[topicmapmail] Latest RM?

[Murray Altheim]

Jan Algermissen wrote:
> Murray Altheim wrote:
>>Jack Park wrote:
>>
>>>12 May 2003
>>>http://www.isotopicmaps.org/tmmm/TMMM-latest.html
>>>Still banging.
>>>Jack
>>>
>>>Murray Altheim wrote:
>>>
>>>>Folks,
>>>>
>>>>Just wondering if the 12 November 2002 is the latest available
>>>>online version of the RM. I'm assuming not. Jack has been banging
>>>>me upside the head for a week now and I think it's time I played
>>>>a little catch-up.
>>
>>Thanks very much. My god. You guys are firmly in the territory
>>of knowledge representation now. This goes a *bit* beyond
>>what we used to call "association templates". 
> 
> Murray--
> 
> great that you read it! (and thanks to Jack for banging ;-)
> 
> I don't understand what you mean with your comment, though. Can you explain
> how it relates to the RM? (what in it is the cause for your comment?)
> 
> Jan

Jan,

First, "that I'm reading it." It's a big document.

A little bit of history, as abbreviated as possible. In 1979 I
started by university career interested in robotics and AI. I
visited SRI as a potential Stanford student and saw an early
version of Doug Lenat's Cyc running some navy test data. I of
course had no idea it was his at the time.

Forward to 2000, and I'm involved in the XTM activity. I'm also
actively looking at ways to enhance web markup to handle metadata
and harvesting, just as were the people at the Dublin Core. Upon
seeing Topic Maps, I'm thinking, "this is it!", but TM keep the
"semantics" outside of the document, using instead a more appro-
priate mapping paradigm -- you don't need to pollute the docs
with metadata, you map it externally. Great.

Towards the end of the XTM 1.0 cycle, we all give up on trying
to develop "association templates", with Steve Newcomb giving
it one valiant effort. But things just weren't ready for prime
time. At one of the various conferences I get to meet Doug Lenat,
who knows Steve. I end up having dinner with him next time I'm
in Texas, and we discuss what Cyc is about. And I am thinking
about how to use TM for KR. I make an attempt to translate Cyc
into XTM. An interesting experiment, but I'm operating too low,
no flaps on my wings.

And association templates were simply that: a way of stating a
template upon which associations in a Topic Map had to conform,
like the ubiquitous "marriage" template.

I subsequently quit Sun and moved into spending my time actually
researching the field of KR, with the idea that I could use TM
as a core tool in KR. Well, I was apparently not alone. Looking
at the RM (or TMMM?) there are all sorts of parts of its
territory that are firmly in the area of KR. If we were to
think of Topic Maps as existing at a certain level of modeling,
then association templates were their planned schema. Like RDFS
is to RDF. What the RM is is more like OWL is to RDF. Well,
given that TM is a higher level of modeling than RDF, and OWL
than the RM, perhaps this isn't quite right.

I'm going to include a private message between Jack and I on
the subject. It's long. I haven't asked Jack but I don't think
he'd mind -- I edited out anything sensitive or personal. For
those who don't want to read that much, my basic point is that
"assertions" are a term in KR that doesn't match the term
proposed in the RM, just like TM's use of "facets" is a mismatch
for pretty much all other definitions (close, but not the same).
If we're going to head into describing assertions, we ought
to be sure they match up with KR usage, if what I'm seeing is
the intent of the RM, which is to establish TM's ability to
do KR. I think we need first a schema language, an RDFS-for-TM,
then we can build our OWL. We should try to figure out the
words to create the acronym PUSSYCAT.

Murray Altheim wrote privately to Jack Park (2003-10-18):
 > Jack Park wrote:
 >
 >> I keep trying to remember to talk about this, but then forget. I fell
 >> asleep last night reminding myself and it worked.  Before I go off for
 >> the day, here goes.
 >>
 >> There is something that, unless I missed something in XTM, needs to be
 >> adjusted in XTM.  It goes like this, and it comes from the RM, where
 >> the various properties of topics are enumerated based on "type" of topic.
 >
 > I would strongly disagree with this approach. It's mixing levels. XTM
 > is the serialization syntax for Topic Map documents. What the RM is
 > doing is a level up, at a Topic Map schema language, no longer strictly
 > in the realm of Topic Maps, but for Topic Maps as a framework for creating
 > statements about reality, i.e., something like OWL, though not quite.
 > You'd use statements in the RM coupled with a base logical schema,
 > developed at that level, to be able to work at the model-theoretic
 > level.
 >
 >    model-theoretic:   OWL          ???
 >                        |            |
 >    schema:            RDFS          RM
 >                        |            |
 >    map:                |           XTM
 >                        |            |
 >    graph:             RDF           |
 >                        |            |
 >    syntax:            XML          XML
 >
 > So, we don't need to change XTM. It serves its intended purpose just
 > fine. We need a set of PSIs to build from XTM the RM, using those PSIs
 > to type topics and associations in the creation of a base RM schema.
 > Then from that schema someone writes the FOL-in-XTM topic map that
 > mirrors OWL, and from that you build knowledge based systems.
 >
 >> The RM calls for these topics:
 >>    x-topic -- the roleplayer topic itself where Joe and Sue (who are
 >> married according to an assertion/association) are the subjects
 >>    t-topic -- a "type" topic for an assertion.  The subject of a
 >> t-topic is the assertion type.
 >>    r-topic -- a "type" topic for roles
 >>    a-topic -- the assertion instance itself
 >>    c-topic -- a casting topic, which "casts" an x-topic into an assertion
 >>
 >> Each of these topics have internal pointers which are not delineated
 >> in XTM.  An x-topic has a link to its casting topic, which is, in
 >> effect, the same thing as including the associations that a topic
 >> plays a role in *inside* the topic (which XTM does with occurrence
 >> <which, btw, are really just another type of association>, which XTM
 >> does not do. Of course, as an implementation issue rather than a model
 >> issue, you can just install links to associations (myAssociations)
 >> inside the topic container object.
 >
 > Yes, and I wouldn't want to muck with XTM just to do this, as it would
 > be making the same mistake as RDF does in mixing levels.
 >
 >> There is a mantra in the RM:
 >> "A Topic is t he *one* place you go to find out *everything* that is
 >> knowable about a subject"
 >
 > Yes, and in the model I've described, this is still true.
 >
 >> XTM doesn't do that, because you must look elsewhere (in the code) to
 >> see the associations in which the topic plays a role. The RM does that
 >> because the x-topic has a link (among other things) to a c-topic,
 >> which is the equivalent of an indirect address to an association.
 >
 > You don't look in the code in a Topic Map, because the code (which I
 > assume you mean the XTM syntax) doesn't represent things correctly. You
 > have to deserialize and bring it into a graph representation from its
 > tree-based XML document structure. After in the graph you can use the
 > graph to answer the RM-level questions. This is the proper level at
 > which Topic Maps operate, which is why SRN is concerned at that level,
 > as he should be. But we don't need to discard or modify XTM to do that.
 >
 >> For a long time, I have been arguing that you don't need the c-topic
 >> since you can just link the x-topic to its assertion which, itself,
 >> provides links to roles and assertion types. But, there may be merit
 >> in leaving the c-topic because it provides links to both the assertion
 >> and to the role, so, in one hop from the x-topic (roleplayer), you
 >> know what you play in which association. I know, this is like picking
 >> fly poops out of pepper. The difference I am observing is that XTM --
 >> a legitimate topic map application -- doesn't explicitly model what
 >> the RM says it can model.
 >
 > I don't follow the argument so far, though I see "PSI" in the next
 > paragraph, so I'll hold off.
 >
 >> There's one other thing that the RM addresses which XTM doesn't, and
 >> it is the notion that identity <hot button!> by way of the "built-in"
 >> notion of a PSI is not enough. Identity is context sensitive. SRN is
 >> trying to evolve a means by which identity is "conferred" according to
 >> context (which he calls "situation").  This, I think, is a means to an
 >> end that was discussed by Giancarlo Rota maybe 30 or 40 years ago, the
 >> notion of formalizing "seeing as". His lament: when you see a person
 >> driving by in a car, you are seeing a driver, not necessarily Joe. The
 >> identity of the object is its role at that moment. It's way more
 >> complex and hairy than that, but I think that SRN is, possibly
 >> accidentally, playing around in that sandbox.  The RM calls for
 >> something called an SIDP which, itself, is a *set* of identifiers
 >> which, when taken together, is the subject of the topic. Assertions
 >> can confer values into the SIDP (if I understand the RM's intent).
 >> There's a sloppy mapping between identity and subject in what I just
 >> said, but I think the takehome message is this: you can still have a
 >> PSI to "Joe" or whatever, but the subject, related to Joe isn't always
 >> *Joe*, but maybe *Joe as husband* or *Joe as driver*.
 >
 > This is waaaay beyond the place that either XTM, the RM, or the next
 > level up probably should be concerned with. You don't deal with levels
 > of epistemology at syntax or schema levels. Push McCarthy's situational
 > calculus at Steve, then. He should just stay away from "situations"
 > until he understands it or is convinced that that sandbox is where
 > Topic Maps should be concerned. If TM is about knowledge modeling,
 > it needs both a schema language and a model-theoretic framework that
 > far surpasses the RM (and CL/FOL). Now you're talking not only
 > second-order logic, you're beginning to talk modal and other forms.
 > If we want Topic Maps to be taken seriously, we should consider not
 > playing in the arenas that those like Nicola Guarino and Sowa have
 > been battling for years.
 >
 >> This concept turns out to offer a candidate solution to the nature of
 >> 'subjectness' when roaming about in 'subjectland' in blogspace.  There
 >> may be some inherent subject, say, Iraq, in a blog entry, but the
 >> subject might be 'Iraq as source of oil' . My intuitions are driven by
 >> a guiding thought: topic maps can replace google. Shit, for one day a
 >> couple of weeks ago, google was experimenting. They detected that I
 >> was using Mozilla and slid open the "search" tab on the left and
 >> painted "related hits" in there. Didn't do the same thing in IE, even
 >> though IE has a search tab.  Only saw it for a few hours on one day.
 >> Haven't seen it since.  I'm assuming that they were gathering usage
 >> statistics. I never clicked one of the links because they were
 >> essentially the same ones I could see in the main search results.
 >> They're not there yet. Back to the thought: a search on "Iraq oil"
 >> implies, in the topic map sense, that the subject is a conjunction of
 >> Iraq and oil, which is more than just finding the two words on the
 >> same page.  That's the intuition and the RM seems, whether it knows
 >> this or not, to be orbiting around that idea.
 >
 > I like where you're heading on this, but again, this has nothing to
 > do with Topic Maps, or rather, Topic Maps can't solve this problem. Or,
 > if they can, it will be at least two levels above XTM and probably at
 > least one level above the RM. YOu can't solve this at the XTM or RM
 > level because you need to be able to make statements about logic, which
 > requires the RM as a base, underneath.
 >
 > Steve is having the same fun in epistemology as I am, but I'm not trying
 > to bring Brandom et al into Topic Maps, which I think would be a terrible
 > mistake.
 >
 >> I really suspect (intuitions, again!) that this whole subjectness
 >> space is ripe for authors as well. It seems to me, from fiddling with
 >> my own novel, that there are a bunch of different "Deepa as..."
 >> subjects to cope with.  I could, theoretically speaking, model her as
 >> a collection of "subInstances" of the "Deepa" topic, but that violates
 >> the notion of the "Deepa" topic being the one place I go to learn
 >> everything that is knowable about "Deepa". There ought to be a way to
 >> roam about explicitly and precisely within the space created by the
 >> "Deepa" topic to learn that there are known contexts in which
 >> knowledge about her is organized.  I don't think the RM is the final
 >> answer to that because, afik, there is only one SIDP, and there may
 >> need to be the equivalent of a hashtable of them, hashed on the "as"
 >> property (the role she plays). I don't think that XTM lets me express
 >> and serialize these distinctions. If you can see how, then let's talk.
 >
 > Are you familiar with Eco's "Six Walks in the Fictional Woods"? If not,
 > it's not long and you should read it before we continue much further.
 > There's other things, too. And I have plans to implement context in
 > Ceryle, as part of the Assertions. And part of hierarchical or modular
 > graphs, whatever we call them (there's a couple of names that are
 > overlapping.) If you haven't had a chance to see W.T. Tuttle's "Graph
 > Theory", it's a really good source for anything to do with graphs.
 >
 >> I don't think that the RM takes anything away from XTM. Rather, it
 >> puts back a few things that are missing (my opinion which is evolving
 >> out of studying the RM -- but which may be terribly naive). It's
 >> always possible to export the results of an RM model to XTM, but you
 >> give up some nice stuff, though you can put it back (maybe) on import.
 >
 > That only makes sense. The RM is one part of a schema language for
 > XTM, depending on the kind of schema language we're talking about. If
 > we're up in the clouds of RM or even above, XTM is down on the ground
 > doing the interchange stuff. But as RELAXNG and XML Schema are written
 > in XML, the RM can be represented in an XTM document.
 >
 >> Speaking of Ceryle, as I was roaming the authoring ontology last
 >> night, I recall fiddling with associations, but don't recall seeing
 >> the roles displayed.
 >
 > For the purposes of Ceryle, there's no way I'd want to visualize
 > those roles. They'll show up (if the author wants them) in the
 > graph editor, where you can see all the details of a Topic or
 > Association. But to visualize all the roles played, that would put
 > off pretty much anyone except the most hardcore knowledge modelers.
 > When I saw your Joe Sixpack diagram I thought you'd probably want
 > to be able to display that in Ceryle, but they're designed for very
 > different applications. If you want, I would be willing to create
 > the *Nodes to do that, colors and all, and then you could extend
 > the TopicMapVisualizer API and TopicMapVisualizerImpl to handle it.
 > I haven't looked in detail at how extensible they both are, but I'm
 > certainly open to making them extensible -- hence the API.

Murray

......................................................................
Murray Altheim                    http://kmi.open.ac.uk/people/murray/
Knowledge Media Institute
The Open University, Milton Keynes, Bucks, MK7 6AA, UK               .

   Monkeys use thoughts to control robotic arm
     http://www.sfgate.com/cgi-bin/article.cgi?file=/c/a/2003/10/13/MN2018.DTL
   Bush uses media expertly to push apocalyptic view
     http://truthout.org/docs_03/091403J.shtml