RE [crm-sig] Have endurance!

Aldo Gangemi a.gangemi at istc.cnr.it
Fri Apr 1 19:08:52 EEST 2005


That's a compilation of the discussion between me and Pat Hayes. 
Please ask for interpretation of dead ends :). It all started from 
the problem of having temporalized parts or not, and of course it 
ended up with endurantism/perdurantism (aka 3D vs 4D)
--Aldo

PS the last (unquoted) text is mine
_____________________________________________________________________________

Some more clarification: our discussion seems to imply a strong 
experimental part, which we are trying to carry out only through 
thought experiments, then let's explicit them.

>>>One can accept a pragmatic or natural-kind distinction between 
>>>object and process/event, without thereby buying into the 
>>>philosophical/ontological distinction between perdurant and 
>>>endurant. The latter is a theoretically biassed way of accounting 
>>>for the former.
>>
>>that's good, provided that you give some minimal axiom to formally 
>>discuss the difference (you can dismiss for example the partition 
>>over the universe of part relations: what consequences can result 
>>in a realistic domain modelling task?)
>
>Well, that's why I said a natural-kind distinction. There may be no 
>actual axiom which is a clear basis for making the distinction, and 
>there doesn't need to be one. We can just make it, and tell people 
>to use their common sense (;-) to decide whether something fits 
>under either of the categories (or maybe both). Not a strategy that 
>would appeal to Aristotle or Kant, but so what? We aren't trying to 
>foundational metaphysics, right?

Not foundational metaphysics, but something that clarifies/encodes 
one's commitments. Natural kinds can be axiomatized (with no 
sufficient criteria).
If no axioms are provided, the distinction would blur most probably 
in a short time (have you read Eco's "Kant and the Platypus"?), 
because of discourse dynamics ... and if it blurs, any other one 
could blur as well ... or if you want axioms for many others, why a 
choice at this level should be so primitive that one must be silent 
about?

If you want to *experiment* on sense shifting, that's a good beginning though.
With my colleague Massi Ciaramita, we have been discussing how to 
design an experiment (there is a tradition in psycholinguistics, 
going back at least to Rosch), and there is something that possibly 
works for so-called "basic categories". And there are some 
complications.
For these general notions, I don't think any experiment based on pure 
names can work at all: subjects need to work over some intended 
meaning, otherwise we get into a "family resemblance" chain, which 
biases any such experiment.
E.g. figure out proposing either one of the following two lists of 
words, asking to classify each word with either "object" or "event", 
where the first item of the list is given (sentences in parentheses 
can be considered as 'gedanken' (it's German, but never mind ;)) 
associations for the sake of our discussion, or used in the test 
explicitly, in this case even without giving the first 
classification):

object<-spoon (given)
.........<-water (I can take it with a spoon)
.........<-wave (it's made of water)

vs.

event<-sailing (given)
........<-sea-sickness (happens during sailing)
........<-wave (causes sickness)

I bet that a good percentage will keep object in the first list and 
event in the second (by making similar associations)

notice that this frame effect (if any, I'm making an ideal experiment 
of a test setting, which is not exactly science in this case) might 
occur with most nouns, not only with literature cases like wave, 
wind, fire, etc. E.g. (less intuitive, as expected, but think about 
submitting images instead of words ...):

object<-piece of wood (given)
.........<-sawing machine (I can use it to cut the wood)
.........<-chair (I can produce it out of wood by using a saw)

vs.

event<-work (given)
........<-being tired (state that often occurs during work)
........<-chair (an affordance to take rest)

this could show that we are talking about two different ways of 
partitioning reality, but also that they have a common gestalt. 
Therefore, like the duck/rabbit figure, we can either choose to 
represent ducks and rabbits separately with a relation linking them 
at different times, or to represent temporal slices of a "dubbit" ... 
ducks at t and rabbits at t will be more useful in some applications, while 
dubbits in others.
Of course, if we choose some hard metaphysics addict, the results of 
the test could be peculiar :)

There remains that the "scene settings" created to play those 
linguistic games are definitely different, and much more important 
than the distinction itself.  Here I agree with you.

Nonetheless, with ducks at t and rabbits at t we need some part and 
participation axioms, and temporal indexes, while with dubbits we 
need snapshots and unity axioms to coreference them. Axioms are 
needed anyway.

Are you still sure that a "pragmatic criterion" or "common sense" is 
enough, whatever solution is adopted?

<snip>

>
>>>
>>>Well, it makes very good sense in a lexicon, for essentially 
>>>lexical reasons. I don't think lexica have very much to do with 
>>>ontology design at all, however, at least the kind Im interested 
>>>in.
>>
>>this goes against current best practices for ontology building 
>>lifecycle, though
>
>Ah, that is interesting. Can you elaborate? I wonder if there are 
>distinct methodologies of ontology building suitable for different 
>kinds of application.

of course yes, but I am saying that in case of large corporate or 
organization ontologies, terminologies, thesauri and lexicon are an 
obvious choice. Ontology learning is an alternative. Building from 
scratch or eliciting from experts are better, but can be applied 
mainly for small scale tasks, like core ontology building.

>>>>>But aside from that claim, my point is more basic. I agree one 
>>>>>CAN handle all these cases in a DOLCE-style ontology by having 
>>>>>events and objects everywhere, ie basically by overloading. But 
>>>>>now I want to ask a purely pragmatic question: what is all this 
>>>>>duplication FOR? What purpose is served by distinguishing two 
>>>>>waves, and requiring the motion-across-the-ocean to apply to the 
>>>>>wave-object but the circular-motion-of-water to apply to 
>>>>>(actually to the objects included in) the wave-event, or the 
>>>>>flame-object to be eternal but the combustion to be going on in 
>>>>>the flame-process? None of this seems to me to be useful or 
>>>>>constructive or required for drawing any conclusions, and it 
>>>>>clearly gets in the way of applying the ontology to complex 
>>>>>cases. I wouldnt know where to start in trying to produce a 
>>>>>DOLCE ontology for, say, cellular biochemistry, where one has to 
>>>>>describe things like the Krebs cycle.
>>>>
>>>>From here. Take this as a first contribution to a possible TF Note.
>>>
>>>Interesting, thanks for sending it. And Ive been googling from 
>>>Acetyl CoA, quite fascinating.
>>>
>>>>
>>>>It's complex, just the basic of it (very basic, I'm not a 
>>>>specialist, although I'm working with biochemistry experts for a 
>>>>DOLCE+-based ontology):
>>>>
>>>>Process: Condensation
>>>>Phases (ordered process parts)
>>>
>>>Temporally ordered? As in a sequence?
>>
>>Yes
>>
>>>>: Acetyl CoA condensation (1), Citryl CoA formation (2), Citryl 
>>>>CoA hydrolyzation (3)
>>>>Participant substances: Citrate synthase (overall?), Acetyl CoA 
>>>>(1), Oxaloacetate (1), Citryl CoA (2,3), Citrate (3), CoA (3)
>>>>Roles: Enzyme, played by Citrate synthase
>>>>
>>>>OWL abstract syntax:
>>>>
>>>>Class(Condensation partial
>>>>  restriction(ProperPart someValuesFrom (Acetyl_CoA_condensation))
>>>>  restriction(ProperPart someValuesFrom (Citryl_CoA_formation))
>>>>  restriction(ProperPart someValuesFrom (Citryl_CoA_hydrolyzation))
>>>>  restriction(TotalConstantParticipant someValuesFrom
>>>>   (intersectionOf Citrate_synthase restriction(ClassifiedBy 
>>>>someValuesFrom(oneOf(Enzyme))))
>>>>  restriction(TotalTemporaryParticipant someValuesFrom
>>>>   (intersectionOf Acetyl_CoA restriction(TotalParticipantIn 
>>>>someValuesFrom (Acetyl_CoA_condensation))))
>>>
>>>Interesting pattern. I am slightly suspicious of nested 
>>>restrictions. These all have a kind of circularity about them, 
>>>since they restrict values of a TemporaryParticipant to take 
>>>values in a set formed by restricting a TotalParticipantIn to have 
>>>values in the class being defined. I'm trying to understand this 
>>>intuitively. Is the idea that this process is a kind of 
>>>composition of three subprocesses, and what these restrictions do 
>>>is to 'place' the various reactants involved into the right 
>>>subprocess (?)
>>
>>Right

<snip>

>>
>>>So this can be glossed as: there must be a temporary-participant 
>>>[substance involved which is] [the] Acetyl-CoA which participates 
>>>in the Acetyl_CoA-condensation [part of this process] (??) If so, 
>>>that last 'of this process' gets you out of OWL, right? Which is 
>>>what you are trying to sneak past with the 'total' vs. 'partial' 
>>>distinction? We've found similar issues here in a completely 
>>>different domain; you need something like 'role maps' to say this 
>>>stuff adequately, sigh.
>>
>>Correct. No role value maps ... only reification can deal with that in OWL-DL
>
>Blech, I'd prefer to give up on OWL than try to use reification seriously.

uhm, it can be logically boring, but there are reification ontologies 
that seem to work nicely (btw, you're implying that the n-ary 
relations note is junk ...)

>>>>  restriction(TotalTemporaryParticipant someValuesFrom
>>>>   (intersectionOf Oxaloacetate restriction(TotalParticipantIn 
>>>>someValuesFrom (Acetyl_CoA_condensation))))
>>>>  restriction(TotalTemporaryParticipant someValuesFrom
>>>>   (intersectionOf Citryl_CoA restriction(TotalParticipantIn 
>>>>someValuesFrom (Citryl_CoA_formation))))
>>>>  restriction(TotalTemporaryParticipant someValuesFrom
>>>>   (intersectionOf Citryl_CoA restriction(TotalParticipantIn 
>>>>someValuesFrom (Citryl_CoA_hydrolyzation))))
>>>>  restriction(TotalTemporaryParticipant someValuesFrom
>>>>   (intersectionOf Citrate restriction(TotalParticipantIn 
>>>>someValuesFrom (Citryl_CoA_formation))))
>>>>  restriction(TotalTemporaryParticipant someValuesFrom
>>>>   (intersectionOf CoA restriction(TotalParticipantIn 
>>>>someValuesFrom (Citryl_CoA_formation))))
>>>>  )
>>>>
>>>>not enough anyway, because in OWL-DL you can't strictly express 
>>>>coreference and time indexed properties.
>>>
>>>Right. :-((

<snip>

>>>>>>More practically: do you think it's the same part-of relation applied to:
>>>>>>
>>>>>>  i) PatHayesYesterday -PART-OF-> PatHayesAsTemporalWorm
>>>>>>
>>>>>>as to:
>>>>>>
>>>>>>  ii) PatHayesLiverYesterday -PART-OF-> PatHayesYesterday?
>>>>>>
>>>>>>or even as to:
>>>>>>
>>>>>>  iii) PatHayesLiverAsTemporalWorm -PART-OF-> PatHayesAsTemporalWorm?
>>>>>
>>>>>I think they have enough in common that it is worth drawing out 
>>>>>that commonality and making a single theory of it, yes.
>>>>
>>>>Am I saying the opposite?
>>>
>>>As I understand DOLCE, it renders such a single theory impossible. 
>>>Certainly I know that Simons' view of the distinction would rule 
>>>it out. If I am wrong, maybe you could explain what I would use, 
>>>in DOLCE, as the category of entities that would specialize into 
>>>endants and perdants and which would support the general parthood 
>>>theory (ie its quantifiers would range over that class.)
>>
>>It is called "particular", "entity" if you prefer
>
>But that is the top node of the entire taxonomy. I want to exclude, 
>for example, numbers and sets, which have no temporal extension.

those are in fact outside DOLCE (as datatypes), or inside, but as 
"regions", which are particulars, but with no time; time is not 
mandatory for the top node

<snip>

>>>...objects that endure through time are extended in time, and can 
>>>be naturally described in the same way that anything else that is 
>>>temporally extended can be described. Me-yesterday is not present 
>>>today, of course (so if I go to a doctor and complain about my 
>>>liver, I'm usually taking about my liver at that time) but the 
>>>previous day of the cricket match is not happening today either. 
>>>The two cases seem exactly similar to me. I just don't see what 
>>>intuition or utility the distinction is supposed to capture.
>>
>>you still have alternatives of comparable (?) complexity: temporal 
>>slices, or temporal indexing of relations ... no special preference 
>>from me, in general ... if you give me a case study, I can choose 
>>with some discernment ...
>

<snip>

>>>
>>>It is my belief that what this debate all boils down to in 
>>>ontological/formal practice is whether one prefers temporal 
>>>parameters syntactically attached to atomic propositions, or to 
>>>terms denoting things. The former is simpler, adequate for many 
>>>uses, and more in line with NL, but somewhat less general; the 
>>>latter is often a very neat way to state things like velocities or 
>>>trends (like an increasing osmotic pressure). I think we can allow 
>>>both, as long as we obey some rules of internal consistency.
>>>
>>
>>that's glorious!
>>

<snip>

-- 



Aldo Gangemi
Research Scientist
Laboratory for Applied Ontology
Institute for Cognitive Sciences and Technology
National Research Council (ISTC-CNR)
Via Nomentana 56, 00161, Roma, Italy
Tel: +390644161535
Fax: +390644161513
a.gangemi at istc.cnr.it

*******************
!!! please don't use the old gangemi at ip.rm.cnr.it
address, because it is under spam attack
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://lists.ics.forth.gr/pipermail/crm-sig/attachments/20050401/82d2adf6/attachment-0001.htm 


More information about the Crm-sig mailing list