29 April 2008

What's wrong with Topic Maps people and their tools?

Hoo boy, I'm feeling a bit testy today, but seriously; how many who read this blog know the difference between an association type, an association role type, and a reified topic type? When we humans try to put to "words" - in a Topic Map, as we're dealing with here - our various models, do you know when to create a new topic type, how to use the role type in multi-member association of given types, what a "type" really is, and what an "occurrence" of anything really is all about? And - more importantly! - do you really need to know?

The Topic Maps paradigm takes a good while to sink into your brain. It takes a few goes and lots of patience to get what it is really all about, and when you get there you discover - surprise, surprise! - that "topic maps" are irrelevant to the real nugget of wisdom within; shareable models.

Here's a model of me. (If you can see it properly, ) Does it seem right to you?

No, not the glossy models on the covers of magazines, but the kind of models the way the brain tries to shape an immensely complex reality for a brain than can only hold bits of small distorted representations at a time. Here's a piece of this, a representation of that, held together with some contextual duct tape, and Voila! you've got some "kinda" understanding of how the world works. Kinda.

The world does not want to be modeled, and anyone who has ever attempted ontology work (another phrase that means everything and nothing at the same time, and - of course - I use all the time!) soon discover how impossibly crazy it is to model anything to any degree of correctness. We, as human beings see this infinite tool of modeling our world, and falls promptly into the trap of actually trying it out. In some ways the amazing things you can do with Topic Maps is its own death! Don't try to model anything to any degree of detail; there's crazy in there.

We model all the time, but until I learned all the ins and outs of Topic Maps I didn't have the words for what I was doing, nor did I have the knowledge that I was even doing it. But I was, and I still am. You are, right now even by just reading these words. Just by writing this blog posting I'm modeling my opinion, phrasing it through prose, interacting with my computer, and posting it contextually, all in the hope of you understanding this model. I have a purpose, a model in my head if you will, of what this message is and how it should come across. The fun is, of course, that no matter how accurately and carefully I try to communicate the model in my head to you, every single person who ever comes across these words will have their own unique special - and, dare I say, still correct - version of it. Every single sound, vision, thought and feeling ever expressed is uniquely interpreted.

So, maybe I describe my model to the best of my ability in an attempt to remove tacit knowledge surrounding my message. Maybe I say things like "I use my blog to post this message, choosing a title that might catch your attention, and I'm talking about more important issues than a selected technology; human perception, communication between us in a way that is as close to truth as we can hope to achieve, that what we all do is communicate models from our brain to the hopefully next one. And of course, Topic Maps is one really good way of doing this." That's my mental model for writing this stuff, but you can be quite sure I've forgotten to say bucket loads of stuff, neatly hidden away in prose, innuendo, lost context and the mere fact that my brain is impossible to model, little less understand. These are tremendously complicated things that are, by their very nature, bloody impossible to get to! It just can't be done!

But, because it looks like it has some element of truth to it, we pursue it as if it is the truth. It looks like we can model things well, so we try to model things well. Ouch. Bummer. No, don't do that; you need to have less precision in your models for them to actually be taken seriously, at least by people. But let's talk about computer to computer communication. Surely there we can have precision?

This is the area which for me really brings Topic Maps to its right, where we - as digital communicators - create artificial models which we can attach our data to, and share around. Forget all this human knowledge stuff; if anyone who is to model their idea need to know what an association type is, forget it. Sure, we allude to what it might be, hold a course in it, or explain it on end, and then we throw ourselves heavy into the Topic Maps Data Model and explain that it is just another model that's mapped to the Topic Maps Reference Model, which is a model which is framed in frames theory (or thereabouts), which is a model of key/value pairs in a table setting, which is a model of simplistic systems, which is in a computer model setting, which is a digital model of ... and so on. Models, models, models. And you know what? There's translation and interpretation at every single step of the way.

Steve Pepper claims that Topic Maps are great because Topic Maps are closer to how the brain works than other means of mapping information, and I agree with him, but only in the "sure thing" kind of way, not the "correct" kinda way. The human brain doesn't think about how it models things. In it, there are no kinds, or types, or roles, or occurrences, or reification, or identifiers of any kind. So let's agree that Topic Maps, at the moment, is perhaps the closest we've got standardized right now that somewhat is closer to the way the brain works in computer terms.

Translations. That's what it's all about. I say "hei", you say "hi", she says "hola", he says "yo." Some use Java, some use C#, some PHP or Perl. Business people use business speak, designers talk the talk, programmers code. There's translations up and down and back and forth between them all, and then some. And wherever there is translation, there's a margin of error that gets higher the further the translated models are from each other. No wonder there's problems in the world.

As technologists we care a lot about these things, and from a technical standpoint, Topic Maps kicks ass! Seriously! But I've worked extensively with Topic Maps over the years, and if there's one thing I've learned it is that people don't give a toss about what the underlying technology is, and they certainly don't care what type or association concept any relationship might be. When people perk up about Topic Maps, it's not because of the data model but it's because of some underlying modeling ideas, that promise of sharing models and bring tacit knowledge up from traditional taxonomic and document-centric ways of dealing with "knowledge" in computer systems. They do not care about how to "map" their concepts, and seriously, don't care whether the modeling tools are standardized, shareable or not. They care about the models, for sure, but only as a conceptual thing they wish to share. So let's forget the technology and the data model and Topic Maps, because it really doesn't matter.

What matter is that communication happens, and it doesn't happen on the Topic Maps Data Model level. There is a higher, sloppier, fuzzier level that we humans live on, and that's the model we should try to get closer to. Sure, we can create cool systems using the Topic Maps standard to do all this, but it sucks as a platform of expression! It really, really sucks! Try right now to model the simple concept which is "freedom"; what are your topics, associations, role types, occurrences? One can appreciate that we can try, but there's no one answer to do this. The TMDM does no support human thinking, don't let yourself be fooled, it can only represent some misguided attempt at jotting it down in some machine-exchangeable way.

Topincs, Omnigator and other tools we Topic Mappers give to people who are to model things are a thing that only a technologist can love, and, in addition, a technologist who understand all the ins and outs of the Topic Maps Data Model. This is a huge limitation! People, the real group we have been trying to sell this concept to for years, just can't wrap their head around the Data Model to such a degree as to conceptualize their models! It's lunacy to think so, but of course, the Topic Maps community is chock full of technologists, so that's kinda expected. But I really wished that we had outside help. The Data Model is there for technologists and tool-makers, not people.

Can we rethink this part of the problem? I'm often embarrassed to give these tools to people, as it is extremely counter to claims we lay down to the greatness of Topic Maps. Don't get me wrong; TM is fantastic stuff, but the tools sucks for normal people, and by "normal people" I mean almost anybody but us.

We need to be even more human in our approach to knowledge representation. Topic Maps is a good foundations to build our systems on, but it sucks as a knowledge representation system for humans. What can we do?


  1. Hi Alex,

    I've just finished reading "The Answer to How is Yes" by Peter Block. He makes some interesting suggestions that the questions we ask are driven by an engineering/economist archetype mindset. I'm not sure how this relates but I sense a connection.

    Is the driver to build a topic map the right question to be answered by it?

    Do topic maps need emotions?


    David Buchan

  2. > People, the real group we have been trying to sell
    > this concept to for years, just can't wrap their
    > head around the Data Model to such a degree as to
    > conceptualize their models!

    Well, I taught TMs to Bond University students. And
    a few of them even got it. Sorta.

    > The Data Model is there for technologists and
    > tool-makers, not people.

    Also my point. With all the admiration for the tool
    builders, this is a wrong development. My thesis is
    that we need more lean languages. Such as that:

    "Alexander Johannesen", who authors someURL,
    which isa blog, holds-opinion, that adequacy
    has level low for tools topicmap-editors
    for task "authoring topic maps" .

    This is TM-speak.

  3. The graph, to me, comes under the heading of "Visualizations That Don't Help You Visualize." There should really be a Tufte for the semantic web. As Barta points out below, there's a lot to be said for text. After all, we've been using it for several thousands of years, with a fair measure of success.