[Top] [All Lists]

Re: [ontolog-forum] Current Semantic Web Layer pizza (was ckae)

To: "[ontolog-forum]" <ontolog-forum@xxxxxxxxxxxxxxxx>
From: Ed Barkmeyer <edbark@xxxxxxxx>
Date: Wed, 05 Sep 2007 19:15:24 -0400
Message-id: <46DF388C.9040408@xxxxxxxx>
At the risk of adding nothing to this thread,...    (01)

Duane Nickull wrote:    (02)

> Semantics is always contextual.    (03)

Pat Hayes said:
> Nonsense. Fortunately; because if it were true, semantics would be impossible.
> This assertion is like the idea that all
> assertions are contexual. It sounds plausible,
> but that 'all' makes it (literally) incoherent.    (04)

Duane said:
> In general, I suspect I should have not used the "always" word but I still
> believe that the context in which an instance of something occurs has a
> great deal of impact on the semantics (what the thing means), albeit perhaps
> not in 100% of instances.  My grammar is generally bad.    (05)

First rule: when discoursing with logicians remember that "all" 
and "always" are reserved words.  ;-)  In logic, "not 100%" is 
the *opposite* of "all".  A single counterexample is a disproof.    (06)

More to the point, whether the context of an utterance is 
necessary to its proper interpretation or not is, IMO, part of 
the "semantics" of that utterance.    (07)

If, for example, the intended interpretation of "all" refers to a 
specific universe, which is the/a universe of the discourse in 
which the statement is made, surely the reference to that 
universe is part of the semantics of "all" in that usage.    (08)

I think of "pragmatics" as an element of interpretation that is 
part of determining the intended semantics of the utterance. 
"Pragmatics" determines that the semantics of the utterance 
includes some unstated "antecedents" that were part of the 
context of utterance and implicitly part of the statement.  And 
that may mean that the whole proposition that was the intended 
assertion is true, even though the pure out-of-context semantics 
of the utterance per se is a false proposition.    (09)

OTOH, when a statement is deliberately bald and intentionally 
universal, then the semantics of that statement *by intention* 
requires no contextual constraints.  The intended meaning of that 
statement out-of-context is the intended meaning, and the 
corresponding proposition stands on its own -- it is either true 
or it is false.  (It may in practice be unknown or unknowable, 
which has nothing to do with whether it is true or false.)    (010)

The fact that the speaker didn't have the breadth of knowledge to 
cover certain counterexamples well-known to others, doesn't make 
the statement "true in some context we can presume for the 
speaker".  It makes the proposition he *intended* false.    (011)

> Semantics cannot exist without including the aspect of context.    (012)

> That is simply obviously false, since it does so
> exist and has done for maybe 70 years now.    (013)

> Perhaps I worded this wrong.  Context is perhaps part of inference.      (014)

I would say this was worded wrong twice.  Context may be part of 
determining the semantics of an utterance.  Context, in the sense 
of additional knowledge, may of course support inferences one can 
make from postulating the proposition that is determined to be 
the semantics of the utterance.  But that is "quite a different 
thing entirely".    (015)

> This word "context" has been the cause of more
> wasted time in KR than almost any other. ...
> There are many, many problems with the idea that
> ontology meanings are contextual.    (016)

Amen.    (017)

> Disagree that it is a complete waste of time given there are important
> lessons learned.  Several papers and efforts have declared that it is an
> open ended rathole and computationally too expansive to contemplate.     (018)

Apparently I need more of Duane's presumed context to interpret 
this statement.  Pat didn't say it was a complete waste of time; 
he said it was a cause of wasted time in KR research.  Context is 
obviously an important consideration in interpreting natural 
language utterances into formal knowledge.  The waste arises from 
trying to drag "context" into the captured knowledge.  If you 
understand what was meant, you can capture that formally without 
a "context" crutch.  If you don't understand what was meant, then 
it is not clear what value you will extract from that lack of 
knowledge, no matter how pretty the "context" clothes you dress 
it in.    (019)

> Even
> the UN/CEFACT CCTS project which had a mere 8 context modifiers in its
> framework could potentially have a very large number of semantic values.  It
> is not even known at what level of granularity to determine a difference.
> The CCCTS work could have 8 modifiers, each with values of up to 3000
> enumerations or an extremely large number.    (020)

CCTS is a wonderful example of the misuse of the term 
"semantics".  I am reasonably certain that the authors of ISO 
15000 do not agree on what the term means, and that most of their 
individual definitions of it do not proceed from academic study. 
  It's simply the buzzword used to label whatever level and kind 
of abstraction is done to data elements.    (021)

The idea that the context of use of a *term* can be identified by 
some set of domains of concern -- language, industry, discipline, 
etc. -- makes sense.  It coincides with the ideas for 
characterizing a "speech community", and therefore its 
"vocabulary", in ISO 1087 and other TC37 standards.    (022)

But then CCTS had to tie the bell on the cat's neck and list the 
possible domains of concern.  And in that area, the CCTS approach 
is to standardize a vector of possibly orthogonal notions of 
"domain of concern" some of which are known to have instances 
that do characterize speech communities.  The upper ontology that 
characterizes the vector, however, and the research that 
demonstrates its validity, completeness, or even usefulness is 
sorely lacking.  (First the execution, then the trial.)    (023)

Yes, we can use the vector to label a context
  {n/a, Germany, automotive industry, n/a, n/a, n/a, n/a, n/a},
but for the "n/a"s, it makes a difference whether I use "all", 
"don't know" or "don't care".  In practice, every n/a will really 
be "don't know", and even the ones that get filled in will be 
presumptuous.  After we get past "date" and "address", the 
context labels will just overcharacterize a specific group of 
firms that agreed to use the same term (XML tag) with the same 
meaning.    (024)

> In general, I have no idea how to solve this issue but have worked enough on
> XML artifacts to know that semantic declarations without context do not work
> for all purposes.    (025)

I offer this statement as evidence for the position I take above.    (026)

> An example - imagine a data element named "FirstNameOfParty".    (027)

An enlightening data element to be sure.  Its semantics is that 
it is a conventional part of a conventional referencer that may 
or many not refer to a unique thing.  The complete referencer 
may, however, be presumed to name a unique thing in the context 
of use (which is much narrower than the nominal context of 
definition of the 'term' "FirstNameOfParty").    (028)

> An instance
> is slightly different if it occurs within a hierarchy of
> //PurchaseOrder/BuyerParty vs. //PurchaseOrder/SellerParty even though the
> semantics of the lone data element do not change.    (029)

No, actually, an instance ... is NOT "slightly different".
The semantics of the data element, such as they are, do not 
change, ever.  The two examples use terms that denote entirely 
different roles in a presumed archetypal business process that is 
labelled by a participating artifact called "PurchaseOrder". 
Each of those roles is always played by a Party.  In a given 
instance of that business process, each of those roles is played 
by a specific instance of Party, and so we come to need a 
referencer that uniquely identifies a Party instance, which takes 
us to a concept called PartyName (whose semantics is purely: 
"agreed upon identifier for a specific Party instance") and then 
to a conventional component of that identifier.    (030)

The relationships:
  Process(archetype) defines Role(archetype)
  Role is played by Party
  Party is identified by PartyIdentifier
are part of pretty much every ontology for Process.  The specific 
case 'PurchaseOrder' Process defines 'BuyerParty' Role is an 
instance of the pattern, which produces classifiers for actual 
Process and Role instances:
  PurchaseOrder is-a-kind-of Process
  BuyerParty is-a-kind-of Role
  PartyName is-a-kind-of PartyIdentifier.
I would have said this was a set of agreed-upon terms for 
relatively well-defined concepts.    (031)

The contexts involved here are many:
  - the business process archetype that gives rise to the roles
  - the speech community that agreed to call the process 
archetype "PurchaseOrder" and the roles "BuyerParty", etc.
  - the speech community that agrees on the mapping of PartyName 
instances to Party instances, for the PartyName instances 
actually used in the discourse (and not necessarily for ANY others)
  - the relationship of the latter speech community to the 
business process instance in question.    (032)

The only part of this "context" that the CCTS vector might be 
able to characterize is the first speech community.  In 
particular, the community that agrees to the name-to-instance 
mapping for Parties might consist of only the players of the 
roles in the process instance.    (033)

> The semantics of each data element were declared in an attempt to allow
> cognitive applications to automatically suggest mappings from one format to
> another (in this case EDI to CBL).  The UBL effort immediately recognized
> that context is utterly essential to the process as the context in which
> something occurs affects many aspects of its semantics.    (034)

I can't argue with this, because I have no idea from what 
dictionary the terms "context" and "semantics" in this paragraph 
are drawn.  But I can argue with some authority that "the context 
in which something occurs" typically involves a great many more 
domains of concern and communities of speech than anything 
captured in UBL.    (035)

Consider what happens when an American automotive manufacturer 
buys electronic parts from a Japanese firm to be delivered to an 
assembly plant in Mexico.  You have referencers from multiple 
language communities, firms and business units, addresses from 
three postal systems, and subject matter from automotive, 
electronics, logistics and finance, at least.    (036)

And this is why it is really all about the set of people who 
agree to communicate, and what they agree to.  At best the 
context labelling assists these people in finding a set of 
information labels they might be willing to agree to use, with 
more or less the definitions given.  The actual interpretation of 
every information unit will be in terms of the process (and maybe 
the process instance) and the role players involved.    (037)

-Ed    (038)

P.S.  I should be careful to say that I think CCTS has merit.  I 
have been calling it the "standard lower ontology" approach to 
interoperability.  But its champions do the project no credit by 
alternately dropping buzzwords and arcana.    (039)

Edward J. Barkmeyer                        Email: edbark@xxxxxxxx
National Institute of Standards & Technology
Manufacturing Systems Integration Division
100 Bureau Drive, Stop 8263                Tel: +1 301-975-3528
Gaithersburg, MD 20899-8263                FAX: +1 301-975-4694    (040)

"The opinions expressed above do not reflect consensus of NIST,
  and have not been reviewed by any Government authority."    (041)

Message Archives: http://ontolog.cim3.net/forum/ontolog-forum/  
Subscribe/Config: 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 Post: mailto:ontolog-forum@xxxxxxxxxxxxxxxx    (042)

<Prev in Thread] Current Thread [Next in Thread>