Semantics - logical vs linguistic
Up to: VoCampNYCJuly2009
Q: I would like a detailed discussion about how the Semantic Web got from a natural language sentence to s,p,o without any time reference (e.g. was). The Semantic Web is about structured data. However, there has to be a detailed justification about how it is related the makeup of a text document. David Jensen
A: The Semantic Web is based on a logical rather than an linguistic notion of semantics. In description logics, concepts may have individuals as members. Individuals may have named properties (once called "roles") that relate them to other individuals. If a concept is asserted to have a property with a particular value, then it is valid to infer (the logic part) that all individual members of the concept have that value for that property. The subject,predicate,object tuple that you refer to is the result of a reduction of all concerns to those expressible in description logics. This was done in order to canonicalize the notation, and to make it possible to prove the consistency of the logic being proposed for reasoning about assertions in the notation.
In order to express that a certain assertion is valid at a certain time in a triple notation, the assertion has to be reified, meaning that it has to be described in the tuple language, not just stated as a tuple. So "Jack holdsOffice mayor" becomes "n123 isAssertion yes", "n123 subject Jack", "n123 predicate holdsOffice", "n123 object mayor", "n123 holdsStartingOn 20090701". If Jack is still mayor, then that's all you would say. And notice that now you're talking about statements and names of individuals, not about the individuals and their relations.
From the point of view of providing consistency of a reasoning system, being able to quote statements and talk about them is a big wildcard that can only be proven safe if certain other aspects of expressive power in the language is limited. So it's not only the format that prevents mentioning time, but also something essential in the logic. Assertions about when an assertion is valid are usually made outside the logical system.
In linguistic semantics, a lot of the battle is to determine the correspondence between terms and senses. Relations expressed linguistically among terms need to be mapped to relations in the semantics among particular senses of those terms. The sense that is chosen as the interpretation of the term should be consistent with the context of the term. When the task of interpretation is done well, inferences that are based on the interpretation can be drawn with (comparatively good) confidence.
