooting brake’ is it’s own synset in 3.0, but only differs as a regional word usage.  The 3.0 mapping is maintained for completeness, but not hyperlinked because an RDF representation does not exist.

It’s a fact, GUIDs are ugly.  So why use them for a common URI?  There are a number of reasons.

• First of all, RDF documents are meant to consumed by machines.  These have no problem processing GUIDs; they actually have an easier time because of the fixed length.
• Second, most existing descriptive URIs have an English bias.  This puts non-English speaking people at an unnecessary disadvantage.
• Third, a descriptive URI gives a false sense of being able to identify a subject by URI alone.  This is dangerous, since /George_Bush could refer to the 41st U.S. president, the 43rd U.S. president, or the guy down the street.  Various disambiguation techniques such as parenthetical suffixes have limited scope.
• Fourth, GUIDs don’t change.  It’s possible for a company or the George Bush down the street to experience a name change.  It would be unrealistic to expect all linked content to change associations to reflect this update, when a mere descriptive change would suffice.
• Finally, GUIDs can be generated offline and still be unique.  This allows asynchronous association to a common Open GUID that specifically identifies a subject.

For a truly global identifier, it is necessary to have multilingual descriptions to disambiguate a subject.  It is a straightforward task to develop tools that associate identifiers with local descriptions to help link content.

A question that keeps arising is why define a new URI for existing OWL classes, SKOS concepts, or existing named entities?

The answer is to provide a way to link classes, concepts, and individuals across ontologies with an idea rooted in the human mind.

Take, for example, organic food. The idea is known to exist in the human mind because it exists as a linguistic concept, i.e. you can see signs for it at the grocery store. However, depending on your needs, you might represent this idea very differently in an ontology.

A food industry ontology might make food:Organic and food:NonOrganic subclasses of food:Food to help reason about the nature of specific items. The publication industry might make a SKOS concept topic:OrganicFood to categorize it’s articles. A company database might consider making organic food an instance of a Product class:

ex:ExampleFarms ex:produces product:OrganicFood

Each concept has a specific representation in it’s ontology with it’s own URI. Declaring these as similar with owl:sameAs would confuse a logical reasoner trying to instantiate the objects.

However, there is still a benefit to linking these concepts across the semantic web. A search engine could easily find related resources, and a reasoner could instantiate each with local semantics depending on it’s linguistic context…and still know facts about it in the others.

The oguid:identical property provides this loose linkage and the Open GUID URI provides a common referent as the concept perceived by humans.  Paraphrased:

<food:Organic> oguid:identical <oguid:OrganicFood>
<topic:OrganicFood> oguid:identical <oguid:OrganicFood>
<product:OrganicFood> oguid:identical <oguid:OrganicFood>