Hi Patrick --
Thanks for your thoughtful reply, attached below. I'll take up just
one point for now, if that's OK with you.
You wrote....
If your argument that widespread use of formal ontologies or other
methods will require better interfaces for the average user, I don't
think you will find much disagreement on this list or elsewhere. But,
and it is an important but, unless the underlying principles which are
used by such an interface can be specified and discussed with
precision, then the results of any interface are not going to be
particularly useful.
The key here is that if one is doing "semantics" in the real world, it
is no longer enough to produce a formal notation and hope that someone
will produce a non-geek-human understandable user interface to it.
That approach leaves open likely semantic disconnects between the human
meaning and the system meaning.
What we have done is essentially to integrate 3 kinds of
semantics in one design and system [1]:
- Semantics1 is "Data Semantics". as in RDB, XML or RDF
- Semantics2 specifies what a reasoning engine should do (in our
case via a model- and fixpoint-theory [2])
- Semantics3 concerns the Application Semantics in the meaning of
English concepts at the author- and user-interface.
In the system, the link between Semantics3 and Semantics2 is of course
automatic, and two way. You can verify that it works by using a
browser to view, run and change the examples at [3], and by writing and
running your own examples.
In principle, the way that this works is by automatically mapping
English sentences, with place holders (variables) for values, to- and
from- predicates. There are some complexities our implementation of
this, but it is conceptually quite simple. It is in no way a
contribution to research in natural language processing [4], just a
minimalist and robust way of grounding Semantics 1 and 2 out to
something meaningful to nontechnical humans (Semantics3).
I hope this makes sense. Thanks in advance for your further comments.
--
Adrian Walker
[1] http://www.semantic-conference.com/program/sessions/S2.html
[2] Backchain Iteration: Towards a Practical Inference Method that is
Simple Enough to be Proved Terminating, Sound and Complete. Journal of
Automated Reasoning, 11:1-22.
[3] www.reengineeringllc.com
[4] Footnote -- I have worked in NLP research, and I really like it.
But making full NL UIs practical and robust for non-technical authors
and users is a hard, maybe "AI-complete" problem. In the meantime, we
need something that works, even if it is not very ambitious NLP-wise.
--
Internet Business Logic (R)
Executable open vocabulary English
Online at www.reengineeringllc.com
Shared use is free
Reengineering, PO Box 1412, Bristol, CT 06011-1412, USA
Phone 860 583 9677 Mobile 860 830 2085 Fax 860 314 1029
Patrick Durusau wrote:
Adrian,
Internet Business Logic wrote:
Adam, Patrick, All --
Yes, indeed, quite a lot of meaning is in the relations (=>, ->,
&, or...) between atomic items in formal notations.
But there's more, much more!
Err, I was trying to understand Adam's position, not necessarily
endorse it. ;-)
BTW, your paper makes much of the need for "human-friendly sentences"
but that isn't really a fair criticism of formal ontologies or other
approaches. (Having written my share of human unfriendly prose and
syntax I feel compelled to defend formal ontologies a bit.)
As was noted in KIF 3.0, there was no intention that KIF be used as a
language for human users to compose ontologies. It certainly could be
used that way but there was no requirement that it be used in that
manner.
By analogy, PostScript and PDF are pretty user unfriendly as well, if
you want to write them by hand, yet users manage to produce documents
using them everyday.
If your argument that widespread use of formal ontologies or other
methods will require better interfaces for the average user, I don't
think you will find much disagreement on this list or elsewhere. But,
and it is an important but, unless the underlying principles which are
used by such an interface can be specified and discussed with
precision, then the results of any interface are not going to be
particularly useful.
To see this, consider slides 14-17, 35-43,
and finally slide 52 of
http://www.reengineeringllc.com/Internet_Business_Logic_e-Government_Presentation.pdf
These slides argue that, if we limit "meaning" to technical notations
(combinations of symbols like =>, ->, &, or..., RDF,
URIs...), then there is no grounding in the world outside the computer.
That in turn severely limits the usefulness of the "meanings" for real
world tasks.
As I said above, part of your complaint (and it is a valid one) is
about user interfaces and not the notations.
BTW, limiting "meaning" (an accusation of a limitation of a formal
notation) and the grounding problem are different problems. Granted,
given a limited enough notation I concede that it would prevent solving
the grounding problem but your paper doesn't not make that case. True
enough, all the things you say about formal notations being difficult
for the average user to use are true, but that is not the same as
proving a limitation in terms of what the formal notation can say.
In other words, to prove your point about the grounding problem you
would have to show:
1. An _expression_ limitation of the formal notation, and
2. That the limitation prevents "grounding" in the world outside of
computers.*
*Note that 2 presumes proof of a solution to the grounding problem and
that a particular notation does not support that solution due to a lack
of expressiveness.
Sure, we all know and love our succinct logic
and programming notations, but they were invented before Von Neumann,
and they deliberately leave out the pragmatics of natural language.
I am not sure what you mean by the "pragmatics of natural language?"
Granted that logics and programming notations are more limited than
natural languages but then they never pretended to be natural
languages.
BTW, historically speaking, there have been a number of logics
"invented" post von Neumann.
That said, full natural language
understanding is of course the "AI-complete problem [1]". However,
there is a minimalist approach that appears to work, and that does
document executably the things that English sentences mean to people.
It's in the implemented system online at www.reengineeringllc.com, and
shared use of the system to write and run examples via a browser is
free. The system functions as a kind of Wiki for executable English
content, and there are already many ontological and other examples.
By "minimalist" approach do you mean entering rules that provide an
English language interface?
Note that the system relying upon external entry of information and
rules for processing is in part the definition of the grounding
problem. A pocket calculator neither knows nor cares about the
"grounding" of the number system in the real world. It is simply
following instructions and data entered externally. How precise or
imprecise the data may be really has no impact on the grounding
problem.
Hope you are having a great day!
Patrick
Thanks in advance for comments.
-- Adrian Walker
[1] A reference to NP-completeness, the key open problem for nearly all
of complexity theory.
|
|