we are not living in a entirely schema-less world but are for sure more relaxed about schemata than many people with an ontology engineering background.

There is a general tendency in the LOD community to reuse as many existing properties and classes from well-known vocabularies like FOAF, DC, SKOS or SIOC as possible. So the usual procedure is to start with these terms and only define additional dataset-specific terms if no fitting terms are found in well-known vocabularies.

The schemata of many datasets within the LOD cloud are defined using RDF-S (hardly any OWL) and there are also some datasets in the cloud that do not even define subclass/subproperty relationships, but are still considered useful by many applications (for instance DBpedia).

What is in contrast considered as very important in the LOD community is that properties and classes can be looked up on the Web. Many applications use this mechanism to retrieve labels for properties. Some brave applications (for instance the FalconS search engine) also do subclass inferencing on Web data after applying some trust heuristics.

I personally have the feeling that the usefulness of Linked Data does not depend too much on the capability to do sophisticated inferences on top of the data, but more on the possibility to discover related data by following RDF links between data sources.

I don’t understand your proposal about using schema/ontologies to improve the linking between the datasets. How do schemata help to find out that two records in different databases talk about the same thing? I think only very indirectly though a combination of schema mapping and identify resolution algorithms?

Domains where more sophisticated schemata level information would for sure be very useful for the Linked Data setting are in my opinion data visualization, schema mapping and data fusion.

It would be great if schema authors would start annotating their schemata with hints on how instances should be rendered by Semantic Web clients. The Fresnel display vocabulary does for instance provide for this.

It would also be great if schema authors would start to publish more mappings between their terms and existing terms from other schemata on the Web. The Neologism vocabulary publishing tool is already a first good step into this direction, as it provides for defining subclass and subproperty links between different vocabularies (
http://www.semanticscripting.org/SFSW2008/slides/sfsw2008-neologism-slides.pdf).

Let’s also hope that the new RIF rule language will provide for publishing more sophisticated schema/term mappings on the Web.

Cheers

Chris