Definitions in Information Management

During the past few weeks I have been finalizing my new book Definitions in Information Management (available via www.data-definition.com) and I have been reflecting on what I have learned about definitions since I began work on the book.
 

Perhaps the most surprising thing was just how much I had to unlearn. Data modelers get relatively little support when it comes to forming the content of definitions, so we tend to use examples from our experience - and the most common example is dictionaries. From this we get ideas such as a definition should be only one sentence long. That makes sense for a dictionary, which has to keep its size and printing costs under control, but it does not make sense for a data model where these issues are not relevant. How can anyone expect the each data object to be defined in only one sentence? It is simply not possible.

Another myth that I had to unlearn is that we need to have definitions clear at the start of any process or discussion. How many times have we all heard that one? In reality, it is the opposite. We get to know things gradually, like a picture slowly coming into focus. Thus definitions cannot be expected to be accurate when we begin to work with them. This is closely connected to another myth about definitions, which is that they are done once and never change. But that cannot be true. If we gradually get to know something better over time, then we will be continuously improving its definition. When will this end? Why should the process ever end? This perspective also implies that definitions cannot be part of projects where they are worked on for a relatively short period of time and then left alone.

Something else I had to unlearn was the notion that definitions must never contain anything that is negative. That is, a definition must never say what a thing is not - only what it is. I agree that a definition cannot be composed solely of statements about what something is not, but definitions are all about setting boundaries - about saying what is in and what is out. Therefore, it can be very important to include statements about what a thing is not.

Perhaps the most important thing that I learned is that definition is as much process as product. We all understand that we need to produce definitions in models, but the process is important too. For instance how are we going to deal with sources, and who will take responsibility for ensuring definitions are correct? This implies that governance is very important in the processes for managing definitions.

For me it has been a lot to think about, but I realize that with initiatives like the Semantic Web the role of definitions is only going to get more important, and data modelers are well placed to take a leading role. We may have to get rid of some old notions and perhaps work in a different way - but the opportunity is there.