Issue 326: Resolving inconsistences between E2, E4, E52 and E92

In the 37th joined meeting of the CIDOC CRM SIG and ISO/TC46/SC4/WG9 and the 30th   FRBR - CIDOC CRM Harmonization meeting,  the crm-sig made a list of problems/realizations needed to be discussed more and concluded that next steps should be:

a)       To find examples of  E2  that are not spacetime volumes (ie non spatial)

b)       To find examples of  declarative time spans and places

c)       To prepare a practice guide experience (decision from CRMgeo) where to use declarative place

Homework for the next meeting is assigned to GB to write a text about this situation and then this text should be reviewed by Wolfgang and CEO

Berlin, December 2016

posted by Martin on 30/3/2017

Dear Christian-Emil,

Your diagrams are extremely helpful. Would be nice to have a tool that creates such graphs on the fly.

Your arguments are mathematical so far:
"If yes then E4 Period equals E2 Temporal Entity" ...means if we do not find another property actually making the distinction;-).

I'd suggest to add to the discussion the question of what is observable/phenomenal and what not.

Merging E2-E52 makes E2 a phenomenal/observable entity. I like that. But that does not mean, that P160 needs to exist as is. If E92 is actually prior to the phenomenal/declarative divide, we may question that the temporal projection of E92 is phenomenal. Rather, we should look at the approximations P82/83.

May be we misunderstand the ontological nature of the temporal projection.

Or we need a pure time-span without phenomenal interpretation on top of E2. 

Posted on 31/3/2017

Dear Christian-Emil

On 30/3/2017 9:34 μμ, Christian-Emil Smith Ore wrote:
> Dear Martin,
> Thank you, but there is always some pedagogical aspects connected to how one organize diagrams. An automatic tool can be a helper
>
> I am in this respect the matematician, you are the physisist. When we introduced the cardinalities/quantification and made the fol representation I, at least, got a tool to reason about the model qua model. We clearly see what the model is today.
I absolutely appreciate the mathematical reasoning, it is the sine qua non in this work. I just made the point, always to make clear the distinction of the kind of argument.
>
> As correctly stated: If we find another property making the distinction, then they are distict. However, today  we don't have this property.
Yes. My point is, logical deductions have always these aspects:

A) They confirm your experience. => Increase trust in your model => Use your model to simulate reality
B) They appear odd to your experience => reexamine your experience
                        => C) experience was wrong, model confirmed (e.g. the chaotic solutions of Volterra
                                        Equation describing that population growth against resource can result in
                                        population suicide, which was first not believed by mathematicians, then proven
                                        by mathematicians, then not believed by biologists, and then experimentally
                                        confirmed by biologists.)
                        => D) Axioms do not fit reality (missing parameter etc.)

I may have given the impression in the past that I do not give a high value to logic. This is absolutely not the case. I only like to stress that logical truths are relative to their premises , whereas empirical truths suffer from insufficient observation.
> 
> A discussion will be good.
>
> Should I distribute these diagrams to the list as well?
Oh yes! 

Posted by Martin on 31/3/2017

Dear Christian-Emil,

On 31/3/2017 8:39 μμ, Christian-Emil Smith Ore wrote:
> Dear all,
> I have little knowledge about (theoretical) physics. My impression is that the development of mathematical tools is necessary for the development of theoretical physics. It is perhaps unmodest to compare our model development with that, but it is easier to see (in)consistencies and consequences  with fol, models, sets and perhaps also category theory.
I am confused why you repeat that. I have really not said or meant anything else:-) ("sine qua non" means absolutely mandatory).
>
> The relation between E2 and E4. As long as there is an isomorphism between the two, they can be considered to be equal.
Sure, as long as we are in a Closed World, the stated properties are the only properties, and then the isomorphism holds. Isn't it?
 

Posted by Christian Emil on 31/3/2017

I see that it is more to this issue. I am used to think in closed world. I will think more about this.

In the 41st joined meeting of the CIDOC CRM SIG and ISO/TC46/SC4/WG9 and the 34th FRBR - CIDOC CRM Harmonization meeting, the sig discussed about the inconsistencies of E2, E4, E52 and E92. CEO suggested intensifying the discussion in order to move towards a solution for next meeting. MD and CEO agree to push forward. MD will send link to CEO on restricted IsA. There is a possibility to merge P4 with P7 or P10   but it would require this restricted IsA.

Lyon, May 2018

Posted by Martin on 2/3/2019

We consider the following properties:

P4 has time-span (is time-span of)
Domain:              E2 Temporal Entity
Range:                E52 Time-Span
Quantification:    many to one, necessary, dependent (1,1:1,n)

P160  has temporal projection (is temporal projection of)
Domain: E92 Spacetime Volume
Range: E52 Time-Span
Quantification: one to one (1,1:1,1)

In FOL:

P4(x,y) ⊃ E2(x), P4(x,y) ⊃ E52(y)

P160(x,y) ⊃ E92(x), P160(x,y)⊃ E52(y)

The problem comes from this: E4 Period being a spacetime volume and a temporal entity.

E4(x) ⊃ E2(x), E4(x) ⊃ E92(x)

I now propose to: declare P4, to imply P160  from E4 Period "downwards":

(P4(x,y) ∧E4(x)) ⊃ P160(x,y), (P160(x,y) ∧E4(x)) ⊃ P4(x,y).

We may then recommend to use only P4 from E4 Period downwards.

I do not know, if we would also need (P160(x,y) ∧E4(x)) ⊃ P4(x,y) in order to make them identical from E4 downwards.

================================================

Further:

P7 took place at (witnessed)
Domain:              E4 Period
Range:                E53 Place

Quantification:    many to many, necessary (1,n:0,n)

"The related E53 Place should be seen as a wider approximation of the geometric area within which the phenomena that characterize the period in question occurred, see below."

P161 has spatial projection (is spatial projection of)
Domain: E92 Spacetime Volume
Range: E53 Place
Superproperty of: E18 Physical Thing. P156 occupies (is occupied by): E53 Place
Quantification: one to many, necessary, dependent (1,n:1,1)

Firstly, I believe the quantification of P161 must be Quantification:    many to many, necessary (1,n:0,n). A place needs not be the projection of a Spacetime Volume.

Then, in FOL:

P7(x,y) ⊃ E4(x), P7(x,y) ⊃ E53(y)

P161(x,y) ⊃ E92(x), P161(x,y) ⊃ E53(y)

I propose to add: The spatial projection of an E4 Period is a "took place at".

(P161(x,y) ∧E4(x)) ⊃ P7(x,y).
Opinions? 

Posted by Martin on 9/3/2019

Dear Robert,

In the first place, E2 has a substance of "phenomena" something "becoming" "changing" "moving", "interacting". In addition, we interpret it now also more statically as including a sort of maintaining something. It is necessarily connected to some "things" on which such interactions, changes or temporary, non-essential formation of properties happen, but we have seen so far no good general way to describe the ways of involvement at the level of E2.

E92 is nothing of that kind. It is just spacetime, the generalized space in which we live and think, not what is there not what happens there. It is just a "where". It is further a volume in that space, i.e., it must have some inner part, and a surface as fuzzy as it may be, and a way to identify it.

We connect E4 and E18 with E92 as second superclass in order to describe a necessary one-to-one combination, in order to save the trivial links between them. We could do that with E2 too, but the space in which things like "being married" occur can hardly be seen as volumes with a surface. In contrast, I can be in the meeting (E4) or outside, in the battle or outside, even though the fuzziness between being inside and outside is very high.

Therefore, I would exclude both, E2 being subclass of E92 or superclass.

The discussion to which degree we should regard any E18 as ongoing interactions in spacetime is old and endless. We have so far rather preferred to think of a fundamental difference between "becoming" and "being" as a psychological and linguistic phenomenon, because this is the most adequate to the way people document things. The problem now is that by introducing E92 we are again confronted with the borderlines between the change itself and the changing thing, the thing that persists over time, but yet is limited in time, the things that are somewhere, but constitute a "where" for others.

Would that make sense?

Posted by Steve Stead on 11/3/2019

I am with George on this.

The fact that substantial things have a 1:1 relationship with an STV does not warrant the E92 superclass status IMHO.

It makes for horrible confusion and lots of “special case” rules and ………….

Please let us avoid this.

Posted by Robert Sanderson 0n 12/3/2019

Dear Martin, all,

I agree with your assessment into the four categories, and that the first three are met, and the last is more complicated.

I also agree with the formalism for E4. It moves some of the complexity around, and doesn’t introduce inconsistency for the temporal side of things for subclasses of E2.

However, I agree with George that this does not hold true for the other sub class of E92, being E18 Physical Thing.  With this subclass assertion, we can partition physical things based on time and then make assertions about those partitions using all of the sub-classes of E18.  For example, to say that the Nightwatch had a width of 17 feet between its production in 1642 and 1715 when it was trimmed to fit on a wall in the Amsterdam town hall, we could have an E22 for the painting throughout time, and use P10 to reference further E22s, each of which were clarified with P160 as to their temporal projection. These projections could then have different dimensions.

<Nightwatch> a E22_Man-Made_Object ;

    P10i_contains <Large_Nightwatch> , <Small_Nightwatch> .

<Large_Nightwatch> a E22_Man-Made_Object ;

    P160_has_temporal_projection [

         a E52_Time-Span ;

         P81a_begin_of_the_begin “1642-01-01”

         P82a_end_of_the_end “1715-12-31”  ] ;

     P43_has_dimension [

        P2_as_type <width-type> ;

        P90_has_value 17 ;

        P91_has_unit <feet-unit> ]

(and the same for Small_Nightwatch, starting 1715 with 14.3 feet as width)

This seems antithetical to the intent of the model (as I understand it) where activities (such as Modification in this case) are kept separate from the entities that they affect.

This particular pattern could be prevented by having E92 not be a sub class of E18, without affecting the P160 / P4 discussion.  However, I note some issues with making only this split:

·         It would still be valuable to have the STV of a physical thing, in order to calculate the intersection between the STV that a physical object projects with Periods (that are themselves STVs). So it would be valuable to introduce a relationship between E18 and E92, introducing pattern inconsistency.

·         While Period and Event seem to share the identity conditions with STV, Activity and below start to seem less identical. I worry that I become the space-time volume of the sum of my activities… and then I am a STV again, even though we removed it from E18 for just this reason.

·         The same issue for P160 / P4 would apply for P161 / P53 – the spatial projection of the object is its former or current location, as they have the same identity currently.

So overall, I think my position is that for consistency of the model, E92 should not be a subclass of either E4 or E18, but instead related via a property.

Hope that helps!

Posted by Martin on 12/3/2019

Dear Christian-Emil, All,

Thank you for your explanations. I agree with you technically completely.

I propose to have cardinality (1,1:0,1) for P4 and P160.

The scope note of P4 must be modified, as you say. We have discussed already, that alternative opinions are questions of the knowledge base, and not of the ontology.

I only disagree with:

"The model will .... and hopefully be more intuitive for the lay persons." with the additional link:

Firstly, I have made the argument, and got no response, that introducing a link between E4 and E92 does not solve the problem of equivalence of P4 and P160. It is still exactly the same, only more complex to formulate: we have to equate a path with a single link.

Secondly, the lay person will see a knowledge graph of instances, and not a theory. A knowledge graph with a lot of trivial links in my opinion makes the model less intuitive to use. Moreover, all our RDF databases are still very bad following links. Any additional join has high cost. Still, most CRM implementations materialize a huge number of paths to increase performance.
.......

I still do not see, why we should reduce performance because we find it difficult to explain the theory

Basically, as you say, we repeat old arguments here, long before decided, without new insight. 

Posted by George on 12/3/2019

Dear Martin et al.,

Although we are on a well trodden path, it seems to still engender questions so perhaps worthy meandering on for a bit longer, not necessarily to belabour any previous decisions but at least to ponder them some more.

I wonder about the following:

1) did STV not enter CRMbase when we still thought to include all extension top classes and relations to base? Now that this is no longer the case (we will specify in a separate document all top level classes and relations, thereby alleviating Base of having to hold all possible top level classes and properties), is there a case for reassessing whether STVs belong in base at all? I do not of course question their utility, but they did arise because of questions in CRMgeo. Are they a necessary feature of base?

> Secondly, the lay person will see a knowledge graph of instances, and not a theory. A knowledge graph with a lot of trivial links in my opinion makes the model less intuitive to use.

and

>> I believe we cannot avoid entering some complexity here in our discussions, and resolve it giving priority to the end-user schema.
>> I think the first arguments should be, if the final schema is confusing, and if the alternative is less confusing.
>

2) I think we all agree that that understandability and teachability is not at all an unimportant factor, especially for something that is supposed to be applied by domain specialists. If the concept cannot be communicated to the community it is supposed to serve, is it serving the community? The lay person doing mapping and modelling (using it) will need to understand these constructs and apply them correctly. We argue that it is this domain/lay person who should ultimately translate or validate their data. Therefore, the concept should be understandable and clear to them (without having to retrain in another discipline). They should more or often than not create intended models.

3) For those people who want and know how to use an STV, the fact that they would populate the knowledge base distinctly would not be a disadvantage but an advantage, as you could query it distinct from the entity of which it was an STV. For everyone else, they simply would not instantiate STVs and so would not have to worry about how they work or what their properties are or if they cause an extra join.

> A knowledge graph with a lot of trivial links in my opinion makes the model less intuitive to use. Moreover, all our RDF databases are still very bad following links. Any additional join has high cost. Still, most CRM implementations materialize a huge number of paths to increase performance.

4) If we change the ontology based on the present technological capacities, are we not violating a modelling principle?

>> The alternative you are advocating for is:
>> a) Fill the database with a very large number of necessary 1:1 links: events are some of the the most frequent items we have.

5) In practice, I have never seen a knowledge base of CRM that automatically creates for example the STV to an E2 (or something similar). If someone (most CH users?) isn’t interested in STVs, they don’t use them.

I am most intrigued though about Rob’s example. It seems like a potentially reasonable way of interpreting E18 when it is a subclass of E92 by a user of the ontology. I also think we do not actually want to allow or encourage this modelling pattern by a user of the ontology. I believe we do not want to do so because an E18 and its E92 are actually distinct, and we cannot say the same thing about the E18 and an E92 (but our model leads one to believe they are the same). We can’t say that an STV is married, but we can say Rob is married. If this doesn’t hold (Rob’s STV can be married), then it’s not clear to me why Rob’s modelling is wrong, although the result is of course a bevy of weird instances. I think the point here is that Rob’s interpretation is not out to lunch. It is one which one could assume would be an intended model. It is not, however, and why it is not cannot clearly be specified possibly because we have said E18 isA E92.

Posted by Martin on 13/3/2019

Dear George et al.

I completely follow your arguments. Teachability is an important concern. At the moment, we make a convoluted discussion.
Let us separate:

Question a) Is the logical formulation as I sent it consistent, and does it solve the logical problem. (I believe yes).

Question a1) Would the link make any difference logically (I believe not, and expect objective arguments in support of), and would it reduce or increase complexity.

Question b) Does the model describe adequately reality, independent from complexity concerns. (I believe yes, and I believe Robert's examples show a valid point. Think of reliquary caskets etc.)

Question c) To whom can it be taught as such. (I believe we have a serious problem here)

Question d) Which is the user community of the spatiotemporal concepts (I believe it is distinct, archaeologists, possibly conservators)

I suggest first to finish the issue by answering a) and b).

It was an immense work to clarify all these relationships. So we should keep this separate.

Then I suggest to reconsider where the whole bunch of spatiotemporal topology belongs to, and how to maintain the consistency we have arrived at.

Therefore we need to identify to total of concepts making up this part. We may separate the standard into two chapters, or we move the thing out, tightly interlaced with CRMbase, into an extension. Then we have a question of maintaining identifiers, and people that would maintain the distinct docs in sync.

Dividing into two chapters may be a way to teach CRM without the STVs for those not interested in these deductions.

Posted by Christian Emil on 14/3/2019

I will not continue the discussion of identifying an instance of  E18 Physical Thing with its spacetimevolume.

The intention with my “explanations” was to see whether the current model support an equality of P4 and P160 from E4 and downwards. In a FOL description of CRM the cardinality constraints are not just guidelines, they have strong implications on the model. For example:

1)      P160 (E92 <-> E52) has the cardinality constraint (1,1:1,1)  and is a 1-1 mapping (isomorphism) between E92 and E52

2)      P4 (E2 <-> E52) has the cardinality constraint (1,1:1,1) and is a mapping between E2 and  E52 where all instances of E52 is related to one or more instances of E2.

1 & 2 say that all time-spans are a time-span both for an instance of E92 and at least one instance of E2. There are no instance of E2, which does not share a time-span with a STV. This, I believe, is not the intention.

If we change the cardinality of  P4 (E2 <-> E52)  from (1,1:1,n) to (1,1:1,1) which is required if P2 and P160 should be equal when lowered to E4 <-> E52, the result is even worse. Then E2 and P52 is in a 1-1 correspondence.

Therefore the cardinality of P4 and P160 should be (1,1: 0,1) allowing instances of E52 to be a time-span for an instance of  E2 without being a time-span for an instance of E52 and vice versa.

Still there is nothing in the formal FOL definition stating that P4 and P160 seen as properties between E4 and E52 are equal here Martin’s additional FOL expression comes in:

(P4(x,y) ∧E4(x)) ⊃ P160(x,y) ∧ (P160(x,y) ∧E4(x)) ⊃ P4(x,y)

This expression states that P4 and P160 is equivalent when the domain is restricted to P4.  This is the only such restriction in CRM. The question is where to put this FOL expression. One could perhaps introduce multiple inheritance for properties and then introduce a subproperty of P4 and P160?

In any case, the 326 issue requires that the cardinalities are changed and that the scopenote of P4 is changed to

“This property describes the temporal confinement of an instance of an E2 Temporal Entity. The related E52 Time-Span is understood as the real Time-Span during which the phenomena were active, which make up the temporal entity instance. It does not convey any other meaning than a positioning on the “time-line” of chronology. The Time-Span in turn is approximated by a set of dates (E61 Time Primitive). Time-Spans may have completely unknown dates but other descriptions by which we can infer knowledge.”

Posted by George Bruseker on 14/3/2019

I seem to have lost Martin’s in between email, but I agree with the general point that there are multiple questions arising and that this issue thread is pointing to one particular problem. This issue in particular is about the formal representation and whether it is correctly formulated (I think yes). The extension of the discussion is related to this issue but does not pertain directly to the particular decision here.

The other questions arose because this thread is parallel in interest to questions of modelling phases or states in CRMSoc and elsewhere and the potential implications and or conflicts in intended models.

Therefore, I would say that we could mark this discussion as relevant to both existing, open phase issues and the issue on specifying the top level classes and properties required to return a complete CRM and family graph.

I don’t think we need to open a new issue for either because these discussions are on-going and already marked for further investigation. If we want to create a separate email thread to distinguish from the FOL representation under question and the related questions is another matter. We could if desired in order to keep things clear.

Posted by Martin on 14/3/2019

Dear Christian-Emil,

On 3/14/2019 11:15 AM, Christian-Emil Smith Ore wrote:
>
> I will not continue the discussion of identifying an instance of  E18 Physical Thing with its spacetimevolume.
>
> The intention with my “explanations” was to see whether the current model support an equality of P4 and P160 from E4 and downwards. In a FOL description of CRM the cardinality constraints are not just guidelines, they have strong implications on the model. For example:
>

> 1)      P160 (E92 <-> E52) has the cardinality constraint (1,1:1,1)  and is a 1-1 mapping (isomorphism) between E92 and E52
>
> 2)      P4 (E2 <-> E52) has the cardinality constraint (1,1:1,1) and is a mapping between E2 and  E52 where all instances of E52 is related to one or more instances of E2.
>
> 1 & 2 say that all time-spans are a time-span both for an instance of E92 and at least one instance of E2. There are no instance of E2, which does not share a time-span with a STV. This, I believe, is not the intention.

>
> If we change the cardinality of  P4 (E2 <-> E52)  from (1,1:1,n) to (1,1:1,1) which is required if P2 and P160 should be equal when lowered to E4 <-> E52, the result is even worse. Then E2 and P52 is in a 1-1 correspondence.
>
> Therefore the cardinality of P4 and P160 should be (1,1: 0,1) allowing instances of E52 to be a time-span for an instance of  E2 without being a time-span for an instance of E52 and vice versa.
Yes!
>
> Still there is nothing in the formal FOL definition stating that P4 and P160 seen as properties between E4 and E52 are equal here Martin’s additional FOL expression comes in:
>
> (P4(x,y) ∧E4(x)) ⊃ P160(x,y) ∧ (P160(x,y) ∧E4(x)) ⊃ P4(x,y)
>
> This expression states that P4 and P160 is equivalent when the domain is restricted to P4.

Would that mean that each instance of E4 must have two time-spans, one for the P4 and one for the P160?

Another solution could be the infamous "punning": Declare P4 to have two domains, E2 and E92. Makes life easy, until we know the "true" superclass of both. Isn't it? That is actually what we mean but do not dare to say.

Posted by martin on 18/3/2019

Dear Robert,

Yes, this is up to the point. An example for E93 would be nice: typically, some reported situations, such as a museum object having been in display case xxx at least for this time-span, etc.

Another important case are the declarative STVs (CRMgeo) we need in great number for approximating "places", such as "Rome (city)" or "Poland(state)" in gazetteers, typically by outer bounds. (Franco Niccolucci and Sorin Hermon propose smaller STV boxes filling up "places").

There are enough high-level classes that need not normally be instantiated, but axiomatizing them as "abstract classes", as semantic network models in the 1980'ies proposed, appeared in the end to be misleading, because it fixes a level of specificity which does not exists for a general ontology as the CRM. In a way, the specificity of any class is arbitrary, because an instance has much more characteristics than any class we use. For instance, if we are interested in genotypes, the class "Person" becomes completely abstract. On the other side, in an Open World, any "abstract class" may have a direct instance, because the respective subclass has not been formulated in this ontology.

Nevertheless, we used the term informally in the CRM. We could name E92 as "abstract".

Posted by Christian Emil on 19/3/2019

Hi Dan, all
The recommended way to model historical periods is as instances of E4 Period. This recommendation is older than the introduction of E92.

The current model with E2 Temporal Entity, E52 Time-Span and E92 STV is not optimal and E92 is not the reason eventhough it is pretty abstract. In my opinion, E52 Time-Span is redundant and can be replaced by E2 Temporal Entity with an adjusted scope-note and change the domain of

P79 beginning is qualified by: E62 String
P80 end is qualified by: E62 String
P81 ongoing throughout: E61 Time Primitive
P82 at some time within: E61 Time Primitive
P83 had at least duration (was minimum duration of): E54 Dimension
P84 had at most duration (was maximum duration of): E54 Dimension
P160 has  temporal projection

P86 falls within (contains): E52 Time-Span can be replaced by P117 occurs during (includes): E2 Temporal Entity

If E52 is removed the property P160 has temporal projection should be an injection into E2 and have the cardinality (1,1:0,1)

(in any case: P78 is identified by (identifies) should be deprecated since E49 Time Appellation is already deprecated.)

See also http://cidoc-crm.org/sites/default/files/issue%20326%20overview%20and%20... for a graphical overview.

Posted by Martin on 20/3/2019

Dear Christian-Emil,

I am not sure this is a good idea. Moving time-span properties onto E2 just washes about the fact that the temporal projection of E2 and of E92 are equivalent. If we make the paths more dissimilar, we just hide that there are two alternative ways to formulate it. If the formal rules of KR do not foresee the case, we just change the cardinality of P4 / P160 to (1,1:0,n) or (1,1:0,2).

I am hesitating to propose a common superclass of E92 and E2 just to have one domain for P4/P160. It becomes an extremely abstract thing.

Opinions?
 

Posted by Francesco Beretta on 21/3/2019

Posted by Francesco Beretta on 21/3/2019

Dear Dan, Franco, all,

in a nutshell:

Period – E4

P4 has time-span E52 Time-Span

P7 took place at E53 Place

Spacetime Volume – E92

P160 has temporal projection E52 Time-Span

P161 has spatial projection E53 Place

Period – E4 (phenomenal) Pxx has projection in Spacetime Volume – E92 (‘region’)

If we keep Spacetime Volume – E92 in the model we should get rid of P4 has time-span and P7 took place at because they are redundant with P160 has temporal projection and P161 has spatial projection, or apply the logical mechanism proposed by Martin which is under discussion.

If we get rid of E92 (and properties) and clearly explain E4 is a spacetime volume by definition, with temporal and spatial projection (P4/P7), then the issue seems to be solved.

E4 being a subclass of E92 is in my opinion (and other’s also as we know) inconsistent with the traditional modelling method, and also misleading.

If E4 can be merged with E2 (E2 would always have a projection in space, at least virtually, be this my brain the ‘place’ for my belief), then E2 is a STV with projection in time and space.

This synthesis may be too simple not to be simplistic and I miss some crucial point ?

Posted by Francesco on 21/3/2019

Dear Franco,

you find me in full agreement with your vision of things.

All the best

Posted by Christian Emil on 21/3/2019

Some additional comments:
First some  tentative answers to the three first questions and then a longer comment on the forth:

1) which E2 is not an E4, even in a broad sense?
Condition states

2) which E94 (relevant, and not just purely abstract) is not also an E4?

A physical object

3) Does the scope note of E94 (E92 I assume?) allow the existence of some instances that are also E4, as implied by the subclass condition and described in Dan’s examples?

Bronze Age in Scandinavia

4) What is the difference between P4/P7 and P160/P162? In the second pair I assume P162 is a typo and the correct pair is P160/P161.

P4 has time-span (is time-span of)
Domain:  E2 Temporal Entity
Range:  E52 Time-Span
Quantification: many to one, necessary, dependent (1,1:1,n)

P7 took place at (witnessed)
Domain:  E4 Period
Range:  E53 Place
Quantification: many to many, necessary (1,n:0,n)

P160  has temporal projection (is temporal projection of)
Domain: E92 Spacetime Volume
Range: E52 Time-Span
Quantification: one to one (1,1:1,1)

P161 has spatial projection (is spatial projection of)
Domain: E92 Spacetime Volume
Range: E53 Place
Superproperty of: E18 Physical Thing. P156 occupies (is occupied by): E53 Place
Quantification: one to many, necessary, dependent (1,n:1,1)

If we assume there are instances of E2 which have only a time component and no spatial component then the cardinality of P4 must be (1,1:0,1) and P160 must be (1,1:0,1) since this implies the existence of instances of  E52 Time-Span which are not  simultaneously a timespan for an instance of E2 and E92. As discussed earlier P4 and P160 should be considered equal from E4 and downwards.  But as long as E2 is a proper superclass of E4 we need P4. The question is if there exists an instance of E2 which is not an instance of E4.

P7 has the cardinality (1,n:0,n) and P161 (1,n:1,1).  The intention of P7 is to link a temporal entity (read E4 Period) to one or more places, cf the last part of the scopenote for P7 “Something happening at a given place can also be considered to happen at a larger place containing the first. For example, the assault on the Bastille July 14th 1789 took place in the area covered by Paris in 1789 but also in the area covered by France in 1789. “
P161 is the projection to the smallest and unique place a STV occupies during its lifetime.  There is only one such place. (It is unclear to me why this instance cannot be shared by to spacetime volumes existing in a row.)

This necessity for a STV to have a unique spatial projection implies that for E4 and down, there will always be an instance of P53 that is the range of an instance of P7 and at the same time the range of an instance of P161, which is fine and which is the reason P7 is a shortcut (cf. the comment by Francesco: “If we keep Spacetime Volume – E92 in the model we should get rid of /P4 has time-span/ and /P7 took place/ at because they are redundant with /P160 has temporal projection/and /P161 has spatial projection/, or apply the logical mechanism proposed by Martin which is under discussion”)

Since E4 is a subclass of E92, there may in principle be many instances of P53 Place that are not the range of P7 (but only the range of E161). The question is analogues to the question above : Can we identify at least some instances of E92 that are not an instance of E4?

Posted by Martin on 22/3/2019

On 3/21/2019 12:49 AM, Francesco Beretta wrote:
> The last issue I see is the one related to modelling of a STV for a E18 Physical thing. The difference between the volume occupied by a physical body as such, be it moving or not, and the volume occupied by the body moving from one place to another, is not clearly defined if you make a E18 Physical thing a subclass of E92 Spacetime volume. In the well known case of the vessels’ fight in Trafalgar, the TSV of the fight can be treated as projection in time and space of the whole event, using P4/P7 and, if needed, the correspondent declarative properties : we have a phenomenon, the fight (E5 as temporal entity), and a document driven approximation of its STV using in SP10/SP2 (without necessarily the need of using a E92 Spacetime volume/SP7 STV classes as separated entities with an own identity).

Dear Francesco,

I think it is unambiguous:

"An instance of E18 Physical Thing occupies not only a particular geometric space, but in the course of its existence it also forms a trajectory through spacetime, which occupies a real, that is phenomenal, volume in spacetime. We include in the occupied space the space filled by the matter of the physical thing and all its inner spaces, such as the interior of a box."

The inner space of the ship is part of the STV.

Correct me if I am wrong.

Posted by Martin on 23/3/2019

Dear Francesco,

Here a more complete comment. I widely agree anyway.

On 3/21/2019 12:49 AM, Francesco Beretta wrote:
>
> Dear Martin, all,
>
> Applying the principles we generally use in conceptual modelling, a model of « CRM top hierarchy and space&time » like the one you’ll find in attachment seems plausible. I’ll comment the slides from the top to the bottom, for the sake of clarity.
>
>
> Let’s start from the basic modelling principles Martin expressed on 9/3/2019 :
>
> « In the first place, E2 has a substance of "phenomena" something "becoming" "changing" "moving", "interacting". In addition, we interpret it now also more statically as including a sort of maintaining something. It is necessarily connected to some "things" on which such interactions, changes or temporary, non-essential formation of properties happen, but we have seen so far no good general way to describe the ways of involvement at the level of E2.
>
> E92 is nothing of that kind. It is just spacetime, the generalized space in which we live and think, not what is there not what happens there. It is just a "where". It is further a volume in that space, i.e., it must have some inner part, and a surface as fuzzy as it may be, and a way to identify it. »
>
>
> As you can see in the attached slides (the more relevant being the first one), E77 Persistent Item and E2 Temporal entity are ‘phenomenal’ classes. In contrast, Place, Time-Span, STV, Dimension are ‘regions’ in a reference ‘space’, be this spatial, temporal or quantitative.
There is a curious difference of the STV to Place, TimeSpan and Dimensions. Points in spacetime are absolute in the world in which we live, at least in modern physics, and in our intuition. They are the "where this happened". If we describe them by coordinates, we use different reference 'spaces' in the physical space, and in time. But their identity does not depend on them. Therefore, descriptions in one reference 'space' can be transformed into descriptions of the same points in another, if the relations are known. This makes a fundamental difference.
>
>
> As we know, E77 Persistent Item instances are in principle not directly related to time (using properties) but they live in time : we model this using temporal entities related to these Persistent item instances « on which such interactions, changes or temporary, non-essential formation of properties happen » (Martin). Temporal entities are phenomena we usually perceive or observe in relation to some persistent item, be this a conceptual or physical one. A very general property ‘Pxx involves/concern’ would clearly express this basic phenomenon : persistent items live in time (and space) and we model this using temporal entities (Events for dynamic moments, Phases for static characteristics, both phenomenal).
Indeed, but it violates the bottom-up methodology to define a property which is not well-distinguished from others, and possibly the Open World assumption, trying to cover all future relations between two classes. It satisfies the theoretical understanding, but is error-prone for data modeling. We should only accept if good definitions are found.
>
> E2 Temporal entities (having a phenomenal substance) are projected in a region in time and these contribute to define their identity (we stress here the phenomenal aspect, not the epistemological). This projection in time is modelled as an instance of E52 Time span.
Yes. Without theory or relativity, this projection is unique except for the zero point. With theory of relativity, it depends on the reference 'space'. With GPS satellites, the correction makes up for an error of about one km on the surface of Earth, if I remember correctly.
>
> Insofar as we are modelling conceptually we need to keep these two classes (E2, E52) separate in the model regardless their cardinality because they are expressions of significantly different substance. The implementation in an information system, if the (1,1;1,1) cardinality is chosen, can merge the two classes but this is about implementation not about the conceptual model : here we must keep both classes separate for the sake of clarity and consistency with the identity principles.
I agree.
>
> How do we model, in the next step, projection in physical space ? The crucial question is: are there any temporal entities without such a projection ? If yes (e.g. I2 Belief), we have two ways of modelling this : 1) with a (1,1:1,1/n) cardinality for P7 took place at associating it to E4 Period (slide 1) or 2)
Besides that P7 allows for defining wider areas, spacial projections are not unique. They depend on reference frames, which are in general moving, be it geologically slow or faster. Therefore (1,1: is wrong).
>
> with a (0,1:1,1/n) cardinality associating it directly to E2 Temporal entity (slide 2). This second model would imply in some cases there is not a projection of a Temporal entity instance in a region in physical space, although fundamentally there can be one. In the perspective of simplicity, the second solution would be the preferable one.
I can only repeat, that if we accept this argument, for questions of simplicity all properties should be at CRM Entity. It is really a trap, because the ways an E2 may occur in physical space needs not qualify as an "extent in space".
>
> But for the sake of consistency with earlier versions of the CRM and for making the conceptual model more clear and explicit, the first modelling choice (using E4 Period) would probably be the best one.
This is not a question of the CRM, but a fundamental modelling principle of ontology. We are committed to understand the "οντα". So far, it has saved us from many modelling errors and pseudo-solutions, causing incompatibilities with other "simplified" high-level theories.
>
> Insofar as it has a projection in time and space, an instance of E4 Period (which is composed by Events and Phases) is associated to a Spacetime Volume. As phenomenon, an instance of E4 Period makes a STV to be virtually present : if we want to make this explicit, and especially if we want to explicitly associate a time span with a place, providing this association with a specific identity, independently from any E4 Period instance, we need a STV instance (and the E92 STV class). For this we would then need to have a property Pxx has spacetime volume (1,1:1,1/n) modelled similarly to P4/P7.
In contrast to Place and Time-Span, this volume is unique. It is (1,1:1,1). It has exactly the same projections to Place and Time as the E4, and therefore is definitely not similar to P4/P7. Please refer to texts about CRMgeo.
>
> Incidentally, for all these properties we have to decide if the maximum cardinality on the side of a Exx Region subclass has to be 1 or n. If we choose ‘n’ we provide a specific identity to the Exx Region instance, independently from the identity of the related Phenomenal Class instance. E.g. different E4 Period instances could be located in the same spatial region, i.e. E53 Place.
Yes, this is correct for P7, but not for the projection. The projection is unique per reference system. No other phenomenon will have exactly the same extent. See CRMgeo for more detailed argumentation.
>
> If we think an autonomous identity of E92 STV is not given, and time and space, and virtual spacetime volume (= time + space) are always related to at least one temporal entity or period (i.e. to a phenomenon),
This is not the case, because we have the declarative STVs in cRMgeo.
>
> then we could deprecate the E92 Spacetime Volume class and use E4 Period instead as common point of meeting of time and space, in the phenomenal sense. A E53 Place being the ‘surface’ of a phenomenon during a given timespan. E93 Presence (as subclass of S4 Observation?) would be an intersection of time and space in the epistemological sense, providing an arbitrarily defined snapshot of a Period. As such, it would have a specific identity and would be modelled as a distinct class.
Correct. We could move the whole stuff to CRMgeo, and allow it to have a huge superclass out of CRMbase. That was what we wanted to avoid, but we could change that.
>
>
> This way of modelling seems to be more robust and consistent with the domain : we model phenomena in cultural life associated to persistent items, phenomena having a projection in time and space, not time/space/STV as such, independently from phenomena.
If we make multiple ISA, it means that both concepts, E2 AND E92 are specialized! The specialization of E92 in E4 is to be in addition phenomenal. See CRMgeo.
>
> Also, this would avoid the issue of the redundancy of properties which was the starting point of this discussion : P4/P160 ; P7/P161. For these reasons I would advocate to abandon the E92 class, knowing that it is virtually present in E4 Period as its implicit spatio-temporal surface.
See above. The redundancy of property here is more crying out for a common superclass on top of E2 and E92.
>
> The last issue I see is the one related to modelling of a STV for a E18 Physical thing. The difference between the volume occupied by a physical body as such, be it moving or not, and the volume occupied by the body moving from one place to another, is not clearly defined if you make a E18 Physical thing a subclass of E92 Spacetime volume.In the well known case of the vessels’ fight in Trafalgar, the TSV of the fight can be treated as projection in time and space of the whole event, using P4/P7 and, if needed, the correspondent declarative properties : we have a phenomenon, the fight (E5 as temporal entity), and a document driven approximation of its STV using in SP10/SP2 (without necessarily the need of using a E92 Spacetime volume/SP7 STV classes as separated entities with an own identity).

Basically, this is a misunderstanding of the nature of an STV. The STV is not the result of a backprojection of a space and a time.

Just the opposite. The moving around of the spatial volume of a physical body IS the STV. It is all spatial points where the volume was one instant of time after another. he TSTV of the fight cannot be treated as projection in time and space of the whole event. It is the STV of the event that has such projections. The resulting box of the two projections is containing/approximating the event's STV. We need to understand that!
>
> On the one side, Nelson’s ship itself, as an instance of E18 Physical thing, would have a time related volume (STV), even being moored in a port without moving, and the wreck of it on the seafloor has a STV different from the navigating ship. But this is not about the position in the fight but about the volume of the Persistent item instance itself (this volume being a E53 Place instance, as specific « extent in space » : the 3D ‘surface’ of the ship) and can be modelled using a Exx Volume/Surface class, modelled as subclass of E3 Condition state.
Well, this is just inventing a new term for the phenomenal STV, if I am not completely mistaken?
>
> E3 would be modelled directly as subclass of E4 Period or as subclass of the new Exx Phase class, expressing « phases during the existence and evolution of an instance of E18 Physical Thing characterized by a substantial appearance » (Martin). In this case the ‘appearance’ is the volume of the physical thing understood as a « surface » with a precise form. The movement of the ship, on the other side, can be modelled as an instance of E5 Event and associated to the fight E5 Event instance using P9 consists of.
See above. Note, that the spatial projection of the STV of the ship on the ship as reference space stays within the ship all time, but the projection of the STV on the seafloor is a long path. The spatial projection of Nelson's body on the seafloor is within this path., and within the ship in the other projection.
>
>
> This approach would allow to have a concise and straightforward model and avoid inconsistency with the CRM well established way of defining identity criteria, which isn’t the case if E4 Period and E18 Physical thing are modelled as subclasses of E92 Spacetime Volume (insofar as this is not a phenomenon but a ‘region’ in space and time).

Just let me repeat, the inconsistency we are discussing is that the projection of the event points in an E4 on time occurs as P4 and as P160, regardless whether we make E4 an STV or not, or if we rename E92 to something else.

Posted by Robert Sanderson on 14/6/2019

HW from 44th SIG is for me to send a new example demonstrating my continued objection to E4 and E18 being subclasses of E92, rather than that they Pxx_have a E92 STV.

The new objection described was that Periods (being STVs) can P10 fall within Physical Things (also being STVs), as the range and domain of P10 is E92 STV.

Thus, I might document the fact that I had a fever as simply as:

  Rob a E21 Person ;

    rdfs:label “The person that is Rob” ;
    …

    P10i contains  [

      a P5 Event ;

      rdfs:label “The event during which Rob has a fever” ;

      … ] .

This was unanimously considered to be complete lunacy and at the same time quite legitimate given the current class hierarchy in which both Events and People are Space Time Volumes.

This example avoids the identity condition issue that the previous example (of Rob having a part that was also a Person), as the Event’s identity is unchanged.