if this works it will be helpful for me, too!
The emphasis is on links, to enable annotation by third parties --
including annotation of links. In "typical" semantic network systems
(e.g. RDF, OMCSNet, or the first prototype of Arxana I got working)
edges are not first-class objects, and annotating them seems like it
would be hard to do.
If I was working on a semantic network with a group of other people,
and someone asserted:
a lizard is a mammal
I'd like to be able to say this is wrong.
Suppose I write a piece of text and several people want to mark it up.
An annotation model allows this, <joe-emph>if</joe-emph> everyone's
comments <ian-asserts-possible>can get integrated into one source
repository</ian-asserts-possible>. In an annotation-only model, there
isn't any particularly good way to annotate existing annotation, for
example, if I want to voice agreement with your assertion.
But the real problem with the annotation model is that documents are
tree-like and so their interpretation is relatively fixed. I want
instead to build a model in which documents are graph-like. A
marginal note on Text A can shift to be the central object of study
without the permission of the owner of Text A.
For me, that provides the "oh cool" response. For one thing, this
captures the essense of postmodern literary theory.
Applications, say, on PlanetMath, would include features like this:
Different users have different sets of incoming messages that they
want to read. Imagine messages being routed by a combination of a
subscription service and a syndication service. If I am the owner of
a given message feed, I may assert that these messages should get
syndicated to XYZ other feeds, so I open the valve to those feeds. In
the current management of email, a given user can turn off the
messages I send by routing them to the trash. In the model where
links can be annotated, other users could close the valve so that my
messages never get routed into their list in the first place.
Different users also have different kinds of objects that they think
are acceptable. Links enable users to categorize objects
independently of other users. But maintaining these links as
stand-alone objects makes it easier for categorizations to "inherit"
(e.g. if I trust user Ray to know whether some article is
mathematically correct).
A commonality between these applications is that they provide a way
for users to encode or modify side-effects (i.e. side-effects to the
action of making a link).
(I'm CCing a small mailing list where we've been talking about the
costs and benefits of using standard systems versus this
Thing/Triple/Theory based system. Maybe this will kick off some
further discussion.)
0 comments:
Post a Comment