From newshub.ccs.yorku.ca!ists!helios.physics.utoronto.ca!news-server.csri.toronto.edu!rpi!zaphod.mps.ohio-state.edu!wupost!uunet!mcsun!uknet!ox-prg!oxuniv!mc703 Tue Nov 19 11:09:53 EST 1991
Article 1295 of comp.ai.philosophy:
Path: newshub.ccs.yorku.ca!ists!helios.physics.utoronto.ca!news-server.csri.toronto.edu!rpi!zaphod.mps.ohio-state.edu!wupost!uunet!mcsun!uknet!ox-prg!oxuniv!mc703
>From: mc703@vax.oxford.ac.uk
Newsgroups: comp.ai.philosophy
Subject: Semantics in Semantic Networks
Message-ID: <1991Nov12.190138.2729@vax.oxford.ac.uk>
Date: 12 Nov 91 19:01:38 GMT
Organization: Oxford University VAXcluster
Lines: 60

A selection of abridged replies to my query about semantic networks:

  "Are researchers as keen on semantic networks as they used to be?"

Scott_Fahlman@edu.CMU.CS.SLISP.SEF-PMAX isn't:

> I've gone off in the direction of neural networks.  It seems to me that NETL
> [a semnet system] was really an implementation of clean, symbolic AI.
> I was more interested in getting at the fuzzy, messy, not-very-symbolic kinds
> of knowledge that seem to occupy all but the top 1% of intelligence.

and interestingly:

> Another reason I quit working on semantic nets is that the field of
> knowledge representation was hijacked by the formalists.  Everything must
> be explained in terms of logic and precise semantic models, and everything
> that doesn't fit into a clean formalism is simply thrown out.  I was always
> more interested in what a piece of knowledge should DO in a given context,
> rather than what it MEANS.

For what it's worth I agree entirely about precise semantic models; I think
Tarskian truth-conditional semantics and its descendents (cf the Davidson of
"Radical Interpretation" etc) are overvalued.  Frege, I submit, never intended
his formal logic to be used for a theory of meaning for natural language - it
was Russell who started the rot :-).  There are good reasons why logic is a bad
tool for the job, and the result is that we are nowhere near a "theory of
meaning" either in philosophy or in AI.  The Wittgenstein of the Philosophische
Untersuchungen is a lot more interesting in my opinion - a theory of language
without truth-conditional semantics.

> Perhaps the reason that nobody much talks about semantic nets any more is
> that once you've boiled it all down to some form of logic, the
> pretty pictures don't buy you much.  It's all just "knowledge
> representation".

An equal and opposite reaction from Manfred Stede <mstede@edu.toronto.cs>:

> Semantic Networks are still fashionable, but the stuff is now
> (following the general AI trend) much more formal and better
> worked out than in the early days.

And dld@au.edu.monash.cs.dec18 (David L Dowe) writes:

> Some work on this stuff is being done here in my department (Dept of Comp Sci,
> Monash University, Clayton, Victoria 3168, Australia) by (the Head of Dept)
> Prof Les Goldschlager and one of his 4th Year Hons students, Brendan
> M(a)cMillan.   Les and Brendan look at associations (although their stuff
> seems a little different to sematic nets).

I wonder if these "associations" are even closer to the old Lockean "connexions
& disagreements"?

My other question was

  "are the inheritance/sharing ideas in object-oriented programming descended
   from AKO and IS-A?"

Scott Fahlmann thinks not:

> Nope, this is a pure case of parallel development, like the marsupial wolf.


