Subject:
Re: SUO: Re: One Stone Fits All
From:
Stefano Borgo <stborgo@indiana.edu>
Date:
Thu, 17 Jul 2003 19:39:45 -0500
To:
"John F. Sowa" <sowa@bestweb.net>
CC:
standard-upper-ontology@ieee.org, nicola.guarino@ladseb.pd.cnr.it


Dear John,
it looks like you are inviting me to a party but don't put much meat on the grill...

There are two ways to look at your arguments. On the one hand, I find your seven points quite weak (if applicable at all) from the perspective of DOLCE. On the other hand, these have some relevance to the whole SUO project. I'll try to answer from both the perspectives labeling them [DOLCE] and [SUO], respectively.

These should read as follows:
[DOLCE] -> Stefano's answer from the point of view of DOLCE
[SUO] -> Stefano's answer in the context of SUO.


> Following are some aspects of the DOLCE ontology, which
> I believe are seriously flawed:
>
>  1. The notion of identity, which implies that different
>     ways of characterizing the same spatiotemporal extent
>     reflect a multiplicity of collocated entities -- e.g.,
>     a vase and the clay of which it is composed are two
>     distinct entities that occupy the same space.  I realize
>     that the problems of identity are extremely difficult,
>     and for that reason, I believe that identity should be
>     treated as something inferred, not as something considered
>     as fundamental as DOLCE tries to make it.  Similarity is
>     somewhat less problematical than identity, and I would
>     treat identity as something inferred from multiple
>     observations of appropriate similarities.


[DOLCE] This shows that you have a different bias, that's all. Unless you mean that our treatment of identity is formally contradictory (which I doubt very much).

[SUO] It seems you are confusing ontology and epistemology. The very idea of using observations to establish identity cannot be but an epistemological stand. Furthermore, on the logical side, I don't see how you can claim that similarity is more fundamental than identity. Do you really believe the latter is obtained from the first? Please, explain to me how one can formally characterize similarity without *any* notion of identity (neither explicit nor implicit).

>  2. The terminology of universals and particulars, which has
>     a long and honorable history in philosophy, but which has
>     had far too many conflicting interpretations to be used
>     in a serious ontology project today.  As an alternative,
>     I would recommend the terms used in logic:  talk about
>     predicates and the instances for which a predicate may
>     be true or false.  But even then, I believe you should
>     make a three-way distinction between the the symbols
>     of the logic, the models in terms of which they are true
>     or false, and the aspects of the world for which the
>     models are good, fair, or poor approximations.  See the
>     attached diagram, mthworld.gif.


[DOLCE] You must be joking! We are not talking about a 2-line email here. Please, read the paper about DOLCE and you'll find that we state what the terms mean. All you need to do is to use an algorithm very similar to something *you* already described to the SUO list:
BEGIN
1) substitute universal by DOLCE-universal and particular by DOLCE-particular (and so for other terms you don't like)
2) read the DOLCE paper looking for the clarification of their meaning
http://www.isib.cnr.it/infor/ontology/Papers/DOLCE2.1-FOL.doc
3) if this is not enough, look at the cited references that further clarify these meanings.
END (of the algorithm)

Regarding the three-way distinction between symbols, models, and the aspects of the world.
The syntax  is given fairly explicitly, the semantics is the standard semantics, the aspects of the worlds are described in the introduction (and the description is even too detailed when considering most of the ontologies available nowadays). What do you want more? Can you give me a detailed example?

[SUO] This is an old issue in SUO. I understand that there is a general agreement to do something along the following line: each term is enriched with a prefix that points to the ontology describing the intended meaning in the given context. There has been some discussion on this already and you gave some suggestions in this sense too. I thought this solved the problem (although there might be some doubt about how to implement this idea).

>  3. The lack of a serious semeiotic foundation (along the
>     lines of Peirce's work).  I have no quarrel with people
>     who might claim that they are somehow going "beyond"
>     what Peirce did, but only if they can show that they
>     have carried his analysis to a deeper level.  In the
>     case of DOLCE, however, I don't believe they have even
>     scratched the surface.  An important aspect of Peirce's
>     semeiotic is his theory of truth, which treats it as
>     the end result of a process of approximation, along the
>     lines of the diagram mthworld.gif:  any theory is true
>     only in terms of some model, which is a more or less
>     adequate approximation to some aspect of the world
>     for some specific purpose.  As the statistician
>     George Box said "All models are wrong, but some are
>     useful."   That is a good summary of pragmatism.


[DOLCE] If you want something along the lines of Peice's work, you are perhaps the best person to do it. Why do you ask people at LOA to do the job for you?
Anyway, now I get a better idea of what your lattice of theories is about. It seems to include only theories based on Peirce's ideas since nothing else is worth according to you.

Why weren't you clear on this last month at the time of voting for your motion?

[SUO] Disregarding the religious claims, it might be important to evaluate how candidate ontologies treat semiotic aspects. You might want to give a list of distinctions (and their characterizations) that could/should be used to evaluate ontologies from the semiotic point of view. Of course, one needs to show the relevance of these very distinctions (this should not be hard though).

>  4. The assumption that DOLCE is limited to a "commonsense" or
>     midworld or mesoscopic view.  I would emphasize that there
>     is no common agreement whatever on what constitutes common
>     sense.  One generation's science is the next generation's
>     common sense.  Two neighbors or co-workers with different
>     educational backgrounds may have radically different notions
>     of common sense.  And there is a continuum between science
>     and whatever common sense might be:  two scientists can
>     freely move from one level to another within the same
>     conversation or even the same sentence.


[DOLCE] Substitute "commonsense" with "reality" and your criticism goes as well. By your own argument, we should not try to describe reality (or what we like to think reality is at this point in time), right? Then, what is exactly that we are doing in SUO?

[SUO] Of course there is no common agreement on what constitutes commonsense. However, we want to provide ontologies for people working in the information era. Thus, we should try to provide ontologies that can handle the descriptions people do use in this world.
Whatever comes out (if any), can be taken to be an ontology that describe (aspects of) "commonsense".
Commonsense is not different from the notion of "average man" used in law (I'm not sure this is the right translation in english, anyway I'm sure you get my point). Have you ever seen the "average man" walking downtown? But without this very notion you cannot make sense of the civil and social laws in our society. If you don't want to get dirty analyzing notions like commonsense and average man, how can you hope to understand what depends on them?
Concepts we use in everyday life (commonsense concepts) are hard to deal with. Often one has to add subtle distinctions to get a consistent theory. But this clarification of the concepts is what (I think) helps *real* people, the same people we are building the ontology for.

>  5. The sharp distinction between continuants (or endurants) and
>     occurrents.  Some things like diamonds are more enduring, and
>     other things like flames are more occurring.  But most things
>     in the world -- including living organisms -- are somewhere
>     in between.  For many things, like tornadoes, glaciers, and
>     even rocks, the difference between an object view and a
>     process view is one of purpose, perspective, and time scale.


[DOLCE] No problem with that. You tend to forget that DOLCE is not the *true ontology*. Nobody claimed that! It is biased and it says how. If you need to describe a perspective that does not fit the DOLCE assumptions, don't use DOLCE.

[SUO] This distinction makes sense in some context. Not always. That's why we need more than one ontology. At least, this seems the way to go at this point of the SUO enterprise. We may find a better solution on the way to paradise.

>  6. And most importantly, the absence of any treatment of purpose
>     in DOLCE.  No classification of any kind -- not even ontology --
>     is possible without recognizing purpose.  Whether a vase is
>     considered to be identical to a lump of clay depends on your
>     purpose.  When a physician is describing a disease, the
>     terminology depends on whether he or she is talking to another
>     physician or to a patient who has never studied medicine.
>     Whether a glacier is considered an object or a process depends
>     on purpose.  The applicability of any ontological category to
>     any aspect of the world depends on purpose.


[DOLCE] At a closer look, here we are dealing with two distinct issues.
One issue is the notion of purpose described by the ontology at stake.
The other is the purpose an ontology has been built for (which is related to the bias taken in building the ontology.)
On this latter issue, I agree that the purpose should be explicitly stated. Indeed, the purpose of DOLCE is given by describing its bias. Something we have spelled out as much as we could. So, this point of yours is not a criticism, on the contrary. We have a theory (ontology) which we think is a useful description of reality (whatever that is) based on premises that some people take for granted. If you don't share those premises, you don't have the same purpose and then you should look for another ontology. It is that simple!
Regarding the notion of purpose within DOLCE, you are right. This notion is missing and I'm glad you put this issue on the table. There are several areas that are not covered in DOLCE and this is one of them. Since it is connected to several other concepts that are not easily captured, this extension requires some time. We are working on it.

[SUO] Keeping the distinction above, I tend to agree with you on both issues.
On the one hand, purpose seems to be an important concept and any upper ontology should say something about it.
On the other hand, it is important to find out the bias of any ontology one might wants to consider. In particular, I believe this is mandatory for those that are part the SUO project. The recent discussion that pointed out the philosophical stand of SUMO was very interesting. I'd like to see some more discussion on this aspect about OpenCyc. For what I could follow, Mathew West, Patrick Cassidy, and Chris Partridge have been asking a lot of important questions to move forward in this project.

>  7. Finally, there are better foundations for ontology, which I have
>     been telling Nicola about for years and which he has consistently
>     ignored.  As I said in a recent note, Whitehead did a masterful
>     job of defining an ontology that reconciles the language of modern
>     science (e.g. relativity and quantum mechanics) with the language
>     of everyday speech.  And I also believe that Peirce provided the
>     most outstanding treatment of semeiotic in relating language, logic,
>     and the world.  I made an attempt to synthesize their work in my
>     Knowledge Representation book.  I won't claim that I got it right.
>     But I do claim that anyone who tries to do ontology without coming
>     to grips with the work of Peirce and Whitehead is doomed to fail.


[DOLCE] I thought we agreed that better and worse is based on purpose.
The people at LOA believe that what is needed is a plurality of ontologies (not infinitely many though, you know, about your lattice...) with a deep analysis on how they relate.

[SUO] I personally welcome other initiatives based on Peirce, Whitehead, or the Bible. If you like to see this as a competition among schools of thought, be my guest. I couldn't care less.
I think some ontology will turn out to be better for some purposes but inevitably worse for others.
All I'm asking is the following: let's judge the theories on how good they perform for the intended tasks and not on how much they match our preferred teachers.

> JS> This, by the way, leads to one of my major criticisms of DOLCE.
>
>>> They use an unqualified modal operator to mark what they claim is
>>> "essential" to some type.  As I have pointed out many times, every
>>> claim of necessity rests on some implicit "law" that makes something
>>> necessary or essential.  Instead of an unqualified "essence marker",
>>> we should require an explicit statement of some law (or axiom) that
>>> states the principle for whatever essence is being assumed.
>>
>>
> SB> At first, it seems you are criticizing DOLCE because it does not give
>
>> the laws characterizing the modal operator. On this issue, you might want to reread section 2.1 of the paper WonderWeb Deliverable D17 (where DOLCE is introduced and motivated):
>> http://www.isib.cnr.it/infor/ontology/Papers/DOLCE2.1-FOL.doc
>>
>> Here you can verify that DOLCE has been developed primarily as an ontology of particulars.
>> The essence of a type (and universals in general) is *not* something DOLCE aims at describing (at least at this point in time.) Let me cite a small part of this section that foresees your argument (pp. 8-9):
>
>
> As I said above, I believe it is a mistake to use the terminology of
> universals and particulars.  The end result of DOLCE (or any other
> ontology we may be discussing) is a collection of axioms and definitions
> of some predicates (or relations) stated in some version of logic.
> What I was criticizing is the version of logic and the use of
> modal operators in stating axioms and definitions.


One of the points in your paper is that these systems are inter-translatable. Now, you are complaining about the choice of the logic itself. I can follow you when you talk about laws, this other issue seems contradictory to me (at least taking for granted the results of your own paper, see my comments below).

> SB> Perhaps, instead of criticizing the "unqualified" modal operator, you
>
>> are criticizing this very restriction that we assume in DOLCE ? In the formalization, we list a minimal set of axioms for the modality (S5 plus the Barcan formulas). This allows people to take DOLCE, characterize the modality further according to the laws they like, and use the resulting system for their applications. In other terms, DOLCE is not (and it has never been presented as) a complete upper ontology. You might consider it a *framework* for such an ontology.
>
>
> You have just touched on one of my very hot hot-buttons:  System S5
> plus the Barcan formula is definitely *not* minimal:  it is a
> ridiculously maximal version of modal logic that has no application
> to anything in the real world.  I realize that many logicians love
> S5 plus the Barcan formula because it is so strong that it enables
> them to prove lots and lots of theorems.  But those theorems are
> pure fantasy -- they don't apply to anything.


It is minimal because it does not include natural, biological, physical laws and the like. To get the right model you should (at least) specify any extra law you consider correct for your purposes.
As I said before, on this regard DOLCE should be considered a framework that you (or anybody else) can adapt to your conceptualization of the world.

Regarding your bold claim that S5 + BF is too strong and applies to nothing, I'll wait for some decent argument before even considering discussing it.

> In terms of the laws and facts semantics (which is equivalent to
> Kripke's possible worlds), S5 implies that every possible world
> has exactly the same set of laws, and the Barcan formula implies
> that all worlds have exactly the same entities.  What that implies,
> depends on who makes the laws:


Have you noticed that there is a temporal parameter in DOLCE and a existence predicate? (ops, I meant DOLCE-predicate). You might want to look again at section 3.1 in DOLCE and at Fitting and Mendelshon book "First-Order Modal Logic" (btw, this is a really good book on modal logic.)

>  1. If the laws are laws of God or nature, then S5 would imply
>     that physics is immutable.  I'm not sure about that, but I
>     wouldn't quarrel with it as a first approximation.
>
>  2. But if the laws are laws of biology, S5 would imply that
>     evolution is impossible since newly emerging species violate
>     the laws that governed their ancestors.
>
>  3. If the laws are the laws of science as currently known,
>     then S5 would imply that no new discoveries are possible.
>
>  4. If the laws are the rules and regulations of any human
>     organization, S5 would imply that no addition, extension,
>     or modification is possible.
>
>  5. If the modal operators are used for temporal logic, the
>     Barcan formula would imply that no entity can come into
>     existence, pass out of existence, or be absorbed into
>     some other entity (when it is eaten, for example).


It seems clear to me that you don't make a distinction between ontology and epistemology.
There are things that do not change no matter what. The simplest example are the laws of logic (assuming we agree on FOL, etc.), arithmetical truths like 1+1=2, etc.
In the same way, what is essential to some type cannot change. Not even when it happens that we become aware of something else or discover we were wrong on some issue.
Of course, your doubts would make perfect sense if we were talking about epistemology. Perhaps, you ended up in the wrong room?
In an ontology with possible worlds, if something is essential to something else (or it is not), it has to be so in all possible world.

>  SB> OTOH, it might be that you have (or know of) a good set of laws that may
>
>> characterize the notion of essence. Then, your point is that DOLCE is not justified in leaving these laws out. If so, please, let the SUO group discuss these laws. I'll be happy to add these to the list of possible extensions of DOLCE.
>
>
> I recommend Section 7 of the paper, which outlines Peirce's
> classification of modalities and how they relate to the choice
> of laws.  And a very important application of that approach is
> to formalize all the "intensional" verbs and speech acts, of
> which there are many thousands in natural languages -- each
> with its implicit set of axioms that are interpreted as the
> laws of that modality:
>
> JS>>    http://www.jfsowa.com/pubs/laws.htm


I must confess that I don't know what to think about the paper and since you keep showing the link, I suggest you to revise it asap. Section 2, at least, seems built on sand.
One gets suspicious when references to specific systems like deontic logic (see sect. 1) and bold statements like (see sect. 2):

(*)
"any Kripke model structure (K,R,\phi)
can be converted to one of Dunn's model
structures"

are paired with strong results that one gets "for free" (see sect 2):

"System T .  The two axioms Nec(p) ->p and
p-> Poss(p) require every world to be accessible
from itself. That property follows from Dunn's
definition because the laws L of any world are a
subset of the facts:  L included in M."

The point is that, in logic, whenever you get something "for free" you cannot refuse it (you cannot throw it away, so to speak). In this case, following your claim, you always *have* axiom T in your system (or better the law corresponding to it). However, the Kripke frames satisfing axiom T are only a tiny fraction of the class of Kripke frames.
This contradicts your claim (*).
Furthermore, the Kripke frames for deontic logic are not included in this tiny fraction, which makes your observations on deontic logic totally irrelevant.

I guess the problem lies in the very definition of pair (M,L), see sect. 2:

"For every Kripke world w, Dunn assumed an
ordered pair ( M,L), where M is a Hintikka-style
model set called the facts of w and L is a subset
of M called the laws of w."

Why do you need to assume that L is a subset of M?
You state this restriction explicitly although the conversion at the end of section 2 does not guarantee it...

> SB> Unfortunately, this paper is largely irrelevant to the issue.
>
>> All it says (in this regard) is that it is better to know the laws characterizing the modalities one has in the formalism. That is why, if you want to use DOLCE for your own applications, you should add the laws that better characterize your domain. Again, this very idea is given in the DOLCE document. Perhaps you missed it.
>
>
> One of the motivations of that paper is to define the many thousands
> of verbs, such as want, need, hope, fear, seek, believe, promise,
> plan, decide, attempt, avoid, prevent, require, etc.  Each of those
> verbs creates a modal-like context, and the axioms associated with
> them are the laws that govern the context.  Those verbs cut across
> every domain of engineering, science, business, and life.


You should show in detail a couple of examples providing the laws (or some approximation) in FOL for some of these verbs. If your examples are convincing, the SUO can get much benefit discussing the advantages of the approach.

> SB> You might be saying something stronger, though. That is, your point
>
>> could be that one should *never* use a modality without characterizing it completely.
>
>
> I would never say "never", but I would say that words like
> "necessary" or "essential" are highly ambiguous because they
> don't say whether something is logically necessary, physically
> necessary, economically necessary (for you or for Bill Gates),
> or legally necessary by the laws of the US, France, Italy,
> London, or your mother.


I answered this already

> SB> But then, why is this claim restricted to modalities
>
>> only? Isn't it applicable to any notion in an ontology (relations, predicates, terms, whatever)? This doesn't seem a criticism specific to
>> DOLCE. Indeed, it would be interesting to analyze the consequences of such a position.
>
>
> I raised it about DOLCE because Nicola uses the word "essential"
> as if it meant something.  But when you replace that word with an
> explicit law, you can ask where the law came from and how immutable
> the supposed "essence" might be.  S5 might be OK for the "true" laws
> of physics, but it certainly is not acceptable for the laws that are
> currently known.


I answered this already

> If DOLCE drops the word "essence" in favor of laws, then we have
> one more point of agreement.  That leaves only the seven points
> of disagreement at the beginning of this note.


If the term "essence" is bothering you that much, what about DOLCE-essence (not sure this changes the substance...)
On the other seven issues, I hope I gave reasonable answers. Anyway, you have my point of view (which I don't claim is shared by all people at LOA, we too come from different schools 

If I wasn't clear, please say so. My first aim in this discussion is to understand your reasons and show our motivations. Agreement is not at stake and should not be forced (but there is always some hope, right?)
Stefano

> John
>