Wordnet Metamodel and Web Service

Updated: 08 March 2004

Introduction

It is critical to be able to link terms in any taxonomy in the taxonomy framework lexically to a synonym set in WordNet.  This provides a baseline "lexical glue" between taxonomies.  Towards this end, I have been examining different ways to model WordNet (both existing and new).  Here is my first attempt at a WordNet schema representation (also parseable as RDF) for a Wordnet Web Service. 

Current work completed and remaining issues:

• Wordnet Sample XML Format.
• Wordnet XML Schema.
• Wordnet OWL representation considerations.
• Wordnet addressing.

Wordnet XML Sample.

Here is a partial sample instance document for the word "dog".  It is important to understand that this sample can be validated as XML via an XML Schema (see below) and parsed as RDF (parsed successfully with the W3C RDF Validator).  Here is the complete xml file.

Wordnet XML Schema.

Here is an XML Schema for the above instance document.  This has been validated via XML Spy 2004.

Wordnet OWL Representation Considerations.

Now that I have an RDF/XSD representation of WordNet.  I need to determine if that is suitable or whether I should normalize this with my other knowledge structures that are in OWL.  Specifically, I am working on a taxonomy representation in OWL (more on that later); however, since the taxonomies will be in OWL and they will be linked to Wordnet, it seems logical that there would be benefit to normalizing my entire search space into OWL.  I plan on experimenting with this over the next week.  Your thoughts/comments and suggestions are appreciated.

Wordnet Addressing.

Regardless of the chosen representation, I need a robust addressing scheme to map a Topic to a Wordnet Synonym (or maybe to a Concept/Synset).  I may need the ability to link to either a Wordnet Synonym or a Wordnet Concept.  Since concepts have unique IDs in wordnet that should be easy; though I need to make sure they are actually unique.  The most obvious solution is a URI with the namespace, and a fragment identifier of the concept id or the concept id/synonym lemma.  If you have a better solution, please email me.