[topicmapmail] Conceptual Graphs are Step 6

[Murray Altheim]

Thomas B. Passin wrote:
> Dan Corwin wrote:
[...]
>>I agree, except CGs really do not *translate* NL.  They *express* 
>>assertions in predicate calculus about the role players that fill 
>>verb-specific association templates based on case frames.
> 
> Yes, but let me quote what Sowa says in his "knowledge Representation"
> book (I just found these passages today) -
> 
> (p. 41) "Since textbooks use natural languages to define everything in
> mathematics, the FOL translations of natural language must also be
> sufficient to define all of mathematics."
> 
> and
> 
> (p. 14) "When the graph is translated to predicate logic, ..."
> 
> I consider this to be good authority for using the word "translate" in
> this context.

This is a leap of intuition or faith, and has no grounds in actual
practice. Yes, *of course* everything we know is expressed in natural
language, but that hardly justifies belief in a straightforward
"translation" being either possible or the result being reasonable
(in both senses of the word). I think John sometimes waxes too
optimistic, tending to gloss over the actual difficulties, e.g.,
yes, theoretically possible, no, never done before. His latest
is Common Logic Controlled English (CLCE), which is a declarative
language for expressing statements in an English-like syntax:

    http://www.jfsowa.com/clce/specs.htm

I have no means of evaluating the efficacy of this approach, other
than to express some skepticism over people's willingness to use it,
especially since it is Anglocentric. There are some languages in
which the English vocabulary and syntax doesn't have demonstrably
equivalent translation, such as Chinese, Arabic or Hindi. But this
is really a different matter than we're talking about here.

>>Good news:  NLP software is getting better good at resolving such 
>>ambiguities.  With heuristics, public lexical data, and sneaky
>>tricks, software encapsulating steps 1-5 can now guess about such
>>things and in some circumstances generate low error rates.
>>
>>My MODELER design can do this.  In fact, if you configure it
>>properly, and interact with it about noticed problems (like
>>misspellings), it will cull ambiguities from your paragraphs with
>>*virtually no* errors, and dump an XTM chart which says in *your*
>>ontology what they asserted:

Then you're ahead of most of the research in computational linguistics
that I've read about. But I've not delved that deeply, since my own
work doesn't attempt to do linguistic analysis.

> Well, you are so far ahead of me - I have never done any machine NL work
> - so there's probably not much I can offer here.
>
>>>I consider Topic Maps to be essentially a subset of Conceptual
>>>Graphs (with a few additional wrinkles).  Some CGs can be expressed
>>>as TMs and some cannot.  The ones that can be so expressed are
>>>nearly identical except for some syntax details.
>>
>>I believe TMs can hold graph structures fully equivalent to those of 
>>any CG, but TMs have no standard inferencing model.  CGs do: some FOL
>>engine that can infer things by using predicate calculus.

I don't think of Topic Maps as any kind of subset of CG, but neither
do I think them incapable of representing CG semantics by the addition
of a suitable set of PSIs. I don't know how there could be some CGs
that are unexpressible as TMs, as the underlying graph structure of
topics (as concepts) and associations (as relations) should be, with
the addition of PSIs to establish CG semantics, identically expressive.
The problem I've found is in tying down exactly those CG semantics.
A Topic Map graph structure just acts as a container for whatever
formalism one desires to express.

> TM is missing universal quantification and negation,in addition of the
> inferencing prescriptions.  To be able to add negation we need to be
> able to use unary (1 role-player) associations.  So I oppose efforts
> being made in some of the modeling efforts to requie an association to
> have at least two role-players.

I agree with this. Topic Maps should be left very open-ended,
precisely for this kind of reason. Back when I did my little
foray into Cyc-in-XTM, I'd developed a Topic Map containing
those FOL semantics. I think it'd be a cool graduate student
project to build an inference engine that worked directly on
XTM, but it *might* be considered kinda pointless, too, since
it's really more a matter of translating the TM expressions
into something an existing engine can understand. OTOH, being
able to plug an inference engine into TM4J and having it perform
these functions without requiring translation would be very
handy. My own project has something like this, about half-done:

   http://www.altheim.com/ceryle/api/org/ceryle/tm/Inferencer.html

I've got unused code to support AND, OR, and NOT. BTW, I'd be happy
to work with anyone on further developing this API. The basic idea
is to (like most engines) return a truth value on each function. I
don't think XTM is the most efficient means of expressing FOL, but
it's not really a competition with existing languages given that any
logical expressions are also *in a Topic Map*, which adds a lot of
value. I see plenty of possibilities here.

>>So, Tom, okay - if you really do think CGs in volume are a good idea,
>> here's a partial proposal for a 2-part project able to produce them:
>>
>>1. On this public thread, we debate XTM models that would be ideal 
>>for a chart, and craft a few prototypes by hand in LTM.  Iff we get a
>>consensus, we will then have a CG-compatible base ontology in TM
>>terms, which MODELER will adapt to use for step 5 output.
>>
>>2. To guide such work in part 1, we can also imagine a file-to-file 
>>converter that takes that XTM chart and produces CGIF.  We should 
>>limit it to use *only* the XTM input syntax, then notice and feed 
>>back requirements on what data that part 1 output must contain.
>>
>>Seriously, I have no concrete plans yet in MODELER for a CGIF output 
>>module, because I do not know CGs or FOL in depth; because I sense
>>they are mostly a tool for teaching logic than a standard; and
>>because it is not yet clear to me how many people actually care about
>>CGIF.
> 
> Murray Altheim is very unhappy about the state of CG and CGIF
> specifications - he says they are not precise enough to build 
> interoperating software with.  He's looked into it a lot more than I 
> have, so I expect he's right.  That leaves things uncertain about how to 
> proceed.

Well, I don't claim to be right, only that I've expressed my
opinion on the matter. Peter Becker and Finnegan Southey have
spent a lot more time trying to make some sense of CG (i.e.,
trying to gain some sense of an underlying data and processing
model), and they are very skeptical. I tried to build a markup
language, and in the ground work for that was unable to discern
enough to build its atoms without doing what every CG tool has
done, which is interpret the spec where it doesn't specify
certain things. While John Sowa is a very thoughtful scholar
in the field of logic, I don't know that CGIF demonstrates
very good language design. He's resisted all attempts by Peter
and I to try to look into the matter further, and apparently
is now spending most of his energy on Common Logic and CLCE.

I don't personally have any plans to work with CGIF, writing
importers or exporters. I think I'd rather extend LTM in some
manner, if that is even necessary (I think it is, in that we
need to be able to return a value from a function). They're
both similarly expressive (with LTM's syntax being influenced
by CGIF, if I remember correctly). Just having a standard set
of PSIs for FOL (likely based on CL) may be enough, and may be
all that's really feasible. After much pleading, both John and
Pat Hayes have told me that the CL spec will have URIs for this
purpose (something good has come from all my harassment).

Everyone is always looking for some kind of authority in
knowledge representation, a Grand Field Theory, but the more
I've investigated the less I feel there is (a) any possibility
of such a thing, or (b) any reason to want it. The idea of
universal meaning is pretty bogus, and without universal
meaning no one representational formalism will do. We decry
the imminent death of thousands of natural languages, why
should we believe in One True Representation? I've just been
discussing this subject recently on the CommunityWiki:

   http://www.emacswiki.org/cgi-bin/community/PlatonicCategories

I don't mean to say that creating specific "semantic toolsets"
isn't valuable, just that I don't share some gnawing desire to
have them be universal, only solve a specific problem.

Murray

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

    The Resume of George W. Bush
    http://www.buzzflash.com/contributors/03/04/23_resume.html

                     The sleep of reason brings forth monsters -- Goya