[Person-ontology] Methodology using an upper ontology

[Pat Hayes]

>Pat,
>   It's hard to avoid responding to this.  But I'll try to avoid much 
>follow up so that we don't rehash the SUO debates.  It's important 
>for other folks on this list at least to know there is another 
>opinion.

Oh well, my view is in the minority, I took this as obvious. But Jim 
did ASK, so...

>   First, you've picked a very unrealistic and philosophically 
>challenging concept as your example.

But look, it isn't. What could be more natural, more everyday, than 
the notion of a wave? Every child knows what it is, everyone who has 
been in a boat or on a sea-shore knows what it is like, most people 
could draw you one. Its only 'philosophically challenging' if you 
insist on cleaving to some rigid (and, yes, PHILOSOPHICAL) a priori 
distinction like 'thing vs. process': just the kind of thing, in 
fact, which tends to be set into stone in an Upper Level Ontology. 
Other examples of the same kind of thing: flames; biological 
processes inside cells; atmospheric things such as weather fronts and 
storms; physical processes which exert forces, such as combustion 
events in a gas engine; hydraulic processes...

(The Web resource/DOLCE example was taken from recent work by some of 
the DOLCE team.)

>  If we pick something more practical, like the concept of "Human"

Human is easier than ocean wave??

>there is a wealth of boring, practical content in an upper ontology, 
>such as SUMO, that helps in defining what a &%Human is (an &%Agent, 
>that can initiate &%IntentionalProcess(es) etc)

Good example. Now, is that in fact a good assumption about all human 
beings? It seems to be false of people when they are asleep, in a 
coma, dead, drugged or in intensive care. If Im writing a legal 
ontology it might be a good axiom to grab (though I'd better be 
careful to distinguish Person from LegalPerson) but if Im writing an 
ontology relevant to, say, human cardiac medicine, its much less 
likely to be relevant, and if I do re-use it I'll almost certainly 
have to modify it, perhaps completely change it, e.g. by introducing 
a notion of 'life period' and qualifying attributions of intentional 
state by the 'period'; a change which will percolate through the 
entire SUMO framework and have all kinds of repercussions for other 
axioms.

But I guess we may be talking at cross purposes. I won't deny that 
things like SUMO are often a handy 'library' of ready-made axiom sets 
embodying frequently-true assumptions. Rather like a software library 
or, better, a coding pattern guide. But as with such a guide, one has 
to be ready and willing to adapt the 'code' (I hope Paula is reading 
this :-) to one's own purposes, not just copy and paste it. But this 
isn't what I mean by an 'upper' ontology. SUMO is more like a 
mini-Cyc: it has stuff in it from many levels. Strip off (or ignore) 
the top three or four categories, and SUMO is indeed a useful 
compilation. But its those top three or four that I'm objecting to.

>and what relations it has to the world. Maybe a long time ontologist 
>such as yourself can churn out in a few hours all the content that a 
>CRM system might need to define a Human and its attributes, 
>participation in actions, constraints on values and so forth, but 
>for the rest of us, there's so much uncontroversial content in SUMO 
>that reuse makes sense.

*Nothing* in SUMO (or any other upper ontology: I don't mean to 
single out SUMO) is uncontroversial.  That is my point. NONE of it 
can be just grabbed off the shelf and re-used, without careful 
thought. You always have to think about whether the SUMO axioms 
really do apply to your particular case, your particular notion of 
'human'. And this checking isn't very different from writing your own 
axioms in the first place; though OK, it might be faster for some 
people.

>  You'll always be able to find difficult boundary cases

Indeed. Except these cases are the NORM, not the exception. Its 
almost all boundary.

>, and maybe the reuse won't be perfect, but as with software 
>engineering (I know you challenge this analogy as well) reuse is 
>proven to save time and increase quality.

PROVEN? Is this really true, Adam? If you can cite studies showing it 
is, I'll stop arguing about it. But until then, I will treat this 
claim like most other claims from 'management studies', i.e. as 
optimistic folklore.

BTW, I find this notion of 'imperfect re-use' rather like that of 
'slightly pregnant'. If I "re-use" some SUMO but change it so that 
its now inconsistent with the SUMO version, what exactly have I done?

Pat


>
>Adam
>
>Pat Hayes wrote:
>>>All,
>>>
>>>It was posted (not sure by who):
>>>
>>>	"Construct an ontology. It almost does not matter exactly where
>>>you
>>>	start, but either the very top (most general) or the very bottom
>>>	(data) is not usually a good place."
>>
>>Me (Pat Hayes). Though the advice comes originally from Doug Lenat.
>>
>>>Question: Why not use an existing upper ontology?
>>
>>Well, question back: Why use it? What would it be useful for? Here 
>>we are, say, trying to formalize a mid-level concept, say 'ocean 
>>wave' (Im making this up), using BOF, say. BOF requires us to 
>>classify ocean wave 'properly'. In particular, we must decide 
>>whether or not it is a continuant. Well, is it or isn't it? Do you 
>>know? I don't. Moreover, I don't CARE. Nothing turns on this 
>>decision other than making BOF happy. THe high-level upper ontology 
>>is just getting in the damn way, forcing me to make decisions I 
>>don't care about, are irrelevant to my concerns and that have no 
>>obvious answer. Suppose, moreover, I toss a coin and decide that my 
>>wave is indeed a continuant. Is this any help? No: all it does is 
>>make it harder for me to say some natural things (since continuants 
>>aren't allowed to have temporal parts).
>>
>>Suppose instead I am using DOLCE and formalizing, say, the W3C TAG 
>>notion of a 'web resource', and I decide that a web resource is 
>>kind of like a 'place' where some information can be stored. So I 
>>look in DOLCE and I find a high-level notion of 'region'. The only 
>>way to decide whether or not I can put my notion under this heading 
>>is to go into the actual axioms of DOLCE (which is a bit like 
>>taking a bath in nitric acid, by the way) and see what it actually 
>>says about 'regions', paying attention to detail and asking 'does 
>>this axiom apply to my idea'? And this is just writing axioms, only 
>>way slower.
>>
>>>  Wouldn't it provide a
>>>good-enough top-down structure?  Wouldn't it also provide many (maybe
>>>half) of the needed concepts?
>>
>>No, it almost certainly would not provide any of them in the form I 
>>am likely to need them, since the people who built it the upper 
>>ontology, no matter how smart or well-meaning they were, weren't 
>>thinking about the issues that arise at my level of detail: so 
>>whatever they put into the UO, its very unlikely to be anything 
>>more than a provisional sketch or draft of what I am going to need. 
>>Mereologists often take it as obvious that partOf is transitive, 
>>because they are thinking about lumps of clay: but if I'm making an 
>>ontology for a parts catalog, parthood isn't transitive. So for me, 
>>all that the UO has done is stolen the term 'part' and forced me to 
>>use a different term for my concept. Im going to have to write my 
>>own axioms for it anyway.
>>
>>Upper ontologies, by and large, are no actual use to practical 
>>ontologizing at all. They warp intuitions and interfere with the 
>>process, like an exasperating micro-manager. They add nothing 
>>computationally useful, since almost nothing useful can be said at 
>>a very high level. They usually embody some a priori philosophical 
>>perspective which likely has no relevance to the actual topic under 
>>discussion. Whoever wrote them wasn't thinking about the topic, for 
>>sure.
>>
>>>Of course, in parallel, gather
>>>requirements and work from the bottom up.
>>
>>The danger is that these two parallel tracks might not actually 
>>meet anywhere. And the worse danger is that when this happens, the 
>>top-down track is usually treated as authoritative, since to change 
>>an entrenched 'upper' ontology is seen as harmful to 
>>interoperation. So we end up with a situation where the entire 
>>shape of the whole ontology has been formed by people who don't 
>>know squat about the actual subject-matter: kind of a pre-emptive 
>>Peter Principle at work.
>>
>>Pat
>>
>>>Jim Schoening
>>>_______________________________________________
>>>Person-ontology mailing list
>>>Person-ontology at idcommons.net
>>>http://idcommons.net/cgi-bin/mailman/listinfo/person-ontology


-- 
---------------------------------------------------------------------
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