Frequently Asked Questions

What is the difference between integration rules and ReDer rules? What about creation rules?

Both integration and ReDer rules transform knowledge, and we use SPARQL construct queries as default implementation for both, for convenience reasons. However, integration rules convert knowledge from external knowledge bases into FrameBase, while ReDer (Reification-Dereification) rules convert knowledge between different knowledge representation forms inside FrameBase. It may be confusing because some integration rules do also reify external knowledge and the structure can be similar in these cases.

To make things a bit more confusing, some of our papers also talk about creation rules. These are a limited set of rules that create ReDer rules from annotated sentences in FrameNet, but they do not operate over RDF.

What is reification and why is it so important?

Reification is a central concept throughout our papers, but it comes with different though related meanings, so it is worth explaining. In general, to reify (from Latin rēs ‎,“thing” +‎ -ify) means to regard something abstract as if it were a concrete material thing. In the context of knowledge bases, this usually means declaring a new entity (usually coining a IRI but it can also be an abstract node) to stand for something that was present before in the knowledge base, not as an explicit entity, but as some sort of implicit concept represented by a triple pattern. The different kinds of reification that one can find across the literature correspond to different triple patterns. In the case of RDF reification, it corresponds to a triple and in the case of FrameBase, to a frame. Other forms of reification exist, though they are not usually named as such: named graphs, structured values and even entities referring to whole ontologies are some examples. In the context of FrameBase, cluster-microframes are also reifications of the set of near-similar elements in the cluster.

We believe that reification amounts to a great part of the semantic heterogeneity in the Linked Open Data cloud, because knowledge may come either reified or unreified, or reified to different degrees or in different ways. In these cases, equivalence or subsumption links like those currently used in Linked Data are not enough. Therefore, it is important to discuss, investigate and categorize it.