Jack Park
Street address
Palo Alto, CA Zip



Memorandum

Date: Fri, 30 Jun 2000 07:50:20 -0700

From:   Jack Park
Reply-To: unrev-II@egroups.com

To:    
..
Subject:   KM Architecture

[Jack cites content from a letter contributed by Joe Willians; commenting on Jack's letter dated June 23, 2000; I think Jack is absolutely right on this - the original never gets changed. I would say we transcode the information into what is called 'canonical XML' for use by collaborators. Esstentially this means we forget style and just store and map content. A standard or user specified style is added back when the information is used.

Thanks,

Joe
Actually, Joe, I kindof have a slightly different view from what you say here.
..
Yup, don't touch original manuscripts. Just start a dialog on some particular aspect of some particular manuscript (here, I am using *manuscript* as a proxy for the ever-present *document*). This is where I think Eric's vision on IBIS is really important (one of the web links to gIbis now appears to be dead, so perhaps we are on our own doing an Ibis implementation). As another view of Ibis, do a google on *argumentation*.
..
For kicks, add *xml* to the search.
..
What you find is that there exists a contingent of folks out there who believe that argumentation as a process is worthy of implementation, particularly, *scholarly* argumentation. I would propose that the DKR fall in line with that notion. I am thinking here of just one of the many ways that the middle layer (what you may have been thinking as *canonical xml*) will grow as the arguments themselves are codified. This process, IMHO, is just one way, and perhaps a very powerful way to build a knowledge base.
..
In some sense, Rod's web site, though it is rather one-sided (just one commentator, not arguing with anyone and not codified in a machine-inferencable form), moves in that direction. Rod brings to the table just one of possibly many ways to organize that effort.
..
No need to transcode manuscripts into some *canonical* form. The effort will (eventually) be better spent actually *mapping* manuscripts into the knowledge base (middle layer), using the upper layer (OHS interface) to view, navigate, and discuss what's there.
..
What's interesting (at least, for me) in this scenario is that topics emerge in the knowledge base as user's interest in them grows.
..
The knowledge base will likely grow in the same manner as this email list bounces around various *threads*, occasionally jumping from one to another. This non-linearity in human behavior mirrors, again, IMHO, the nature of knowledge itself. I would like to think that the DKR will be a kind of playground in which non-linear behavior is encouraged. What's nifty here is that you never have to wade into a flood of emails just go play in your favorite sandbox. You can always surf with the OHS to see what others are up to (read: what areas of the KB are growing). Indeed, akin to siliconinvestor.com, there can be statistics on the DKR *home page* that show what's hot, and so forth. In fact, the system could model each user such that when one discussion references some user's stuff, and that user is not part of the discussion, that user could be automatically invited into the flow. Just imagine how much Bill Clinton would just *love* to hear that somebody is talking about Monica, or that Bill Gates on Innovation.
..
Ultimately, I suspect that as some topic area emerges, the work of the automated mapping will be facilitated since the discussion forms a *context* in which the mapping engine can begin its work. As the mapping engine adds fuel to the fire (so to speak), the dialog has the opportunity to become even more non-linear.
..
We know that the *neural nets* in our bodies involve, at once, stimulus, feedback, and decay. If we didn't have decay (forgetting), we'd be in a helluva mess. If we didn't have feedback, we might never get interested in anything. The architecture I am proposing involves all of these. We all remember the loud speakers that holler when somebody turns a microphone on. That's feedback. The mike and the speaker, acting through the amplifier, rapidly get very interested in each other. They appear to love to yell about it. Believe it or not, the DKR should facilitate similar behavior. IMHO, the DKR is that amplifier.
..
On rereading this before sending, I am struck by the vast number of metaphors I have used. I must now point out that, given the idea that we will be using some form of natural language in the dialogs (argumentation), in the beginning we will be forced to use a restricted (Lofti Zadeh calls it a *precisified*) natural language. NLP (natural language processing) technology is nowhere near up to the task of handling the vast number of ways in which we presently communicate. There exists on the web a project www.alicebot.org that attempts to build templates (using a fairly nifty XML markup language) to capture the essence of some conversation and render a response. Thus far, they are up to some 21,000 templates, headed for some 200,000. While I am not convinced that this represents the technology we will ultimately use, it does have the potential for an immediate jump into an implementation to get things started. I know this to be true given my experience with The Scholar's Companion, my discovery engine. I built a pattern-matching parser for it that allowed a high-school kid to *teach* TSC about a marine ecosystem. A couple of high schools in Dayton have used the same system as a query engine to study aspects of immunology in their biology classes. We can, indeed, IMHO, get a start on this process.
..
Film at 11...

Sincerely,



Jack Park