Steven, Rich, Gregg, Pat C, Matthew, and Joel, (01)
I'll start with points by Steven, Rich, and Gregg, which cover
the basic issues. Those issues would clarify the discussions
with Pat about primitives, definitions, and ontology. (02)
SEZ
> A minimum core set of ideas has little to do with language and
> has much more to do with mathematics. (03)
I strongly agree that we need to distinguish the formal ideas
(logical and mathematical) from the content (the subject matter). (04)
SEZ
> It's not that words hold any intrinsic "meaning," it is that words
> produce a range of biophysical behaviors. It is these behaviors, and
> by them their "grounding," that may rightly be described as "meaning." (05)
I agree, but I prefer Peirce's phrase "perception and action" to
"biophysical behavior". Consider a medical ontology for physicians,
nurses, pharmacists, hospital administrators (managers & staff),
accounting, insurance companies, building & grounds personnel,
first responders, and patients. (06)
All those people have to communicate, but they associate different
perceptions and actions with each term. And they may use different
words in different syntactic patterns for the same referents. (07)
RC
> On slide 21 [of http://www.jfsowa.com/talks/goal3.pdf ], you describe
> Peirce's rules of FOL... Do you agree that... [1]English rewriting
> systems can perform full FOL? [2] What would keep it from full FOL? (08)
Short answers: [1] Yes. [2] Nothing. (09)
Longer: Every aspect of any version of math and logic can be
expressed in any NL. The issues are complex, and I suggest the
following for starters: http://www.jfsowa.com/pubs/eg2cg.pdf (010)
GR
> Have you looked at Natural Semantic Metalanguage [NSM]?
> http://en.wikipedia.org/wiki/Natural_semantic_metalanguage (011)
I have a high regard for the work by Anna Wierzbicka and her students
and colleagues. They have done important *empirical* studies of the
semantics across multiple languages, which they express in a notation
(NSM) based on about 60 "semantic primes". Basic issues: (012)
1. AW said that NSM is *not* a logic, and I agree. I once tried to
map her notation to FOL. But any such mapping breaks down for
reasons #2, #3, and #4 below. (013)
2. Nearly all NSM primes are "accordion words" that can be stretched
and squeezed to fit an open-ended range of applications. Examples:
LIKE (for any and all varieties of similarity); SAY, WORD, TRUE
(for any verbal communication); GOOD, BAD (evaluators)... (014)
3. The primes are mapped to the simplest syntax and core vocabulary
of every language they analyzed. The "symbol grounding" of those
primes is the human body with its modes of perception and action. (015)
4. But every one of those primes can be and has been analyzed in much
greater detail in the past few millennia of philosophy, science,
logic, and math. In fact, any language that has a long literary
history has many words for expressing much finer distinctions. (016)
5. Nevertheless, I believe that it would be a useful exercise to
define other proposed "primitives" in terms of NSM. (017)
PC
> if one wants to reproduce human-level language understanding one has
> to remember that people don't understand each other perfectly - even
> those who are well acquainted with their native language, and
> particularly when emotions are engaged and debating points are being
> made. The computer doesn't have to be perfect, just as good as people. (018)
On that point, we agree completely. (019)
PC
> For natural language, we expect inaccuracies, even among people
> using the same language. (020)
That is an understatement. Every word in every NL is an accordion word.
Tokens of the "same" word in two different documents typically have
different "microsenses". Even professionals in lexicography and the
subject matter *frequently* disagree on more than just microsenses. (021)
PC
> but a finite set of well-specified ontology elements **can** have
> a "fixed, precise set of mappings" to a finite set of databases,
> the meanings of whose elements are determined by the operations
> and goals of their applications. (022)
NO!!! The best you can have is a *loose* mapping -- except in the
very rare case when *every* application is explicitly designed to
exactly the same precise specifications and *every* update to *every*
application is thoroughly tested for compliance with those spec's. (023)
PC
> Of course, people stretch meanings of words in new situations... (024)
Everybody stretches every meaning of every word in every situation. (025)
PC
> We have to explain new uses to other people, as well as to our
> computers, by providing clarifying definitions. (026)
Have you noticed the *absence* of manuals for the computers and
software you buy? Almost nobody reads definitions. Those who
do rarely remember them. And those who remember them frequently
make mistakes. (027)
PC
> I will be very disappointed if, after a hundred applications have
> been mapped with the ontology, *every* new application still
> requires some new primitives to be added. (028)
I believe that you can define every word of English in the COSMO
terms. I also believe that you can define every COSMO term in
Wierzbicka's NSM. But that is only because every term in NSM,
COSMO, and English is an accordion word can be stretched and
squeezed to fit any application that may arise. (029)
MW
> So you need to prove that there is a number of primitive concepts,
> n, where n is (a lot) less than infinity, such that it is not
> possible to come up with some useful set of all these objects that
> is not the intersection of some of those n concepts.
>
> It does not seem credible to me that there is any such number,
> but I look forward to seeing any proof that such a number exists. (030)
Spatial patterns are extremely difficult to define in words.
Dictionaries usually include a picture (for plants and animals)
or a diagram (for artifacts). (031)
Colors, odors, and feelings are impossible to define in words --
except by vague comparisons or by accordion words that are just
as vague. For examples, read the descriptions on the back of
a wine bottle and ask somebody to identify the wine based on
such a description. (032)
MW
> As long as you do not equate words and phrases with concepts
> (even though it is concepts they are describing the intended
> interpretation of) then I think you are OK. (033)
I agree. But I would replace the word 'concept' with some
term in logic, such as predicate, relation, type, or class. (034)
JLC
> Stevan Harnad... was the first author that has explicitly
> discussed the so called "symbol grounding problem". (035)
Aristotle recognized that issue and solved it quite well.
He didn't name the problem because it wasn't a problem. With
his semiotics, Peirce developed the solution in greater depth.
For a summary, see http://www.jfsowa.com/pubs/rolelog.pdf (036)
By the way, the term 'accordion word' is not one I invented.
With a bit of googling, I found that Wilfrid Sellars (1969)
had used it without suggesting that it was novel. Its origin
is lost in the mists of pre-WWW epochs. (037)
John (038)
_________________________________________________________________
Message Archives: http://ontolog.cim3.net/forum/ontolog-forum/
Config Subscr: http://ontolog.cim3.net/mailman/listinfo/ontolog-forum/
Unsubscribe: mailto:ontolog-forum-leave@xxxxxxxxxxxxxxxx
Shared Files: http://ontolog.cim3.net/file/
Community Wiki: http://ontolog.cim3.net/wiki/
To join: http://ontolog.cim3.net/cgi-bin/wiki.pl?WikiHomePage#nid1J (039)
|