Wikipedia has fairly standard search functionality - full-text keyword matches, really. To enhance this functionality, some researchers in Germany have extracted various types of structured information from Wikipedia and used RDF to build semantic models to represent this information, organizing them for more intelligent retrieval. The result of this is a semi-Semantic-Web construct called dbpedia.

The benefit with this is that now we can make semantic searches. Some of the sample queries include searches to generate a list of notable scientists and their doctoral advisors. Another finds a list of notable films lasting longer than 5 hours.

The model has the potential to be extended with domain knowledge, meaning that it can be made aware (for some definition of aware) of meanings of words, via ontologies. The query engine would then be able to understand simple relationships.

These are toy queries of course. The query language is slightly arcane, and I’m not seeing much domain knowledge in there. However, I can’t help but think that perhaps this would be the way to the Semantic Web, or at least a significant subset of semantically searchable content. Not to force everyone to write semantic markup, but to have some means of generating semantic content from data sources. Above all, you can see a slight hint of the great power and promise of queries against a semantic database.

On a less dreamy note, as mentioned on New Media Hack, given its free license and large data set, Wikipedia seems to be an interesting corpus for various natural language and semantic processing projects. Especially since you can download snapshots of the Wikipedia database to your local machine, in SQL and XML formats. A dump of the current versions of English pages typically clocks in at a couple of gigabytes, which is very manageable for the processing power available to plain old personal PCs. Encyclopedias are generally pretty structured, formally written text, and there are categories to distinguish various types of content. MIght be an interesting resource to use, especially for NLP or information extraction tasks.