FACTS: A fact extraction framework

One thing to note, for the experienced Semaphore user, is that whilst the framework might look like a skinning of the underlying rules, this is in fact not quite the case (although it would be fair to say that they are strongly aligned). A lot of work has gone on “under-the-hood” to make the extractors and facts work as you would expect them to work from a common-sense point of view. Sometimes this is addressing idiosyncrasies in how CS implements rules or how rules align (or not) with a common-sense reading of a document – with others it is just a natural progression of how the rules need to be written as per the rules’ logic.

As said already, the idea behind the framework is to bring a user-friendly and Semaphore-supported methodology to fact extraction. That methodology uses a basic framework to represent the different elements of the extraction methodology. That framework is largely as per the following hierarchy of elements:

  Document Type
      Document Metadata
          Context
              Fact / Context
              

That is, the modelling of the different elements is hierarchical for ease of representation and understanding. So, if we consider the Document Type as the top element, then it has one or more children of a subtype of Document Fact, either Document Fact or Document Anchor. Each Document Fact can then have one or more children of type Context. Each Context can then have one or more children of type Fact or another Context.

We will look at each of these types in turn.