Thomas B. Passin wrote:
Teresa Silio wrote:
What I think is that if it is possible to descibe a text using only a
set of primitive roles, then it should be possible to describe a set
of texts indexed by topic maps using the same primitives roles.
Once you have an Action, an Agent, a Pacient or an Objetc, the roles
are defined 'per se', you don't need to set out who is the buyer, as
there is an action and an agent. You may define roles in a
standarized way:
Action: buy Agent: buyer Object: what was bought
You won't need to create a role for 'buyer' another one for the thing
that was bought and so on. You just need to define an action, and
then set out which is the agent, and which is the object. This means
you save roles, labels and extra metainformation.
Sure, it is very feasible, but of course your choice of the roles
depends on the source material and what you want to do or say. Since
thematic roles are closely linked to natural language, they seem natural
for analyzing a series of texts.
Murray didn't say so just now, but he has created a set of subject
indicators for a more-or-less standard collection of thematic roles.
I'm not sure whether he considers them to be publicly released at the
moment, but perhaps he would comment.
Tom,
The specific roles you're talking about are the 23 Conceptual
Graphs relations as originally published by John Sowa, then
documented nicely as part of a CG course by Ulrik Petersen, Henrik
Schärfe, and Peter Øhrstrøm [CG]. There was some discussion of a
collaboration between myself, yourself and a few others, but this
did not work out. As you know, I've gained permission to use Ulrik's
text and will be publishing the PSI set as part of the rest of the
sets I have planned, which form modules in a larger ontology. If
it becomes valuable to others for me to release any or all of the
ontology prior to my intended schedule, I'm certainly willing
insofar as I'm able, timewise.
Teresa and I have begun a correspondence on this issue since at
first glance it seems that our respective research has a lot of
overlap. We both are interested in collaboration, so perhaps some
publication will come out of this... too early to tell of course.
My feeling is that yes, these do form a good basis for a prototype
of roughly two dozen roles, but it's hard for me to understand how
the complexity of natural language can be *computationally*
devolved to one of these 23 roles, given that in any instance a
specific word can conceivably have one or more meanings influenced
by an enormous variety of contexts, inflections, metaphors, etc.
The set is fine for general KR (declarative) modeling, which is
what CG was designed for, but for indexing entire texts via
linguistic analysis (where each key word playing a role would be
computationally assigned one of these 23 roles), this seems like
quite an oversimplification. But then again, the viability of
this approach depends a lot on the specific requirements of the
project. CL has done some very interesting things with texts
taken en toto, but I've not seen anything to suggest that a
discrete sentence can be "understood" very accurately -- too many
variables. I probably need to better understand what Teresa's
trying to accomplish (even if I were to assume I'd be a good
judge) before making any further foot-in-mouth statements.
That said, I'm pretty enthusiastic about the whole idea. It
reminds me of another unfinished project, that of a PSI set
for an Order Relation, which would be a primitive I believe
necessary as the foundation of the 23. I think Bernard and I
have been a bit too busy to tackle this, but this serves as
a reminder. I've got a very rough but functional version of
this as part of Ceryle, but in order for this to be useful
it needs to be much better defined and documented.
We have a lot to do... but best in the process not to follow
the altogether too-common tendency to reinvent another wheel
or widget.
Murray
[CG] http://www.huminf.aau.dk/cg/
......................................................................
Murray Altheim http://kmi.open.ac.uk/people/murray/
Knowledge Media Institute
The Open University, Milton Keynes, Bucks, MK7 6AA, UK .
"Instead of buckin' your head against a stone
wall, sit quietly with hands folded and wait
for the walls to crumble." -- Henry Miller
|
