[Top] [All Lists]

Re: [ontolog-forum] Visual Complexity

To: paola.dimaio@xxxxxxxxx
Cc: "[ontolog-forum]" <ontolog-forum@xxxxxxxxxxxxxxxx>
From: Pat Hayes <phayes@xxxxxxx>
Date: Tue, 30 Jan 2007 12:24:19 -0600
Message-id: <p06230905c1e52c021c6a@[]>
>I opened a page, although I am reserving some rights there
>so please feel free to evolve (for progress' sake)
>Look forward to contribution    (01)

Some comments. Overall, as stated, it is an odd 
combination of asking for technology which is 
already routinely deployed, and asking for the 
impossible. Details.    (02)

1. I really don't think it makes sense to ask for    (03)

"a ... set of agreed terms... that embodies and 
represents and synthesizes all available, valid 
knowledge that is deemed to pertain to a given 
domain"    (04)

It is the 'all's here that make this impossible. 
One can never get ALL the available, valid 
knowledge about anything. One can only hope to 
get a workable amount, and attempt to keep it 
unpolluted by falsehood and reasonably up to date 
and so forth. Be less ambitious, and we might 
have a hope of realizing the dream.    (05)

Second, though, what does it mean to say that a 
set of terms - what I would call a vocabulary - 
can "embody" knowledge? Terms are just, well, 
terms. The knowledge is represented by larger 
structures - axioms, sentences, diagrams, texts, 
ontologies, topic maps, whatever - which 
themselves contain and use the terms and, in the 
final analysis, give the terms meaning. 
Identifying a set of terms is just a first - in 
practice, indeed, often a preliminary - stage of 
representing useful knowledge, not the final goal 
or endpoint.    (06)

2. Just an aside, but this sentence seems to 
indicate a misunderstanding about how ontologies 
are actually built these days:    (07)

"Among the barrier to adoption for Ontology, 
current research identifies not only different 
linguistic, conceptual and cultural differences, 
but also knowledge and point of view differences 
that set apart academics  who generally develop 
ontologies and related tools and methodologies  
from experts  who understand lingo and the 
dynamics - system developers  programmers, 
systems designers and end users at large."    (08)

Both real ontologies and ontology standards such 
as RDF and OWL are actually mostly the product of 
collaborations between teams which are dominated 
by 'experts' and 'system developers' rather than 
'academics' (although we dwellers in the ivory 
towers do play a marginal role here and there.) 
Some of the most widely used ontologies are 
entirely the work of 'domain experts'. In any 
case, the boundaries between the academy, 
business enterprise and 'end users' such as 
medical researchers, the intelligence community 
or weather forecasters is increasingly blurred 
and indeterminate, and people move back and forth 
across it with ease. So I think this particular 
'barrier' is a figment of your imagination, 
frankly.    (09)

Back to the details.    (010)

3. You want GPL or public licencing. But semantic 
web ontologies are just like Web pages: they are 
open to all. You can copy them using HTTP. Why do 
you think that licencing is even an issue on the 
Web?    (011)

4. Ontologies should "declare what high-level 
knowledge it references". Again, this is a 
non-issue. By design, OWL ontologies may 
reference ("import") other ontologies, and these 
references are part of the ontology, by 
definition. So yes, of course they "declare" in 
this way. Do you have some other mechanism in 
mind? (The "named graph" proposal allows 
ontologies to make explicit assertions about 
other ontologies, such as agreeing with it, 
disagreeing, basing itself on it, warranting the 
truth of it, etc.. ; is this what you have in 
mind?)    (012)

5. It should "declare what kind of 
reasoning/inference supports/it is based on" . 
Again, a non-issue. This is like asking that a 
bridge should have a label on it saying what kind 
of bridge it is. Of *course* any ontology will be 
written in some language which supports some 
kinds of inference. That is why such language 
specifications include a detailed semantics. 
Given this, then, what this amounts to is that 
the ontology should identify what language it is 
written in. Which is a good idea, but again a 
solved problem, so a non-issue, at least if it is 
written using XML; since the XML spec provides 
for just such declarations using the XML header.    (013)

6. It should "support queries via natural 
language as well as machine language" Whoah 
there. Supporting queries in natural language is, 
at the present time, close to science fiction. At 
best it is a research ambition which is at the 
cutting edge of AI research. And in practice, it 
doesn't work very well (ask CyCorp about their 
experiences.) It is, in any case, well beyond 
what it is reasonable to ask of any kind of 
standardized protocols. This is way too ambitious.    (014)

(By the way, what exactly do you mean by "machine 
language" here? Do you mean formal language? 
Humans can learn to use formal notations.)    (015)

7. "It should be 'easy to understand' by generic 
users without specialized skills"  Again, way too 
ambitious. I'm not sure it even makes sense. If 
you can't read or understand L, you won't be able 
to read a text written in L. This seems obvious 
whether L is English, Spanish or OWL. Is having a 
grasp of Spanish a 'specialized skill'? 
Personally I find OWL easier than, say, Russian.    (016)

But the central point is that an ontology, by its 
very nature, is ultimately a text written in some 
language; and so to understand it, you have to 
know that language. (And to ward off possible 
misunderstanding, I'm here using 'language' 
broadly to include, eg, map-making and 
diagrammatic conventions; so that for example 
circuit diagrams or flowcharts or social networks 
displayed as graphs are all kinds of language. 
The basic point still applies.)    (017)

So trying to draw a contrast between 'generic 
users' and 'academics' or whatever isn't helpful, 
seems to me. What might be more use is to ask, 
how long does it take to learn the relevant 
language? Can we find ways of displaying 
ontological content to make it easier to learn? 
(We have been trying to do this in the COE system 
for OWL, for example, and VivoMind are focusing 
on CLSE 'structured English'. But you still have 
to learn to use COE - it takes about a day - and 
its a lot easier to read CLSE than to write it.)    (018)

8. "It should include instructions on how to 
relate such 'high level knowledge' to standard 
knowledge representation artifacts used in 
software and systems engineering, such as 
entities, attributes, classes, objects, 
properties, sub-properties, values and 
relationships"    (019)

Well, sure. But take a look at OWL: it is all 
about entities (it calls them 'resources', 
following a W3C nomenclature), classes, 
properties, values and relationships. It has 
owl:subProperty as one of its primitives.  What 
could be more clearly related to your list of 
terms?  Its primary documentation includes a 
tutorial. What more do you want?    (020)

But to be less rhetorical for a second, there is 
a danger lurking here. The class/property/value 
terminology is also used in OO programming, and 
although they two uses are closely similar, there 
are important distinctions which need to be got 
clear. Ontology class hierarchies need not be 
taxonomic, for example. So the ontology world is 
not the same as the software/systems engineering 
world, but it is close enough for the 
similarities to sometimes obscure the differences.    (021)

9.  "It should be implementation independent; 
this means not only usable by OWL/DAML model but 
also reusable by alternative ontology languages"    (022)

What does this even mean? An OWL ontology (for 
example) isn't "usable" by the "OWL model", it is 
*written in* OWL. It *is* an OWL text. There is a 
BNF grammar for such things, so a machine can 
check if it is legal OWL or not. Similarly, an 
ontology written in, say, Common Logic is a CL 
text. There isn't such a thing as a 
language-independent representation of a language 
text. (Even diagrams need to be parsable if they 
are to be used by machines.) So this requirement 
seems to me to be nonsensical as it stands.
What might make sense is to require that ontology 
languages be inter-translatable. I agree, and 
indeed a great deal of my effort over the past 
few years has been towards bringing about this 
desirable state of affairs, culminating in the 
IKL notation. But this is a requirement on 
ontology language standards, not on ontologies. 
And it is basically impossible to enforce, other 
than by exhortation, since what can stop a 
person, company or working group from inventing a 
new, incompatible, notation? (As for example the 
business rules community recently did, with a 
standard notation SBVR based, regrettably, on 
modal logic.)    (023)

10."it should support one view of the world if 
required, and allow for simultaneous multiple 
views, meaning that it should aim to be perfectly 
elastic, flexible and adaptable,"
I'm not sure what this means, but it sounds 
either trivial or impossible. Clearly, an 
ontology may be written with one world-view in 
mind, as it were, and so make distinctions (or 
fail to make them), assign classifications, 
quantify over a universe, etc., based on this 
'world-view', that of its composer. In fact, it 
is difficult to see how this can be avoided, or 
what it would mean for it to be false. That is 
the trivial sense. More ambitiously, one might 
claim that an ontology *completely defines* one 
particular world-view, in the sense that it says 
so much that it allows for only one possible 
interpretation (this is obviously related to the 
first point about containing all the information 
about some topic). And this is impossible; in 
fact, in most cases, *provably* impossible.    (024)

It may be that what you mean here is something in 
between these two extremes, but in that case I 
don't know what it is that you do mean, and some 
more exposition would be very useful. (Experience 
with being on committees writing standards has 
shown that loosely written or imprecise 
requirements can be very harmful to progress, as 
they fail to serve their most useful purpose: to 
limit endless discussion about what exactly 
people mean, and focus effort on getting 
technical issues clear.)    (025)

--------    (026)

In spite of all the above, I think it is a very 
good idea to try to get the requirements clear, 
and that this is a good start: thanks!    (027)

Pat Hayes    (028)

IHMC            (850)434 8903 or (650)494 3973   home
40 South Alcaniz St.    (850)202 4416   office
Pensacola                       (850)202 4440   fax
FL 32502                        (850)291 0667    cell
phayesAT-SIGNihmc.us       http://www.ihmc.us/users/phayes    (029)

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    (030)

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