Friday, December 23, 2011

On Levels of Definitions and the Semantic Web

Stijn made a couple of sharp points in a comment on the post How is a Definition Different from an Explanation? (Part 1) ( 

He notes that there is a need for definitions to be short in certain circumstances, as when a user is scanning through a list.  I think this is a good point.  Users may be more in search mode when they are doing something like this.  They want to know if the definition is close to some target they have in mind.  Obviously, a full definition is not fit for such a purpose.

So we might have three levels of definition: (a) a one-liner, suitable for lists; (b) a one-paragraph, suitable for a quick read with some detail - and display on a screen with scarce real estate; and (c) the full definition, as an authoritative reference.  I have no problem with the last one being very long and including pictures - certainly more than half a page.   

The second point Stijn makes is about the Semantic Web, where there is a need for a "general purpose" description.  I agree with Stijn that there is a problem here.  The definition of a concept must include something about the concept system the concept is located in - such as relationships to proximate concepts. So if one concept can be placed in different concept systems, then the definition will change.  A mortgage loan in a servicing system is not the same as in a loan origination system, and is not the same as in a securitization system.  The Semantic Web may have an unspoken assumption of a single model of reality.  This will cause problems if, as I maintain, one concept can be placed in many concept systems.  It will lead to the frustration Stijn describes.  

Where I disagree with Stijn is his equating explanation with description.  These are quite distinct.  There can certainly be descriptive definitions, and these are very common in natural science.  But explanations are different.

1 comment:

  1. I see you have come across the same realization that I did in one of my first posts (2006), namely, the Semantic web and Object Oriented Programming in general (ala C++/Java/etc) have the unspoken built in assumption that a single model of the world will suffice (since that is all that it lets you use in the normal way of doing things).

    This of course makes it very difficult to make systems/models work together that were developed independently of each other.