Transformation and Derivation Relations - interpretation and consequences
michel.d...@gmail.com
Am I wrong in thinking that the OBO/BFO relations "transforms_into"
or "transforms_from" is non-monotonic?
Take the example "an instance of Child transforms_into an instance of
Adult, thereby changing their classification while retaining their
identity" ... if an individual is an instance of the class Child at
time t, and it loses this class membership at time t', then doesn't
this count as non-monotonic behavior? Unless the transformation leads
to additional properties that don't violate previous inferences, we'll
have to be careful in how we apply this relation for OWL-DL reasoning.
Also, I'm concerned about the discussion of "derives_from" or
"derives_into" in the GB relations paper. In particular, "a significant
[biological] portion of matter of the earlier continuant is inherited
by the later". How does one assess what is meant by "significant"??? I
don't think there is any room for these kinds of subjective
assessments. Whether the continuant is affected by a simple
modification or a complete conversion, it still remains that a
continuant at time t' derives_from the continuant at time t. Also, this
definition, in contrast to "transforms_into", is monotic in its
behavior.
What do you think?
-=Michel=-
Cristian Cocos
RO paper have long been exposed as rather "uninspired" (see M. Brochhausen's
paper, which I have no handy reference to at the moment :-) ). They are on
the list of relations to be fixed. Formalizing such relations, however,
raises deep philosophical issues with no definite answer--at least not to my
knowledge. What could reasonably be expected at this point is a list
constraints/axioms governing their behavior, constraints that do not,
unfortunately, pinpoint them unequivocally, but offer some degree of
"precision." This means, on the other hand, that some degree of subjectivity
will persist.
CC
_________________________________
"I don't want to achieve immortality through my work. I want to achieve it
through not dying." -- Woody Allen
Andrew D Spear
William Bug
Smith, Barry
>Hi,
> Am I wrong in thinking that the OBO/BFO relations "transforms_into"
>or "transforms_from" is non-monotonic?
>
>Take the example "an instance of Child transforms_into an instance of
>Adult, thereby changing their classification while retaining their
>identity" ... if an individual is an instance of the class Child at
>time t, and it loses this class membership at time t', then doesn't
>this count as non-monotonic behavior? Unless the transformation leads
>to additional properties that don't violate previous inferences, we'll
>have to be careful in how we apply this relation for OWL-DL reasoning.
OWL-DL has notorious problems in dealing with entities which change
over time. Hence some OWLists deny that there are such entities.
There is no michel.dumontier but only michel.dumontier.at.t1,
michel.dumontier.at.t2, michel.dumontier.at.t3, etc.
You can never step into the same river twice because not only does
the river not exist there is also no you.
BS
Michel Dumontier
Well, as I'm never one to turn down a challenge (read stubborn), find attached an example of temporal modeling for OWL-DL in which the Person "michel" exists, and so do his temporal instances, linked together by "hasTemporalInstance", in which those instances are linked through time by the transitive "transformsInto" (+inverse) and the non-transitive "immediatelyTransformsInto" (+ inverse). Best use Protege 4, run the reasoner and use class queries i put in the ontology comment field. Maybe this would make a nice design pattern for an OWLED paper?
So, in retrospect, I don't mind the transformation property that much anymore...
Cheers!
-=Michel=-
At 07:57 PM 1/9/2007, you wrote:
>Hi,
> Am I wrong in thinking that the OBO/BFO relations "transforms_into"
>or "transforms_from" is non-monotonic?
>
>Take the example "an instance of Child transforms_into an instance of
>Adult, thereby changing their classification while retaining their
>identity" ... if an individual is an instance of the class Child at
>time t, and it loses this class membership at time t', then doesn't
>this count as non-monotonic behavior? Unless the transformation leads
>to additional properties that don't violate previous inferences, we'll
>have to be careful in how we apply this relation for OWL-DL reasoning.
OWL-DL has notorious problems in dealing with entities which change
over time. Hence some OWLists deny that there are such entities.
There is no michel.dumontier but only michel.dumontier.at.t1,
michel.dumontier.at.t2 , michel.dumontier.at.t3, etc.
Chris Mungall
On Jan 10, 2007, at 3:31 AM, Michel Dumontier wrote:
> Ah!!! Thanks for the paper!!! Glad to know I wasn't completely out
> to lunch on that comment!
>
> Well, as I'm never one to turn down a challenge (read stubborn),
> find attached an example of temporal modeling for OWL-DL in which
> the Person "michel" exists, and so do his temporal instances,
> linked together by "hasTemporalInstance", in which those instances
> are linked through time by the transitive
> "transformsInto" (+inverse) and the non-transitive
> "immediatelyTransformsInto" (+ inverse). Best use Protege 4, run
> the reasoner and use class queries i put in the ontology comment
> field. Maybe this would make a nice design pattern for an OWLED
> paper?
What you have done here is essentially what Barry stated that OWL
forces you to do - you're treating yourself as a slice of a 4-
dimensional worm. This is all well and good if you're a perdurantist,
but it's hard to see how this would work with an endurantist ontology
like BFO (or DOLCE for that matter).
In your example, both michel and michel_teen are instances of Person.
The latter stands in a hasTemoralInstance relation to the former.
Is Person intended to be a subtype of BFO:Continuant? Or of
BFO:SpatiotemporalRegion?
Is michel intended to be the mereological sum of michel_
{infant,child,teen,adult}?
As far as I know OWL lacks good documentation on representing this
sort of thing. So there may be room for some kind of OWLED paper on
this - I won't discourage you from this. I do think you need to work
harder on making your ontological commitments explicit - but I am
pretty sure the results will be logically incompatible with BFO, and
thus of limited interest to this list.
On the other hand, I'm still not sure how OWL semantics can be
reconciled with BFO's treatment of endurants and time. For those of
us who are forced to use OWL for certain applications, I can foresee
some fairly ugly design patterns being forced upon us whereby we have
to indirectly refer to SpatiotemporalRegions rather than directly to
continuants.
> <temporal_reasoning.owl>
William Bug
Michel Dumontier
"Is Person intended to be a subtype of BFO:Continuant? Or of BFO:SpatiotemporalRegion?"
a Person (a human being) is a continuant - its instances endure through time. Adulthood might be considered a TemporalRegion. However, what I might really doing in this case is talking about the "Life of Michel", which would be a Process. In OWL, due to monotonic constraints, I don't think we have a choice but to model time slices explicitly, if we want to avoid making potentially inconsistent assertions about our "Person". Even SWRL won't change that requirement.
One option, perhaps, is to start with "michel" as an instance of Person and describe this person's time-independent properties, while chaining the time-dependent instances by the transitive transforms_into.
Does that sound reasonable?
-=Michel=-
--
Michel Dumontier
Assistant Professor of Bioinformatics
Carleton University
http://dumontierlab.com
Alan Ruttenberg
Smith, Barry
legs, your cells and your arteries. Your youth is not a part of you;
it is a temporal part of that occurrent process which is your life.
BS
Alan Ruttenberg
The constraint is that occurrents only have parts that are
occurrents, and continuants only parts that are continuants.
Also, BFO gives status to fiat parts of occurrents, such as the 4-d
part (e.g. my 3rd year of life, a part of my youth) of my 4-d life
which is bounded by two 3-d hyperplanes defined by, e.g. two time
points (the date of my 2nd birthday and the date of my 3rd birthday).
Do I have this right?
If so, perhaps "temporal part" in the definitions should be changed
"occurrent part" to make this more clear.
-Alan
William Bug
So rephrasing:The constraint is that occurrents only have parts that areoccurrents, and continuants only parts that are continuants.
Also, BFO gives status to fiat parts of occurrents, such as the 4-dpart (e.g. my 3rd year of life, a part of my youth) of my 4-d lifewhich is bounded by two 3-d hyperplanes defined by, e.g. two timepoints (the date of my 2nd birthday and the date of my 3rd birthday).
Do I have this right?If so, perhaps "temporal part" in the definitions should be changed"occurrent part" to make this more clear.
Alan Ruttenberg
the problem is with derives_from. The argument is that "inherits a
biologically significant portion of the matter of" is too imprecise
to use in a definition, with the emphasis on "biologically
significant". It then gives an example of a parasite devouring it's
host a legitimate example of derivation, and points out that one
can't distinguish, with derives_from between reproduction and ingestion.
First, I'd point out that the RO paper seems to indicate that that
ingestion is not derivation, undermining the example that is given:
"We can also represent more complex cases where transformation and an
analog of derivation are combined, for example in the case of budding
in yeast [27], where one continuant continues to exist identically
through a process wherein a second continuant floats free from its
host; or in absorption, where one continuant continues to exist
identically through a process wherein it absorbs another continuant,
for example through digestion"
In other words, absorption is an "analog" of derivation, not a kind
of derivation.
I think it comes closest to the correct insight when it says: "The
difference between transformation and derivation depends solely on
the attribution of identity"
To me the problems arise from two issues. The first is a lack of
definitions for key concepts used in the RO definition: "identity",
"inherits". The second is a lack of appropriate care in choosing and
using a consistent choice of granularity. Identity would seem to be
tightly tied to granularity, but needs more than that to be used, for
example some choices about what the role of genetics is. "Inherits"
is another story - this seems like an inappropriate word to use in
such a definition because it has a number of possible meanings, some
of which are related to genetics and so carry biological "baggage".
The "same" progression might be considered a derivation at one
granularity and a transformation at another. I quote "same", however,
because I don't know how to relate continuants to one-another across
granularities. Perhaps having another go at http://
ontology.buffalo.edu/smith/articles/partitions.pdf would make that
clearer. In any case, I hardly see any attempt to control
granularity, despite the fact that BFO takes it as a premise that a
single ontology talks about a single granularity.
-Alan
Smith, Barry
>So rephrasing:
>
>The constraint is that occurrents only have parts that are
>occurrents, and continuants only parts that are continuants.
Exactly
>Also, BFO gives status to fiat parts of occurrents, such as the 4-d
>part (e.g. my 3rd year of life, a part of my youth) of my 4-d life
>which is bounded by two 3-d hyperplanes defined by, e.g. two time
>points (the date of my 2nd birthday and the date of my 3rd birthday).
>
>Do I have this right?
Indeed
>If so, perhaps "temporal part" in the definitions should be changed
>"occurrent part" to make this more clear.
"Temporal part" is a technical term running through the many
philosophical discussions underlying all of this. Also there are
wrinkles -- consider, e.g. a football match M; M has two fiat parts
-- the sum of all processes taking place in the left side of the
pitch; the sum of all processes taking place on the right side of the
pitch. These are both occurrent parts, but they are not temporal
parts of the occurrent M.
BS
>-Alan
>
>
>
>On Jan 11, 2007, at 6:21 PM, Smith, Barry wrote:
>
>>
Alan Ruttenberg
>>>
>>>
>>>> The definition for continuant pulled from BFO is "an entity that
>>>> exists in full at any time in which it exists at all, persists
>>>> through time while maintaining its identity and has no temporal
>>>> parts"
On Jan 12, 2007, at 1:35 PM, Smith, Barry wrote:
> "Temporal part" is a technical term running through the many
> philosophical discussions underlying all of this. Also there are
> wrinkles -- consider, e.g. a football match M; M has two fiat parts
> -- the sum of all processes taking place in the left side of the
> pitch; the sum of all processes taking place on the right side of
> the pitch. These are both occurrent parts, but they are not
> temporal parts of the occurrent M.
> BS
But these occurrent parts can also not be parts of a continuant. So
saying that a continuant has no temporal parts is true, but
incomplete. Since a temporal part seems to be by definition an
occurrent part, saying that a continuant can't have any occurrent
part implies what is currently said.
Unless there can be also be temporal parts which are not occurrents.
Can there?
BTW, are temporal parts dependent entities - your example would seem
to imply that they are.
-Alan
Cristian Cocos
>
> ... consider, e.g. a football match M; M has two fiat parts
> -- the sum of all processes taking place in the left side of the
> pitch; the sum of all processes taking place on the right side of the
> pitch. These are both occurrent parts, but they are not temporal
> parts of the occurrent M.
Why not use (a) (spatial) section_of a process and (b) subproces when
talking about the diferent ways to partition processes (i.e. using
spatial criteria vs. temporal criteria); both would have to be
regarded as species of part_of. That would enable precise reference to
all subprocesses of a given process, hence it would give us formal
tools to access all parts of a process.
(My interest in such topic comes via my desire to be able to speak of
(a) a section and (b) a subfunction of a function of a continuant.)
Cristian
--
Pierre Grenon
spatiotemporal parts, temporal parts are very special spatiotemporal
parts.
Pierre Grenon
> But these occurrent parts can also not be parts of a continuant. So
> saying that a continuant has no temporal parts is true, but
> incomplete. Since a temporal part seems to be by definition an
> occurrent part, saying that a continuant can't have any occurrent
> part implies what is currently said.
>
> Unless there can be also be temporal parts which are not occurrents.
> Can there?
No, only occurrents have temporal parts and temporal parts are all
occurrents, there is a principle of "type conservation" via parthood
at this level. I think this has to be given as an axiom as it has been
in BFO.
> BTW, are temporal parts dependent entities - your example would seem
> to imply that they are.
Don't be confused by terminology here. All occurrents are
"existentially dependent" on some continuant. Since temporal parts are
occurrents, they too are existentially depedent on some continuant.
However, occurrents do not inhere (and so neither do temporal parts of
occurrents) in continuants, they are not "dependent entities" in the
sense in which this expression is a label for one of the SNAP
categories.
> -Alan
>
>
> >
>
Pierre Grenon
This is precisely a four dimensionalist treatment of persons (your
temporal instances are nothing but temporal parts) and this is indeed
something which is straightforward to do in OWL. (this said, nice
file)
cheers
pierre
Pierre Grenon
> BFO:SpatiotemporalRegion?"
>
> The definition for continuant pulled from BFO is "an entity that exists in
> full at any time in which it exists at all, persists through time while
> maintaining its identity and has no temporal parts"
>
> a Person (a human being) is a continuant - its instances endure through
> time.
as I said about your file, your temporal instances are temporal parts
> Adulthood might be considered a TemporalRegion. However, what I might
> really doing in this case is talking about the "Life of Michel", which would
> be a Process.
Yes, very good, this is the way you need to think when using BFO.
> In OWL, due to monotonic constraints, I don't think we have a
> choice but to model time slices explicitly, if we want to avoid making
> potentially inconsistent assertions about our "Person".
We suggest transfering temporally sensitive information to the life of
continuants and their parts.
(isa Michel Person)
(life_of Michel Life_M)
(has_part Life_M Adulthood_M)
(isa Adulthood_M Adulthood)
When Michel's adulthood starts and ends is tied directly to
Adulthood_M and not to Michel.
During the time during which Adulthood_M exists, Michel falls under
the (junkish) category of persons who's life is currently adulthood,
this is what it means for him to be an adult.
> Even SWRL won't
> change that requirement.
> One option, perhaps, is to start with "michel" as an instance of Person and
> describe this person's time-independent properties, while chaining the
> time-dependent instances by the transitive transforms_into.
Maybe the transforms_into relation can be used at the class level to
link junkish classes sucha s described above.
Alan Ruttenberg
>
>> At 07:57 PM 1/9/2007, Michel Dumontier wrote:
>> Hi,
>> Am I wrong in thinking that the OBO/BFO relations "transforms_into"
>> or "transforms_from" is non-monotonic?
I don't think that non-monotonic is the thing to worry about here.
What non-monotonicity guarantees for the current OWL is that
operations on the ontology will yield the same result independently
of the order in which the assertions are processed. However, within
the current specification there is no notion of change or update to
an OWL ontology.
>> Take the example "an instance of Child transforms_into an instance of
>> Adult, thereby changing their classification while retaining their
>> identity" ... if an individual is an instance of the class Child at
>> time t, and it loses this class membership at time t', then doesn't
>> this count as non-monotonic behavior? Unless the transformation leads
>> to additional properties that don't violate previous inferences,
>> we'll
>> have to be careful in how we apply this relation for OWL-DL
>> reasoning.
If you want a "instance of" b in this definition to mean a rdf:type
b, OWL does not support this. rdf:type has no time index. It has
nothing to do with monotonicity - you just can't say this in the
current OWL.
If you decide to try to model "instance of" using something other
than rdf:type, then there might be some ability to play around with
this sort of thing, but you won't have to worry about "violating"
monotonicity. You will, however, need to define what you expect to
happen when you do this, and that your solution, along with the OWL
reasoning meets those expectations. Monotonicity is a property of the
logic, not what you do with it.
> OWL-DL has notorious problems in dealing with entities which change
> over time.
Compared to what? I'm not sure what "dealing with" means here. If you
mean "stating", then I'm not sure there is really a problem.
However, OWL is designed with the goal of being decidable, sound, and
complete - computing, in a finite amount of time,
answers to a certain class of questions which are correct, and which
don't miss parts of the answer, wrt the axioms that are given.
Comparing OWL with alternatives which don't have this property
doesn't really make sense.
On the other hand, we could usefully argue about what the criteria
for a representation system for use in, e.g. biology, should be. I
propose we start by defining what "dealing with" means.
> Hence some OWLists deny that there are such entities.
> There is no michel.dumontier but only michel.dumontier.at.t1,
> michel.dumontier.at.t2, michel.dumontier.at.t3, etc.
Setting aside whether any "OWLists" believe this by virtue of their
association with OWL, other than noting I find this hard to believe,
I still don't really understand the nature of this criticism. The
kinds of representation you are referring to are approximations in
the same sense that SNAP ontologies are approximations in
representing only a snapshot of the true story of e.g. continuants
which persist through time. A lone SNAP ontology doesn't talk about
time - it needs an associated SPAN ontology to do that.
Such systems *are* typically concerned with entities like the
persisting michel.dumontier, the entity which unites the various time
slices/instance/.... They differ by their ability to reason about the
various representations of the entities. Certainly a reasoner that
attempted to simultaneously compute across a constellation of SNAP
and SPAN ontologies with their temporal indices and granularities
would suffer similar problems.
> You can never step into the same river twice because not only does
> the river not exist there is also not you.
?
I'll reiterate that I don't think that this line of criticism serves
our shared goals. In http://ontology.buffalo.edu/medo/Subsumption.pdf
you write:
> "In and of themselves, however, DLs do not systematically ensure
> compliance with the principles of classification required if
> reasoning is to be performed accurately."
I couldn't agree more, and can't imagine any "OWList" who would
disagree, where "reasoning" is interpreted as generating answers that
agree with reality. But there, as well, follows a comparison which is
between different kinds of things that I don't see as comparable:
> "More than the use of any formalism, we believe that compliance
> with sound ontological principles is what guarantees the accuracy
> of reasoning."
It's a little like asking which is more important for having a
sensible conversation across a distance: The communications system
that carries the audio, or the sense to not be so intoxicated that
one is incoherent. Both are essential.
Best regards,
Alan
Michel Dumontier
>> Take the example "an instance of Child transforms_into an instance of
>> Adult, thereby changing their classification while retaining their
>> identity" ... if an individual is an instance of the class Child at
>> time t, and it loses this class membership at time t', then doesn't
>> this count as non-monotonic behavior? Unless the transformation leads
>> to additional properties that don't violate previous inferences,
>> we'll
>> have to be careful in how we apply this relation for OWL-DL
>> reasoning.
If you want a "instance of" b in this definition to mean a rdf:type
b, OWL does not support this. rdf:type has no time index. It has
nothing to do with monotonicity - you just can't say this in the
current OWL.
You can make any individual an instance of (rdf:type) any number of classes. The OWL-DL reasoner will complain if they are disjoint.
If you decide to try to model "instance of" using something other
than rdf:type, then there might be some ability to play around with
this sort of thing, but you won't have to worry about "violating"
monotonicity. You will, however, need to define what you expect to
happen when you do this, and that your solution, along with the OWL
reasoning meets those expectations. Monotonicity is a property of the
logic, not what you do with it.
OWL-DL cannot handle the BFO ontological commitment where one individual may in fact be the instance of two disjoint classes during the full of its existence. The real modeling problem, as far as OWL goes, is that to meet BFO's ontological commitment, one has to decide a priori whether to represent a person's life (an occurrent which has continuant participants) or a person (a continuant that exists in its entirety during some time, and has no temporal parts). If we know that the thing we want to model has temporal parts, then we must consider instead the process that glues it together. Barry and others didn't like that because I would just be slicing up an individual over time. But I don't see that we have a choice with this knowledge representation technology.
As an alternative, i offered the scenerio where I describe a continuant "michel" linked to time-dependent continuants by virtue of some property. However, in retrospect, it became more obvious to me that to adopt the ontological commitments of the BFO, one should instantiate an occurrent, say "life_of_michel" which has the time sliced continuants "michel_t1", etc. as participants, with a transforms_into (although i'm still debating between this and derives_into) between them. Of course, I can still give each of them the same name though some object property.
Cheers,
Alan Ruttenberg
> You can make any individual an instance of (rdf:type) any number of
> classes. The OWL-DL reasoner will complain if they are disjoint.
Suppose c instance of C1 at t1 and c instance of C2 at t2. Further
suppose that C1 and C2 are not disjoint, so you
represent this as c rdf:type (C1 intersection C2) - equivalent to
the two separate rdf:type statements (there's no qualifying them by
the time, currently)
How do you ask: What type is c at t2?
Suppose you have C2 transformation_of C1 represented in your KB. How
do you not trigger the part of the transformation_of definition:
"there is no t2 such that c instantiates C at t2 and c instantiates
C' at t2", from the definition in the RO.
How do you differentiate the above case from that where c is instance
of (C1 intersection C2) always (not so I can read it, but so I can
query or classify it).
etc.
You can do what you say, but from the current OWL-DL point of view,
there is no progression. c will be an instance of both C1 and C2 in
every model. All the axioms are processed as if at once.
Not to mention the fact that C1 and C2 will be disjoint in most
interesting biological cases.
> OWL-DL cannot handle the BFO ontological commitment where one
> individual may in fact be the instance of two disjoint classes
> during the full of its existence.
In BFO terminology, "individual" => "particular", "during the full of
its existence" => "In different SNAP ontologies". OWL-DL does not
have the concept of "full of existence" in the sense you mean.
> The real modeling problem, as far as OWL goes, is that to meet
> BFO's ontological commitment, one has to decide a priori whether to
> represent a person's life (an occurrent which has continuant
> participants) or a person (a continuant that exists in its entirety
> during some time, and has no temporal parts). If we know that the
> thing we want to model has temporal parts, then we must consider
> instead the process that glues it together.
If the thing we want to model has temporal parts then it *is* a
process. (well, occurant). BFOs commitment, as I understand it, is
that both the person and the person's life "exist", that there is a
way to represent their relationship, and that there are different
sorts of things that can be said about each type of entity (and you
should be careful about doing so)
"If we know...": Every living thing has a life. It's not it's not
whether we know. It's whether we want to say/infer something that has
to do with the life versus the living thing. I'd argue that for the
purposes of advancing the discussion we should discuss what should be
inferred, given what we say.
> Barry and others didn't like that because I would just be slicing
> up an individual over time. But I don't see that we have a choice
> with this knowledge representation technology.
And I still don't understand why. As I said, this looks to be the
same as representing the same particular in different SNAP ontologies.
> As an alternative, i offered the scenerio where I describe a
> continuant "michel" linked to time-dependent continuants by virtue
> of some property. However, in retrospect, it became more obvious
> to me that to adopt the ontological commitments of the BFO, one
> should instantiate an occurrent, say "life_of_michel" which has the
> time sliced continuants "michel_t1", etc. as participants.
In BFO it is the participation relationship that is time indexed,
rather than the continuant. But maybe this amounts to the same thing.
It would be nice to have a way to tell.
> , with a transforms_into (although i'm still debating between this
> and derives_into) between them. Of course, I can still give each
> of them the same name though some object property.
The distinction should be made based on some question which gets
answered one way if transformation_of, and answered a different way
if derives_from. Do you have such a question to guide your choice?
Best,
Alan
Michel Dumontier
You can do what you say, but from the current OWL-DL point of view,
there is no progression. c will be an instance of both C1 and C2 in
every model. All the axioms are processed as if at once.
That's my point exactly! That's why I want to figure out a way to model the progression as a transitive property in OWL-DL. The penalty is that it necessarily destroys the continuity that is described by BFO over the whole of existence for that individual. i.e. we might say that c1 at t1 is not the same individual as c2 at t2.
"If we know...": Every living thing has a life. It's not it's not
whether we know. It's whether we want to say/infer something that has
to do with the life versus the living thing. I'd argue that for the
purposes of advancing the discussion we should discuss what should be
inferred, given what we say.
Oh - what i mean by "if we know" is whether you anticipate to be making time-sensitive statements about some continuant which requires you to differentiate relations at different times. In many cases, it could be argued that the information that is to be represent is static during the lifetime of that continuant. This will conform just fine with BFO. It's the case where the knowledge about the individual must be time differentiated and _because_ they lead to inconsistencies.
> Barry and others didn't like that because I would just be slicing
> up an individual over time. But I don't see that we have a choice
> with this knowledge representation technology.
And I still don't understand why. As I said, this looks to be the
same as representing the same particular in different SNAP ontologies.
As already discussed, It's because OWL doesn't have the intrinsic notion of time.
> As an alternative, i offered the scenerio where I describe a
> continuant "michel" linked to time-dependent continuants by virtue
> of some property. However, in retrospect, it became more obvious
> to me that to adopt the ontological commitments of the BFO, one
> should instantiate an occurrent, say "life_of_michel" which has the
> time sliced continuants "michel_t1", etc. as participants.
In BFO it is the participation relationship that is time indexed,
rather than the continuant. But maybe this amounts to the same thing.
It would be nice to have a way to tell.
> , with a transforms_into (although i'm still debating between this
> and derives_into) between them. Of course, I can still give each
> of them the same name though some object property.
The distinction should be made based on some question which gets
answered one way if transformation_of, and answered a different way
if derives_from. Do you have such a question to guide your choice?
It has to do with identity conservation. Does the thing transform_into the same thing? Or does it derive_into a different thing? Because of the way I'm modelling time in OWL, it looks like transformation into a different thing (notwithstanding the lack of UNA). So if i can link all the instances to some functional identifier, perhaps i can use transformation_into.
Cheers,
-=Michel=-
Cristian Cocos
> spatiotemporal parts, temporal parts are very special spatiotemporal
> parts.
Good to know that. My posting was actually concerning a terminological
matter. I take it, hence, you do not have any issues with section_of
and subprocess for processes and, possibly, neither with section_of
and subfunction_of for functions...
Pierre Grenon
>
> > This distinction has been made for ages, we have been speaking of
> > spatiotemporal parts, temporal parts are very special spatiotemporal
> > parts.
>
> Good to know that. My posting was actually concerning a terminological
> matter.
sorry, didn't realize (hey does that make me a realizable entity, I
wonder), must have been a bit cranky too
> I take it, hence, you do not have any issues with section_of
> and subprocess for processes and, possibly, neither with section_of
> and subfunction_of for functions...
The problem with 'subprocess' is that technically, given a chosen
level of granularities, parts of processes are not processes and
'subprocess' sounds like 'smaller process'. Although I admit this
might be merely my terminological fancy, it is bound to cause
troubles.
Use the terms you like, but, since you ask, the terms 'temporal part'
and 'spatiotemporal part' are good enough in many cases for
processuals. I used in the past the term 'component' for process-like
parts of processes (i.e. which may count as processes on another level
of granularity) and 'phase' for salient temporal parts (e.g. first
half of the game, is it what you want to call 'section'?) but it's
neither trademarked nor has anybody ever gave a damn about this
terminology. Have a look at section 3.3.11, page 67 and see whether
it's useful.
In general I would think 'sub*_of' is a relation between two instances
of *, otherwise I would use variations on 'part', you do what you wnat
with this remark, of course.
Is a subfunction a function whose functioning is a spatiotemporal part
(whether temporal or not also) of the functioning of a
(super)function? Or is it another relation you have in mind? In the
first case, 'subfunction' sounds like a fine name, Amen.
cheers,
pierre
> >
>
Pierre Grenon
> > and subprocess for processes and, possibly, neither with section_of
> Have a look at section 3.3.11, page 67 and see whether
> it's useful.
forgot the link:
http://www.ifomis.uni-saarland.de/Research/IFOMISReports/IFOMIS%20Report%2005_2003.pdf
Michel Dumontier
Good to know that. My posting was actually concerning a terminological
matter. I take it, hence, you do not have any issues with section_of
and subprocess for processes and, possibly, neither with section_of
and subfunction_of for functions...
Why not just refer to part_of?
-=Michel=-
William Bug
You can never step into the same river twice because not only does the river not exist there is also not you.
William Bug
Bill BugSenior Research Analyst/Ontological EngineerLaboratory for Bioimaging & Anatomical InformaticsDepartment of Neurobiology & AnatomyDrexel University College of Medicine2900 Queen LanePhiladelphia, PA 19129215 991 8430 (ph)610 457 0443 (mobile)215 843 9367 (fax)Please Note: I now have a new email - Willi...@DrexelMed.edu
Michel Dumontier
Now we are trying move ahead across the field (not just in singular studies that require a huge amount technical investment and the results of which have limited re-usability) to provide support for:* support for monotonic reasoning on these entities* the ability do so in a way that approaches the real world scenarios where representation of continuants and occurents must be made inter-operable in the algorithmic sense.
I completely agree.
In regards to the particular points in this discussion, two points strike me as relevant - though I'm not certain my thinking on either of these issues is sufficiently rigorous - so I'd appreciate some feedback on these issues.1) Perception vs. formalized, ontological description of reality
This is far too philosophical.
2) Contradictions and/or inability for OWL-DL to represent and DL reasoners to process continuants and occurents expressed as tightly coupled entitiesThis is a critical task. Most of us cannot move forward in a big way without having some way of addressing it. Fabian pointed this out to me almost a year ago, but I must admit, it took a few months later as I tried to truly wrestle with this issue in the context of neuroanatomy and related biological domains that seriousness of the problem became real for me. That's when I started casting about for existing work in OWL that could help address the issue.
aye. so did i.
Am I wrong in assuming the work done by researchers at USC/ISI on the OWL Time ontology ( http://www.isi.edu/~pan/SWBP/time-ontology-note/time-ontology-note.html) is not relevant to this issue.
My question is whether or not there isn't a significant overlap in the ultimate requirements bioinformaticists have in this domain and the requirements being address in this OWL-Time ontology, and whether this resource - already in use in specific domain (web services expressed using OWL-S) - might not be able to help us address this task in semantically-formal bioinformatics.
I think so... it's the direction that i'm leaning towards. The question is, how do we integrate the notion of time to make it compatible in spirit with BFO and compatible in representation for OWL-DL. Ideally, we'd like one common design pattern for all such cases, that would allow interoperability between knowledge providers. Your suggestions here would be greatly appreciated.
Cheers,
-=Michel=-
Smith, Barry
>
>>OWL-DL has notorious problems in dealing with entities which change
>>over time.
>
>Compared to what? I'm not sure what "dealing with" means here. If you
>mean "stating", then I'm not sure there is really a problem.
>However, OWL is designed with the goal of being decidable, sound, and
>complete - computing, in a finite amount of time,
>answers to a certain class of questions which are correct, and which
>don't miss parts of the answer, wrt the axioms that are given.
>Comparing OWL with alternatives which don't have this property
>doesn't really make sense.
>
>On the other hand, we could usefully argue about what the criteria
>for a representation system for use in, e.g. biology, should be. I
>propose we start by defining what "dealing with" means.
I was thinking of first order logic: it deals with time rather
easily, by incorporating the 'at t' in for example 'this cell is part
of that body at t' as an additional term of the part-relation, which
turns into a 3-place relation.
The advantages of this move are that it is easy to understand and
thus to reproduce, and thus it represents a stable solution which
serves the goal of ontology to support wide communication,
(it is also part of a logic which is itself stable (has been so for
more than a century), highly generalizable, etc., etc.).
The disadvantage is undecidability -- however, there are ways round
this, and effective computational systems which reason well with FOL.
>>Hence some OWLists deny that there are such entities.
>>There is no michel.dumontier but only michel.dumontier.at.t1,
>>michel.dumontier.at.t2, michel.dumontier.at.t3, etc.
>
>Setting aside whether any "OWLists" believe this by virtue of their
>association with OWL, other than noting I find this hard to believe,
>I still don't really understand the nature of this criticism.
OWL, given its expressibility constraints, cannot directly capture
the 'at t'; several types of solutions to this problem can be
distinguished, e.g.:
1. deny that there are any such relations because everything is an
occurrent and so everything holds timelessly
2. identify every continuant which is the corresponding occurrent
which is its life
3. assert that there is no michel, but only michel.dumontier.at_t1,
michel.dumontier.at_t2, etc.
(for more see Pierre Grenon FOIS 2006 paper)
The problem is that all of these solutions are fixes, which means
that they fall short when measured against the scale of being easy to
understand and thus to reproduce. Thus they all fail to represent
some one stable solution which can serve the goal of ontology to
support wide communication.
>The
>kinds of representation you are referring to are approximations in
>the same sense that SNAP ontologies are approximations in
>representing only a snapshot of the true story of e.g. continuants
>which persist through time. A lone SNAP ontology doesn't talk about
>time - it needs an associated SPAN ontology to do that.
Each SNAP ontology is like a snapshot -- not a fix, but a pretty good
representation of the corresponding portion of reality, which humans
find easy to understand; that's why they buy cameras.
The question is: is OWL-DL essential? If yes, then we need a czar who
will impose a single fix for the 'at_t' problem (and, I suppose, for
a range of similar problems), so that OWL-DL does not spawn the very
tangled ghettoization which ontology is designed to eliminate. We
presumably would also need lots of training, to ensure that everyone
understands what these fixes are, and when to use them (such training
will be hard, because fixes are by definition unintuitive).
BS
Robert Hoehndorf
Hi,
BS> OWL, given its expressibility constraints, cannot directly
BS> capture the 'at t'; several types of solutions to this problem
BS> can be distinguished, e.g.:
BS> 1. deny that there are any such relations because everything
BS> is an occurrent and so everything holds timelessly 2. identify
BS> every continuant which is the corresponding occurrent which is
BS> its life 3. assert that there is no michel, but only
BS> michel.dumontier.at_t1, michel.dumontier.at_t2, etc.
Another solution would be to model persistance through time by some
abstract individual as in [1], combining your proposals 2 and 3: then
michel could be an abstract individual (something like an individual
concept in [2]) capturing that which remains identical in all
michel.at_ti (concrete objects), where the michel.at_ti are connected
by the life of michel (a process).
Something like that appears to me already implied by BFO, but I may
have misunderstood: when michel in BFO is a continuant persisting
through different SNAPs, having different properties at t1 and t2,
then what kind of entity is "michel" without any time-index? If this
is the continuant, then michel-without-time-index does not have
properties. Then I wonder what kind of entities michel-at-t1 and
michel-at-t2 are. Both entities may have different properties, and
should therefore not be logically identical (and therefore not
continuants?). But if michel-without-time-index is the continuant
which is exemplified by some entities at t1 and t2, then
michel-without-time-index should be something abstract, or am I wrong
there?
Rob.
[1] http://www.onto-med.de/en/theories/gfo/part1/index.html
[2] http://www-formal.stanford.edu/jmc/concepts.ps
Smith, Barry
>>>>>>"BS" == Smith, Barry <phis...@buffalo.edu> writes:
>
>Hi,
>
> BS> OWL, given its expressibility constraints, cannot directly
> BS> capture the 'at t'; several types of solutions to this problem
> BS> can be distinguished, e.g.:
>
> BS> 1. deny that there are any such relations because everything
> BS> is an occurrent and so everything holds timelessly 2. identify
> BS> every continuant which is the corresponding occurrent which is
> BS> its life 3. assert that there is no michel, but only
> BS> michel.dumontier.at_t1, michel.dumontier.at_t2, etc.
>
>Another solution would be to model persistance through time by some
>abstract individual as in [1], combining your proposals 2 and 3: then
>michel could be an abstract individual (something like an individual
>concept in [2]) capturing that which remains identical in all
>michel.at_ti (concrete objects), where the michel.at_ti are connected
>by the life of michel (a process).
This solution, too, has its problems, in addition to the fact that it
is yet one more workaround to add to the growing tangle: For an
ontology to support (for example) biomedical research we do not want
our patients, biosamples, portions of body fluid, etc., to be
abstract individuals, because if we have to teach biologists to work
with an ontology which treats them this way, they will likely be
turned away. Ontologies need human experts who will maintain and use them.
Moreover, are we sure that there is only one stable way of treating
Michel as an abstract individual?
>Something like that appears to me already implied by BFO, but I may
>have misunderstood: when michel in BFO is a continuant persisting
>through different SNAPs, having different properties at t1 and t2,
>then what kind of entity is "michel" without any time-index?
Same sort of entity as Robert. A human being. I can photograph you
today. I can photograph you tomorrow. You are the identical concrete
thing on each occasion, even though your properties change. There is
no underlying unchanging abstract Roberty thing to account for this identity.
>If this
>is the continuant, then michel-without-time-index does not have
>properties. Then I wonder what kind of entities michel-at-t1 and
>michel-at-t2 are.
They are logical constructions created by OWLists (and others)
>Both entities may have different properties, and
>should therefore not be logically identical (and therefore not
>continuants?). But if michel-without-time-index is the continuant
>which is exemplified by some entities at t1 and t2, then
>michel-without-time-index should be something abstract, or am I wrong
>there?
You are wrong. Instead of michel.at.tX and michel-without-time index,
there is just the one changing Michel, with all those portions of body fluid.
BS
Michel Dumontier
Yes, those are interesting thoughts. However, I don't think that BFO will consider anything that is "abstract".
-=Michel=-
Carleton University
http://dumontierlab.com